| 1 |
- {"config":{"lang":["en"],"separator":"[\\s\\-]+","pipeline":["stopWordFilter"]},"docs":[{"location":"","title":"Turn LMs into SWE agents (and more)!","text":"<p>SWE-agent turns LMs (e.g. GPT-4) into agents that can \ud83d\udc1b fix issues in real GitHub repositories, \u26f3\ufe0f solve coding challenges, and \ud83d\udd25 crack offensive cybersecurity challenges (EnIGMA mode).</p> <ul> <li> <p> Background & goals</p> <p>Learn more about the project goals and academic research.</p> <p> Learn more</p> </li> <li> <p> Installation</p> <p>Three different ways to get started, including running installation-free in your browser.</p> <p> Get started</p> </li> <li> <p> Usage</p> <p>Learn how to make the most out of SWE-agent.</p> <p> Tutorials, tips and tricks</p> </li> <li> <p> Configuration</p> <p>SWE-agent can be tweaked extensively without modifying the code.</p> <p> Modify SWE-agent behavior</p> </li> <li> <p> Development</p> <p>Dig into SWE-agent's code and build your own agent!</p> <p> Development information</p> </li> <li> <p> EnIGMA</p> <p>EnIGMA turns SWE-agent into an offensive cybersecurity expert.</p> <p> Learn more</p> </li> <li> <p> Changelog</p> <p>See what's new in SWE-agent.</p> <p> Read the changelog</p> </li> </ul>"},{"location":"_footer/","title":"footer","text":"<ul> <li> <p> Something broken? Report bug</p> </li> <li> <p> Something unclear? Ask question</p> </li> </ul>"},{"location":"faq/","title":"Frequently Asked Questions","text":"<p>How can I change the demonstrations given to SWE-agent?</p> <p>At the start of each run, we feed the agent a demonstration trajectory, showing it how to solve an example issue. This substantially improves the agent's abilities to solve novel issues. If you'd like to modify or totally change this demonstration, to better fit your use case, see this.</p> <p>Does SWE-agent run on windows?</p> <p>You can run it in a docker container, though this is not our first choice for running SWE-agent. SWE-agent runs best on Mac and Linux, as these are the environments we use for SWE-agent development. We're open to merge simple fixes to make the development setup work on Windows.</p> <p>Which LMs do you support?</p> <p>Currently our model support is limited, mostly focused on GPT-4 and Claude 3. SWE-agent will not perform well with small or local models. More information on models.</p>"},{"location":"background/","title":"Learn more about the project.","text":"<p>This section of the documentation talks about the architecture and research goals of SWE-agent and EnIGMA.</p> <p>Just want to run SWE-agent or EnIGMA? Skip ahead to our installation notes.</p>"},{"location":"background/#swe-agent","title":"SWE-agent","text":"<p>SWE-agent turns LMs (e.g. GPT-4) into software engineering agents that can fix issues in GitHub repositories.</p> <p>On SWE-bench, SWE-agent resolves 12.29% of issues, achieving the state-of-the-art performance on the full test set.</p> <p>We accomplish our results by designing simple LM-centric commands and feedback formats to make it easier for the LM to browse the repository, view, edit and execute code files. We call this an \ud83e\udd16 Agent-Computer Interface (ACI). Read more about the ACI here.</p> <p>SWE-agent is built and maintained by researchers from Princeton University.</p> <p>For a quick introduction, watch the following video:</p> <p>A longer lecture touching on the project's motivation, research findings, as well as providing a hands-on tutorial on how to install, use, and configure SWE-agent is provided here:</p> <p>For in-depth information, read our paper. If you found this work helpful, please consider using the following citation:</p> <pre><code>@misc{yang2024sweagent,\n title={SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering},\n author={John Yang and Carlos E. Jimenez and Alexander Wettig and Kilian Lieret and Shunyu Yao and Karthik Narasimhan and Ofir Press},\n year={2024},\n}\n</code></pre>"},{"location":"background/#swe-agent-enigma","title":"SWE-agent EnIGMA","text":"<p>SWE-agent EnIGMA adds advanced offensive cybersecurity capabilities.</p> <p>On the NYU CTF benchmark, EnIGMA solves 13.5% of the capture the flag (CTF) challenges, achieving the state-of-the-art performance on the full test set of 200 challenges, surpassing previous agents by more than 3x (leaderboard).</p> <p>We accomplish our results by extending the \ud83e\udd16 ACIs concept first introduced in SWE-agent, to the cybersecurity domain. We establish the novel Interactive Agent Tools (IATs) concept, which enables our agent to use interactive tools such as a debugger, in a multitasking way such that the agent still has access to the main shell while using the debugger.</p> <p>We also use a new Summarizer concept integrated into the agent to deal with long context. Read more about our different summarizers here.</p> <p>Specific demonstrations were built per each CTF category (cryptography, reverse-engineering, forensics, ...), to enhance the model ability to solve new tasks from the same category.</p> <p>EnIGMA is built and maintained by researchers from Tel-Aviv University, New York University and Princeton University.</p> <p>For a quick introduction, watch the following video:</p> <p>For all the details, read our paper. If you found this work helpful, please consider using the following citation:</p> <pre><code>@misc{abramovich2024enigmaenhancedinteractivegenerative,\n title={EnIGMA: Enhanced Interactive Generative Model Agent for CTF Challenges},\n author={Talor Abramovich and Meet Udeshi and Minghao Shao and Kilian Lieret and Haoran Xi and Kimberly Milner and Sofija Jancheska and John Yang and Carlos E. Jimenez and Farshad Khorrami and Prashanth Krishnamurthy and Brendan Dolan-Gavitt and Muhammad Shafique and Karthik Narasimhan and Ramesh Karri and Ofir Press},\n year={2024},\n eprint={2409.16165},\n archivePrefix={arXiv},\n primaryClass={cs.AI},\n url={https://arxiv.org/abs/2409.16165},\n}\n</code></pre>"},{"location":"background/aci/","title":"Agent Computer Interface (ACI)","text":"<p>We accomplish our results by designing simple LM-centric commands and feedback formats to make it easier for the LM to browse the repository, view, edit and execute code files. We call this an Agent-Computer Interface (ACI) and build the SWE-agent repository to make it easy to iterate on ACI design for repository-level coding agents.</p> <p>Just like how typical language models requires good prompt engineering, good ACI design leads to much better results when using agents. As we show in our paper, a baseline agent without a well-tuned ACI does much worse than SWE-agent.</p> <p>SWE-agent contains features that we discovered to be immensely helpful during the agent-computer interface design process:</p> <ol> <li>We add a linter that runs when an edit command is issued, and do not let the edit command go through if the code isn't syntactically correct.</li> <li>We supply the agent with a special-built file viewer, instead of having it just <code>cat</code> files. We found that this file viewer works best when displaying just 100 lines in each turn. The file editor that we built has commands for scrolling up and down and for performing a search within the file.</li> <li>We supply the agent with a special-built full-directory string searching command. We found that it was important for this tool to succinctly list the matches- we simply list each file that had at least one match. Showing the model more context about each match proved to be too confusing for the model.</li> <li>When commands have an empty output we return a message saying \"Your command ran successfully and did not produce any output.\"</li> </ol> <p>Read our paper for more details here. A recent extension of our ACI are Interactive Agent Tools.</p>"},{"location":"background/architecture/","title":"Architecture","text":"<p>This page walks you through the general architecture of the SWE-agent package. Want to just run it? Jump ahead to the installation or usage sections.</p> <p></p> <p>The central entry point to SWE-agent is the <code>run.py</code> script (1). It initializes the <code>SWEEnv</code> instance (2) that manages the environment. Upon initialization (and by default for every new instance), it starts a docker container together with a shell session (3). This shell session will be kept alive throughout the task. All installation commands and actions from the model will be executed therein.</p> <p><code>SWEEnv</code> then installs the dependencies of the repository to which the task instance belongs into a new conda environment (4).</p> <p>The second class that is initialized by <code>run.py</code> is the <code>Agent</code> class (5). It can be configured with a yaml file (see config). It's most important method is <code>forward()</code> which prompts the model and executes its action.</p> <p>To prompt the model, the history (all prompts to the model together with actions and outputs) need to be sent to the LM. In order to make the best use of the context window of the model, the history gets compressed by a <code>HistoryProcessor</code> (7). The model output (8) is then interpreted by the <code>Agent</code> class and executed in the Shell session via <code>SWEEnv</code>.</p> <p>The ACI elements are implemented as custom commands (9) that are available to the shell session.</p>"},{"location":"background/iat/","title":"Interactive Agent Tools (IATs)","text":"<p>Interactive Agent Tools were first developed as part of our EnIGMA project, but are compatible with all uses of SWE-agent. This page focuses on their usage for capture the flag (CTF) challenges.</p> <p>Tools useful for debugging (<code>gdb</code>, <code>radare2</code>), remote server interaction (<code>netcat</code>, <code>socat</code>) and penetration testing (<code>metasploit</code>) are widely used during CTF problem-solving and by cybersecurity experts. These tools are all interactive, i.e., they wait for user input, evaluate it, and print the results (read-eval-print loop, REPL). Current LM agents, which build their ACI around a running shell as central REPL, lack the ability to nest REPLs or start separate REPLs for interactive tools.</p> <p>In EnIGMA, we build Interactive Agent Tools (IATs) that use the same principles from the ACIs introduced in SWE-agent, but are also able to nest new REPL of the desired tool for agent interaction.</p> <p>Just like how typical programmer or cybersecurity expert uses multiple programs concurrently to build software or defend against cyber attacks, our IAT design enables the agent to use the interactive tools while still having the ability to access the main shell to run other commands.</p> <p>As we show in our paper, a baseline agent without our IATs does worse than EnIGMA.</p> <p>EnIGMA contains interactive tools that we discovered to be immensely helpful for CTF solving during the agent-computer interface design process:</p> <ol> <li>We add a debugger (based on <code>gdb</code>) that have basic commands for controlling (<code>start</code>, <code>stop</code>, <code>step</code>, <code>add_breakpoint</code>), similar to how graphic interfaces for debuggers provide buttons for frequently used operations. We also have a generic command that lets the agent perform arbitrary <code>gdb</code> command inside the interactive debugging session.</li> <li>We supply the agent with a special-built server connection tool, instead of having it just <code>netcat</code> to servers. The server connection utility that we built has commands for connecting to a server and sending a line with bytes or unicode strings to the server.</li> </ol> <p>Read our paper for more details here.</p>"},{"location":"config/commands/","title":"Command Configuration","text":"<p>In this document, we describe how to implement your own commands for the SWE-agent ACI. To see examples of command implementations, open the <code>.sh</code> and <code>.py</code> files in the <code>config/commands</code> folder.</p>"},{"location":"config/commands/#scaffolding","title":"Scaffolding","text":"<p>Every command subscribes to the following skeleton code.</p> <pre><code># @yaml\n# signature: [command] [argument(s)]\n# docstring: [Brief description of what your command does.]\n# arguments:\n# [argument 1 name]:\n# type: [type (i.e. integer, string)]\n# description: [Brief description of this argument]\n# required: [true|false]\n# [argument 2 name]:\n# ...\n[command]() {\n # Implementation here\n}\n</code></pre> <ul> <li>If a command takes in arguments, reference them via positional parameters notation (i.e. <code>$1</code>).</li> <li>If there are no arguments, omit the <code>arguments</code> section.</li> <li>The implementation for your command is unconstrained. There are no limitations on the form of the underlying command code.</li> <li>The minimal documentation requirements are <code>signature</code> and <code>docstring</code>.</li> <li>If you'd like multiple commands to make modifications to a similar body of functions, we recommend using global variables.<ul> <li>For instance, in <code>config/commands/default.sh</code>, you'll see we define the <code>CURRENT_LINE</code> variable for the file viewer. This variable is modified across multiple commands, including <code>open</code>, <code>goto</code>, <code>scroll_up</code>, <code>scroll_down</code>, and <code>edit</code>.</li> <li>You can also leverage third party libraries (check out how we do linting enabled <code>edit</code> in <code>config/commands/edit_linting.sh</code>).</li> </ul> </li> <li>To show effects of the command, print to standard output (i.e. <code>echo</code>). SWE-agent is implemented such that it does not look for a return value from these commands.</li> <li>The following environment variables are used to persist information between commands:<ul> <li><code>CURRENT_FILE</code>: File that is currently open</li> <li><code>CURRENT_LINE</code>: First line of the window that is currently being shown/edited</li> <li><code>WINDOW</code> (start line to end line): Part of the file that is currently shown/edited</li> <li><code>START_CURSOR</code>, <code>END_CURSOR</code>: Only used for the <code>cursors_*</code> commands.</li> </ul> </li> <li>You can also use environment variables to set parameters for your commands. For this, edit the <code>env_vars</code> section of the config. For example, the <code>WINDOW</code> setting controls the number of context lines shown when editing a file.</li> </ul>"},{"location":"config/commands/#displaying-the-command-to-swe-agent","title":"Displaying the Command to SWE-agent","text":"<p>After you define a command, there are a small set of additional steps to making it available for the agent to use.</p> <p>First, within your config file...</p> <ul> <li>Add <code>config/commands/<file name>.sh</code> file to the <code>command_files</code> field.</li> <li>Set the <code>parse_command</code> field to <code>ParseCommandBash</code> or <code>ParseCommandDetailed</code>. This key points to the functionality that generates how command documentation is shown to the agent.</li> <li>Decide which template(s) you want to show the <code>{command_docs}</code> in.<ul> <li>We strongly recommend including <code>{command_docs}</code> in the <code>system_template</code>, which is the first message shown to the agent for every task instance episode.</li> <li>You might also consider adding <code>{command_docs}</code> to the <code>format_error_template</code>, which is shown if the response provided by a model is malformed.</li> </ul> </li> <li>(Optional) Including a demonstration that uses a command is helpful to showcase proper use + increases the frequency with which the agent uses the command. If you'd like to add a demonstration...<ul> <li>Create a demonstration manually (i.e. <code>python run.py --model human_thought ...</code>) or automatically (i.e. <code>python run_replay --traj_path ...</code>)</li> <li>Add/Update the demonstration to the <code>demonstrations</code> argument.</li> <li>Update <code>demonstration_template</code> to control how the demonstration is displayed to the agent.</li> </ul> </li> </ul> <p>Config files</p> <p>If you're not familiar with how SWE-agent configuration files work, we recommend checking out the <code>config</code> documentation.</p> <p>Next, run your configuration and see how your agent uses the commands! <pre><code>python run.py --config_file config/[your config].yaml ...\n</code></pre></p> <ul> <li> <p> Something broken? Report bug</p> </li> <li> <p> Something unclear? Ask question</p> </li> </ul>"},{"location":"config/config/","title":"Configuration","text":"<p>This page contains details describing how to write your own configurations to control how agents can interact with the <code>SWEEnv</code> environment.</p> <p>A configuration is represented as a single <code>.yaml</code> file, specified by the <code>--config</code> flag in the command line interface, allowing you to...</p> <ul> <li>Define the commands that agents may use to traverse + modify a codebase (see here for more details)</li> <li>Write prompts that are deterministically/conditionally shown to the agent over the course of a single trajectory.</li> <li>Control the input/output interface that sits between the agent and <code>SWEEnv</code>.</li> </ul> <p>Default config files</p> <p>Our default config files are in the <code>config/</code> directory.</p>"},{"location":"config/config/#configuration-file-fields","title":"Configuration File Fields","text":"<p>The configuration is a <code>.yaml</code> file that consists of several fields. They are fully represented in this following outline:</p> <pre><code># Prompt Templates: Control how observations of environment are shown to agent\nsystem_template: | # .yaml syntax for multi-line string value\n First `system` message shown to agent\ninstance_template: |- # .yaml syntax for multi-line string value w/ no new line\n Instance prompt, contains task instance-specific content\nnext_step_template: |-\n Format template of per-turn observation (Contains standard output from agent's action)\nnext_step_no_output_template: |-\n Format template of observation when there is no standard output from the agent's action\nformat_error_template: |-\n Format template of error message (Used when agent's action causes an error)\ndemonstration_template: |\n Format template for showing a demonstration to the agent\ndemonstrations:\n- `trajectories/<username>/<experiment folder>/*.traj`\n- File is a demonstration of how to solve a task. This could an agent generated trajectory.\n- You can include 1+ demonstrations\n\n# Environment States: Define features of the SWEEnv environment\nenv_variables:\n# Default variables for SWEEnv at the beginning of each instance\n CURRENT_FILE: 0\n CURRENT_LINE:\n OVERLAP:\n SEARCH_FILES:\n SEARCH_INDEX:\n SEARCH_RESULTS:\n WINDOW_SIZE:\n START_INDEX:\n END_INDEX:\n START_CURSOR:\n END_CUROSR:\n START_CURSORS_MARK:\n END_CURSOR_MARK:\nstate_command: |\n# `state_command` allows you to update state variables to reflect any aspect of the environment (e.g. current working directory)\n name: state\n code: |\n state() { echo '{\"pwd\": \"'$PWD'\"}';\n\n# Action Interface: Define how an agent interacts with the SWEEnv environment\ncommand_files:\n- path/to/bash_file.sh\n- Each file contains a list of commands implemented in bash\n- You can include 1+ command files\nparse_command: Reference to functionality for defining command documentation\nhistory_processor: Reference to functionality for controlling agent's message history\nparse_function: Parser run on agent output\n</code></pre> <p>In the <code>config/</code> directory, we recommend looking at...</p> <ul> <li><code>configs/</code> for examples of properly formatted configuration files. Each configuration differs in its set of commands, input/output format, demonstrations, etc.</li> <li><code>commands/</code> for the bash implementations of the custom commands that SWE-agent uses to navigate + edit the codebase. More information here.</li> </ul> <p>Relative paths</p> <p>Relative paths in config files are resolved to the <code>SWE_AGENT_CONFIG_ROOT</code> environment variable (if set) or the SWE-agent repository root.</p> <code>default_from_url.yaml</code> <pre><code>system_template: |-\n SETTING: You are an autonomous programmer, and you're working directly in the command line with a special interface.\n\n The special interface consists of a file editor that shows you {WINDOW} lines of a file at a time.\n In addition to typical bash commands, you can also use the following commands to help you navigate and edit files.\n\n COMMANDS:\n {command_docs}\n\n Please note that THE EDIT COMMAND REQUIRES PROPER INDENTATION.\n If you'd like to add the line ' print(x)' you must fully write that out, with all those spaces before the code! Indentation is important and code that is not indented correctly will fail and require fixing before it can be run.\n\n RESPONSE FORMAT:\n Your shell prompt is formatted as follows:\n (Open file: <path>) <cwd> $\n\n You need to format your output using two fields; discussion and command.\n Your output should always include _one_ discussion and _one_ command field EXACTLY as in the following example:\n DISCUSSION\n First I'll start by using ls to see what files are in the current directory. Then maybe we can look at some relevant files to see what they look like.\n ```\n ls -a\n ```\n\n You should only include a *SINGLE* command in the command section and then wait for a response from the shell before continuing with more discussion and commands. Everything you include in the DISCUSSION section will be saved for future reference.\n If you'd like to issue two commands at once, PLEASE DO NOT DO THAT! Please instead first submit just the first command, and then after receiving a response you'll be able to issue the second command.\n You're free to use any other bash commands you want (e.g. find, grep, cat, ls, cd) in addition to the special commands listed above.\n However, the environment does NOT support interactive session commands (e.g. python, vim), so please do not invoke them.\ninstance_template: |-\n We're currently solving the following issue within our repository. Here's the issue text:\n ISSUE:\n {issue}\n\n INSTRUCTIONS:\n Now, you're going to solve this issue on your own. Your terminal session has started and you're in the repository's root directory. You can use any bash commands or the special interface to help you. Edit all the files you need to and run any checks or tests that you want.\n Remember, YOU CAN ONLY ENTER ONE COMMAND AT A TIME. You should always wait for feedback after every command.\n When you're satisfied with all of the changes you've made, you can submit your changes to the code base by simply running the submit command.\n Note however that you cannot use any interactive session commands (e.g. python, vim) in this environment, but you can write scripts and run them. E.g. you can write a python script and then run it with `python <script_name>.py`.\n\n NOTE ABOUT THE EDIT COMMAND: Indentation really matters! When editing a file, make sure to insert appropriate indentation before each line!\n\n IMPORTANT TIPS:\n 1. Always start by trying to replicate the bug that the issues discusses.\n If the issue includes code for reproducing the bug, we recommend that you re-implement that in your environment, and run it to make sure you can reproduce the bug.\n Then start trying to fix it.\n When you think you've fixed the bug, re-run the bug reproduction script to make sure that the bug has indeed been fixed.\n\n If the bug reproduction script does not print anything when it successfully runs, we recommend adding a print(\"Script completed successfully, no errors.\") command at the end of the file,\n so that you can be sure that the script indeed ran fine all the way through.\n\n 2. If you run a command and it doesn't work, try running a different command. A command that did not work once will not work the second time unless you modify it!\n\n 3. If you open a file and need to get to an area around a specific line that is not in the first 100 lines, say line 583, don't just use the scroll_down command multiple times. Instead, use the goto 583 command. It's much quicker.\n\n 4. If the bug reproduction script requires inputting/reading a specific file, such as buggy-input.png, and you'd like to understand how to input that file, conduct a search in the existing repo code, to see whether someone else has already done that. Do this by running the command: find_file \"buggy-input.png\" If that doesn't work, use the linux 'find' command.\n\n 5. Always make sure to look at the currently open file and the current working directory (which appears right after the currently open file). The currently open file might be in a different directory than the working directory! Note that some commands, such as 'create', open files, so they might change the current open file.\n\n 6. When editing files, it is easy to accidentally specify a wrong line number or to write code with incorrect indentation. Always check the code after you issue an edit to make sure that it reflects what you wanted to accomplish. If it didn't, issue another command to fix it.\n\n 7. It may be necessary to install the repository from source before you can run code. Please think about how to install the environment from the repository directory if you need to do so.\n\n\n (Open file: {open_file})\n (Current directory: {working_dir})\n bash-$\nnext_step_template: |-\n {observation}\n (Open file: {open_file})\n (Current directory: {working_dir})\n bash-$\nnext_step_no_output_template: |-\n Your command ran successfully and did not produce any output.\n (Open file: {open_file})\n (Current directory: {working_dir})\n bash-$\ndemonstration_template: |\n Here is a demonstration of how to correctly accomplish this task.\n It is included to show you how to correctly use the interface.\n You do not need to follow exactly what is done in the demonstration.\n --- DEMONSTRATION ---\n {demonstration}\n --- END OF DEMONSTRATION ---\nstate_command:\n name: state\n code: |\n state() {\n local working_dir=\"$PWD\";\n if [ -z \"$CURRENT_FILE\" ]; then\n echo '{\"open_file\": \"n/a\", \"working_dir\": \"'$working_dir'\"}';\n else\n echo '{\"open_file\": \"'$(realpath \"$CURRENT_FILE\")'\", \"working_dir\": \"'$working_dir'\"}';\n fi\n };\nparse_function: ThoughtActionParser\nenv_variables:\n WINDOW: 100\n OVERLAP: 2\n CURRENT_LINE: 0\n CURRENT_FILE: ''\n SEARCH_RESULTS: ()\n SEARCH_FILES: ()\n SEARCH_INDEX: 0\ncommand_files:\n- config/commands/defaults.sh\n- config/commands/search.sh\n- config/commands/edit_linting.sh\n- config/commands/_split_string.py\n- config/commands/submit.sh\nparse_command: ParseCommandDetailed\nhistory_processor: Last5Observations\ndemonstrations:\n- trajectories/demonstrations/replay__marshmallow-code__marshmallow-1867__default__t-0.20__p-0.95__c-2.00__install-1___install_from_source/marshmallow-code__marshmallow-1867.traj\n</code></pre>"},{"location":"config/config/#how-a-configuration-file-is-processed","title":"How a Configuration File is Processed","text":"<p>Some notes on processing that occurs on config fields when SWE-agent is run:</p> <ul> <li>Commands specified in <code>command_files</code> will be parsed into a single block of documentation text that can be referenced as <code>{command_docs}</code>.</li> <li><code>env_variables</code> are the default variables for the bash environment at the beginning of each instance.</li> <li><code>state_command</code> is used to extract state information from the bash environment (formatted as json) to be used in the templates given to the agent.</li> </ul> <p>Possible variables that can be used in templates are: - <code>{command_docs}</code> (an automatically compiled collection of available commands + their docstrings) - any variable given in <code>env_variables</code> (same spelling), e.g., <code>{WINDOW_SIZE}</code> - any variable extracted as json as part of the <code>state_command</code> function - the last observation <code>{observation}</code> - ... this list will grow as we implement more features!</p>"},{"location":"config/config/#template-workflow","title":"Template Workflow","text":"<p>The following diagram illustrates where each template is shown within a single episode of solving one task instance.</p> <p></p> <p>One of three templates can be shown per turn:</p> <ul> <li>\"Next Step\" (<code>next_step_template</code>): Displayed if the model's action successfully runs. The output and a prompt for the next action is shown</li> <li>\"Next Step (No Output)\" (<code>next_step_no_output_template</code>): Displayed if the model's action successfully runs, but does not produce any standard output (e.g. <code>rm</code>, <code>cd</code>)</li> <li>\"Format Error\" (<code>format_error_template</code>): Displayed if the model's response is malformed. Over the next two turns...</li> <li>If one of the model's next response is correct, the message history is updated such that the \"Format Error\" turn is not kept. The episode continues.</li> <li>If the model's next two responses are both malformed, the episode terminates.</li> </ul> <ul> <li> <p> Something broken? Report bug</p> </li> <li> <p> Something unclear? Ask question</p> </li> </ul>"},{"location":"config/demonstrations/","title":"Changing the demonstrations","text":"<p>An important way to show LMs how to use commands and interact with the environment is through providing a demonstration - which is basically a completed trajectory that the LM can learn from.</p> <p>For simplicity we only ingest demonstrations in the from of a trajectory file. However, since trajectory files are usually JSON, you can convert them to yaml using the <code>make_demos/convert_traj_to_demo.py</code> script to be more human-readable and easier to edit.</p> <p>Demo (yaml) files are stored in the <code>make_demos/demos</code> directory by default and consist primarily of the sequence of actions that an LM would need to take to complete a task. It's important that your demo have the proper format to be parsed by SWE-agent and your config.</p>"},{"location":"config/demonstrations/#manually-creating-a-custom-trajectory","title":"Manually creating a custom trajectory","text":"<p>You can manually generate a trajectory by running the agent with <code>--model_name human_thought</code>. This lets you input, at each turn, the thought (ending with END_THOUGHT) and then the action (a single command).</p> <p>You should then convert that trajectory into a demonstration as shown below.</p> <p>To edit text in <code>human_thought</code> mode:</p> <ol> <li>Run the command <code>edit edit_start_line:edit_end_line</code></li> <li>Write the text you want to insert. Feel free to write the text across multiple lines.</li> <li>Press <code>return</code> then write <code>end_of_edit</code> and then press <code>return</code> again to submit the edit.</li> </ol> <p>If you would like to run <code>human_thought</code> mode without having to type in a thought at each turn (for debugging for example), use <code>--model_name human</code>.</p>"},{"location":"config/demonstrations/#converting-an-existing-trajectory-into-a-demonstration","title":"Converting an existing trajectory into a demonstration","text":"<p>Here's how you can make a demo from an existing trajectory file (like the one created from the previous step):</p> <ol> <li>Find a basic trajectory that you already like and want to use as the basis for your demo. For instance, consider the <code>.traj</code> files in the <code>trajectories/demonstrations/</code> folder or find the trajectory from the previous step (the path will be printed at the bottom).</li> <li>Run <code>python convert_traj_to_demo.py <path to trajectory file.traj></code> to convert the trajectory to a demo. This demo will be saved as a readable yaml file in the <code>make_demos/demos</code> directory.</li> <li>Edit the demo by hand to make it work for your particular use case and configuration.</li> <li>(Optional) Run <code>python run_replay.py --traj_path <path to demo> --config_file <path to config file></code> to execute the actions of the demo, have the system generate the execution output, and ensure that it works as expected.</li> <li>Inspect the resulting trajectory to ensure it was executed correctly.</li> <li>Specify the path to your demonstration in your config file</li> </ol> <ul> <li> <p> Something broken? Report bug</p> </li> <li> <p> Something unclear? Ask question</p> </li> </ul>"},{"location":"config/docker/","title":"Build your own agent Docker image","text":"<p>This section is about modifying the Docker image in which we run the commands suggested by the agent.</p> <p>There are two reasons to build your own Docker image</p> <ol> <li>You need a very specific environment to install your package (anything that you cannot simply install inside of a conda environment)</li> <li>You want to pre-install your package for speedup.</li> </ol> <p>There are three steps involved:</p> <ol> <li>Modify the <code>swe.Dockerfile</code> Dockerfile (also shown below). We provide some extended explanation of the Dockerfile here.</li> <li>Build the image. One way is to simply run <code>./setup.sh</code>. Alternatively, especially if you want to change the default tag (<code>sweagent/swe-agent:latest</code>), run <pre><code>docker build -t \"YOUR TAG HERE\" -f docker/swe.Dockerfile \\\n --build-arg TARGETARCH=$(uname -m) .\n</code></pre></li> <li>Make sure you use the new image by passing the <code>--image_name</code> flag to <code>run.py</code>.</li> </ol> <p>Default Dockerfile:</p> <pre><code>FROM ubuntu:jammy\n\nARG TARGETARCH\n\n# Install third party tools\nRUN apt-get update && \\\n apt-get install -y bash gcc git jq wget g++ make && \\\n apt-get clean && \\\n rm -rf /var/lib/apt/lists/*\n\n# Initialize git\nRUN git config --global user.email \"sweagent@pnlp.org\"\nRUN git config --global user.name \"sweagent\"\n\n# Environment variables\nENV ROOT='/dev/'\nRUN prompt() { echo \" > \"; };\nENV PS1=\"> \"\n\n# Create file for tracking edits, test patch\nRUN touch /root/files_to_edit.txt\nRUN touch /root/test.patch\n\n# add ls file indicator\nRUN echo \"alias ls='ls -F'\" >> /root/.bashrc\n\n# Install miniconda\nENV PATH=\"/root/miniconda3/bin:${PATH}\"\nARG PATH=\"/root/miniconda3/bin:${PATH}\"\nCOPY docker/getconda.sh .\nRUN bash getconda.sh ${TARGETARCH} \\\n && rm getconda.sh \\\n && mkdir /root/.conda \\\n && bash miniconda.sh -b \\\n && rm -f miniconda.sh\nRUN conda --version \\\n && conda init bash \\\n && conda config --append channels conda-forge\n\n# Cache python versions\nRUN conda create -y -n python3.9 python=3.9\nRUN conda create -y -n python3.10 python=3.10\n\n# Install python packages\nCOPY docker/requirements.txt /root/requirements.txt\nRUN pip install -r /root/requirements.txt\n\nWORKDIR /\n\nCMD [\"/bin/bash\"]\n</code></pre>"},{"location":"config/env/","title":"Environment variables","text":"<p>Persisting environment variables</p> <p>All environment variables can also be added to <code>keys.cfg</code> instead. See here for more information.</p> <p>This page details all environment variables that are currently in use by SWE-agent.</p> <ul> <li>All API keys (for LMs and GitHub) can be set as an environment variable. See here for more information.</li> <li><code>SWE_AGENT_CONFIG_ROOT</code>: Used to resolve relative paths in the config</li> <li><code>SWE_AGENT_ENV_LONG_TIMEOUT</code> (default: 500): Timeout in seconds used for commands that install instance environment.</li> <li><code>SWE_AGENT_ACTION_TIMEOUT</code> (default: 25): Timeout in seconds used for commands issued by the agent</li> <li><code>SWE_AGENT_ACTION_NO_OUTPUT_TIMEOUT</code> (default: equal to <code>SWE_AGENT_ACTION_TIMEOUT</code>): Timeout in seconds used when no output is produced for the defined duration for commands issued by the agent</li> <li><code>SWE_AGENT_MODEL_MAX_RETRIES</code> (default: 10): Maximum retries when querying the model</li> </ul> <p>The following three variables can only be set as environment variables, not in the config file</p> <ul> <li><code>SWE_AGENT_LOG_TIME</code>: Add timestamps to log</li> <li><code>SWE_AGENT_LOG_STREAM_LEVEL</code>: Level of logging that is shown on the command line interface (<code>TRACE</code> being a custom level below <code>DEBUG</code>)</li> <li><code>SWE_AGENT_LOG_FILE_LEVEL</code>: Like <code>SWE_AGENT_LOG_STREAM_LEVEL</code> but for the log file</li> </ul> <p>Unstable</p> <p>The following variables might still be subject to change</p> <ul> <li><code>SWE_AGENT_COMMUNICATE_METHOD</code>: Determines how SWE-agent communicates with the running process in the docker container: <code>end-marker</code> (default, fast) or <code>processes</code> (legacy, slow, more tested)</li> <li><code>SWE_AGENT_CLONE_METHOD</code>: <code>shallow</code> (default, only retrieves relevant commit) or <code>full</code> (clones repository including full history). When using persistent containers or running over multiple problem statements, we fall back to <code>full</code>.</li> <li><code>SWE_AGENT_DOCKER_START_UP_DELAY</code>: Number of seconds to wait after starting a docker container</li> </ul>"},{"location":"config/summarizers/","title":"Summarizers Configuration","text":"<p>LMs perform best if given concise inputs; superfluous context can degrade performance while increasing costs. Because agents require LMs to process entire trajectories, compressing context is of particular importance.</p> <p>We designed two summarizers to handle long output commands. The first, a simple summarizer, saves the command output to a file if it exceeds a certain configurable line count. We show an indicative warning to the agent and tell it to open the saved command output using the built-in SWE-agent utility <code>open</code>.</p> <p>The second, an LM summarizer, integrates with the main agent to enhance its problem-solving efficacy and avoid exceeding input length limitations. It employs the same model (though can be configured differently) as the main agent; it receives a configurable prompt containing some context about the problem, the most recent action taken by the main agent, and any observations from that action exceeding a configurable line count threshold. The LM summarizer then generates a concise summary of the observation that is directly relevant to the ongoing problem. This summary is sent to the main agent, accompanied by a warning message indicating that the command output was summarized due to exceeding the line count limit.</p> <p>The summarizer configuration is defined within the main configuration <code>.yaml</code> file, and has the following structure:</p> <pre><code>summarizer_config:\n function: Reference to functionality of the summarizer. Can be one of SimpleSummarizer, LMSummarizer or Identity\n window_length: Threshold of the line count limit. Observation output exceeding these number, will be summarized.\n system_template: |-\n First `system` message shown to the LMSummarizer.\n This has no effect in other summarizer functionalities.\n instance_template: |-\n Instance prompt, contains task instance-specific content,\n the most recent action taken by the main agent,\n and any observations from that action exceeding the line count threshold.\n This has effect only for the LM Summarizer functionality.\n model: [Optional configuration of a different model,\n if not configured the same model as the main agent model will be used.\n This has effect only for the LM Summarizer functionality.]\n model_name: Name of the model to use\n per_instance_cost_limit: Cost limit for every instance (task)\n total_cost_limit: Total cost limit for summarizer only\n temperature: Sampling temperature\n top_p: Sampling top-p\n host_url: Host URL when using Ollama model\n</code></pre>"},{"location":"dev/contribute/","title":"Contribute to SWE-agent","text":"<p>Formatting change</p> <p>We've recently added automated formatting to our code base. If you are dealing with merge-conflicts when opening a PR or updating your fork, please first install <code>pre-commit</code> and run <code>pre-commit run --all-files</code> and try again.</p> <p>The easiest way to contribute is to give us feedback.</p> <ul> <li>Something isn't working? Open a bug report. Rule of thumb: If you're running something and you get some error messages, this is the issue type for you.</li> <li>You have a concrete question? Open a question issue.</li> <li>You are missing something? Open a feature request issue</li> <li>Open-ended discussion? Talk on discord. Note that all actionable items should be an issue though.</li> </ul> <p>Wanna do more and actually contribute code? Great! Please see the following sections for tips and guidelines!</p>"},{"location":"dev/contribute/#development-repository-set-up","title":"Development repository set-up","text":"<p>Please install the repository from source, following our usual instructions but add the <code>[dev]</code> option to the <code>pip</code> command (you can just run the command again):</p> <pre><code>pip install -e '.[dev]'\n</code></pre> <p>Then, make sure to set up <code>pre-commit</code>:</p> <pre><code># cd to our repo root\npre-commit install\n</code></pre> <p><code>pre-commit</code> will check for formatting and basic syntax errors before your commits.</p> <p>Autofixes</p> <p>Most problems (including formatting) will be automatically fixed. Therefore, if <code>pre-commit</code>/<code>git commit</code> fails on its first run, simply try running it a second time.</p> <p>Some more autofixes can be enabled with the <code>--unsafe-fixes</code> option from <code>ruff</code>:</p> <pre><code>pipx run ruff check --fix --unsafe-fixes\n</code></pre>"},{"location":"dev/contribute/#running-tests","title":"Running tests","text":"<p>We provide a lot of tests that can be very helpful for rapid development. Run them with</p> <pre><code>pytest\n</code></pre> <p>Some of the tests might be slower than others. You can exclude them with</p> <pre><code>pytest -m \"not slow\"\n</code></pre>"},{"location":"dev/contribute/#tips-for-pull-requests","title":"Tips for pull requests","text":"<ul> <li>If you see a lot of formatting-related merge conflicts, please see here.</li> <li>Please open separate PRs for separate issues. This makes it easier to incorporate part of your changes.</li> <li>It might be good to open an issue and discuss first before investing time on an experimental feature.</li> <li>Don't know where to get started? Look for issues marked \ud83d\udc4b good first issue or \ud83d\ude4f help wanted</li> <li>When changing the behavior of the agent, we need to have some indication that it actually improves the success rate of SWE-agent. However, if you make the behavior optional without complicating SWE-agent (for example by providing new commands), we might be less strict.</li> <li>Please add simple unit tests or integration tests wherever possible. Take a look in the tests directory for inspiration. We emphasize simple easy-tow-rite tests that get a lot of coverage.</li> </ul>"},{"location":"dev/contribute/#building-the-documentation","title":"Building the documentation","text":"<p>Simply run</p> <pre><code># cd repo root\nmkdocs serve\n</code></pre> <p>and point your browser to port 8000 or click one of the links in the output.</p>"},{"location":"dev/contribute/#diving-into-the-code","title":"Diving into the code","text":"<ul> <li> <p> Code structure and reference</p> <p>Read the reference for more information on our code.</p> <p> Read more</p> </li> </ul> <ul> <li> <p> Something broken? Report bug</p> </li> <li> <p> Something unclear? Ask question</p> </li> </ul>"},{"location":"dev/formatting_conflicts/","title":"Formatting conflicts","text":"<p>On May 28th, 2024, we introduced automated formatting with <code>ruff-format</code> and <code>pre-commit</code>. This changed almost every file in the project. If you forked or branched off before these changes and now try to synchronize your fork/branch with <code>princeton-nlp/SWE-agent:main</code>, you will see a lot of merge conflicts.</p> <p>To solve this, you need to apply the same formatting to your code. Here's how you can do it.</p> <p>First let's add the official remote (if it exists, you've probably already added it and you can ignore the warning).</p> <pre><code>git remote add upstream https://github.com/princeton-nlp/SWE-agent.git\ngit fetch upstream\n</code></pre> <p>Now, you need the updated <code>pyproject.toml</code> and <code>.pre-commit-config.yaml</code> files. We can get them from <code>princeton-nlp/SWE-agent:main</code>:</p> <pre><code>git checkout upstream/main -- .pre-commit-config.yaml pyproject.toml\ngit commit -m \"Update formatting instructions\" --no-verify\n</code></pre> <p>Let's assume that your changes are on branch <code>FEATURE_BRANCH</code>, for example, if you've committed to <code>main</code>:</p> <pre><code>export FEATURE_BRANCH=\"main\"\n</code></pre> <p>Next we create a copy of this branch (so we don't further modify it):</p> <pre><code>git branch \"${FEATURE_BRANCH}\" \"${FEATURE_BRANCH}_REBASED\"\n</code></pre> <p>And now comes the tricky bit: We rebase your changes on top of <code>upstream/mean</code>, while applying the formatting fixes at every step:</p> <pre><code>git rebase upstream/main \"${FEATURE_BRANCH}_REBASED\" \\\n -Xtheirs \\\n --exec 'git reset --soft HEAD^; pre-commit run; pipx run ruff check --fix --unsafe-fixes; git add -u; git commit -C HEAD@{1} --no-verify'\n</code></pre> <p>Understanding the last command</p> <p>Here's what is happening:</p> <ul> <li><code>git rebase upstream/main \"${FEATURE_BRANCH}_REBASED\"</code> applies every commit from <code>\"${FEATURE_BRANCH}_REBASED\"</code> on top of <code>upstream/main</code>.</li> <li><code>-Xtheirs</code> tells git to always take your changes for merge conflicts (rather than the format changes).</li> <li>After every commit, the command from <code>--exec</code> is being called.<ul> <li><code>git reset --soft HEAD^</code> undos the <code>git commit</code> action (while leaving the changes staged),</li> <li>then we apply the formatting, and</li> <li>finally we commit the formatted changes again.</li> </ul> </li> </ul> <p>Still merge conflicts?</p> <p>It's possible that there are non-formatting-related merge conflicts that you are encountering. In this case, <code>git rebase</code> will stop every time it cannot resolve the conflict. Simply fix the merge conflicts as you would normally do (edit the file, commit once done), and then run <code>git rebase --continue</code>.</p> <p>You can now open a PR from <code>${FEATURE_BRANCH}_REBASED</code> or make it your new default branch.</p> <ul> <li> <p> Something broken? Report bug</p> </li> <li> <p> Something unclear? Ask question</p> </li> </ul>"},{"location":"installation/","title":"Setting up SWE-agent","text":"<ul> <li> <p> All in browser</p> <p>Run SWE-agent using GitHub codespaces in an in-browser VSCode environment. Best for a quick first peek.</p> <p> Get started</p> </li> <li> <p> Install from source</p> <p>Install SWE-agent locally from source using <code>pip</code>. This is the default option.</p> <p> Get started</p> </li> <li> <p> Run in docker</p> <p>Pull a docker container and directly run SWE-agent. This is our fallback solution if the local installation does not work for you.</p> <p> Get started</p> </li> <li> <p> Changelog</p> <p>See what's new in SWE-agent!</p> <p> Read the changelog</p> </li> </ul>"},{"location":"installation/changelog/","title":"Changelog","text":""},{"location":"installation/changelog/#070-2024-09-23","title":"0.7.0 (2024-09-23)","text":""},{"location":"installation/changelog/#added","title":"Added","text":"<p>The main new feature is the EnIGMA mode, which included additions like support for Interactive Agent Tools and Summarizers.</p> <ul> <li>Add filemap command in the spirit of repomap by @samuela in #619</li> <li>Create config to run human eval style challenges by @ofirpress in #658</li> <li>Add claude 3.5 sonnet to models by @carlosejimenez in #601</li> <li>Enh: Warn if scrolling >= 3 times by @klieret in #626</li> <li>feat: support deepseek-coder LLM by @jcraftsman in #638</li> <li>Enh: Make timeout for agent commands configurable by @klieret in #674</li> <li>Add support for new gpt-4o-mini model by @ivan4722 in #693</li> <li>Groq Models Integration by @MohammedNagdy in #721</li> <li>Make log level configurable; add TRACE level by @klieret in #612</li> </ul>"},{"location":"installation/changelog/#fixes","title":"Fixes","text":"<ul> <li>Compatibility with SWE-bench 2.0 by @klieret in #671</li> <li>ensure variables work in special command docstring by @forresty in #628</li> <li>Important fix: Catch CostLimitExceeded in retry because of format/block by @klieret in #682</li> <li>Fix: Handle empty traj in should_skip by @klieret in #616</li> <li>Fix for end-marker communicate: Exit status always 0/invalid by @klieret in #644</li> <li>Fix: Insufficient quoting of git commit message by @klieret in #646</li> <li>Fix nonsensical trajectory formatting for PRs by @klieret in #647</li> <li>Fix: sweunexpected keyword 'python_version' by @klieret in #692</li> <li>Fix: Use LONG_TIMEOUT for pre_install commands by @klieret in #695</li> <li>Fix: UnboundLocalError when catching decoding issue by @klieret in #709</li> <li>Also create empty patch files for completeness by @klieret in #725</li> <li>Fix: Raise ContextWindowExceeded instead of exit_cost by @klieret in #727</li> <li>Fix: Deal with non-utf8 encoded bytes in comm by @klieret in #731</li> <li>Fix: Handle spaces in repo names by @klieret in #734</li> <li>Fix: Ensure utils is part of package by @klieret in #742</li> <li>Fix: Submitting ' ' in human mode crashes container by @klieret in #749</li> <li>Fix: Block su as command by @klieret in #752</li> <li>Fix: SWE_AGENT_MODEL_MAX_RETRIES needs casting by @klieret in #757</li> </ul>"},{"location":"installation/changelog/#new-contributors","title":"New Contributors","text":"<p>\ud83c\udf89 @talorabr, @udiboy1209, @haoranxi, @NickNameInvalid, @rollingcoconut joined the team to build EnIGMA \ud83c\udf89</p> <ul> <li>@carlosejimenez made their first contribution in #601</li> <li>@samefarrar made their first contribution in #606</li> <li>@hubstrauss made their first contribution in #625</li> <li>@samuela made their first contribution in #619</li> <li>@forresty made their first contribution in #628</li> <li>@jcraftsman made their first contribution in #638</li> <li>@ivan4722 made their first contribution in #693</li> <li>@JoshuaPurtell made their first contribution in #703</li> <li>@MohammedNagdy made their first contribution in #721</li> <li>@pdemro made their first contribution in #729</li> </ul>"},{"location":"installation/changelog/#061-2024-06-20","title":"0.6.1 (2024-06-20)","text":"<p>All new commits</p> <p>This is (mostly) a patch release, in particular fixing several issues that had been introduced by the speed improvements of v0.6.0. We also solve a bug where existing linter errors in a file left SWE-agent unable to edit (because of our lint-retry-loop).</p>"},{"location":"installation/changelog/#breaking-changes","title":"Breaking changes","text":"<ul> <li>Change: sparse clone method is now correctly called \"shallow\" by @klieret in #591</li> </ul>"},{"location":"installation/changelog/#improved","title":"Improved","text":"<ul> <li>Enh: Show commands when encountering timeout error by @klieret in #582</li> <li>Enh: Configuration option to show time in log by @klieret in #583</li> <li>Enh: Allow to configure LONG_TIMEOUT for SWEEnv by @klieret in #584</li> <li>Enh: Always write log to traj directory by @klieret in #588</li> </ul>"},{"location":"installation/changelog/#fixed","title":"Fixed","text":"<ul> <li>fix <code>docker.errors.NotFound</code> by @klieret in #587</li> <li>Fix: Revert to full clone method when needed by @klieret in #589</li> <li>Fix: Refresh container_obj before querying status by @klieret in #590</li> <li>Fixed #571 - show message that model arg is ignored in case of using Azure OpenAI by @jank in #592</li> <li>Fix: Linting blocks for existing lint errors by @klieret in #593</li> <li>Fix: Process done marker not found in read with timeout by @klieret in #596</li> </ul>"},{"location":"installation/changelog/#060-2024-06-05","title":"0.6.0 (2024-06-05)","text":"<p>All new commits</p> <p>We sped up SWE-agent by 2x (timed with GPT4o). This is mostly due to faster communication with the running processes inside of the Docker container and other container setup & installation related improvements. Here are a few relevant PRs:</p> <ul> <li>Switch to fast communicate and shallow clone by default by @klieret in #530</li> <li>Change: Only wait 1s for docker to start by @klieret in #541</li> <li>Feat: experimental shallow cloning by @klieret in #498</li> <li>Enh: Start from clone of python conda environment for speedup by @klieret in #548</li> <li>Enh: Use uv for editable install by default by @klieret in #547</li> </ul>"},{"location":"installation/changelog/#improved_1","title":"Improved","text":"<ul> <li>Improve scrolling behavior in web UI by @anishfish2 in #420</li> <li>Web UI: Render Markdown in agent feed messages. by @kwight in #486</li> <li>Enh: Remove redundant 'saved traj to X' messages by @klieret in #528</li> <li>Allow to disable config dump to log by @klieret in #537</li> <li>Resolve relative paths to demonstrations and commands by @klieret in #444</li> </ul>"},{"location":"installation/changelog/#fixed_1","title":"Fixed","text":"<ul> <li>Web UI: Remove -n option to wait by @klieret in #487</li> <li>Web UI: Kill the Flask server on exit. by @kwight in #479</li> <li>Web UI: Avoid proxy errors on MacOS by @klieret in #506</li> <li>Ensure container_name is reset for non-persistent containers by @klieret in #463</li> <li>Fix: Do not allow persistent container with cache task imgs by @klieret in #551</li> </ul>"},{"location":"installation/changelog/#050-2024-05-28","title":"0.5.0 (2024-05-28)","text":"<p>All new commits</p> <p>\u2728 The big news is our brand new documentation \u2728</p> <p>Secondly, @ollmer added a new flag <code>--cache_task_images</code> that will significantly speed up SWE-agent when running on the same environment/repository multiple times (no more waiting for cloning and installation!)</p>"},{"location":"installation/changelog/#breaking-changes_1","title":"Breaking changes","text":"<ul> <li>We have reformatted our codebase. If you create a PR based on a previous commit, make sure you install our <code>pre-commit</code> hook to avoid merge-conflicts because of formatting. See our docs for more information.</li> <li>Remove direct imports in <code>__init__.py</code> (you can no longer <code>from sweagent import Agent</code> by @klieret in #436</li> </ul>"},{"location":"installation/changelog/#added_1","title":"Added","text":"<ul> <li>Running the web UI is now supported when running swe-agent completely in docker</li> <li>Speed up evaluation by caching task environments as docker images by @ollmer in #317</li> </ul>"},{"location":"installation/changelog/#improved_2","title":"Improved","text":"<ul> <li>Add gpt-4o model by @raymyers in #344</li> <li>Web: Allow to specify commit hash by @klieret in #358</li> <li>Add default environment_setup config by @klieret in #351</li> <li>Enh: Suppress openai logging; improve formatting of stats by @klieret in #416</li> <li>Remove signal dependency by @klieret in #428</li> <li>Do not use select if running on Windows by @klieret in #429</li> <li>Use custom Config class to support env and keys.cfg (this allows passing keys as environment variables) by @klieret in #430</li> </ul>"},{"location":"installation/changelog/#fixed_2","title":"Fixed","text":"<ul> <li>Web: Fix script_path input by @klieret in #334</li> <li>Fix: Don't print patch msg for exit_cost patch by @klieret in #343</li> <li>Fix: Do not request job control in bash by @klieret in #345</li> <li>Fix: --base_commit not used for gh urls by @klieret in #346</li> <li>Fix: Separate data path/traj dir cause exception by @klieret in #348</li> <li>Add docker-py lower bound by @klieret in #406</li> <li>Fix: IndexError when replaying incomplete trajectories by @klieret in #410</li> </ul>"},{"location":"installation/changelog/#040-2024-05-09","title":"0.4.0 (2024-05-09)","text":"<p>All new commits</p>"},{"location":"installation/changelog/#added_2","title":"Added","text":"<p>We\u2019re excited to launch the SWE-agent web UI! Specify a bug, press start and watch SWE-agent do the magic.</p>"},{"location":"installation/changelog/#030-2024-05-02","title":"0.3.0 (2024-05-02)","text":""},{"location":"installation/changelog/#added_3","title":"Added","text":"<ul> <li>Run SWE-agent in the cloud using GitHub Codespaces</li> <li>Add GPT4-turbo model by @zgrannan in #252</li> <li>feat: Amazon Bedrock support (Claude models) by @JGalego in #207</li> </ul>"},{"location":"installation/changelog/#fixed_3","title":"Fixed","text":"<ul> <li>Better error handling for --open_pr by @klieret in #239</li> <li>Fixed a potential error by @DanjieTang in #242</li> <li>fix: TARGETARCH not set on some OS/docker setups by @mspronesti in #249</li> <li>Pass Python version to get_environment_yml by @waterson in #271</li> <li>Fix Together model validation error by @mikanfactory in #236</li> <li>Doc: Avoid invalid github token by @klieret in #292</li> </ul>"},{"location":"installation/changelog/#020-2024-04-15","title":"0.2.0 (2024-04-15)","text":"<p>All new commits</p>"},{"location":"installation/changelog/#added_4","title":"Added","text":"<ul> <li>Allow to run on local repos (new flag: <code>--repo_path</code>) in #193</li> <li>Patch files are now saved separately to a patch directory in #126</li> <li>Allow to supply custom installation commands when running on gh issues or locally (<code>--environment_setup</code>) in #153</li> <li>Allow to specify openapi base url in <code>keys.cfg</code> in #118</li> </ul>"},{"location":"installation/changelog/#improved_3","title":"Improved","text":"<ul> <li>Improve error handling of docker issues in #165</li> <li>Make github token fully optional in #189</li> </ul>"},{"location":"installation/changelog/#fixed_4","title":"Fixed","text":"<ul> <li>Fix opening PR from fork in #229</li> <li>Fix: Choosing TogetherAI models in #130</li> </ul>"},{"location":"installation/codespaces/","title":"Running SWE-agent in your browser","text":"<p>Running SWE-agent in your browser is the easiest way to try out our project.</p> <ol> <li>Click </li> <li>Add your API keys to <code>keys.cfg</code> (find the file in the left sidebar and fill out the template). More information on the keys here.</li> <li>Make sure to wait until the <code>postCreateCommand</code> in the terminal window at the bottom is finished</li> <li>Enter your SWE-agent command, see using the web interface or using the command line.</li> </ol>"},{"location":"installation/codespaces/#running-the-web-ui","title":"Running the Web UI","text":"<p>Go to the terminal and enter</p> <pre><code>./start_web_ui.sh\n</code></pre> <p>After a while, you should see a popup offering you to forward port <code>3000</code>. Click <code>Open in Browser</code>.</p> <p></p> <p>If you instead only see the offer to forward port <code>8000</code>, do not click it (this is the port that's being used by the backend).</p> <p>Instead, click on the <code>Ports</code> tab, and click on the globe next to port <code>3000</code>:</p> <p></p> <p>More information</p> <p>See running the web UI for more information about the web UI and additional hints for how to solve problems with the starting it.</p> <ul> <li> <p> Something broken? Report bug</p> </li> <li> <p> Something unclear? Ask question</p> </li> </ul>"},{"location":"installation/docker/","title":"Fallback: Usage with docker","text":"<p>Instead of installing SWE-agent from source (the preferred option), you can also run the software directly using Docker.</p> <p>Limited support</p> <p>Wherever possible, the installation from source is preferred.</p> <ol> <li>Install Docker (follow the docs or use the get-docker.sh script for linux), then start Docker locally. Problems? See docker issues.</li> <li>Run <code>docker pull sweagent/swe-agent:latest</code>. Optional: If you want to use EnIGMA run also <code>docker pull sweagent/enigma:latest</code>.</li> <li>Add your API tokens to a file <code>keys.cfg</code> as explained here or pass them as environment variables.</li> </ol>"},{"location":"installation/docker/#running-the-command-line-interface","title":"Running the command line interface","text":"<p>Assuming that you create <code>keys.cfg</code> in the current directory, run</p> <pre><code>docker run --rm -it -v /var/run/docker.sock:/var/run/docker.sock \\\n -v $(pwd)/keys.cfg:/app/keys.cfg \\\n sweagent/swe-agent-run:latest \\\n python run.py --image_name=sweagent/swe-agent:latest \\\n --model_name gpt4 \\\n --data_path https://github.com/pvlib/pvlib-python/issues/1603 \\\n --config_file config/default_from_url.yaml \\\n --skip_existing=False\n</code></pre> <p>For EnIGMA, change the <code>run.py</code> arguments as usual:</p> <pre><code>docker run --rm -it -v /var/run/docker.sock:/var/run/docker.sock \\\n -v $(pwd)/keys.cfg:/app/keys.cfg \\\n sweagent/swe-agent-run:latest \\\n python run.py --image_name=sweagent/enigma:latest \\\n --ctf \\\n --model_name gpt4 \\\n --data_path /path/to/challenge.json \\\n --repo_path /path/to/repo \\\n --config_file config/default_ctf.yaml \\\n --skip_existing=False\n</code></pre> <p>For more information about running EnIGMA please read our usage instructions.</p> Output <pre><code>INFO \ud83d\udcd9 Arguments: actions:\n apply_patch_locally: false\n open_pr: false\n push_gh_repo_url: ''\n skip_if_commits_reference_issue: true\n agent:\n config:\n _commands:\n - arguments:\n line_number:\n description: the line number to move the window to (if not provided, the\n window will start at the top of the file)\n required: false\n type: integer\n path:\n description: the path to the file to open\n required: true\n type: string\n code: 'open() { if [ -z \"$1\" ] then echo \"Usage: open <file>\" return fi #\n Check if the second argument is provided if [ -n \"$2\" ]; then #\n Check if the provided argument is a valid number if ! [[ $2 =~ ^[0-9]+$\n ]]; then echo \"Usage: open <file> [<line_number>]\" echo\n \"Error: <line_number> must be a number\" return # Exit if the line\n number is not valid fi local max_line=$(awk ''END {print NR}''\n $1) if [ $2 -gt $max_line ]; then echo \"Warning: <line_number>\n ($2) is greater than the number of lines in the file ($max_line)\" echo\n \"Warning: Setting <line_number> to $max_line\" local line_number=$(jq\n -n \"$max_line\") # Set line number to max if greater than max elif\n [ $2 -lt 1 ]; then echo \"Warning: <line_number> ($2) is less than\n 1\" echo \"Warning: Setting <line_number> to 1\" local\n line_number=$(jq -n \"1\") # Set line number to 1 if less than 1 else local\n OFFSET=$(jq -n \"$WINDOW/6\" | jq ''floor'') local line_number=$(jq\n -n \"[$2 + $WINDOW/2 - $OFFSET, 1] | max | floor\") fi else local\n line_number=$(jq -n \"$WINDOW/2\") # Set default line number if not provided fi if\n [ -f \"$1\" ]; then export CURRENT_FILE=$(realpath $1) export\n CURRENT_LINE=$line_number _constrain_line _print elif [ -d\n \"$1\" ]; then echo \"Error: $1 is a directory. You can only open files.\n Use cd or ls to navigate directories.\" else echo \"File $1 not found\" fi}'\n docstring: opens the file at the given path in the editor. If line_number is\n provided, the window will be move to include that line\n end_name: null\n name: open\n signature: open <path> [<line_number>]\n - arguments:\n line_number:\n description: the line number to move the window to\n required: true\n type: integer\n code: 'goto() { if [ $# -gt 1 ]; then echo \"goto allows only one line\n number at a time.\" return fi if [ -z \"$CURRENT_FILE\" ] then echo\n \"No file open. Use the open command first.\" return fi if [ -z\n \"$1\" ] then echo \"Usage: goto <line>\" return fi if\n ! [[ $1 =~ ^[0-9]+$ ]] then echo \"Usage: goto <line>\" echo\n \"Error: <line> must be a number\" return fi local max_line=$(awk\n ''END {print NR}'' $CURRENT_FILE) if [ $1 -gt $max_line ] then echo\n \"Error: <line> must be less than or equal to $max_line\" return fi local\n OFFSET=$(jq -n \"$WINDOW/6\" | jq ''floor'') export CURRENT_LINE=$(jq -n\n \"[$1 + $WINDOW/2 - $OFFSET, 1] | max | floor\") _constrain_line _print}'\n docstring: moves the window to show <line_number>\n end_name: null\n name: goto\n signature: goto <line_number>\n - arguments: null\n code: scroll_down() { if [ -z \"$CURRENT_FILE\" ] then echo \"No file\n open. Use the open command first.\" return fi export CURRENT_LINE=$(jq\n -n \"$CURRENT_LINE + $WINDOW - $OVERLAP\") _constrain_line _print}\n docstring: moves the window down {WINDOW} lines\n end_name: null\n name: scroll_down\n signature: scroll_down\n - arguments: null\n code: scroll_up() { if [ -z \"$CURRENT_FILE\" ] then echo \"No file\n open. Use the open command first.\" return fi export CURRENT_LINE=$(jq\n -n \"$CURRENT_LINE - $WINDOW + $OVERLAP\") _constrain_line _print}\n docstring: moves the window down {WINDOW} lines\n end_name: null\n name: scroll_up\n signature: scroll_up\n - arguments:\n filename:\n description: the name of the file to create\n required: true\n type: string\n code: \"create() { if [ -z \\\"$1\\\" ]; then echo \\\"Usage: create <filename>\\\"\\\n \\ return fi # Check if the file already exists if [ -e \\\"\\\n $1\\\" ]; then echo \\\"Error: File '$1' already exists.\\\"\\t\\topen \\\"$1\\\"\\\n \\ return fi # Create the file an empty new line printf \\\"\\\\\\\n n\\\" > \\\"$1\\\" # Use the existing open command to open the created file \\\n \\ open \\\"$1\\\"}\"\n docstring: creates and opens a new file with the given name\n end_name: null\n name: create\n signature: create <filename>\n - arguments: null\n code: 'submit() { cd $ROOT # Check if the patch file exists and is non-empty if\n [ -s \"/root/test.patch\" ]; then # Apply the patch in reverse git\n apply -R < \"/root/test.patch\" fi git add -A git diff --cached > model.patch echo\n \"<<SUBMISSION||\" cat model.patch echo \"||SUBMISSION>>\"}'\n docstring: submits your current code and terminates the session\n end_name: null\n name: submit\n signature: submit\n - arguments:\n dir:\n description: the directory to search in (if not provided, searches in the\n current directory)\n required: false\n type: string\n search_term:\n description: the term to search for\n required: true\n type: string\n code: 'search_dir() { if [ $# -eq 1 ]; then local search_term=\"$1\" local\n dir=\"./\" elif [ $# -eq 2 ]; then local search_term=\"$1\" if\n [ -d \"$2\" ]; then local dir=\"$2\" else echo \"Directory\n $2 not found\" return fi else echo \"Usage: search_dir\n <search_term> [<dir>]\" return fi dir=$(realpath \"$dir\") local\n matches=$(find \"$dir\" -type f ! -path ''*/.*'' -exec grep -nIH -- \"$search_term\"\n {} + | cut -d: -f1 | sort | uniq -c) # if no matches, return if [ -z\n \"$matches\" ]; then echo \"No matches found for \\\"$search_term\\\" in $dir\" return fi #\n Calculate total number of matches local num_matches=$(echo \"$matches\" |\n awk ''{sum+=$1} END {print sum}'') # calculate total number of files matched local\n num_files=$(echo \"$matches\" | wc -l | awk ''{$1=$1; print $0}'') # if num_files\n is > 100, print an error if [ $num_files -gt 100 ]; then echo \"More\n than $num_files files matched for \\\"$search_term\\\" in $dir. Please narrow\n your search.\" return fi echo \"Found $num_matches matches for\n \\\"$search_term\\\" in $dir:\" echo \"$matches\" | awk ''{$2=$2; gsub(/^\\.+\\/+/,\n \"./\", $2); print $2 \" (\"$1\" matches)\"}'' echo \"End of matches for \\\"$search_term\\\"\n in $dir\"}'\n docstring: searches for search_term in all files in dir. If dir is not provided,\n searches in the current directory\n end_name: null\n name: search_dir\n signature: search_dir <search_term> [<dir>]\n - arguments:\n file:\n description: the file to search in (if not provided, searches in the current\n open file)\n required: false\n type: string\n search_term:\n description: the term to search for\n required: true\n type: string\n code: 'search_file() { # Check if the first argument is provided if [\n -z \"$1\" ]; then echo \"Usage: search_file <search_term> [<file>]\" return fi #\n Check if the second argument is provided if [ -n \"$2\" ]; then #\n Check if the provided argument is a valid file if [ -f \"$2\" ]; then local\n file=\"$2\" # Set file if valid else echo \"Usage: search_file\n <search_term> [<file>]\" echo \"Error: File name $2 not found. Please\n provide a valid file name.\" return # Exit if the file is not valid fi else #\n Check if a file is open if [ -z \"$CURRENT_FILE\" ]; then echo\n \"No file open. Use the open command first.\" return # Exit if no\n file is open fi local file=\"$CURRENT_FILE\" # Set file to the\n current open file fi local search_term=\"$1\" file=$(realpath \"$file\") #\n Use grep to directly get the desired formatted output local matches=$(grep\n -nH -- \"$search_term\" \"$file\") # Check if no matches were found if [\n -z \"$matches\" ]; then echo \"No matches found for \\\"$search_term\\\" in\n $file\" return fi # Calculate total number of matches local\n num_matches=$(echo \"$matches\" | wc -l | awk ''{$1=$1; print $0}'') # calculate\n total number of lines matched local num_lines=$(echo \"$matches\" | cut -d:\n -f1 | sort | uniq | wc -l | awk ''{$1=$1; print $0}'') # if num_lines is\n > 100, print an error if [ $num_lines -gt 100 ]; then echo \"More\n than $num_lines lines matched for \\\"$search_term\\\" in $file. Please narrow\n your search.\" return fi # Print the total number of matches and\n the matches themselves echo \"Found $num_matches matches for \\\"$search_term\\\"\n in $file:\" echo \"$matches\" | cut -d: -f1-2 | sort -u -t: -k2,2n | while\n IFS=: read -r filename line_number; do echo \"Line $line_number:$(sed\n -n \"${line_number}p\" \"$file\")\" done echo \"End of matches for \\\"$search_term\\\"\n in $file\"}'\n docstring: searches for search_term in file. If file is not provided, searches\n in the current open file\n end_name: null\n name: search_file\n signature: search_file <search_term> [<file>]\n - arguments:\n dir:\n description: the directory to search in (if not provided, searches in the\n current directory)\n required: false\n type: string\n file_name:\n description: the name of the file to search for\n required: true\n type: string\n code: 'find_file() { if [ $# -eq 1 ]; then local file_name=\"$1\" local\n dir=\"./\" elif [ $# -eq 2 ]; then local file_name=\"$1\" if\n [ -d \"$2\" ]; then local dir=\"$2\" else echo \"Directory\n $2 not found\" return fi else echo \"Usage: find_file\n <file_name> [<dir>]\" return fi dir=$(realpath \"$dir\") local\n matches=$(find \"$dir\" -type f -name \"$file_name\") # if no matches, return if\n [ -z \"$matches\" ]; then echo \"No matches found for \\\"$file_name\\\" in\n $dir\" return fi # Calculate total number of matches local\n num_matches=$(echo \"$matches\" | wc -l | awk ''{$1=$1; print $0}'') echo\n \"Found $num_matches matches for \\\"$file_name\\\" in $dir:\" echo \"$matches\"\n | awk ''{print $0}''}'\n docstring: finds all files with the given name in dir. If dir is not provided,\n searches in the current directory\n end_name: null\n name: find_file\n signature: find_file <file_name> [<dir>]\n - arguments:\n end_line:\n description: the line number to end the edit at (inclusive)\n required: true\n type: integer\n replacement_text:\n description: the text to replace the current selection with\n required: true\n type: string\n start_line:\n description: the line number to start the edit at\n required: true\n type: integer\n code: 'edit() { if [ -z \"$CURRENT_FILE\" ] then echo ''No file open.\n Use the `open` command first.'' return fi local start_line=\"$(echo\n $1: | cut -d: -f1)\" local end_line=\"$(echo $1: | cut -d: -f2)\" if [\n -z \"$start_line\" ] || [ -z \"$end_line\" ] then echo \"Usage: edit\n <start_line>:<end_line>\" return fi local re=''^[0-9]+$'' if\n ! [[ $start_line =~ $re ]]; then echo \"Usage: edit <start_line>:<end_line>\" echo\n \"Error: start_line must be a number\" return fi if ! [[ $end_line\n =~ $re ]]; then echo \"Usage: edit <start_line>:<end_line>\" echo\n \"Error: end_line must be a number\" return fi # Bash array starts\n at 0, so let''s adjust local start_line=$((start_line - 1)) local end_line=$((end_line)) local\n line_count=0 local replacement=() while IFS= read -r line do replacement+=(\"$line\") ((line_count++)) done #\n Create a backup of the current file cp \"$CURRENT_FILE\" \"/root/$(basename\n \"$CURRENT_FILE\")_backup\" # Read the file line by line into an array mapfile\n -t lines < \"$CURRENT_FILE\" local new_lines=(\"${lines[@]:0:$start_line}\"\n \"${replacement[@]}\" \"${lines[@]:$((end_line))}\") # Write the new stuff\n directly back into the original file printf \"%s\\n\" \"${new_lines[@]}\" >|\n \"$CURRENT_FILE\" # Run linter if [[ $CURRENT_FILE == *.py ]]; then lint_output=$(flake8\n --isolated --select=F821,F822,F831,E111,E112,E113,E999,E902 \"$CURRENT_FILE\"\n 2>&1) else # do nothing lint_output=\"\" fi # if there\n is no output, then the file is good if [ -z \"$lint_output\" ]; then export\n CURRENT_LINE=$start_line _constrain_line _print echo\n \"File updated. Please review the changes and make sure they are correct (correct\n indentation, no duplicate lines, etc). Edit the file again if necessary.\" else echo\n \"Your proposed edit has introduced new syntax error(s). Please read this error\n message carefully and then retry editing the file.\" echo \"\" echo\n \"ERRORS:\" _split_string \"$lint_output\" echo \"\" # Save\n original values original_current_line=$CURRENT_LINE original_window=$WINDOW #\n Update values export CURRENT_LINE=$(( (line_count / 2) + start_line\n )) # Set to \"center\" of edit export WINDOW=$((line_count + 10)) # Show\n +/- 5 lines around edit echo \"This is how your edit would have looked\n if applied\" echo \"-------------------------------------------------\" _constrain_line _print echo\n \"-------------------------------------------------\" echo \"\" #\n Restoring CURRENT_FILE to original contents. cp \"/root/$(basename \"$CURRENT_FILE\")_backup\"\n \"$CURRENT_FILE\" export CURRENT_LINE=$(( ((end_line - start_line + 1)\n / 2) + start_line )) export WINDOW=$((end_line - start_line + 10)) echo\n \"This is the original code before your edit\" echo \"-------------------------------------------------\" _constrain_line _print echo\n \"-------------------------------------------------\" # Restore original\n values export CURRENT_LINE=$original_current_line export WINDOW=$original_window echo\n \"Your changes have NOT been applied. Please fix your edit command and try\n again.\" echo \"You either need to 1) Specify the correct start/end line\n arguments or 2) Correct your edit code.\" echo \"DO NOT re-run the same\n failed edit command. Running it again will lead to the same error.\" fi #\n Remove backup file rm -f \"/root/$(basename \"$CURRENT_FILE\")_backup\"}'\n docstring: replaces lines <start_line> through <end_line> (inclusive) with the\n given text in the open file. The replacement text is terminated by a line\n with only end_of_edit on it. All of the <replacement text> will be entered,\n so make sure your indentation is formatted properly. Python files will be\n checked for syntax errors after the edit. If the system detects a syntax error,\n the edit will not be executed. Simply try to edit the file again, but make\n sure to read the error message and modify the edit command you issue accordingly.\n Issuing the same command a second time will just lead to the same error message\n again.\n end_name: end_of_edit\n name: edit\n signature: |-\n edit <start_line>:<end_line>\n <replacement_text>\n end_of_edit\n _subroutines: {}\n blocklist:\n - vim\n - vi\n - emacs\n - nano\n - nohup\n - git\n blocklist_error_template: Interactive operation '{name}' is not supported by this\n environment\n blocklist_standalone:\n - python\n - python3\n - ipython\n - bash\n - sh\n - exit\n - /bin/bash\n - /bin/sh\n - nohup\n - vi\n - vim\n - emacs\n - nano\n command_docs: |+\n open:\n docstring: opens the file at the given path in the editor. If line_number is provided, the window will be move to include that line\n signature: open <path> [<line_number>]\n arguments:\n - path (string) [required]: the path to the file to open\n - line_number (integer) [optional]: the line number to move the window to (if not provided, the window will start at the top of the file)\n\n goto:\n docstring: moves the window to show <line_number>\n signature: goto <line_number>\n arguments:\n - line_number (integer) [required]: the line number to move the window to\n\n scroll_down:\n docstring: moves the window down {WINDOW} lines\n signature: scroll_down\n\n scroll_up:\n docstring: moves the window down {WINDOW} lines\n signature: scroll_up\n\n create:\n docstring: creates and opens a new file with the given name\n signature: create <filename>\n arguments:\n - filename (string) [required]: the name of the file to create\n\n submit:\n docstring: submits your current code and terminates the session\n signature: submit\n\n search_dir:\n docstring: searches for search_term in all files in dir. If dir is not provided, searches in the current directory\n signature: search_dir <search_term> [<dir>]\n arguments:\n - search_term (string) [required]: the term to search for\n - dir (string) [optional]: the directory to search in (if not provided, searches in the current directory)\n\n search_file:\n docstring: searches for search_term in file. If file is not provided, searches in the current open file\n signature: search_file <search_term> [<file>]\n arguments:\n - search_term (string) [required]: the term to search for\n - file (string) [optional]: the file to search in (if not provided, searches in the current open file)\n\n find_file:\n docstring: finds all files with the given name in dir. If dir is not provided, searches in the current directory\n signature: find_file <file_name> [<dir>]\n arguments:\n - file_name (string) [required]: the name of the file to search for\n - dir (string) [optional]: the directory to search in (if not provided, searches in the current directory)\n\n edit:\n docstring: replaces lines <start_line> through <end_line> (inclusive) with the given text in the open file. The replacement text is terminated by a line with only end_of_edit on it. All of the\n <replacement text> will be entered, so make sure your indentation is formatted properly. Python files will be checked for syntax errors after the edit. If the system detects a syntax error, the edit will\n not be executed. Simply try to edit the file again, but make sure to read the error message and modify the edit command you issue accordingly. Issuing the same command a second time will just lead to the\n same error message again.\n signature: edit <start_line>:<end_line>\n <replacement_text>\n end_of_edit\n arguments:\n - start_line (integer) [required]: the line number to start the edit at\n - end_line (integer) [required]: the line number to end the edit at (inclusive)\n - replacement_text (string) [required]: the text to replace the current selection with\n\n command_files:\n - /Users/fuchur/Documents/24/git_sync/SWE-agent/config/commands/defaults.sh\n - /Users/fuchur/Documents/24/git_sync/SWE-agent/config/commands/search.sh\n - /Users/fuchur/Documents/24/git_sync/SWE-agent/config/commands/edit_linting.sh\n - /Users/fuchur/Documents/24/git_sync/SWE-agent/config/commands/_split_string.py\n demonstration_template: |\n Here is a demonstration of how to correctly accomplish this task.\n It is included to show you how to correctly use the interface.\n You do not need to follow exactly what is done in the demonstration.\n --- DEMONSTRATION ---\n {demonstration}\n --- END OF DEMONSTRATION ---\n demonstrations:\n -\n /Users/fuchur/Documents/24/git_sync/SWE-agent/trajectories/demonstrations/replay__marshmallow-code__marshmallow-1867__default__t-0.20__p-0.95__c-2.00__install-1___install_from_source/marshmallow-code__ma\n rshmallow-1867.traj\n env_variables:\n CURRENT_FILE: ''\n CURRENT_LINE: '0'\n OVERLAP: '2'\n SEARCH_FILES: ()\n SEARCH_INDEX: '0'\n SEARCH_RESULTS: ()\n WINDOW: '100'\n format_error_template: |\n Your output was not formatted correctly. You must always include one discussion and one command as part of your response. Make sure you do not have multiple discussion/command tags.\n Please make sure your output precisely matches the following format:\n DISCUSSION\n Discuss here with yourself about what your planning and what you're going to do in this step.\n\n ```\n command(s) that you're going to run\n ```\n history_processor: {}\n history_processor_args: {}\n instance_template: |-\n We're currently solving the following issue within our repository. Here's the issue text:\n ISSUE:\n {issue}\n\n INSTRUCTIONS:\n Now, you're going to solve this issue on your own. Your terminal session has started and you're in the repository's root directory. You can use any bash commands or the special interface to help\n you. Edit all the files you need to and run any checks or tests that you want.\n Remember, YOU CAN ONLY ENTER ONE COMMAND AT A TIME. You should always wait for feedback after every command.\n When you're satisfied with all of the changes you've made, you can submit your changes to the code base by simply running the submit command.\n Note however that you cannot use any interactive session commands (e.g. python, vim) in this environment, but you can write scripts and run them. E.g. you can write a python script and then run it\n with `python <script_name>.py`.\n\n NOTE ABOUT THE EDIT COMMAND: Indentation really matters! When editing a file, make sure to insert appropriate indentation before each line!\n\n IMPORTANT TIPS:\n 1. Always start by trying to replicate the bug that the issues discusses.\n If the issue includes code for reproducing the bug, we recommend that you re-implement that in your environment, and run it to make sure you can reproduce the bug.\n Then start trying to fix it.\n When you think you've fixed the bug, re-run the bug reproduction script to make sure that the bug has indeed been fixed.\n\n If the bug reproduction script does not print anything when it successfully runs, we recommend adding a print(\"Script completed successfully, no errors.\") command at the end of the file,\n so that you can be sure that the script indeed ran fine all the way through.\n\n 2. If you run a command and it doesn't work, try running a different command. A command that did not work once will not work the second time unless you modify it!\n\n 3. If you open a file and need to get to an area around a specific line that is not in the first 100 lines, say line 583, don't just use the scroll_down command multiple times. Instead, use the\n goto 583 command. It's much quicker.\n\n 4. If the bug reproduction script requires inputting/reading a specific file, such as buggy-input.png, and you'd like to understand how to input that file, conduct a search in the existing repo\n code, to see whether someone else has already done that. Do this by running the command: find_file \"buggy-input.png\" If that doesn't work, use the linux 'find' command.\n\n 5. Always make sure to look at the currently open file and the current working directory (which appears right after the currently open file). The currently open file might be in a different\n directory than the working directory! Note that some commands, such as 'create', open files, so they might change the current open file.\n\n 6. When editing files, it is easy to accidentally specify a wrong line number or to write code with incorrect indentation. Always check the code after you issue an edit to make sure that it\n reflects what you wanted to accomplish. If it didn't, issue another command to fix it.\n\n 7. It may be necessary to install the repository from source before you can run code. Please think about how to install the environment from the repository directory if you need to do so.\n\n\n (Open file: {open_file})\n (Current directory: {working_dir})\n bash-$\n next_step_no_output_template: |-\n Your command ran successfully and did not produce any output.\n (Open file: {open_file})\n (Current directory: {working_dir})\n bash-$\n next_step_template: |-\n {observation}\n (Open file: {open_file})\n (Current directory: {working_dir})\n bash-$\n parse_command: {}\n parse_function: {}\n put_demos_in_history: false\n state_command:\n arguments: null\n code: |\n state() {\n local working_dir=\"$PWD\";\n if [ -z $CURRENT_FILE ]; then\n echo '{\"open_file\": \"n/a\", \"working_dir\": \"'$working_dir'\"}';\n else\n echo '{\"open_file\": \"'$(realpath $CURRENT_FILE)'\", \"working_dir\": \"'$working_dir'\"}';\n fi\n };\n docstring: null\n end_name: null\n name: state\n signature: null\n strategy_template: null\n submit_command: submit\n subroutine_types: []\n system_template: |-\n SETTING: You are an autonomous programmer, and you're working directly in the command line with a special interface.\n\n The special interface consists of a file editor that shows you {WINDOW} lines of a file at a time.\n In addition to typical bash commands, you can also use the following commands to help you navigate and edit files.\n\n COMMANDS:\n {command_docs}\n\n Please note that THE EDIT COMMAND REQUIRES PROPER INDENTATION.\n If you'd like to add the line ' print(x)' you must fully write that out, with all those spaces before the code! Indentation is important and code that is not indented correctly will fail and\n require fixing before it can be run.\n\n RESPONSE FORMAT:\n Your shell prompt is formatted as follows:\n (Open file: <path>) <cwd> $\n\n You need to format your output using two fields; discussion and command.\n Your output should always include _one_ discussion and _one_ command field EXACTLY as in the following example:\n DISCUSSION\n First I'll start by using ls to see what files are in the current directory. Then maybe we can look at some relevant files to see what they look like.\n ```\n ls -a\n ```\n\n You should only include a *SINGLE* command in the command section and then wait for a response from the shell before continuing with more discussion and commands. Everything you include in the\n DISCUSSION section will be saved for future reference.\n If you'd like to issue two commands at once, PLEASE DO NOT DO THAT! Please instead first submit just the first command, and then after receiving a response you'll be able to issue the second\n command.\n You're free to use any other bash commands you want (e.g. find, grep, cat, ls, cd) in addition to the special commands listed above.\n However, the environment does NOT support interactive session commands (e.g. python, vim), so please do not invoke them.\n util_functions:\n - arguments: null\n code: '_print() { local total_lines=$(awk ''END {print NR}'' $CURRENT_FILE) echo\n \"[File: $(realpath $CURRENT_FILE) ($total_lines lines total)]\" lines_above=$(jq\n -n \"$CURRENT_LINE - $WINDOW/2\" | jq ''[0, .] | max | floor'') lines_below=$(jq\n -n \"$total_lines - $CURRENT_LINE - $WINDOW/2\" | jq ''[0, .] | max | round'') if\n [ $lines_above -gt 0 ]; then echo \"($lines_above more lines above)\" fi cat\n $CURRENT_FILE | grep -n $ | head -n $(jq -n \"[$CURRENT_LINE + $WINDOW/2, $WINDOW/2]\n | max | floor\") | tail -n $(jq -n \"$WINDOW\") if [ $lines_below -gt 0 ];\n then echo \"($lines_below more lines below)\" fi}'\n docstring: null\n end_name: null\n name: _print\n signature: _print\n - arguments: null\n code: _constrain_line() { if [ -z \"$CURRENT_FILE\" ] then echo \"No\n file open. Use the open command first.\" return fi local max_line=$(awk\n 'END {print NR}' $CURRENT_FILE) local half_window=$(jq -n \"$WINDOW/2\" |\n jq 'floor') export CURRENT_LINE=$(jq -n \"[$CURRENT_LINE, $max_line - $half_window]\n | min\") export CURRENT_LINE=$(jq -n \"[$CURRENT_LINE, $half_window] | max\")}\n docstring: null\n end_name: null\n name: _constrain_line\n signature: _constrain_line\n config_file: config/default_from_url.yaml\n model:\n host_url: localhost:11434\n model_name: azure:gpt4\n per_instance_cost_limit: 2.0\n replay_path: null\n temperature: 0.0\n top_p: 0.95\n total_cost_limit: 0.0\n environment:\n base_commit: null\n cache_task_images: false\n container_name: null\n data_path: https://github.com/SWE-agent/test-repo/issues/1\n environment_setup: null\n image_name: sweagent/swe-agent:latest\n install_environment: true\n no_mirror: false\n repo_path: ''\n split: dev\n timeout: null\n verbose: true\n instance_filter: .*\n print_config: true\n raise_exceptions: false\n skip_existing: true\n suffix: ''\n\nINFO Base commit reference None resolved to commit hash 8c179cd2be750cd9f2bb91b21adb39934311e9b8\nINFO \ud83d\udcbd Loaded dataset from https://github.com/SWE-agent/test-repo/issues/1\nINFO Found image sweagent/swe-agent:latest with tags: ['sweagent/swe-agent:latest'], created: 2024-06-05T01:13:45.176471384Z for linux arm64.\nDEBUG Starting container with command: docker run -i --rm --name sweagent-swe-agent-latest-01edf87adc sweagent/swe-agent:latest /bin/bash -l\nINFO \ud83c\udf31 Environment Initialized\nDEBUG Environment initialization took 2.09 seconds\nINFO \u25b6\ufe0f Beginning task 0\nINFO Trying to clone from non-mirror...\nWARNING install_environment is set to True, but the data path is a GitHub URL without an environment config file (environment_config key/flag). Skipping conda environment installation.\nINFO Initializing agent settings for container 26cd13d1f31252475cde7e1ae1981d11f43e88d2066c2532611f5f2182d42737\nINFO Resetting model stats\nINFO SYSTEM (primary)\n SETTING: You are an autonomous programmer, and you're working directly in the command line with a special interface.\n\n The special interface consists of a file editor that shows you 100 lines of a file at a time.\n In addition to typical bash commands, you can also use the following commands to help you navigate and edit files.\n\n COMMANDS:\n open:\n docstring: opens the file at the given path in the editor. If line_number is provided, the window will be move to include that line\n signature: open <path> [<line_number>]\n arguments:\n - path (string) [required]: the path to the file to open\n - line_number (integer) [optional]: the line number to move the window to (if not provided, the window will start at the top of the file)\n\n goto:\n docstring: moves the window to show <line_number>\n signature: goto <line_number>\n arguments:\n - line_number (integer) [required]: the line number to move the window to\n\n scroll_down:\n docstring: moves the window down {WINDOW} lines\n signature: scroll_down\n\n scroll_up:\n docstring: moves the window down {WINDOW} lines\n signature: scroll_up\n\n create:\n docstring: creates and opens a new file with the given name\n signature: create <filename>\n arguments:\n - filename (string) [required]: the name of the file to create\n\n submit:\n docstring: submits your current code and terminates the session\n signature: submit\n\n search_dir:\n docstring: searches for search_term in all files in dir. If dir is not provided, searches in the current directory\n signature: search_dir <search_term> [<dir>]\n arguments:\n - search_term (string) [required]: the term to search for\n - dir (string) [optional]: the directory to search in (if not provided, searches in the current directory)\n\n search_file:\n docstring: searches for search_term in file. If file is not provided, searches in the current open file\n signature: search_file <search_term> [<file>]\n arguments:\n - search_term (string) [required]: the term to search for\n - file (string) [optional]: the file to search in (if not provided, searches in the current open file)\n\n find_file:\n docstring: finds all files with the given name in dir. If dir is not provided, searches in the current directory\n signature: find_file <file_name> [<dir>]\n arguments:\n - file_name (string) [required]: the name of the file to search for\n - dir (string) [optional]: the directory to search in (if not provided, searches in the current directory)\n\n edit:\n docstring: replaces lines <start_line> through <end_line> (inclusive) with the given text in the open file. The replacement text is terminated by a line with only end_of_edit on it. All of the\n <replacement text> will be entered, so make sure your indentation is formatted properly. Python files will be checked for syntax errors after the edit. If the system detects a syntax error, the edit will\n not be executed. Simply try to edit the file again, but make sure to read the error message and modify the edit command you issue accordingly. Issuing the same command a second time will just lead to the\n same error message again.\n signature: edit <start_line>:<end_line>\n <replacement_text>\n end_of_edit\n arguments:\n - start_line (integer) [required]: the line number to start the edit at\n - end_line (integer) [required]: the line number to end the edit at (inclusive)\n - replacement_text (string) [required]: the text to replace the current selection with\n\n\n\n Please note that THE EDIT COMMAND REQUIRES PROPER INDENTATION.\n If you'd like to add the line ' print(x)' you must fully write that out, with all those spaces before the code! Indentation is important and code that is not indented correctly will fail and\n require fixing before it can be run.\n\n RESPONSE FORMAT:\n Your shell prompt is formatted as follows:\n (Open file: <path>) <cwd> $\n\n You need to format your output using two fields; discussion and command.\n Your output should always include _one_ discussion and _one_ command field EXACTLY as in the following example:\n DISCUSSION\n First I'll start by using ls to see what files are in the current directory. Then maybe we can look at some relevant files to see what they look like.\n ```\n ls -a\n ```\n\n You should only include a *SINGLE* command in the command section and then wait for a response from the shell before continuing with more discussion and commands. Everything you include in the DISCUSSION\n section will be saved for future reference.\n If you'd like to issue two commands at once, PLEASE DO NOT DO THAT! Please instead first submit just the first command, and then after receiving a response you'll be able to issue the second command.\n You're free to use any other bash commands you want (e.g. find, grep, cat, ls, cd) in addition to the special commands listed above.\n However, the environment does NOT support interactive session commands (e.g. python, vim), so please do not invoke them.\nINFO DEMONSTRATION:\n /Users/fuchur/Documents/24/git_sync/SWE-agent/trajectories/demonstrations/replay__marshmallow-code__marshmallow-1867__default__t-0.20__p-0.95__c-2.00__install-1___install_from_source/marshmallow-code__ma\n rshmallow-1867.traj\nINFO Trajectory will be saved to trajectories/fuchur/azure-gpt4__SWE-agent__test-repo__default_from_url__t-0.00__p-0.95__c-2.00__install-1/SWE-agent__test-repo-i1.traj\nINFO \ud83e\udd16 MODEL INPUT\n We're currently solving the following issue within our repository. Here's the issue text:\n ISSUE:\n SyntaxError: invalid syntax\n I'm running `missing_colon.py` as follows:\n\n ```python\n division(23, 0)\n ```\n\n but I get the following error:\n\n ```\n File \"/Users/fuchur/Documents/24/git_sync/swe-agent-test-repo/tests/./missing_colon.py\", line 4\n def division(a: float, b: float) -> float\n ^\n SyntaxError: invalid syntax\n ```\n\n\n INSTRUCTIONS:\n Now, you're going to solve this issue on your own. Your terminal session has started and you're in the repository's root directory. You can use any bash commands or the special interface to help you.\n Edit all the files you need to and run any checks or tests that you want.\n Remember, YOU CAN ONLY ENTER ONE COMMAND AT A TIME. You should always wait for feedback after every command.\n When you're satisfied with all of the changes you've made, you can submit your changes to the code base by simply running the submit command.\n Note however that you cannot use any interactive session commands (e.g. python, vim) in this environment, but you can write scripts and run them. E.g. you can write a python script and then run it with\n `python <script_name>.py`.\n\n NOTE ABOUT THE EDIT COMMAND: Indentation really matters! When editing a file, make sure to insert appropriate indentation before each line!\n\n IMPORTANT TIPS:\n 1. Always start by trying to replicate the bug that the issues discusses.\n If the issue includes code for reproducing the bug, we recommend that you re-implement that in your environment, and run it to make sure you can reproduce the bug.\n Then start trying to fix it.\n When you think you've fixed the bug, re-run the bug reproduction script to make sure that the bug has indeed been fixed.\n\n If the bug reproduction script does not print anything when it successfully runs, we recommend adding a print(\"Script completed successfully, no errors.\") command at the end of the file,\n so that you can be sure that the script indeed ran fine all the way through.\n\n 2. If you run a command and it doesn't work, try running a different command. A command that did not work once will not work the second time unless you modify it!\n\n 3. If you open a file and need to get to an area around a specific line that is not in the first 100 lines, say line 583, don't just use the scroll_down command multiple times. Instead, use the goto 583\n command. It's much quicker.\n\n 4. If the bug reproduction script requires inputting/reading a specific file, such as buggy-input.png, and you'd like to understand how to input that file, conduct a search in the existing repo code, to\n see whether someone else has already done that. Do this by running the command: find_file \"buggy-input.png\" If that doesn't work, use the linux 'find' command.\n\n 5. Always make sure to look at the currently open file and the current working directory (which appears right after the currently open file). The currently open file might be in a different directory\n than the working directory! Note that some commands, such as 'create', open files, so they might change the current open file.\n\n 6. When editing files, it is easy to accidentally specify a wrong line number or to write code with incorrect indentation. Always check the code after you issue an edit to make sure that it reflects what\n you wanted to accomplish. If it didn't, issue another command to fix it.\n\n 7. It may be necessary to install the repository from source before you can run code. Please think about how to install the environment from the repository directory if you need to do so.\n\n\n (Open file: n/a)\n (Current directory: /SWE-agent__test-repo)\n bash-$\nINFO input_tokens=10,194, output_tokens=56, instance_cost=0.10, cost=0.10\nINFO total_tokens_sent=10,194, total_tokens_received=56, total_cost=0.10, total_api_calls=1\nINFO \ud83d\udcad THOUGHT (primary)\n First, let's locate the file `missing_colon.py` mentioned in the issue to understand the syntax error and fix it. We'll use the `find_file` command to search for this file in the current directory.\n\nINFO \ud83c\udfac ACTION (primary)\n find_file \"missing_colon.py\"\n\nINFO \ud83e\udd16 MODEL INPUT\n Found 1 matches for \"missing_colon.py\" in /SWE-agent__test-repo:\n /SWE-agent__test-repo/tests/missing_colon.py\n (Open file: n/a)\n (Current directory: /SWE-agent__test-repo)\n bash-$\nINFO input_tokens=10,314, output_tokens=45, instance_cost=0.21, cost=0.10\nINFO total_tokens_sent=20,508, total_tokens_received=101, total_cost=0.21, total_api_calls=2\nINFO \ud83d\udcad THOUGHT (primary)\n Now that we've located the file `missing_colon.py` in the `tests` directory, let's open it to inspect the syntax error mentioned in the issue.\n\nINFO \ud83c\udfac ACTION (primary)\n open tests/missing_colon.py\n\nINFO \ud83e\udd16 MODEL INPUT\n [File: /SWE-agent__test-repo/tests/missing_colon.py (10 lines total)]\n 1:#!/usr/bin/env python3\n 2:\n 3:\n 4:def division(a: float, b: float) -> float\n 5: return a/b\n 6:\n 7:\n 8:if __name__ == \"__main__\":\n 9: print(division(123, 15))\n 10:\n (Open file: /SWE-agent__test-repo/tests/missing_colon.py)\n (Current directory: /SWE-agent__test-repo)\n bash-$\nINFO input_tokens=10,484, output_tokens=59, instance_cost=0.31, cost=0.11\nINFO total_tokens_sent=30,992, total_tokens_received=160, total_cost=0.31, total_api_calls=3\nINFO \ud83d\udcad THOUGHT (primary)\n The syntax error is due to the missing colon at the end of the function definition on line 4. Let's add the missing colon to correct the syntax error.\n\nINFO \ud83c\udfac ACTION (primary)\n edit 4:4\n def division(a: float, b: float) -> float:\n end_of_edit\n\nINFO \ud83e\udd16 MODEL INPUT\n [File: /SWE-agent__test-repo/tests/missing_colon.py (10 lines total)]\n 1:#!/usr/bin/env python3\n 2:\n 3:\n 4:def division(a: float, b: float) -> float:\n 5: return a/b\n 6:\n 7:\n 8:if __name__ == \"__main__\":\n 9: print(division(123, 15))\n 10:\n File updated. Please review the changes and make sure they are correct (correct indentation, no duplicate lines, etc). Edit the file again if necessary.\n (Open file: /SWE-agent__test-repo/tests/missing_colon.py)\n (Current directory: /SWE-agent__test-repo)\n bash-$\nINFO input_tokens=10,698, output_tokens=47, instance_cost=0.42, cost=0.11\nINFO total_tokens_sent=41,690, total_tokens_received=207, total_cost=0.42, total_api_calls=4\nINFO \ud83d\udcad THOUGHT (primary)\n The syntax error has been corrected by adding the missing colon at the end of the function definition. Now, let's run the script to ensure that it executes without any syntax errors.\n\nINFO \ud83c\udfac ACTION (primary)\n python tests/missing_colon.py\n\nINFO \ud83e\udd16 MODEL INPUT\n 8.2\n (Open file: /SWE-agent__test-repo/tests/missing_colon.py)\n (Current directory: /SWE-agent__test-repo)\n bash-$\nINFO input_tokens=10,790, output_tokens=45, instance_cost=0.53, cost=0.11\nINFO total_tokens_sent=52,480, total_tokens_received=252, total_cost=0.53, total_api_calls=5\nINFO \ud83d\udcad THOUGHT (primary)\n The script executed successfully and printed the result of the division function, which confirms that the syntax error has been resolved. Now that the issue is fixed, we can submit the changes to the\n code base.\n\nINFO \ud83c\udfac ACTION (primary)\n submit\n\nINFO Found submission:\n diff --git a/tests/missing_colon.py b/tests/missing_colon.py\n index 20edef5..5857437 100755\n --- a/tests/missing_colon.py\n +++ b/tests/missing_colon.py\n @@ -1,7 +1,7 @@\n #!/usr/bin/env python3\n\n\n -def division(a: float, b: float) -> float\n +def division(a: float, b: float) -> float:\n return a/b\n\n\n\nINFO Trajectory saved to trajectories/fuchur/azure-gpt4__SWE-agent__test-repo__default_from_url__t-0.00__p-0.95__c-2.00__install-1/SWE-agent__test-repo-i1.traj\nINFO Saved predictions to trajectories/fuchur/azure-gpt4__SWE-agent__test-repo__default_from_url__t-0.00__p-0.95__c-2.00__install-1/all_preds.jsonl\n\u256d\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500 \ud83c\udf89 Submission successful \ud83c\udf89 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u256e\n\u2502 SWE-agent has produced a patch that it believes will solve the issue you submitted! \u2502\n\u2502 Use the code snippet below to inspect or apply it! \u2502\n\u2570\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u256f\n\n # The patch has been saved to your local filesystem at:\n PATCH_FILE_PATH='/Users/fuchur/Documents/24/git_sync/SWE-agent/trajectories/fuchur/azure-gpt4__SWE-agent__test-repo__default_from_url__t-0.00__p-0.95__c-2.00__install-1/patches/SWE-agent__test-repo-i1.patch'\n # Inspect it:\n cat \"${PATCH_FILE_PATH}\"\n # Apply it to a local repository:\n cd <your local repo root>\n git apply \"${PATCH_FILE_PATH}\"\n</code></pre> <p>Windows</p> <p>If you're using docker on Windows, use <code>-v //var/run/docker.sock:/var/run/docker.sock</code> (double slash) to escape it (more information).</p> <p>More tips</p> <p>See the docker issues section for more help if you run into trouble.</p> <p>If you instead want to pass the keys as environment variables, use</p> <pre><code>docker run --rm -it -v /var/run/docker.sock:/var/run/docker.sock \\\n -e GITHUB_TOKEN=\"yourgithubtoken\" \\\n -e OPENAI_API_KEY=\"youropenaikey\" \\\n sweagent/swe-agent-run:latest \\\n # rest of the command above\n</code></pre> <p>Getting updates</p> <p>Even though the image <code>sweagent/swe-agent:latest</code> has the tag <code>latest</code>, it is not automatically updated every time you run <code>docker run</code>. Instead, you need to manually run</p> <pre><code>docker pull sweagent/swe-agent-run:latest\ndocker pull sweagent/swe-agent:latest\n</code></pre> <p>periodically.</p> <p>Retrieving generated files</p> <p>The optional <code>--rm</code> flag removes the docker container after the command has terminated. Therefore, to retrieve files (like generated patch files) from the container, please remove this flag.</p>"},{"location":"installation/docker/#running-the-web-server","title":"Running the web server","text":"<p>Tip</p> <p>Please also read the previous section for tips on passing environment variables and staying up to date.</p> <p>To run the web server, make sure to forward port 3000:</p> <pre><code>docker run -p 3000:3000 -it -v /var/run/docker.sock:/var/run/docker.sock \\\n -v $(pwd)/keys.cfg:/app/keys.cfg \\\n sweagent/swe-agent-run:latest bash start_web_ui.sh\n</code></pre> <p>More information</p> <p>See running the web UI for more information about the web UI and additional hints for how to solve problems with the starting it.</p> <ul> <li> <p> Something broken? Report bug</p> </li> <li> <p> Something unclear? Ask question</p> </li> </ul>"},{"location":"installation/keys/","title":"Adding your API keys","text":"<p>In order to access the LM of your choice (and to access private GitHub repositories), you need to supply the corresponding keys.</p> <p>There are two options to do this:</p> <ol> <li>Set the corresponding environment variables.</li> <li>Create a <code>keys.cfg</code> file at the root of this repository.</li> </ol> <p>The following <code>keys.cfg</code> example shows you how the keys are named:</p> <pre><code># Remove the comment '#' in front of the line for all keys that you have set\n# GITHUB_TOKEN: 'GitHub Token for access to private repos'\n# OPENAI_API_KEY: 'OpenAI API Key Here if using OpenAI Model'\n# ANTHROPIC_API_KEY: 'Anthropic API Key Here if using Anthropic Model'\n# TOGETHER_API_KEY: 'Together API Key Here if using Together Model'\n# AZURE_OPENAI_API_KEY: 'Azure OpenAI API Key Here if using Azure OpenAI Model'\n# AZURE_OPENAI_ENDPOINT: 'Azure OpenAI Endpoint Here if using Azure OpenAI Model'\n# AZURE_OPENAI_DEPLOYMENT: 'Azure OpenAI Deployment Here if using Azure OpenAI Model'\n# AZURE_OPENAI_API_VERSION: 'Azure OpenAI API Version Here if using Azure OpenAI Model'\n# OPENAI_API_BASE_URL: 'LM base URL here if using Local or alternative api Endpoint'\n</code></pre> <p>See the following links for tutorials on obtaining Anthropic, OpenAI, and Github tokens.</p> <p>Models</p> <p>Some more information about the available models in our usage FAQ.</p>"},{"location":"installation/source/","title":"Installation from source","text":"<p>Installation from source is the preferred way to set up SWE-agent on your machine.</p> <p>Issues on Windows</p> <p>Expect some issues with Windows (we're working on them). In the meantime, use Docker.</p> <ol> <li>Install Docker (follow the docs or use the get-docker.sh script for linux), then start Docker locally. Problems? See docker issues.</li> <li>If you plan on using the web-based GUI: Install <code>nodejs</code>.</li> <li>Clone the repository, for example with <pre><code>git clone https://github.com/princeton-nlp/SWE-agent.git\n</code></pre></li> <li>Run <pre><code>python -m pip install --upgrade pip && pip install --editable .\n</code></pre> at the repository root (as with any python setup, it's recommended to use conda or virtual environments to manage dependencies).</li> <li>Run <pre><code>docker pull sweagent/swe-agent:latest\n</code></pre> Optional. If you want to run EnIGMA for cybersecurity challenges, run also: <pre><code>docker pull sweagent/enigma:latest\ndocker network create ctfnet\n</code></pre> Errors? See docker issues. Alternatively, you can run <code>./setup.sh</code> to create your own <code>swe-agent</code> docker image or run <code>./setup_ctf.sh</code> to create your own EnIGMA docker image.</li> <li>Set up your LM API keys as explained here.</li> </ol> <p>Docker issues</p> <p>If you run into docker issues, see the installation tips section for more help.</p> <p>Updating</p> <p>SWE-agent is still in active development. Features and enhancement are added often. To make sure you are on the latest version, periodically run <code>git pull</code> (there is no need to redo the <code>pip install</code>). You might also want to run <code>docker pull sweagent/swe-agent:latest</code> or <code>./setup.sh</code> periodically (though changes to the container are more rare).</p> <p>Development setup</p> <p>Want to modify SWE-agent? Great! There are a few extra steps and tips: Please check our contribution guide.</p> <ul> <li> <p> Something broken? Report bug</p> </li> <li> <p> Something unclear? Ask question</p> </li> </ul>"},{"location":"installation/tips/","title":"More installation tips","text":""},{"location":"installation/tips/#docker-issues","title":"Docker issues","text":"<p>First, test if you can use docker in general, for example by running</p> <pre><code>docker run hello-world\n</code></pre> <p>If you get an error like</p> <pre><code>docker: permission denied while trying to connect to the Docker daemon socket at unix:///var/run/docker.sock: Head \"http://%2Fvar%2Frun%2Fdocker.sock/_ping\": dial unix /var/run/docker.sock: connect: permission denied.\n</code></pre> <ul> <li>Make sure that you allow the use of the Docker socket. In Docker desktop, click Settings > Advanced > Allow the default Docker socket to be used (requires password).</li> <li>On the command line, you can try <code>sudo chmod 666 /var/run/docker.sock</code> or add your user to the <code>docker</code> linux user group</li> <li>If your docker installation uses a different socket, you might have to symlink them, see this command for example</li> </ul> <p>If you are using any containers from dockerhub (i.e., you ran <code>docker pull ...</code> or you are running <code>docker run ...</code>), please make sure that you are using the latest versions. Just because an image has the <code>latest</code> tag (e.g., <code>sweagent/swe-agent-run:latest</code>) does not mean that it will auto-update. Please run <code>docker pull sweagent/swe-agent-run:latest</code> to make sure you actually have the most recent version!</p> <p>Any remaining issues? Please open a GitHub issue!</p>"},{"location":"reference/","title":"Code structure and reference","text":"<p>This section is in development</p> <p>Some submodules are missing. We also do not strive for completeness, but provide this as an easy entry point for people who want to start reading the code. Also note that SWE-agent is still developed very actively, so the python implementation details are still changing. See the changelog for more information.</p> <p>SWE-agent architecture</p> <p>Please first read the architecture page for an overview of SWE-agent.</p> <p>The core:</p> <ul> <li>The <code>sweagent/agent/</code> submodule implements the agent.<ul> <li>Read about the <code>Agent</code> class</li> <li>Explore the code</li> </ul> </li> <li>The <code>sweagent/environment/</code> submodule handles the communication with the docker container where we execute code.<ul> <li>Read about the <code>SWEEnv</code> class</li> <li>Explore the code</li> </ul> </li> </ul> <p>More subfolders</p> <ul> <li>See the <code>scripts/</code> folder for other useful scripts and details.</li> <li>See the <code>config/</code> folder for details about how you can define your own configuration!</li> <li>See the <code>trajectories/</code> folder for details about the output of <code>run.py</code>.</li> </ul>"},{"location":"reference/agent/","title":"The agent class","text":""},{"location":"reference/agent/#sweagent.agent.agents.Agent","title":"<code>Agent</code>","text":"<p>Agent handles the behaviour of the model and how it interacts with the environment.</p> Source code in <code>sweagent/agent/agents.py</code> <pre><code>class Agent:\n \"\"\"Agent handles the behaviour of the model and how it interacts with the environment.\"\"\"\n\n def __init__(self, name: str, args: AgentArguments):\n self.name = name\n # todo: currently only used to get the model name, so might remove this later\n self._args = args\n self.model = get_model(args.model, args.config._commands + args.config.subroutine_types)\n self.summarizer_model = get_model(\n args.config.summarizer_config.model if args.config.summarizer_config.model is not None else args.model\n )\n self.config = args.config\n assert self.config is not None # mypy\n self.system_args = {\n \"command_docs\": self.config.command_docs,\n **self.config.env_variables,\n }\n self.instance_args = None\n self._parse_command_patterns()\n self.last_container_id = None\n self.hooks = []\n self.logger = get_logger(\"agent\")\n # Requires instance, so is set in `setup` methods\n self._rloop = None\n\n # Set in run method\n self._env: SWEEnv | None = None\n self.traj_dir: None | Path = None\n\n #: Number of attempts to solve the issue when using a review loop\n self._i_attempt: int = 0\n\n #: The following three attributes collect the information about how the agent\n #: solved the problem.\n self._history_by_attempt: dict[int, list] = defaultdict(list)\n self._trajectory_by_attempt: dict[int, Trajectory] = defaultdict(list)\n self._info_by_attempt: dict[int, AgentInfo] = defaultdict(dict)\n\n #: Variables to be referenced in the templates that are forwarded from one\n #: solution attempt to the next\n self._forwarded_vars: dict[str, Any] = {}\n\n @property\n def history(self) -> History:\n \"\"\"History that is passed on to the model.\n Use `_append_history` to modify.\n \"\"\"\n return self._history_by_attempt[self._i_attempt]\n\n @history.setter\n def history(self, value: History):\n self._history_by_attempt[self._i_attempt] = value\n\n @property\n def trajectory(self) -> Trajectory:\n \"\"\"Trajectory of the agent for the current instance. In contrast to `history`,\n this is mostly for the informational value of how the agent interacted with\n the environment and is also what is being used when replaying the trajectory\n \"\"\"\n return self._trajectory_by_attempt[self._i_attempt]\n\n @trajectory.setter\n def trajectory(self, value: Trajectory):\n self._trajectory_by_attempt[self._i_attempt] = value\n\n @property\n def info(self) -> AgentInfo:\n \"\"\"Information about the agent's run\"\"\"\n return self._info_by_attempt[self._i_attempt]\n\n @info.setter\n def info(self, value: AgentInfo):\n self._info_by_attempt[self._i_attempt] = value\n\n @property\n def traj_path(self) -> Path | None:\n \"\"\"Returns path to the trajectory.\n The path is reset for every new instance.\n \"\"\"\n if self.traj_dir and self._env is not None:\n assert self._env.record\n return self.traj_dir / (self._env.record[\"instance_id\"] + \".traj\")\n return None\n\n def add_hook(self, hook: AgentHook) -> None:\n \"\"\"Add hook to agent\"\"\"\n hook.on_init(agent=self)\n self.hooks.append(hook)\n\n def _append_history(self, item: HistoryItem) -> None:\n \"\"\"Adds an item to the history.\"\"\"\n for hook in self.hooks:\n hook.on_query_message_added(**item)\n self.history.append(item)\n\n # todo: klieret: Long term: Might make more sense to reinitialize the agent class for every instance instead of this\n def setup(self, instance_args: dict[str, Any], init_model_stats: APIStats | None = None) -> None:\n \"\"\"Setup the agent for a new instance. This includes\n formatting the system message and adding demonstrations to the history.\n\n Args:\n instance_args: Arguments for the instance\n \"\"\"\n assert self.config is not None # mypy\n self.instance_args = instance_args\n\n self._i_attempt = 0\n self._history_by_attempt = defaultdict(list)\n self._trajectory_by_attempt = defaultdict(list)\n self._info_by_attempt = defaultdict(dict) # type: ignore\n self._forwarded_vars = {}\n if self._rloop is not None:\n self._forwarded_vars = self._rloop.get_forwarded_vars()\n\n self.setup_attempt(init_model_stats=init_model_stats)\n\n for hook in self.hooks:\n hook.on_setup_done()\n\n def setup_attempt(self, *, init_model_stats: APIStats | None = None) -> None:\n \"\"\"Setup the agent for a new attempt. This includes resetting the model stats.\"\"\"\n assert self.config is not None # mypy\n if self._i_attempt > 0 and init_model_stats is not None:\n msg = (\n \"We might be dealing with nested retries, where subroutines are mixed with retries. \"\n \"Currently, this messes up accounting with init_model_stats.\"\n )\n raise ValueError(msg)\n if self._i_attempt > 0:\n assert self._env is not None # mypy\n self._env.reset_for_new_attempt()\n self.model.reset_stats(init_model_stats)\n # self.model = get_model(self._args.model, self.config._commands + self.config.subroutine_types)\n # fixme: This doesn't reset total cost\n system_msg = self.config.system_template.format(**self.system_args, **self.instance_args)\n self.logger.info(f\"SYSTEM ({self.name})\\n{system_msg}\")\n self._append_history(HistoryItem({\"role\": \"system\", \"content\": system_msg, \"agent\": self.name}))\n if \"history_to_messages\" in dir(self.model):\n for demonstration_path in self.config.demonstrations:\n if self.config.demonstration_template is None and not self.config.put_demos_in_history:\n msg = \"Cannot use demonstrations without a demonstration template or put_demos_in_history=True\"\n raise ValueError(msg)\n\n # Load history\n self.logger.info(f\"DEMONSTRATION: {demonstration_path}\")\n demo_history = json.loads(Path(demonstration_path).read_text())[\"history\"]\n demo_history = [\n entry\n for entry in demo_history\n if (\"agent\" not in entry) or (\"agent\" in entry and entry[\"agent\"] == self.name)\n ]\n\n if self.config.put_demos_in_history:\n if self.config.demonstration_template is not None:\n self.logger.warning(\"Demonstration template is ignored for put_demos_in_history=True\")\n # Add demonstration to history directly as separate messages\n for entry in demo_history:\n if entry[\"role\"] != \"system\":\n entry[\"is_demo\"] = True\n self._append_history(entry)\n else:\n # Add demonstration as single message to history\n demo_message = self.model.history_to_messages(\n demo_history,\n is_demonstration=True,\n )\n demonstration = self.config.demonstration_template.format(demonstration=demo_message)\n self._append_history(\n {\n \"agent\": self.name,\n \"content\": demonstration,\n \"is_demo\": True,\n \"role\": \"user\",\n },\n )\n\n @property\n def state_command(self) -> str:\n \"\"\"Return the bash command that will be used to extract the environment state.\"\"\"\n assert self.config is not None\n return self.config.state_command.name\n\n @property\n def local_history(self) -> list[dict[str, str]]:\n \"\"\"Return the history of the agent since the last reset.\"\"\"\n return self.config.history_processor([entry for entry in self.history if entry[\"agent\"] == self.name])\n\n def _get_total_stats(self) -> APIStats:\n \"\"\"Combine model stats of different attempts\"\"\"\n total_stats = APIStats()\n for stats in self._info_by_attempt.values():\n assert \"model_stats\" in stats # mypy\n attempt_stats = APIStats(**stats[\"model_stats\"]) # type: ignore\n total_stats += attempt_stats\n if self._rloop is not None:\n total_stats += self._rloop.model_stats\n return total_stats\n\n def save_trajectory(\n self,\n ) -> None:\n \"\"\"Save the trajectory to disk.\n This includes the history, the environment state, and the model stats.\n \"\"\"\n\n def get_attempt_data(attempt_idx: int) -> dict[str, Any]:\n \"\"\"Get data saved for every attempt\"\"\"\n assert self._env is not None\n # The deepcopy here is important because else the\n # data[\"info\"][\"model_stats\"] update will create havoc!\n return copy.deepcopy(\n {\n \"environment\": self._env.name,\n \"trajectory\": self._trajectory_by_attempt[attempt_idx],\n \"history\": self._history_by_attempt[attempt_idx],\n \"info\": self._info_by_attempt[attempt_idx],\n }\n )\n\n data = {\n **get_attempt_data(0),\n }\n\n assert self.traj_path is not None\n self.traj_path.write_text(json.dumps(data, indent=2))\n\n def _get_first_match(self, action: str, pattern_type: str) -> re.Match | None:\n \"\"\"Return the first match of a command pattern in the action string.\"\"\"\n assert self.config is not None # mypy\n if pattern_type == \"subroutine\":\n patterns = {k: v for k, v in self.subroutine_patterns.items()}\n elif pattern_type == \"multi_line\":\n patterns = {\n k: v\n for k, v in self.command_patterns.items()\n if k in self.config.multi_line_command_endings or k == self.config.submit_command\n }\n patterns += {\n k: v for k, v in self.subroutine_patterns.items() if k in self.config.multi_line_command_endings\n }\n elif pattern_type == \"multi_line_no_subroutines\":\n patterns = {k: v for k, v in self.command_patterns.items() if k in self.config.multi_line_command_endings}\n else:\n msg = f\"Unknown pattern type: {pattern_type}\"\n raise ValueError(msg)\n matches = list()\n for _, pat in patterns.items():\n match = pat.search(action)\n if match:\n matches.append(match)\n if len(matches) == 0:\n return None\n matches = sorted(matches, key=lambda x: x.start())\n return matches[0]\n\n def _guard_multiline_input(self, action: str) -> str:\n \"\"\"Split action by multiline commands, then append the first line in each multiline command with \"<< '{end_name}'\".\n Multiline commands (which are specified by an end_name) are commands that span multiple lines and are terminated by a specific end_name.\n\n Their multi-line argument is sent using a heredoc, which is a way to send a multi-line string to a command in bash.\n \"\"\"\n parsed_action = list()\n rem_action = action\n while rem_action.strip():\n first_match = self._get_first_match(rem_action, \"multi_line_no_subroutines\")\n if first_match:\n pre_action = rem_action[: first_match.start()]\n match_action = rem_action[first_match.start() : first_match.end()]\n rem_action = rem_action[first_match.end() :]\n if pre_action.strip():\n parsed_action.append(pre_action)\n if match_action.strip():\n eof = first_match.group(3).strip()\n if not match_action.split(\"\\n\")[0].strip().endswith(f\"<< '{eof}'\"):\n guarded_command = match_action[first_match.start() :]\n first_line = guarded_command.split(\"\\n\")[0]\n guarded_command = guarded_command.replace(first_line, first_line + f\" << '{eof}'\", 1)\n parsed_action.append(guarded_command)\n else:\n parsed_action.append(match_action)\n else:\n parsed_action.append(rem_action)\n rem_action = \"\"\n return \"\\n\".join(parsed_action)\n\n def split_actions(self, action: str, pattern_type=\"subroutine\") -> list[SubAction]:\n \"\"\"Split an action into a list of actions in a greedy manner, each of which is a subroutine call or a single command.\"\"\"\n parsed_action: list[SubAction] = list()\n rem_action = action\n while rem_action.strip():\n first_match = self._get_first_match(rem_action, pattern_type)\n if first_match:\n pre_action = rem_action[: first_match.start()]\n match_action = rem_action[first_match.start() : first_match.end()]\n rem_action = rem_action[first_match.end() :]\n if pre_action.strip():\n parsed_action.append({\"agent\": self.name, \"action\": pre_action, \"cmd_name\": None, \"args\": \"\"})\n if match_action.strip():\n if match_action.split()[0] == self.config.submit_command:\n parsed_action.append(\n SubAction(\n {\n \"agent\": self.name,\n \"action\": match_action,\n \"cmd_name\": first_match.group(1),\n \"args\": \"\",\n },\n )\n ) # submit command is not a subroutine\n else:\n parsed_action.append(\n SubAction(\n {\n \"agent\": first_match.group(1),\n \"args\": first_match.group(2),\n \"action\": match_action,\n \"cmd_name\": first_match.group(1),\n },\n )\n )\n else:\n parsed_action.append(\n SubAction({\"agent\": self.name, \"action\": rem_action, \"cmd_name\": None, \"args\": \"\"})\n )\n rem_action = \"\"\n return parsed_action\n\n def _parse_command_patterns(self) -> None:\n assert self.config is not None # mypy\n self.command_patterns = dict()\n for command in self.config._commands:\n if command.end_name is not None:\n pat = re.compile(\n rf\"^\\s*({command.name})\\s*(.*?)^({command.end_name})\\s*$\",\n re.DOTALL | re.MULTILINE,\n )\n self.command_patterns[command.name] = pat\n else:\n pat = re.compile(rf\"^\\s*({command.name})\\s*(.*?)$\", re.MULTILINE)\n self.command_patterns[command.name] = pat\n self.subroutine_patterns = dict()\n for _, subroutine in self.config._subroutines.items():\n if subroutine.end_name is None:\n pat = re.compile(rf\"^\\s*({subroutine.name})\\s*(.*?)$\", re.MULTILINE)\n self.subroutine_patterns[subroutine.name,] = pat\n else:\n pat = re.compile(\n rf\"^\\s*({subroutine.name})\\s*(.*?)^({subroutine.end_name})\\s*$\",\n re.DOTALL | re.MULTILINE,\n )\n self.subroutine_patterns[subroutine.name] = pat\n if hasattr(self.config, \"submit_command_end_name\"):\n submit_pat = re.compile(\n rf\"^\\s*({self.config.submit_command})\\s*(.*?)^({self.config.submit_command_end_name})\\s*$\",\n re.DOTALL | re.MULTILINE,\n )\n else:\n submit_pat = re.compile(rf\"^\\s*({self.config.submit_command})(\\s*)$\", re.MULTILINE) # group 2 is nothing\n self.subroutine_patterns[self.config.submit_command] = submit_pat\n self.command_patterns[self.config.submit_command] = submit_pat\n\n def forward(self, observation: str | None, available_actions: list[str], state: str) -> tuple[str, str, str]:\n \"\"\"Forwards the model\n\n Args:\n observation: Observation\n available_actions: Currently not used\n state:\n\n Returns:\n thought: model reasoning\n action: action that the model proposes\n output: raw model output (not output of the action)\n \"\"\"\n thought, action, output = self.forward_with_error_check(observation, state)\n\n self._append_history(\n {\n \"role\": \"assistant\",\n \"content\": output,\n \"thought\": thought,\n \"action\": action,\n \"agent\": self.name,\n },\n )\n\n self.logger.info(f\"\ud83d\udcad THOUGHT ({self.name})\\n{thought}\")\n self.logger.info(f\"\ud83c\udfac ACTION ({self.name})\\n{action}\")\n\n return thought, action, output\n\n def forward_model(self, observation: str | None, state: str) -> str:\n \"\"\"Query the model with the current state and observation with the appropriate template.\n\n Returns:\n output: raw model output (not output of the command)\n \"\"\"\n assert self.config is not None # mypy\n try:\n state_vars = json.loads(state)\n except json.JSONDecodeError as e:\n msg = f\"State {state!r} is not valid json. This is an internal error, please report it.\"\n raise ValueError(msg) from e\n\n templates: list[str] = []\n # Determine observation template based on what prior observation was\n if self.history[-1][\"role\"] == \"system\" or self.history[-1].get(\"is_demo\", False):\n # Show instance template if prev. obs. was initial system message\n templates = [self.config.instance_template]\n if self.config.strategy_template is not None:\n templates.append(self.config.strategy_template)\n elif observation is None or observation.strip() == \"\":\n # Show no output template if observation content was empty\n templates = [self.config.next_step_no_output_template]\n else:\n # Show standard output template if there is observation content\n templates = [self.config.next_step_template]\n\n # Populate selected template(s) with information (e.g., issue, arguments, state)\n messages = []\n for template in templates:\n messages.append(\n template.format(\n **self.instance_args,\n **self.system_args,\n **state_vars,\n observation=(observation if observation is not None else \"\"),\n **self._forwarded_vars,\n ),\n )\n\n message = \"\\n\".join(messages)\n\n self.logger.info(f\"\ud83e\udd16 MODEL INPUT\\n{message}\")\n self._append_history({\"role\": \"user\", \"content\": message, \"agent\": self.name})\n\n for hook in self.hooks:\n hook.on_model_query(query=self.local_history, agent=self.name)\n return self.model.query(self.local_history)\n\n def retry_after_format_fail(self, output: str) -> str:\n \"\"\"Ask the model to correct (without committing to persistent history) after a malformatted model output\"\"\"\n format_error_template = self.config.format_error_template\n\n self.logger.warning(f\"MALFORMED OUTPUT\\n{output}\")\n self.logger.warning(f\"FORMAT ERROR\\n{format_error_template}\")\n\n temp_history = self.local_history + [\n {\"role\": \"assistant\", \"content\": output, \"agent\": self.name},\n {\"role\": \"user\", \"content\": format_error_template, \"agent\": self.name},\n ]\n return self.model.query(temp_history)\n\n def retry_after_blocklist_fail(self, output: str, action: str) -> str:\n \"\"\"Ask the model to correct (without committing to persistent history) after a disallowed command\"\"\"\n name = action.strip().split()[0]\n blocklist_error_message = self.config.blocklist_error_template.format(name=name)\n\n self.logger.warning(f\"BLOCKLISTED OUTPUT\\n{output}\")\n self.logger.warning(f\"BLOCKLIST ERROR\\n{blocklist_error_message}\")\n\n temp_history = self.local_history + [\n {\"role\": \"assistant\", \"content\": output, \"agent\": self.name},\n {\"role\": \"user\", \"content\": blocklist_error_message, \"agent\": self.name},\n ]\n return self.model.query(temp_history)\n\n def should_block_action(self, action: str) -> bool:\n \"\"\"Check if the command should be blocked.\"\"\"\n names = action.strip().split()\n if len(names) == 0:\n return False\n name = names[0]\n if name in self.config.blocklist:\n return True\n if name in self.config.blocklist_standalone and name == action.strip():\n return True\n if name in self.config.block_unless_regex and not re.search(self.config.block_unless_regex[name], action):\n return True\n return False\n\n def check_format_and_requery(\n self,\n output: str,\n ) -> tuple[str, str, str]:\n \"\"\"Query the model with the current state and observation with the appropriate template.\n\n Try to parse the output into a thought and action. Retry if the output is malformatted or the action is blocked.\n\n Returns:\n thought: model reasoning\n action: action that the model proposes\n output: raw model output\n \"\"\"\n # Condition for handling outputs with no thought (just action)\n if self.model.args.model_name == \"human\":\n return \"\", output, output\n elif self.model.args.model_name == \"human_thought\":\n thought, action = ParseFunction.get(\"ThoughtActionParser\")(\n output,\n self.config._commands + self.config.subroutine_types,\n strict=False,\n )\n return thought, action, output\n\n format_fails = blocklist_fails = 0\n\n while format_fails + blocklist_fails <= 2:\n try:\n thought, action = self.config.parse_function(\n output,\n self.config._commands + self.config.subroutine_types,\n strict=False,\n )\n except KeyboardInterrupt:\n raise\n except FormatError:\n format_fails += 1\n output = self.retry_after_format_fail(output)\n continue\n if self.should_block_action(action):\n blocklist_fails += 1\n output = self.retry_after_blocklist_fail(output, action)\n else:\n return thought, action, output\n self.logger.warning(f\"Malformat limit reached: \\n{output}\")\n return \"Exit due to format error\", \"exit_format\", output\n\n def forward_with_error_check(self, observation: str | None, state: str) -> tuple[str, str, str]:\n \"\"\"Wrapper around `self.forward_model` that handles errors and retries\n due to format errors or blocked actions.\n\n Returns:\n thought: model reasoning\n action: action that the model proposes\n output: raw model output\n \"\"\"\n try:\n return self.check_format_and_requery(self.forward_model(observation, state))\n except KeyboardInterrupt:\n raise\n except RuntimeError as e:\n self.logger.warning(f\"Runtime error: {e}\")\n return (\n f\"Exit due to runtime error: {e}\",\n \"exit_error\",\n f\"exit due to runtime error: {e}\",\n )\n except ContextWindowExceededError:\n self.logger.warning(\"Context window exceeded\")\n return \"Exit due to context window\", \"exit_context\", \"Exit due to context window\"\n except CostLimitExceededError:\n self.logger.warning(\"Cost limit exceeded\")\n return \"Exit due to cost limit\", \"exit_cost\", \"Exit due to cost limit\"\n except RetryError as e:\n self.logger.warning(f\"Retry error: {e}\")\n return (\n f\"Exit due to retry error: {e}\",\n \"exit_api\",\n f\"exit due to retry error: {e}\",\n )\n\n def init_environment_vars(self, env: SWEEnv):\n assert self.config is not None\n self.set_environment_vars(env, self.config.env_variables)\n\n def set_environment_vars(self, env: SWEEnv, env_variables: dict[str, Any]) -> None:\n \"\"\"Sets environment variables in the container and for example makes sure\n that all the commands are available in the PATH on the container.\n \"\"\"\n assert self.config is not None # mypy\n commands_to_execute = (\n [self.config.state_command.code]\n +\n # [code for code in self.config.util_functions] +\n # [command.code for command in self.config._commands] +\n [f\"{k}={v}\" for k, v in env_variables.items()]\n )\n commands = \"\\n\".join(commands_to_execute)\n try:\n output = env.communicate(commands)\n if env.returncode != 0:\n msg = f\"Nonzero return code: {env.returncode}\\nOutput: {output}\"\n raise RuntimeError(msg)\n except KeyboardInterrupt:\n raise\n except Exception as e:\n self.logger.warning(f\"Failed to set environment variables: {traceback.format_exc()}\")\n raise e\n command_files = list()\n for file in self.config.command_files:\n datum = dict()\n with open(file) as f:\n contents = f.read()\n datum[\"contents\"] = contents\n filename = Path(file).name\n if not contents.strip().startswith(\"#!\"):\n if filename.endswith(\".sh\"):\n # files are sourced, so they are not executable\n datum[\"name\"] = Path(file).name\n datum[\"type\"] = \"source_file\"\n elif filename.startswith(\"_\"):\n # files are sourced, so they are not executable\n datum[\"name\"] = Path(file).name\n datum[\"type\"] = \"utility\"\n else:\n msg = (\n f\"Non-shell script file {file} does not start with shebang.\\n\"\n \"Either add a shebang (#!) or change the file extension to .sh if you want to source it.\\n\"\n \"You can override this behavior by adding an underscore to the file name (e.g. _utils.py).\"\n )\n raise ValueError(msg)\n else:\n # scripts are made executable\n datum[\"name\"] = Path(file).name.rsplit(\".\", 1)[0]\n datum[\"type\"] = \"script\"\n command_files.append(datum)\n env.add_commands(command_files)\n\n def get_environment_vars(self, env: SWEEnv) -> dict[str, Any]:\n \"\"\"Get environment variables inside of the container\"\"\"\n assert self.config is not None # mypy\n env_vars = dict()\n for var in self.config.env_variables:\n env_vars[var] = env.communicate(f\"echo ${var}\").strip()\n return env_vars\n\n def call_subroutine(self, agent_name: str, sub_action: SubAction, env: SWEEnv):\n \"\"\"Call subroutine\"\"\"\n assert self.config is not None # mypy\n env_vars = self.get_environment_vars(env)\n cwd = env.communicate(\"pwd -P\").strip()\n init_observation = self.config._subroutines[agent_name].init_observation\n if init_observation is not None:\n obs, _, _, _ = env.step(init_observation.format(args=sub_action[\"args\"]))\n else:\n obs = None\n if env.returncode != 0:\n self._append_history(HistoryItem({\"role\": \"user\", \"content\": obs, \"agent\": agent_name}))\n msg = f\"Nonzero return code: {env.returncode} for init_observation in {agent_name}.\\n{obs}\"\n raise RuntimeError(msg)\n return_type = self.config._subroutines[agent_name].return_type\n sub_agent = Agent(agent_name, self.config._subroutines[agent_name].agent_args)\n sub_agent_output = sub_agent.run(\n {\"issue\": sub_action[\"args\"]},\n env,\n observation=obs,\n return_type=return_type,\n init_model_stats=self.model.stats,\n )\n self.history += sub_agent.history\n self.set_environment_vars(env, env_vars)\n env.communicate(f\"cd {cwd}\")\n self.model.stats.replace(sub_agent.model.stats)\n return sub_agent_output\n\n def _update_summarizer_stats(self, cost: APIStats):\n \"\"\"Update stats for summarizer\"\"\"\n self.model.stats += cost\n if \"summarizer\" not in self.info:\n self.info[\"summarizer\"] = {\n \"model_stats\": APIStats().to_dict(),\n \"n_calls\": 0,\n }\n total_cost = APIStats(**self.info[\"summarizer\"][\"model_stats\"])\n total_cost += cost\n self.info[\"summarizer\"][\"model_stats\"] = total_cost.to_dict()\n self.info[\"summarizer\"][\"n_calls\"] += 1\n\n def _run_sub_action(self, sub_action: SubAction) -> tuple[str | None, bool]:\n \"\"\"Execute a sub-action. If the sub-action is a command, execute it.\n If it is a subroutine, call the subroutine.\n\n Returns:\n observation: Observation\n done: Whether `submit` or another exit reason was called\n \"\"\"\n assert self._env is not None\n assert self.config is not None\n if sub_action[\"agent\"] == self.name or sub_action[\"cmd_name\"] == self.config.submit_command:\n # Normal command, not a subroutine\n for hook in self.hooks:\n hook.on_sub_action_started(sub_action=sub_action)\n observation, _, done, _info = self._env.step(sub_action[\"action\"])\n observation, additional_cost = self.config.summarizer_config.function( # type: ignore\n sub_action[\"action\"], observation, self._env, self.summarizer_model\n )\n self._update_summarizer_stats(additional_cost)\n self.info.update(_info)\n for hook in self.hooks:\n hook.on_sub_action_executed(obs=observation, done=done)\n if sub_action[\"cmd_name\"] == self.config.submit_command:\n done = True\n else:\n agent_name = sub_action[\"agent\"]\n sub_agent_output = self.call_subroutine(agent_name, sub_action, self._env)\n observation = sub_agent_output\n assert isinstance(observation, str) or observation is None\n done = False\n return observation, done\n\n def _run_step(self, observation: str | None) -> tuple[str | None, bool]:\n \"\"\"Run a step of the agent (forward, execute, and save).\n\n Returns:\n observation: Observation\n done: Whether `submit` or another exit reason was called\n \"\"\"\n\n assert self.config is not None # mypy\n assert self._env is not None\n\n for hook in self.hooks:\n hook.on_step_start()\n\n # fixme: This will probably fail if the state command is not set\n state = self._env.communicate(self.state_command) if self.state_command else None\n thought, action, output = self.forward(observation, self._env.get_available_actions(), state)\n for hook in self.hooks:\n hook.on_actions_generated(thought=thought, action=action, output=output)\n run_action: str = self._guard_multiline_input(action)\n\n # Loop over sub-actions (if any)\n done = False\n observations: list[str | None] = list()\n execution_t0 = time.perf_counter()\n for sub_action in self.split_actions(run_action):\n observation, done = self._run_sub_action(sub_action)\n # If the last sub-action is done, the observation is not\n # appended.\n if done:\n break\n observations.append(observation)\n observation = \"\\n\".join([obs for obs in observations if obs is not None])\n execution_time = time.perf_counter() - execution_t0\n\n trajectory_step = TrajectoryStep(\n {\n \"action\": action,\n \"observation\": observation,\n \"response\": output,\n \"state\": state,\n \"thought\": thought,\n \"execution_time\": execution_time,\n },\n )\n self.trajectory.append(trajectory_step)\n model_stats: APIStats = self.model.stats\n self.info[\"model_stats\"] = model_stats.to_dict()\n for hook in self.hooks:\n hook.on_step_done(trajectory_step=trajectory_step, model_stats=model_stats)\n return observation, done\n\n def run(\n self,\n setup_args: dict[str, Any],\n env: SWEEnv,\n observation: str | None = None,\n traj_dir: Path | None = None,\n return_type: str = \"info_trajectory\",\n init_model_stats: APIStats | None = None,\n ):\n \"\"\"\n Run the agent on an environment.\n Return the final value of the specified return type.\n\n Args:\n setup_args: Arguments to pass to the agent's setup method.\n env: The environment to run the agent on.\n observation: Output from environment setup\n traj_dir: Directory to save the trajectory to\n return_type: Controls what to return.\n This should be left at `info_trajectory`, the\n other values are for internal usage with subroutines.\n init_model_stats: Initial model stats to use for the run.\n\n Returns:\n If return_type is \"info_trajectory\", returns a tuple of\n the info dictionary and the trajectory (list of dictionaries).\n \"\"\"\n assert env.record is not None\n assert env.container_obj is not None\n if env.container_obj.id != self.last_container_id:\n self.logger.info(f\"Initializing agent settings for container {env.container_obj.id}\")\n self.init_environment_vars(env)\n self.last_container_id = env.container_obj.id\n # Re-initialize primary\n self.setup(setup_args, init_model_stats)\n self.config.summarizer_config.function.setup(setup_args, self.config)\n\n # Save/reset some attributes\n self.trajectory = Trajectory()\n self._env = env\n self.info = AgentInfo()\n self.traj_dir = traj_dir\n\n self.logger.info(\"Trajectory will be saved to %s\", self.traj_path)\n\n # Run action/observation loop\n for hook in self.hooks:\n hook.on_run_start()\n done = False\n while not done:\n observation, done = self._run_step(observation)\n self.save_trajectory()\n if done:\n done = True\n for hook in self.hooks:\n hook.on_run_done(trajectory=self.trajectory, info=self.info)\n\n self.logger.info(\"Trajectory saved to %s\", self.traj_path)\n\n if return_type == \"info\":\n return self.info\n if return_type == \"info_trajectory\":\n return self.info, self.trajectory\n return self.trajectory[-1][return_type]\n</code></pre>"},{"location":"reference/agent/#sweagent.agent.agents.Agent.history","title":"<code>history: History</code> <code>property</code> <code>writable</code>","text":"<p>History that is passed on to the model. Use <code>_append_history</code> to modify.</p>"},{"location":"reference/agent/#sweagent.agent.agents.Agent.info","title":"<code>info: AgentInfo</code> <code>property</code> <code>writable</code>","text":"<p>Information about the agent's run</p>"},{"location":"reference/agent/#sweagent.agent.agents.Agent.local_history","title":"<code>local_history: list[dict[str, str]]</code> <code>property</code>","text":"<p>Return the history of the agent since the last reset.</p>"},{"location":"reference/agent/#sweagent.agent.agents.Agent.state_command","title":"<code>state_command: str</code> <code>property</code>","text":"<p>Return the bash command that will be used to extract the environment state.</p>"},{"location":"reference/agent/#sweagent.agent.agents.Agent.traj_path","title":"<code>traj_path: Path | None</code> <code>property</code>","text":"<p>Returns path to the trajectory. The path is reset for every new instance.</p>"},{"location":"reference/agent/#sweagent.agent.agents.Agent.trajectory","title":"<code>trajectory: Trajectory</code> <code>property</code> <code>writable</code>","text":"<p>Trajectory of the agent for the current instance. In contrast to <code>history</code>, this is mostly for the informational value of how the agent interacted with the environment and is also what is being used when replaying the trajectory</p>"},{"location":"reference/agent/#sweagent.agent.agents.Agent.add_hook","title":"<code>add_hook(hook)</code>","text":"<p>Add hook to agent</p> Source code in <code>sweagent/agent/agents.py</code> <pre><code>def add_hook(self, hook: AgentHook) -> None:\n \"\"\"Add hook to agent\"\"\"\n hook.on_init(agent=self)\n self.hooks.append(hook)\n</code></pre>"},{"location":"reference/agent/#sweagent.agent.agents.Agent.call_subroutine","title":"<code>call_subroutine(agent_name, sub_action, env)</code>","text":"<p>Call subroutine</p> Source code in <code>sweagent/agent/agents.py</code> <pre><code>def call_subroutine(self, agent_name: str, sub_action: SubAction, env: SWEEnv):\n \"\"\"Call subroutine\"\"\"\n assert self.config is not None # mypy\n env_vars = self.get_environment_vars(env)\n cwd = env.communicate(\"pwd -P\").strip()\n init_observation = self.config._subroutines[agent_name].init_observation\n if init_observation is not None:\n obs, _, _, _ = env.step(init_observation.format(args=sub_action[\"args\"]))\n else:\n obs = None\n if env.returncode != 0:\n self._append_history(HistoryItem({\"role\": \"user\", \"content\": obs, \"agent\": agent_name}))\n msg = f\"Nonzero return code: {env.returncode} for init_observation in {agent_name}.\\n{obs}\"\n raise RuntimeError(msg)\n return_type = self.config._subroutines[agent_name].return_type\n sub_agent = Agent(agent_name, self.config._subroutines[agent_name].agent_args)\n sub_agent_output = sub_agent.run(\n {\"issue\": sub_action[\"args\"]},\n env,\n observation=obs,\n return_type=return_type,\n init_model_stats=self.model.stats,\n )\n self.history += sub_agent.history\n self.set_environment_vars(env, env_vars)\n env.communicate(f\"cd {cwd}\")\n self.model.stats.replace(sub_agent.model.stats)\n return sub_agent_output\n</code></pre>"},{"location":"reference/agent/#sweagent.agent.agents.Agent.check_format_and_requery","title":"<code>check_format_and_requery(output)</code>","text":"<p>Query the model with the current state and observation with the appropriate template.</p> <p>Try to parse the output into a thought and action. Retry if the output is malformatted or the action is blocked.</p> <p>Returns:</p> Name Type Description <code>thought</code> <code>str</code> <p>model reasoning</p> <code>action</code> <code>str</code> <p>action that the model proposes</p> <code>output</code> <code>str</code> <p>raw model output</p> Source code in <code>sweagent/agent/agents.py</code> <pre><code>def check_format_and_requery(\n self,\n output: str,\n) -> tuple[str, str, str]:\n \"\"\"Query the model with the current state and observation with the appropriate template.\n\n Try to parse the output into a thought and action. Retry if the output is malformatted or the action is blocked.\n\n Returns:\n thought: model reasoning\n action: action that the model proposes\n output: raw model output\n \"\"\"\n # Condition for handling outputs with no thought (just action)\n if self.model.args.model_name == \"human\":\n return \"\", output, output\n elif self.model.args.model_name == \"human_thought\":\n thought, action = ParseFunction.get(\"ThoughtActionParser\")(\n output,\n self.config._commands + self.config.subroutine_types,\n strict=False,\n )\n return thought, action, output\n\n format_fails = blocklist_fails = 0\n\n while format_fails + blocklist_fails <= 2:\n try:\n thought, action = self.config.parse_function(\n output,\n self.config._commands + self.config.subroutine_types,\n strict=False,\n )\n except KeyboardInterrupt:\n raise\n except FormatError:\n format_fails += 1\n output = self.retry_after_format_fail(output)\n continue\n if self.should_block_action(action):\n blocklist_fails += 1\n output = self.retry_after_blocklist_fail(output, action)\n else:\n return thought, action, output\n self.logger.warning(f\"Malformat limit reached: \\n{output}\")\n return \"Exit due to format error\", \"exit_format\", output\n</code></pre>"},{"location":"reference/agent/#sweagent.agent.agents.Agent.forward","title":"<code>forward(observation, available_actions, state)</code>","text":"<p>Forwards the model</p> <p>Parameters:</p> Name Type Description Default <code>observation</code> <code>str | None</code> <p>Observation</p> required <code>available_actions</code> <code>list[str]</code> <p>Currently not used</p> required <code>state</code> <code>str</code> required <p>Returns:</p> Name Type Description <code>thought</code> <code>str</code> <p>model reasoning</p> <code>action</code> <code>str</code> <p>action that the model proposes</p> <code>output</code> <code>str</code> <p>raw model output (not output of the action)</p> Source code in <code>sweagent/agent/agents.py</code> <pre><code>def forward(self, observation: str | None, available_actions: list[str], state: str) -> tuple[str, str, str]:\n \"\"\"Forwards the model\n\n Args:\n observation: Observation\n available_actions: Currently not used\n state:\n\n Returns:\n thought: model reasoning\n action: action that the model proposes\n output: raw model output (not output of the action)\n \"\"\"\n thought, action, output = self.forward_with_error_check(observation, state)\n\n self._append_history(\n {\n \"role\": \"assistant\",\n \"content\": output,\n \"thought\": thought,\n \"action\": action,\n \"agent\": self.name,\n },\n )\n\n self.logger.info(f\"\ud83d\udcad THOUGHT ({self.name})\\n{thought}\")\n self.logger.info(f\"\ud83c\udfac ACTION ({self.name})\\n{action}\")\n\n return thought, action, output\n</code></pre>"},{"location":"reference/agent/#sweagent.agent.agents.Agent.forward_model","title":"<code>forward_model(observation, state)</code>","text":"<p>Query the model with the current state and observation with the appropriate template.</p> <p>Returns:</p> Name Type Description <code>output</code> <code>str</code> <p>raw model output (not output of the command)</p> Source code in <code>sweagent/agent/agents.py</code> <pre><code>def forward_model(self, observation: str | None, state: str) -> str:\n \"\"\"Query the model with the current state and observation with the appropriate template.\n\n Returns:\n output: raw model output (not output of the command)\n \"\"\"\n assert self.config is not None # mypy\n try:\n state_vars = json.loads(state)\n except json.JSONDecodeError as e:\n msg = f\"State {state!r} is not valid json. This is an internal error, please report it.\"\n raise ValueError(msg) from e\n\n templates: list[str] = []\n # Determine observation template based on what prior observation was\n if self.history[-1][\"role\"] == \"system\" or self.history[-1].get(\"is_demo\", False):\n # Show instance template if prev. obs. was initial system message\n templates = [self.config.instance_template]\n if self.config.strategy_template is not None:\n templates.append(self.config.strategy_template)\n elif observation is None or observation.strip() == \"\":\n # Show no output template if observation content was empty\n templates = [self.config.next_step_no_output_template]\n else:\n # Show standard output template if there is observation content\n templates = [self.config.next_step_template]\n\n # Populate selected template(s) with information (e.g., issue, arguments, state)\n messages = []\n for template in templates:\n messages.append(\n template.format(\n **self.instance_args,\n **self.system_args,\n **state_vars,\n observation=(observation if observation is not None else \"\"),\n **self._forwarded_vars,\n ),\n )\n\n message = \"\\n\".join(messages)\n\n self.logger.info(f\"\ud83e\udd16 MODEL INPUT\\n{message}\")\n self._append_history({\"role\": \"user\", \"content\": message, \"agent\": self.name})\n\n for hook in self.hooks:\n hook.on_model_query(query=self.local_history, agent=self.name)\n return self.model.query(self.local_history)\n</code></pre>"},{"location":"reference/agent/#sweagent.agent.agents.Agent.forward_with_error_check","title":"<code>forward_with_error_check(observation, state)</code>","text":"<p>Wrapper around <code>self.forward_model</code> that handles errors and retries due to format errors or blocked actions.</p> <p>Returns:</p> Name Type Description <code>thought</code> <code>str</code> <p>model reasoning</p> <code>action</code> <code>str</code> <p>action that the model proposes</p> <code>output</code> <code>str</code> <p>raw model output</p> Source code in <code>sweagent/agent/agents.py</code> <pre><code>def forward_with_error_check(self, observation: str | None, state: str) -> tuple[str, str, str]:\n \"\"\"Wrapper around `self.forward_model` that handles errors and retries\n due to format errors or blocked actions.\n\n Returns:\n thought: model reasoning\n action: action that the model proposes\n output: raw model output\n \"\"\"\n try:\n return self.check_format_and_requery(self.forward_model(observation, state))\n except KeyboardInterrupt:\n raise\n except RuntimeError as e:\n self.logger.warning(f\"Runtime error: {e}\")\n return (\n f\"Exit due to runtime error: {e}\",\n \"exit_error\",\n f\"exit due to runtime error: {e}\",\n )\n except ContextWindowExceededError:\n self.logger.warning(\"Context window exceeded\")\n return \"Exit due to context window\", \"exit_context\", \"Exit due to context window\"\n except CostLimitExceededError:\n self.logger.warning(\"Cost limit exceeded\")\n return \"Exit due to cost limit\", \"exit_cost\", \"Exit due to cost limit\"\n except RetryError as e:\n self.logger.warning(f\"Retry error: {e}\")\n return (\n f\"Exit due to retry error: {e}\",\n \"exit_api\",\n f\"exit due to retry error: {e}\",\n )\n</code></pre>"},{"location":"reference/agent/#sweagent.agent.agents.Agent.get_environment_vars","title":"<code>get_environment_vars(env)</code>","text":"<p>Get environment variables inside of the container</p> Source code in <code>sweagent/agent/agents.py</code> <pre><code>def get_environment_vars(self, env: SWEEnv) -> dict[str, Any]:\n \"\"\"Get environment variables inside of the container\"\"\"\n assert self.config is not None # mypy\n env_vars = dict()\n for var in self.config.env_variables:\n env_vars[var] = env.communicate(f\"echo ${var}\").strip()\n return env_vars\n</code></pre>"},{"location":"reference/agent/#sweagent.agent.agents.Agent.retry_after_blocklist_fail","title":"<code>retry_after_blocklist_fail(output, action)</code>","text":"<p>Ask the model to correct (without committing to persistent history) after a disallowed command</p> Source code in <code>sweagent/agent/agents.py</code> <pre><code>def retry_after_blocklist_fail(self, output: str, action: str) -> str:\n \"\"\"Ask the model to correct (without committing to persistent history) after a disallowed command\"\"\"\n name = action.strip().split()[0]\n blocklist_error_message = self.config.blocklist_error_template.format(name=name)\n\n self.logger.warning(f\"BLOCKLISTED OUTPUT\\n{output}\")\n self.logger.warning(f\"BLOCKLIST ERROR\\n{blocklist_error_message}\")\n\n temp_history = self.local_history + [\n {\"role\": \"assistant\", \"content\": output, \"agent\": self.name},\n {\"role\": \"user\", \"content\": blocklist_error_message, \"agent\": self.name},\n ]\n return self.model.query(temp_history)\n</code></pre>"},{"location":"reference/agent/#sweagent.agent.agents.Agent.retry_after_format_fail","title":"<code>retry_after_format_fail(output)</code>","text":"<p>Ask the model to correct (without committing to persistent history) after a malformatted model output</p> Source code in <code>sweagent/agent/agents.py</code> <pre><code>def retry_after_format_fail(self, output: str) -> str:\n \"\"\"Ask the model to correct (without committing to persistent history) after a malformatted model output\"\"\"\n format_error_template = self.config.format_error_template\n\n self.logger.warning(f\"MALFORMED OUTPUT\\n{output}\")\n self.logger.warning(f\"FORMAT ERROR\\n{format_error_template}\")\n\n temp_history = self.local_history + [\n {\"role\": \"assistant\", \"content\": output, \"agent\": self.name},\n {\"role\": \"user\", \"content\": format_error_template, \"agent\": self.name},\n ]\n return self.model.query(temp_history)\n</code></pre>"},{"location":"reference/agent/#sweagent.agent.agents.Agent.run","title":"<code>run(setup_args, env, observation=None, traj_dir=None, return_type='info_trajectory', init_model_stats=None)</code>","text":"<p>Run the agent on an environment. Return the final value of the specified return type.</p> <p>Parameters:</p> Name Type Description Default <code>setup_args</code> <code>dict[str, Any]</code> <p>Arguments to pass to the agent's setup method.</p> required <code>env</code> <code>SWEEnv</code> <p>The environment to run the agent on.</p> required <code>observation</code> <code>str | None</code> <p>Output from environment setup</p> <code>None</code> <code>traj_dir</code> <code>Path | None</code> <p>Directory to save the trajectory to</p> <code>None</code> <code>return_type</code> <code>str</code> <p>Controls what to return. This should be left at <code>info_trajectory</code>, the other values are for internal usage with subroutines.</p> <code>'info_trajectory'</code> <code>init_model_stats</code> <code>APIStats | None</code> <p>Initial model stats to use for the run.</p> <code>None</code> <p>Returns:</p> Type Description <p>If return_type is \"info_trajectory\", returns a tuple of</p> <p>the info dictionary and the trajectory (list of dictionaries).</p> Source code in <code>sweagent/agent/agents.py</code> <pre><code>def run(\n self,\n setup_args: dict[str, Any],\n env: SWEEnv,\n observation: str | None = None,\n traj_dir: Path | None = None,\n return_type: str = \"info_trajectory\",\n init_model_stats: APIStats | None = None,\n):\n \"\"\"\n Run the agent on an environment.\n Return the final value of the specified return type.\n\n Args:\n setup_args: Arguments to pass to the agent's setup method.\n env: The environment to run the agent on.\n observation: Output from environment setup\n traj_dir: Directory to save the trajectory to\n return_type: Controls what to return.\n This should be left at `info_trajectory`, the\n other values are for internal usage with subroutines.\n init_model_stats: Initial model stats to use for the run.\n\n Returns:\n If return_type is \"info_trajectory\", returns a tuple of\n the info dictionary and the trajectory (list of dictionaries).\n \"\"\"\n assert env.record is not None\n assert env.container_obj is not None\n if env.container_obj.id != self.last_container_id:\n self.logger.info(f\"Initializing agent settings for container {env.container_obj.id}\")\n self.init_environment_vars(env)\n self.last_container_id = env.container_obj.id\n # Re-initialize primary\n self.setup(setup_args, init_model_stats)\n self.config.summarizer_config.function.setup(setup_args, self.config)\n\n # Save/reset some attributes\n self.trajectory = Trajectory()\n self._env = env\n self.info = AgentInfo()\n self.traj_dir = traj_dir\n\n self.logger.info(\"Trajectory will be saved to %s\", self.traj_path)\n\n # Run action/observation loop\n for hook in self.hooks:\n hook.on_run_start()\n done = False\n while not done:\n observation, done = self._run_step(observation)\n self.save_trajectory()\n if done:\n done = True\n for hook in self.hooks:\n hook.on_run_done(trajectory=self.trajectory, info=self.info)\n\n self.logger.info(\"Trajectory saved to %s\", self.traj_path)\n\n if return_type == \"info\":\n return self.info\n if return_type == \"info_trajectory\":\n return self.info, self.trajectory\n return self.trajectory[-1][return_type]\n</code></pre>"},{"location":"reference/agent/#sweagent.agent.agents.Agent.save_trajectory","title":"<code>save_trajectory()</code>","text":"<p>Save the trajectory to disk. This includes the history, the environment state, and the model stats.</p> Source code in <code>sweagent/agent/agents.py</code> <pre><code>def save_trajectory(\n self,\n) -> None:\n \"\"\"Save the trajectory to disk.\n This includes the history, the environment state, and the model stats.\n \"\"\"\n\n def get_attempt_data(attempt_idx: int) -> dict[str, Any]:\n \"\"\"Get data saved for every attempt\"\"\"\n assert self._env is not None\n # The deepcopy here is important because else the\n # data[\"info\"][\"model_stats\"] update will create havoc!\n return copy.deepcopy(\n {\n \"environment\": self._env.name,\n \"trajectory\": self._trajectory_by_attempt[attempt_idx],\n \"history\": self._history_by_attempt[attempt_idx],\n \"info\": self._info_by_attempt[attempt_idx],\n }\n )\n\n data = {\n **get_attempt_data(0),\n }\n\n assert self.traj_path is not None\n self.traj_path.write_text(json.dumps(data, indent=2))\n</code></pre>"},{"location":"reference/agent/#sweagent.agent.agents.Agent.set_environment_vars","title":"<code>set_environment_vars(env, env_variables)</code>","text":"<p>Sets environment variables in the container and for example makes sure that all the commands are available in the PATH on the container.</p> Source code in <code>sweagent/agent/agents.py</code> <pre><code>def set_environment_vars(self, env: SWEEnv, env_variables: dict[str, Any]) -> None:\n \"\"\"Sets environment variables in the container and for example makes sure\n that all the commands are available in the PATH on the container.\n \"\"\"\n assert self.config is not None # mypy\n commands_to_execute = (\n [self.config.state_command.code]\n +\n # [code for code in self.config.util_functions] +\n # [command.code for command in self.config._commands] +\n [f\"{k}={v}\" for k, v in env_variables.items()]\n )\n commands = \"\\n\".join(commands_to_execute)\n try:\n output = env.communicate(commands)\n if env.returncode != 0:\n msg = f\"Nonzero return code: {env.returncode}\\nOutput: {output}\"\n raise RuntimeError(msg)\n except KeyboardInterrupt:\n raise\n except Exception as e:\n self.logger.warning(f\"Failed to set environment variables: {traceback.format_exc()}\")\n raise e\n command_files = list()\n for file in self.config.command_files:\n datum = dict()\n with open(file) as f:\n contents = f.read()\n datum[\"contents\"] = contents\n filename = Path(file).name\n if not contents.strip().startswith(\"#!\"):\n if filename.endswith(\".sh\"):\n # files are sourced, so they are not executable\n datum[\"name\"] = Path(file).name\n datum[\"type\"] = \"source_file\"\n elif filename.startswith(\"_\"):\n # files are sourced, so they are not executable\n datum[\"name\"] = Path(file).name\n datum[\"type\"] = \"utility\"\n else:\n msg = (\n f\"Non-shell script file {file} does not start with shebang.\\n\"\n \"Either add a shebang (#!) or change the file extension to .sh if you want to source it.\\n\"\n \"You can override this behavior by adding an underscore to the file name (e.g. _utils.py).\"\n )\n raise ValueError(msg)\n else:\n # scripts are made executable\n datum[\"name\"] = Path(file).name.rsplit(\".\", 1)[0]\n datum[\"type\"] = \"script\"\n command_files.append(datum)\n env.add_commands(command_files)\n</code></pre>"},{"location":"reference/agent/#sweagent.agent.agents.Agent.setup","title":"<code>setup(instance_args, init_model_stats=None)</code>","text":"<p>Setup the agent for a new instance. This includes formatting the system message and adding demonstrations to the history.</p> <p>Parameters:</p> Name Type Description Default <code>instance_args</code> <code>dict[str, Any]</code> <p>Arguments for the instance</p> required Source code in <code>sweagent/agent/agents.py</code> <pre><code>def setup(self, instance_args: dict[str, Any], init_model_stats: APIStats | None = None) -> None:\n \"\"\"Setup the agent for a new instance. This includes\n formatting the system message and adding demonstrations to the history.\n\n Args:\n instance_args: Arguments for the instance\n \"\"\"\n assert self.config is not None # mypy\n self.instance_args = instance_args\n\n self._i_attempt = 0\n self._history_by_attempt = defaultdict(list)\n self._trajectory_by_attempt = defaultdict(list)\n self._info_by_attempt = defaultdict(dict) # type: ignore\n self._forwarded_vars = {}\n if self._rloop is not None:\n self._forwarded_vars = self._rloop.get_forwarded_vars()\n\n self.setup_attempt(init_model_stats=init_model_stats)\n\n for hook in self.hooks:\n hook.on_setup_done()\n</code></pre>"},{"location":"reference/agent/#sweagent.agent.agents.Agent.setup_attempt","title":"<code>setup_attempt(*, init_model_stats=None)</code>","text":"<p>Setup the agent for a new attempt. This includes resetting the model stats.</p> Source code in <code>sweagent/agent/agents.py</code> <pre><code>def setup_attempt(self, *, init_model_stats: APIStats | None = None) -> None:\n \"\"\"Setup the agent for a new attempt. This includes resetting the model stats.\"\"\"\n assert self.config is not None # mypy\n if self._i_attempt > 0 and init_model_stats is not None:\n msg = (\n \"We might be dealing with nested retries, where subroutines are mixed with retries. \"\n \"Currently, this messes up accounting with init_model_stats.\"\n )\n raise ValueError(msg)\n if self._i_attempt > 0:\n assert self._env is not None # mypy\n self._env.reset_for_new_attempt()\n self.model.reset_stats(init_model_stats)\n # self.model = get_model(self._args.model, self.config._commands + self.config.subroutine_types)\n # fixme: This doesn't reset total cost\n system_msg = self.config.system_template.format(**self.system_args, **self.instance_args)\n self.logger.info(f\"SYSTEM ({self.name})\\n{system_msg}\")\n self._append_history(HistoryItem({\"role\": \"system\", \"content\": system_msg, \"agent\": self.name}))\n if \"history_to_messages\" in dir(self.model):\n for demonstration_path in self.config.demonstrations:\n if self.config.demonstration_template is None and not self.config.put_demos_in_history:\n msg = \"Cannot use demonstrations without a demonstration template or put_demos_in_history=True\"\n raise ValueError(msg)\n\n # Load history\n self.logger.info(f\"DEMONSTRATION: {demonstration_path}\")\n demo_history = json.loads(Path(demonstration_path).read_text())[\"history\"]\n demo_history = [\n entry\n for entry in demo_history\n if (\"agent\" not in entry) or (\"agent\" in entry and entry[\"agent\"] == self.name)\n ]\n\n if self.config.put_demos_in_history:\n if self.config.demonstration_template is not None:\n self.logger.warning(\"Demonstration template is ignored for put_demos_in_history=True\")\n # Add demonstration to history directly as separate messages\n for entry in demo_history:\n if entry[\"role\"] != \"system\":\n entry[\"is_demo\"] = True\n self._append_history(entry)\n else:\n # Add demonstration as single message to history\n demo_message = self.model.history_to_messages(\n demo_history,\n is_demonstration=True,\n )\n demonstration = self.config.demonstration_template.format(demonstration=demo_message)\n self._append_history(\n {\n \"agent\": self.name,\n \"content\": demonstration,\n \"is_demo\": True,\n \"role\": \"user\",\n },\n )\n</code></pre>"},{"location":"reference/agent/#sweagent.agent.agents.Agent.should_block_action","title":"<code>should_block_action(action)</code>","text":"<p>Check if the command should be blocked.</p> Source code in <code>sweagent/agent/agents.py</code> <pre><code>def should_block_action(self, action: str) -> bool:\n \"\"\"Check if the command should be blocked.\"\"\"\n names = action.strip().split()\n if len(names) == 0:\n return False\n name = names[0]\n if name in self.config.blocklist:\n return True\n if name in self.config.blocklist_standalone and name == action.strip():\n return True\n if name in self.config.block_unless_regex and not re.search(self.config.block_unless_regex[name], action):\n return True\n return False\n</code></pre>"},{"location":"reference/agent/#sweagent.agent.agents.Agent.split_actions","title":"<code>split_actions(action, pattern_type='subroutine')</code>","text":"<p>Split an action into a list of actions in a greedy manner, each of which is a subroutine call or a single command.</p> Source code in <code>sweagent/agent/agents.py</code> <pre><code>def split_actions(self, action: str, pattern_type=\"subroutine\") -> list[SubAction]:\n \"\"\"Split an action into a list of actions in a greedy manner, each of which is a subroutine call or a single command.\"\"\"\n parsed_action: list[SubAction] = list()\n rem_action = action\n while rem_action.strip():\n first_match = self._get_first_match(rem_action, pattern_type)\n if first_match:\n pre_action = rem_action[: first_match.start()]\n match_action = rem_action[first_match.start() : first_match.end()]\n rem_action = rem_action[first_match.end() :]\n if pre_action.strip():\n parsed_action.append({\"agent\": self.name, \"action\": pre_action, \"cmd_name\": None, \"args\": \"\"})\n if match_action.strip():\n if match_action.split()[0] == self.config.submit_command:\n parsed_action.append(\n SubAction(\n {\n \"agent\": self.name,\n \"action\": match_action,\n \"cmd_name\": first_match.group(1),\n \"args\": \"\",\n },\n )\n ) # submit command is not a subroutine\n else:\n parsed_action.append(\n SubAction(\n {\n \"agent\": first_match.group(1),\n \"args\": first_match.group(2),\n \"action\": match_action,\n \"cmd_name\": first_match.group(1),\n },\n )\n )\n else:\n parsed_action.append(\n SubAction({\"agent\": self.name, \"action\": rem_action, \"cmd_name\": None, \"args\": \"\"})\n )\n rem_action = \"\"\n return parsed_action\n</code></pre>"},{"location":"reference/agent/#sweagent.agent.agents.AgentArguments","title":"<code>AgentArguments</code> <code>dataclass</code>","text":"<p> Bases: <code>FlattenedAccess</code>, <code>FrozenSerializable</code></p> <p>Configure the agent's behaviour (templates, parse functions, blocklists, ...).</p> Source code in <code>sweagent/agent/agents.py</code> <pre><code>@dataclass(frozen=True)\nclass AgentArguments(FlattenedAccess, FrozenSerializable):\n \"\"\"Configure the agent's behaviour (templates, parse functions, blocklists, ...).\"\"\"\n\n model: ModelArguments = None\n\n # Policy can only be set via config yaml file from command line\n config_file: Path | None = None\n config: AgentConfig | None = field(default=None, cmd=False)\n\n def __post_init__(self):\n if self.config is None and self.config_file is not None:\n # If unassigned, we load the config from the file to store its contents with the overall arguments\n config = AgentConfig.load_yaml(self.config_file)\n object.__setattr__(self, \"config\", config)\n assert self.config is not None # mypy\n for subroutine in getattr(self.config, \"subroutines\", {}).values():\n model_args = subroutine.model\n object.__setattr__(\n model_args,\n \"per_instance_cost_limit\",\n self.model.per_instance_cost_limit,\n )\n object.__setattr__(model_args, \"total_cost_limit\", self.model.total_cost_limit)\n</code></pre>"},{"location":"reference/agent/#sweagent.agent.agents.AgentHook","title":"<code>AgentHook</code>","text":"Source code in <code>sweagent/agent/agents.py</code> <pre><code>class AgentHook:\n def on_init(self, *, agent: Agent):\n \"\"\"Note: Depending on the internals of `Agent` should be done with care,\n it's best to use this as little as possible.\n \"\"\"\n\n def on_run_start(\n self,\n ): ...\n\n def on_step_start(self): ...\n\n def on_actions_generated(self, *, thought: str, action: str, output: str): ...\n\n def on_sub_action_started(self, *, sub_action: str): ...\n\n def on_sub_action_executed(self, *, obs: str, done: bool): ...\n\n def on_step_done(self, *, trajectory_step: TrajectoryStep, model_stats: APIStats): ...\n\n def on_run_done(self, *, trajectory: Trajectory, info: AgentInfo): ...\n\n def on_model_query(self, *, query: str, agent: str):\n \"\"\"Actually query the model with the complete history.\"\"\"\n\n def on_query_message_added(\n self,\n *,\n role: str,\n content: str,\n agent: str,\n is_demo: bool = False,\n thought: str = \"\",\n action: str = \"\",\n ): ...\n\n def on_setup_done(self): ...\n</code></pre>"},{"location":"reference/agent/#sweagent.agent.agents.AgentHook.on_init","title":"<code>on_init(*, agent)</code>","text":"<p>Note: Depending on the internals of <code>Agent</code> should be done with care, it's best to use this as little as possible.</p> Source code in <code>sweagent/agent/agents.py</code> <pre><code>def on_init(self, *, agent: Agent):\n \"\"\"Note: Depending on the internals of `Agent` should be done with care,\n it's best to use this as little as possible.\n \"\"\"\n</code></pre>"},{"location":"reference/agent/#sweagent.agent.agents.AgentHook.on_model_query","title":"<code>on_model_query(*, query, agent)</code>","text":"<p>Actually query the model with the complete history.</p> Source code in <code>sweagent/agent/agents.py</code> <pre><code>def on_model_query(self, *, query: str, agent: str):\n \"\"\"Actually query the model with the complete history.\"\"\"\n</code></pre>"},{"location":"reference/env/","title":"The environment class","text":""},{"location":"reference/env/#sweagent.environment.swe_env.EnvHook","title":"<code>EnvHook</code>","text":"<p>Hook to be used in <code>SWEEnv</code>.</p> <p>Subclass this class, add functionality and add it with <code>SWEEEnv.add_hook(hook)</code>. This allows to inject custom functionality at different stages of the environment lifecycle, in particular to connect SWE-agent to a new interface (like a GUI).</p> Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>class EnvHook:\n \"\"\"Hook to be used in `SWEEnv`.\n\n Subclass this class, add functionality and add it with `SWEEEnv.add_hook(hook)`.\n This allows to inject custom functionality at different stages of the environment\n lifecycle, in particular to connect SWE-agent to a new interface (like a GUI).\n \"\"\"\n\n def on_init(self) -> None:\n \"\"\"Gets called when the hook is added\"\"\"\n\n def on_copy_repo_started(self, *, repo_type: str, repo_path: str) -> None:\n \"\"\"Gets called when the repository is being cloned to the container\n\n Args:\n repo_type: Type of repository. Either 'local' or 'github'\n repo_path: Path to the repository\n \"\"\"\n\n def on_install_env_started(self) -> None:\n \"\"\"Called when we start installing the environment\"\"\"\n\n def on_close(self):\n \"\"\"Called when the environment is closed\"\"\"\n</code></pre>"},{"location":"reference/env/#sweagent.environment.swe_env.EnvHook.on_close","title":"<code>on_close()</code>","text":"<p>Called when the environment is closed</p> Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>def on_close(self):\n \"\"\"Called when the environment is closed\"\"\"\n</code></pre>"},{"location":"reference/env/#sweagent.environment.swe_env.EnvHook.on_copy_repo_started","title":"<code>on_copy_repo_started(*, repo_type, repo_path)</code>","text":"<p>Gets called when the repository is being cloned to the container</p> <p>Parameters:</p> Name Type Description Default <code>repo_type</code> <code>str</code> <p>Type of repository. Either 'local' or 'github'</p> required <code>repo_path</code> <code>str</code> <p>Path to the repository</p> required Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>def on_copy_repo_started(self, *, repo_type: str, repo_path: str) -> None:\n \"\"\"Gets called when the repository is being cloned to the container\n\n Args:\n repo_type: Type of repository. Either 'local' or 'github'\n repo_path: Path to the repository\n \"\"\"\n</code></pre>"},{"location":"reference/env/#sweagent.environment.swe_env.EnvHook.on_init","title":"<code>on_init()</code>","text":"<p>Gets called when the hook is added</p> Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>def on_init(self) -> None:\n \"\"\"Gets called when the hook is added\"\"\"\n</code></pre>"},{"location":"reference/env/#sweagent.environment.swe_env.EnvHook.on_install_env_started","title":"<code>on_install_env_started()</code>","text":"<p>Called when we start installing the environment</p> Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>def on_install_env_started(self) -> None:\n \"\"\"Called when we start installing the environment\"\"\"\n</code></pre>"},{"location":"reference/env/#sweagent.environment.swe_env.EnvironmentArguments","title":"<code>EnvironmentArguments</code> <code>dataclass</code>","text":"<p> Bases: <code>FrozenSerializable</code></p> <p>Configure data sources and setup instructions for the environment in which we solve the tasks.</p> Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>@dataclass(frozen=True)\nclass EnvironmentArguments(FrozenSerializable):\n \"\"\"Configure data sources and setup instructions for the environment in which we solve the tasks.\"\"\"\n\n # Source of issue statement/problem statement. To run over a batch of issues: Path to a data file\n # (`json`, `jsonl`) or directory. To run over single issue: github issue url or path to markdown file\n # with problem statement or problem statement as text prefixed with `text://`.\n data_path: str\n # Name of the docker image to use for the environment. Defaults to sweagent/swe-agent:latest\n image_name: str = \"sweagent/swe-agent:latest\"\n # When running over SWE-bench issues: Specify the split to use.\n split: str = \"dev\"\n # Specify a branch name or a commit hash to checkout before running the task.\n # Only used when running over a single problem statement/issue.\n base_commit: str | None = None\n # Use a persistent container with this name. After every task, the container will be paused, but not removed.\n # This is useful for speedup when running multiple tasks from the same repositories in a row, as the repositories\n # will have already been cloned and the conda environments will have been installed.\n container_name: str | None = None\n # Try to install the environment before running the task.\n install_environment: bool = True\n # No effect, kept for backwards compatibility.\n timeout: int | None = None\n # Enable environment logger.\n verbose: bool = False\n # Do not use attempt to use a repository mirror from https://github.com/swe-bench.\n no_mirror: bool = False\n # Cache task images to speed up task initialization. This means that the environment will be saved as a\n # docker image for every repository, base commit, and setup combination. This uses quite a bit of disk space\n # but speeds up task initialization significantly when running over multiple issues from the same repository\n # (or using different models for the same issues).\n cache_task_images: bool = False\n # Custom environment setup. Currently only used when data_path points to a single issue.\n # This needs to be either a string pointing to a yaml file (with yaml, yml file extension)\n # or a shell script (with sh extension).\n # See https://princeton-nlp.github.io/SWE-agent/usage/cl_tutorial#environment-setup\n environment_setup: str | None = None\n # Only used when running on single issue. Path to local repository or github repository.\n repo_path: str = \"\"\n # Interactive command configuration\n interactive_sessions_config: dict[str, InteractiveSessionConfig] = field(\n default_factory=lambda: INTERACTIVE_SESSIONS_CONFIG\n )\n # Container mounts - additional folders to mount into the environment (useful for caching)\n container_mounts: list[str] = field(default_factory=list)\n\n def __post_init__(self):\n if self.timeout is not None:\n default_logger.warning(\"The 'timeout' argument is deprecated and has no effect.\")\n if self.cache_task_images and self.container_name:\n msg = (\n \"Not allowed to use persistent container with caching task images \"\n \"(probably doesn't make sense and takes excessive space).\"\n )\n raise ValueError(msg)\n if self.container_name is not None and self.container_name.strip() == \"\":\n msg = \"Set container_name to None if you don't want to use a persistent container.\"\n raise ValueError(msg)\n</code></pre>"},{"location":"reference/env/#sweagent.environment.swe_env.SWEEnv","title":"<code>SWEEnv</code>","text":"<p> Bases: <code>Env</code></p> <p>Gym environment for SWE-bench. This class should handle all communication with the docker container.</p> Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>class SWEEnv(gym.Env):\n \"\"\"Gym environment for SWE-bench. This class should handle all communication with the docker container.\"\"\"\n\n name = \"swe_main\"\n # This prefix will be prepended to the image name when caching task images\n cached_image_prefix = \"swe-agent-task-env-\"\n\n def __init__(self, args: EnvironmentArguments):\n super().__init__()\n t0 = time.perf_counter()\n self.args = args\n self.base_commit: str | None = None\n self.communicate_output: str | None = None\n self.container_name: str | None = args.container_name\n self.install_environment = args.install_environment\n self.logger = get_logger(\"SWEEnv\")\n self.persistent = args.container_name is not None\n self.container_mounts = args.container_mounts\n self.returncode: None | int = None\n if not self.args.verbose:\n # fixme: This creates problems if we have multiple instances of this class\n self.logger.disabled = True\n\n #: The commit hash of the swe-agent repository\n self.commit_sha = None\n try:\n repo = Repo(REPO_ROOT, search_parent_directories=True)\n self.commit_sha = repo.head.object.hexsha\n except KeyboardInterrupt:\n raise\n except Exception as e:\n self.logger.exception(\"Failed to get commit hash for this repo: %s\", str(e))\n\n self._github_token: str = keys_config.get(\"GITHUB_TOKEN\", \"\") # type: ignore\n\n # Load Task Instances\n self.data_path = self.args.data_path\n self.data = get_instances(\n self.data_path,\n self.args.base_commit,\n self.args.split,\n token=self._github_token,\n repo_path=self.args.repo_path,\n )\n #: Instance we're currently processing. Gets set in self.reset.\n self.record: dict[str, Any] | None = None\n self.logger.info(f\"\ud83d\udcbd Loaded dataset from {self.data_path}\")\n\n # Establish connection with execution container\n self.image_name = args.image_name\n self.container_obj: docker.models.containers.Container | None = None\n self.container: subprocess.Popen | None = None\n self.docker_compose: Path | None = None\n self.challenge: dict[str, Any] | None = None\n self._reset_container()\n\n self.interactive_session: InteractiveSession | None = None\n\n self.idx = 0\n self.clean_multi_line_functions = lambda x: x\n self.hooks: list[EnvHook] = []\n\n self.logger.debug(\"Environment initialization took %.2f seconds\", time.perf_counter() - t0)\n\n def _get_cached_task_image_name(self) -> str:\n assert self.record is not None\n inputs: list[str] = [\n self.record[\"repo\"],\n self.record[\"base_commit\"],\n self.args.environment_setup or \"no_setup\",\n ]\n tag = hashlib.sha256(\"\".join(inputs).encode()).hexdigest()[:50]\n return f\"{self.cached_image_prefix}{tag}\"\n\n def add_hook(self, hook: EnvHook):\n \"\"\"Add `EnvHook` to the environment.\n\n This allows to inject custom functionality at different stages of the environment\n lifecycle, in particular to connect SWE-agent to a new interface (like a GUI).\n \"\"\"\n hook.on_init()\n self.hooks.append(hook)\n\n @property\n def _repo_name(self) -> str:\n \"\"\"Name of the local copy of the repository\"\"\"\n assert self.record is not None\n return self.record[\"repo\"].replace(\"/\", \"__\").replace(\" \", \"-\").replace(\"'\", \"\")\n\n def _copy_repo(self) -> str:\n \"\"\"Clone/copy repository/codebase in container\n\n Returns:\n folder name of clone\n \"\"\"\n assert self.container_obj is not None\n assert self.record is not None # mypy\n for hook in self.hooks:\n hook.on_copy_repo_started(repo_type=self.record[\"repo_type\"], repo_path=self.record[\"repo\"])\n if self.record[\"repo_type\"] == \"local\":\n if \"challenge\" in self.record:\n self.communicate_with_handling(\n input=f\"mkdir {self._repo_name}\", error_msg=f\"Failed to create {self._repo_name} in container\"\n )\n for file_name in self.record[\"challenge\"][\"files\"]:\n self.logger.debug(f\"Copying file {file_name} to container\")\n copy_anything_to_container(\n self.container_obj,\n str(Path(self.record[\"repo\"].removeprefix(\"local://\")) / file_name),\n \"/\" + self._repo_name,\n )\n else:\n copy_anything_to_container(\n self.container_obj,\n self.record[\"repo\"].removeprefix(\"local://\"),\n \"/\" + self._repo_name,\n )\n self.communicate_with_handling(\n input=f\"chown -R root:root {self._repo_name}\",\n error_msg=\"Failed to change permissions on copied repository\",\n )\n return self._repo_name\n assert self.record[\"repo_type\"] == \"github\"\n token_prefix = \"\"\n if self._github_token:\n token_prefix = f\"{self._github_token}@\"\n # fixme: This if statement is brittle and should probably be replaced with better logic\n if not self.args.no_mirror and self.record[\"problem_statement_source\"] == \"swe-bench\":\n self.logger.info(f\"{self._repo_name} not found in container, cloning...\")\n clone_url = f\"https://{token_prefix}github.com/swe-bench/{self._repo_name}.git\"\n else:\n self.logger.info(\"Trying to clone from non-mirror...\")\n clone_url = f\"https://{token_prefix}github.com/{self.record['repo']}.git\"\n clone_method = keys_config.get(\"SWE_AGENT_CLONE_METHOD\", default=\"shallow\", choices=[\"shallow\", \"full\"])\n if len(self.data) > 1 or self.persistent:\n msg = \"Falling back to full cloning method due to multiple instances or persistent container\"\n clone_method = \"full\"\n self.logger.debug(msg)\n if clone_method == \"full\":\n self.communicate_with_handling(\n input=f\"git clone {clone_url} {self._repo_name}\",\n error_msg=\"Failed to clone repository from conservative method\",\n timeout_duration=LONG_TIMEOUT,\n )\n else:\n base_commit = self.record[\"base_commit\"]\n self.communicate_with_handling(\n input=\"&&\".join(\n (\n f\"mkdir {self._repo_name}\",\n f\"cd {self._repo_name}\",\n \"git init\",\n f\"git remote add origin {clone_url}\",\n f\"git fetch --depth 1 origin {base_commit}\",\n \"git checkout FETCH_HEAD\",\n \"cd ..\",\n )\n ),\n error_msg=\"Failed to clone repository with fast method\",\n timeout_duration=LONG_TIMEOUT,\n )\n return self._repo_name\n\n def reset(self, index: int | None = None, apply_test_patch: bool = False) -> tuple[str | None, dict]:\n \"\"\"\n Function to reset container between each task instance.\n\n * Clones instance's repository\n * Cleans repository of prior modifications\n * Resets environment variables\n * Check out base commit\n\n Args:\n index: index of task instance to reset to\n\n Returns:\n observation: output from container\n info: additional information (e.g. debugging information)\n \"\"\"\n info = {}\n info[\"commit_sha\"] = self.commit_sha\n\n # Get task instance\n self.idx = index if index is not None else self.idx\n self.record = self.data[self.idx]\n self.idx += 1\n\n # Set query, gold command\n self.base_commit = self.record[\"base_commit\"]\n self.query = self.record[\"problem_statement\"]\n self.challenge = self.record.get(\"challenge\")\n self.reward = None\n\n ### Reset Container ###\n self._init_docker_compose()\n\n if self.args.cache_task_images:\n cached_image = self._get_cached_task_image_name()\n if image_exists(cached_image):\n self.logger.info(f\"Restore environment from cached image {cached_image}\")\n self.close() # stop current container\n self._init_container(cached_image=cached_image)\n self.communicate(\"export $(xargs </.env)\")\n envs = self.communicate(\"env\")\n self.logger.debug(f\"Environment variables restored from the image:\\n{envs}\\n\")\n if apply_test_patch:\n self._apply_test_patch()\n return None, info\n else:\n self.logger.info(f\"Cached image {cached_image} not found, rebuilding task environment...\")\n\n # Init docker network\n self._init_docker_network()\n\n # Clone repository if not already cloned\n self.communicate(input=\"cd /\")\n folders = self.communicate(input=\"ls\").split(\"\\n\")\n if self._repo_name not in folders:\n self._copy_repo()\n\n self._reset_repository()\n self._reset_environment_variables()\n\n # Set up environment\n self.communicate_with_handling(\n \"source /root/miniconda3/etc/profile.d/conda.sh\",\n error_msg=\"Failed to source conda\",\n )\n\n system = self.communicate(\"uname -s\").strip().lower()\n arch = self.communicate(\"uname -m\").strip().lower()\n if system == \"linux\" and arch == \"x86_64\":\n self.communicate_with_handling(\n \"apt update; apt install build-essential -y\",\n error_msg=\"Failed to install build-essential\",\n timeout_duration=LONG_TIMEOUT,\n )\n\n # Call install environment helper function if specified\n if self.install_environment:\n self.install_env()\n # Install mypy for linting purposes\n self.communicate_with_handling(\"pip install flake8\", error_msg=\"Failed to install flake8 (lint library)\")\n\n if self.args.cache_task_images:\n envs = self.communicate(\"env\")\n self.logger.debug(f\"Environment variables to save:\\n{envs}\\n\")\n self.communicate(\"env >> /.env\")\n assert self.container_obj is not None # mypy\n self.container_obj.commit(cached_image)\n self.logger.info(f\"Container with environment {self.container_obj.id} cached as image {cached_image}\")\n\n if apply_test_patch:\n self._apply_test_patch()\n # Write any metadata to info if necessary\n return None, info\n\n def _reset_repository(self) -> None:\n \"\"\"Clean repository of any modifications + Checkout base commit\"\"\"\n startup_commands = [\n \"echo -n > /root/files_to_edit.txt\",\n f\"cd /{self._repo_name}\",\n \"export ROOT=$(pwd -P)\",\n ]\n if self.challenge is None:\n startup_commands += [\n \"git status\",\n \"git restore .\",\n f\"git reset --hard {self.base_commit}\",\n \"git clean -fdxq\",\n ]\n self.communicate_with_handling(\n input=\" && \".join(startup_commands),\n error_msg=\"Failed to clean repository\",\n )\n\n def _reset_environment_variables(self) -> None:\n \"\"\"Reset environment variables (`CURRENT_FILE`) etc. within container\"\"\"\n cmd = [\n 'export CURRENT_FILE=\"\"',\n \"export CURRENT_LINE=0\",\n \"export SEARCH_RESULTS=()\",\n \"export SEARCH_FILES=()\",\n \"export SEARCH_INDEX=0\",\n ]\n self.communicate_with_handling(\n input=\" && \".join(cmd),\n error_msg=\"Failed to reset environment variables\",\n )\n\n def reset_for_new_attempt(\n self,\n ) -> None:\n \"\"\"Compared to `reset`, which prepares the container for a new instance,\n this prepares the container for taking another shot at the same instance.\n \"\"\"\n self._reset_repository()\n self._reset_environment_variables()\n\n def _apply_test_patch(self):\n \"\"\"\n Apply test patch for oracle setting\n \"\"\"\n assert self.record is not None\n path_to_patch = \"test.patch\"\n with open(path_to_patch, \"w\") as f:\n f.write(self.record[\"test_patch\"])\n subprocess.run(\n f\"docker cp {path_to_patch} {self.container_name}:/root/test.patch\",\n shell=True,\n check=False,\n )\n self.communicate_with_handling(\n input=\"git apply /root/test.patch\",\n error_msg=\"Failed to apply test patch correctly\",\n )\n os.remove(path_to_patch)\n\n def _get_edited_files_with_context(self, patch: str) -> dict[str, str]:\n \"\"\"Get the edited files with context from the patch\"\"\"\n pf = PatchFormatter(patch, read_method=self.read_file) if patch else None\n out = {}\n for context_length in [30, 50, 70]:\n value = \"Empty. No edited files found.\"\n if pf is not None:\n value = pf.get_files_str(original=False, context_length=context_length)\n out[f\"edited_files{context_length}\"] = value\n return out\n\n def _terminate_interactive_session(self, session_name: str):\n if not self.interactive_session:\n # Maybe fixing #772\n return\n try:\n self.interactive_session.session_process.terminate()\n self.communicate(self.interactive_session.config.exit_command)\n except Exception as e:\n msg = (\n f\"Failed to terminate interactive session {session_name}: {e}.\"\n \"\\nHere's the full traceback\\n\" + traceback.format_exc()\n )\n self.logger.warning(msg)\n self.interactive_session = None\n\n def _handle_interactive_commands(self, observation: str) -> str:\n \"\"\"Handle interactive commands in the environment, essentially substituting dummy\n output for the actual output of the interactive commands.\n\n Args:\n observation: Output from running the interactive command wrappers in the\n environment. They will returns some dummy output that will be caught and then\n we will run the actual commands in the interactive session and return the\n actual output.\n\n Returns:\n observation: The observation shown to the model. If no interactive commands\n are detected, this is the same as the input observation.\n Else, only the output from the interactive commands is returned.\n \"\"\"\n session_name, interactive_commands = get_interactive_commands(observation, logger=self.logger)\n if session_name is None:\n return observation\n if (\n session_name is not None\n and self.interactive_session is not None\n and self.interactive_session.name != session_name\n ):\n return self.interactive_session._get_only_one_interactive_error_message_observation()\n\n observation = \"\"\n for command in interactive_commands:\n if command == \"START\":\n # Start the session if previous session does not exist\n if self.interactive_session is not None:\n return self.interactive_session._get_only_one_interactive_error_message_observation()\n assert self.container_name is not None\n _observation, self.interactive_session = get_interactive_session(\n ctr_name=self.container_name,\n ctr_obj=self.container_obj,\n cwd=\"/\" + self._repo_name,\n session_name=session_name,\n config=self.args.interactive_sessions_config[session_name],\n logger=self.logger,\n )\n observation += _observation\n elif command == \"STOP\":\n if self.interactive_session is None:\n observation = f\"Interactive session {session_name!r} is not running, so it cannot be stopped!\"\n else:\n if self.interactive_session.session_process.poll() is None:\n self.logger.warning(\"Session did not quit successfully, terminating.\")\n self.interactive_session.session_process.terminate()\n observation = f\"Interactive session {session_name!r} stopped successfully\"\n self.interactive_session = None\n else:\n if self.interactive_session is None:\n self.logger.warning(\"Tried to run interactive commands without starting session\")\n start_command = self.args.interactive_sessions_config[session_name].start_command\n observation = f\"Interactive session {session_name!r} is not running! please start it first using `{start_command}`\"\n elif self.interactive_session and self.interactive_session.session_process.poll() is not None:\n start_command = self.args.interactive_sessions_config[session_name].start_command\n observation = f\"Interactive session {session_name!r} was unexpectedly closed! Please start it again using `{start_command}`\"\n self._terminate_interactive_session(session_name=session_name)\n else:\n _observation, terminate = self.interactive_session.communicate_with_handling(\n command,\n timeout_duration=AGENT_ACTION_TIMEOUT,\n no_output_timeout_duration=AGENT_ACTION_NO_OUTPUT_TIMEOUT,\n )\n observation += _observation\n if terminate:\n self._terminate_interactive_session(session_name=session_name)\n observation += \"\\n\"\n return observation\n\n def step(self, action: str) -> tuple[str | None, int, bool, AgentInfo]:\n \"\"\"\n Runs an action proposed by the agent in the environment and returns the corresponding output.\n\n Args:\n action: command to run in bash shell\n\n Returns:\n observation: output from container\n reward: Always set to 0\n done: whether task is over\n info: additional information (e.g. debugging information)\n \"\"\"\n info: AgentInfo = {}\n # Make sure to have the right keys even if the submission is missing/empty\n info.update(self._get_edited_files_with_context(patch=\"\")) # type: ignore\n\n observation = \"\"\n # Handle special actions\n action = action.strip()\n if action == \"skip\":\n observation = \"Skipped\"\n info[\"exit_status\"] = \"skipped\"\n return observation, 0, True, info\n if action == \"exit_forfeit\":\n observation = \"Exited\"\n info[\"exit_status\"] = action\n return observation, 0, True, info\n if action in {\"exit_context\", \"exit_cost\", \"exit_error\", \"exit_format\", \"exit_api\"}:\n try:\n observation = self.communicate(input=\"submit\")\n submission = self.get_submission(observation)\n assert submission is not None and submission.strip() != \"\", AssertionError(\"No submission found.\")\n self.logger.info(f\"Found submission: {submission}\")\n info[\"exit_status\"] = f\"submitted ({action})\"\n info[\"submission\"] = submission\n info.update(self._get_edited_files_with_context(patch=submission)) # type: ignore\n observation = \"Exited (autosubmitted)\"\n self.logger.info(\"Exiting with autosubmission\")\n return observation, 0, True, info\n except KeyboardInterrupt:\n raise\n except:\n observation = \"Exited\"\n info[\"exit_status\"] = action\n return observation, 0, True, info\n\n # Attempt to run action in container\n observation = \"\"\n try:\n observation = self.communicate(\n input=action,\n timeout_duration=AGENT_ACTION_TIMEOUT,\n no_output_timeout_duration=AGENT_ACTION_NO_OUTPUT_TIMEOUT,\n set_last_action=True,\n )\n except TimeoutError as e:\n try:\n observation += e.args[1] if len(e.args) > 1 else \"\"\n observation += self.interrupt()\n observation += \"\\nEXECUTION TIMED OUT\"\n observation += (\n f\" BECAUSE NO OUTPUT WAS PRODUCED FOR MORE THAN {AGENT_ACTION_NO_OUTPUT_TIMEOUT} SECONDS.\\nPLEASE REFINE YOUR RUNNING COMMAND SO IT WILL PRODUCE OUTPUT IN THE SPECIFIED TIME FRAME.\"\n if isinstance(e, NoOutputTimeoutError)\n else f\" BECAUSE THE COMMAND WAS RUNNING FOR MORE THAN {AGENT_ACTION_TIMEOUT} SECONDS.\"\n )\n except RuntimeError as e:\n observation += e.args[1] if len(e.args) > 1 else \"\"\n observation += \"\\nEXECUTION TIMED OUT AND INTERRUPT FAILED. RESTARTING PROCESS.\"\n info[\"exit_status\"] = \"early_exit\"\n self.logger.warning(f\"Failed to interrupt container: {e}\\nRESTARTING PROCESS.\")\n self.reset_container()\n return observation, 0, True, info\n except RuntimeError as e:\n observation += e.args[1] if len(e.args) > 1 else \"\"\n observation += \"\\nCOMMAND FAILED TO EXECUTE. RESTARTING PROCESS.\"\n info[\"exit_status\"] = \"early_exit\"\n self.logger.warning(f\"Failed to execute command: {e}\\nRESTARTING PROCESS.\")\n self.reset_container()\n return observation, 0, True, info\n except BrokenPipeError as e:\n observation += \"\\nBROKEN PIPE ERROR. RESTARTING PROCESS.\"\n info[\"exit_status\"] = \"early_exit\"\n self.logger.error(f\"Broken pipe error: {e}\\nRESTARTING PROCESS.\")\n self.reset_container()\n return observation, 0, True, info\n except UnicodeError as e:\n observation += \"\\nCOMMAND PRODUCED TOO MANY NON-UNICODE CHARACTERS. PLEASE TRY ANOTHER COMMAND.\\nIF YOU WANT TO VIEW BINARY FILES, PLEASE USE `xxd` OR `hexdump` INSTEAD.\\n\"\n self.logger.error(f\"Unicode error: {e}\")\n except Exception:\n observation += \"\\nEXECUTION FAILED OR COMMAND MALFORMED\"\n self.logger.exception(\"Unknown exception\")\n\n # Record submission and end episode if `submit` keyword found\n submission = self.get_submission(observation)\n if submission is not None:\n if self.validate_submission(submission):\n self.logger.info(f\"Found submission: {submission}\")\n info[\"exit_status\"] = \"submitted\"\n info[\"submission\"] = submission if submission.strip() != \"\" else None\n info.update(self._get_edited_files_with_context(patch=submission)) # type: ignore\n observation = submission if submission.strip() != \"\" else None\n return observation, 0, True, info\n else:\n # Currently only validating CTF challenges\n assert self.challenge is not None\n self.logger.warning(f\"Wrong submission found: {submission} (real flag is {self.challenge['flag']})\")\n observation = \"Wrong flag!\"\n return observation, 0, False, info\n\n observation = self._handle_interactive_commands(observation)\n\n return observation, 0, False, info\n\n def close(self) -> None:\n \"\"\"\n Handle environment shutdown\n \"\"\"\n self.logger.info(\"Beginning environment shutdown...\")\n try:\n self.communicate(input=\"exit\")\n except KeyboardInterrupt:\n raise\n except:\n self.logger.warning(\"Errors when exiting container\", exc_info=True)\n assert self.container is not None # mypy\n self.container.terminate()\n if self.docker_compose is not None:\n terminate_docker_compose(self.docker_compose)\n if self.interactive_session is not None:\n try:\n self.interactive_session.session_process.terminate()\n except KeyboardInterrupt:\n raise\n except Exception:\n self.logger.warning(\"Failed to stop interactive session: %s\", traceback.format_exc())\n self.interactive_session = None\n else:\n self.logger.info(\"Interactive session stopped\")\n self.interactive_session = None\n if self.container_obj is None:\n pass\n elif self.persistent:\n # stopping is Podman specific, but doesn't hurt to include\n # https://stackoverflow.com/a/32428199/\n # Want to avoid https://github.com/princeton-nlp/SWE-agent/issues/496\n # Note that container_obj.status might not be updated throughout the container\n # lifecycle, so let's get the container_obj again\n assert self.container_name\n try:\n self.container_obj = docker.from_env().containers.get(self.container_name)\n except Exception:\n self.logger.warning(f\"Failed to get fresh container object: {traceback.format_exc()}\", exc_info=True)\n if self.container_obj.status not in {\"paused\", \"exited\", \"dead\", \"stopping\"}:\n try:\n self.container_obj.pause()\n except Exception:\n self.logger.warning(\"Failed to pause container.\", exc_info=True)\n except KeyboardInterrupt:\n raise\n else:\n self.logger.info(\"Agent container paused\")\n else:\n self.logger.info(f\"Agent container status: {self.container_obj.status}\")\n else:\n try:\n self.container_obj.remove(force=True)\n except KeyboardInterrupt:\n raise\n except docker.errors.NotFound:\n # We already tried to exit the container, so it's actually good if\n # it's not found\n pass\n except Exception:\n self.logger.warning(\"Failed to remove container\", exc_info=True)\n else:\n self.logger.info(\"Agent container stopped\")\n for hook in self.hooks:\n hook.on_close()\n\n # MARK: Helper functions #\n\n def _reset_container(self) -> None:\n if self.container is not None:\n try:\n self.container.terminate()\n except KeyboardInterrupt:\n raise\n except:\n self.logger.warning(\"Failed to terminate container\", exc_info=True)\n else:\n self.logger.debug(\"Terminated container\")\n if self.docker_compose is not None:\n try:\n terminate_docker_compose(self.docker_compose)\n except KeyboardInterrupt:\n raise\n except:\n self.logger.warning(\"Failed to terminate docker compose\", exc_info=True)\n else:\n self.logger.debug(\"Terminated docker compose\")\n self._init_container()\n self._init_scripts()\n\n def reset_container(self) -> None:\n self.close()\n self.container = None\n self.container_obj = None\n self._reset_container()\n\n @staticmethod\n def _get_container_name(image_name: str) -> str:\n \"\"\"Return name of container\"\"\"\n process_id = str(os.getpid())\n current_time = str(datetime.datetime.now())\n unique_string = current_time + process_id\n hash_object = hashlib.sha256(unique_string.encode())\n image_name_sanitized = image_name.replace(\"/\", \"-\")\n image_name_sanitized = image_name_sanitized.replace(\":\", \"-\")\n return f\"{image_name_sanitized}-{hash_object.hexdigest()[:10]}\"\n\n # ctf\n def _init_docker_network(self) -> None:\n \"\"\"\n Add the \"ctfnet\" network interface for all the containers used for CTF challenges\n \"\"\"\n assert self.container_name is not None\n if self.challenge is not None:\n attach_network_interface_to_container(self.container_name)\n\n # ctf\n def _init_docker_compose(self) -> None:\n \"\"\"\n Handles docker compose initialization for challenge with docker compose file.\n \"\"\"\n if self.challenge is not None and self.challenge.get(\"docker_compose\") is not None:\n self.docker_compose = get_docker_compose(self.challenge[\"docker_compose\"])\n self.logger.info(\"\ud83c\udf31 Initialized docker compose for challenge\")\n\n def _init_container(self, cached_image: str | None = None) -> None:\n \"\"\"\n Handles container initialization. Defines container name and creates it.\n If cached_image is provided, it will use that image name instead of the default.\n \"\"\"\n image_name = self.image_name\n if cached_image is not None:\n image_name = cached_image\n self.logger.info(f\"Using cached image: {image_name}\")\n if self.persistent:\n assert self.container_name is not None\n else:\n # Make sure that we get a new container name just in case removing didn't work.\n # Might be a fix for https://github.com/princeton-nlp/SWE-agent/issues/451\n self.container_name = self._get_container_name(image_name)\n self.container, self.parent_pids = get_container(\n self.container_name, image_name, persistent=self.persistent, container_mounts=self.container_mounts\n )\n try:\n client = docker.from_env(timeout=600)\n except docker.errors.DockerException as e:\n if \"Error while fetching server API version\" in str(e):\n msg = \"Docker is not running. Please start Docker and try again.\"\n else:\n msg = \"Unknown docker exception occurred. Are you sure docker is running?\"\n raise RuntimeError(msg) from e\n t0 = time.time()\n self.container_obj = None\n while time.time() - t0 < 60:\n try:\n self.container_obj = client.containers.get(self.container_name)\n except docker.errors.NotFound:\n self.logger.debug(\"Couldn't find container. Let's wait and retry.\")\n time.sleep(1)\n else:\n break\n else:\n print(f\"{self.persistent=}\")\n available_containers = client.containers.list(all=True)\n available_containers_info = json.dumps([str(c.attrs) for c in available_containers], indent=2)\n print(available_containers_info)\n msg = \"Failed to get container object.\"\n raise RuntimeError(msg)\n self.logger.info(\"\ud83c\udf31 Environment Initialized\")\n\n def _init_scripts(self):\n \"\"\"\n Initialize custom commands within container\n \"\"\"\n self.communicate_with_handling(\n \"source /root/.bashrc\",\n error_msg=\"Failed to source .bashrc\",\n )\n self.communicate_with_handling(\n \"mkdir -p /root/commands\",\n error_msg=\"Failed to create commands directory\",\n )\n self.communicate_with_handling(\n \"touch /root/commands/__init__.py\",\n error_msg=\"Failed to create __init__.py\",\n )\n self.communicate_with_handling(\n \"export PATH=$PATH:/root/commands\",\n error_msg=\"Failed to add commands directory to PATH\",\n )\n\n def _communicate_experimental(\n self,\n input: str,\n timeout_duration: int | float = 25,\n no_output_timeout_duration: int | float = 25,\n ) -> str:\n \"\"\"Experimental version of `_communicate`\"\"\"\n assert self.container is not None\n # Sleep to ensure that the exit code is in the last line\n # See https://github.com/princeton-nlp/SWE-agent/issues/595\n command_suffix = (\n f'EXITSTATUS=\"$?\"; sleep 0.01; echo {PROCESS_DONE_MARKER_START}$EXITSTATUS{PROCESS_DONE_MARKER_END}\\n'\n )\n try:\n self.returncode = None\n cmd = input if input.endswith(\"\\n\") else input + \"\\n\"\n cmd += command_suffix\n os.write(self.container.stdin.fileno(), cmd.encode()) # type: ignore\n time.sleep(0.03)\n self.container.stdin.flush() # type: ignore\n except BrokenPipeError:\n traceback.print_exc()\n self.logger.error(\"Failed to communicate with container. Check docker logs for more information.\")\n msg = \"Failed to communicate with container\"\n raise RuntimeError(msg)\n\n try:\n buffer, exit_code = read_with_timeout_experimental(\n self.container, timeout_duration, no_output_timeout_duration\n )\n except Exception:\n msg = f\"Read with timeout failed on input:\\n---\\n{input}\\n---\"\n self.logger.error(msg)\n raise\n if exit_code == \"$EXITSTATUS\":\n # this sometimes happens if the command badly fails\n # for example if you just try to run python with no arguments\n # in this case, the error message is usually also garbage, so let's set\n # something new.\n # See https://github.com/princeton-nlp/SWE-agent/issues/630\n buffer = (\n \"Unkknown error occurred when running the command. Please double check syntax \"\n \"and that you're not running an interactive command.\"\n )\n self.logger.warning(\"Couldn't get real exit code. Setting it to 999\")\n exit_code = 999\n elif not exit_code.isdigit():\n # this sometimes happens if the command is being killed, for example radare2\n # we set the error to 998 in that case\n self.logger.warning(\"Couldn't get real exit code. Setting it to 998\")\n exit_code = 998\n self.returncode = int(exit_code)\n return buffer\n\n def _communicate(\n self,\n input: str,\n timeout_duration: int | float = 25,\n no_output_timeout_duration: int | float = 25,\n ) -> str:\n \"\"\"Runs command in container and returns output\n\n Args:\n input: command to run in container\n timeout_duration: duration to wait for output\n no_output_timeout_duration: duration to wait when the process stopped produce any output\n \"\"\"\n assert self.container is not None\n communicate_method = keys_config.get(\n \"SWE_AGENT_COMMUNICATE_METHOD\", default=\"end-marker\", choices=[\"end-marker\", \"processes\"]\n )\n if communicate_method == \"end-marker\":\n return self._communicate_experimental(input, timeout_duration, no_output_timeout_duration)\n try:\n self.returncode = None\n cmd = input if input.endswith(\"\\n\") else input + \"\\n\"\n os.write(self.container.stdin.fileno(), cmd.encode()) # type: ignore\n time.sleep(0.1)\n self.container.stdin.flush() # type: ignore\n except BrokenPipeError:\n traceback.print_exc()\n self.logger.error(\"Failed to communicate with container. Check docker logs for more information.\")\n msg = \"Failed to communicate with container\"\n raise RuntimeError(msg)\n try:\n buffer = read_with_timeout(self.container, self.get_pids, timeout_duration)\n self.container.stdin.write(\"echo $?\\n\") # type: ignore\n time.sleep(0.1)\n self.container.stdin.flush() # type: ignore\n exit_code = read_with_timeout(self.container, self.get_pids, 5).strip()\n except Exception as e:\n self.logger.error(f\"Read with timeout failed on input:\\n---\\n{input}\\n---\")\n raise e\n if not exit_code.isdigit():\n msg = f\"Failed to get exit code. Output:\\n---\\n{buffer}\\n---\"\n raise RuntimeError(msg)\n self.returncode = int(exit_code)\n return buffer\n\n def _check_syntax(self, input: str) -> tuple[str, bool]:\n \"\"\"\n Check syntax of command.\n\n Returns:\n output: Output of the command\n success: whether the exit code was 0\n \"\"\"\n output = self._communicate(f\"/bin/bash -n <<'EOF'\\n{input}\\nEOF\\n\")\n return output, self.returncode == 0\n\n def communicate(\n self,\n input: str,\n timeout_duration: int | float = 25,\n no_output_timeout_duration: int | float | None = None,\n *,\n set_last_action: bool = False,\n ) -> str:\n \"\"\"\n Sends input to container and returns output\n\n Args:\n input: input to send to container\n timeout_duration: duration to wait for output\n set_last_action: whether to set the LAST_ACTION environment variable\n\n Returns:\n output: output from container\n \"\"\"\n assert self.container is not None\n if no_output_timeout_duration is None:\n no_output_timeout_duration = timeout_duration\n if input.strip() != \"exit\":\n self.logger.log(logging.TRACE, \"Input:\\n%s\", input) # type: ignore\n output, valid = self._check_syntax(input)\n if not valid:\n return output # shows syntax errors\n output = self._communicate(\n input,\n timeout_duration=timeout_duration,\n no_output_timeout_duration=no_output_timeout_duration,\n )\n self.logger.log(logging.TRACE, \"Output:\\n%s\", output) # type: ignore\n self.communicate_output = output\n if set_last_action:\n # Cannot merge this with last command, because of multiline command\n # handling.\n last_action_string = shlex.quote(input.strip())\n input = f\"export LAST_ACTION={last_action_string}\"\n self._communicate(input, timeout_duration=5, no_output_timeout_duration=5)\n return output\n else:\n self.container.terminate()\n self.returncode = 0\n self.communicate_output = \"\"\n return \"\"\n\n def communicate_with_handling(self, input: str, error_msg: str, timeout_duration: int | float = 25) -> str:\n \"\"\"\n Wrapper for communicate function that raises error if return code is non-zero\n\n Args:\n input: input to send to container\n error_msg: error message to raise if return code is non-zero\n timeout_duration: duration to wait for output\n\n Returns:\n output: output from container\n \"\"\"\n logs = self.communicate(input, timeout_duration=timeout_duration)\n if self.returncode != 0:\n self.logger.error(f\"{error_msg}: {logs}\")\n self.close()\n msg = f\"{error_msg}: {logs}\"\n raise RuntimeError(msg)\n return logs\n\n def get_available_actions(self) -> list[str]:\n \"\"\"\n Returns list of available actions in current environment state\n\n Currently not in use.\n \"\"\"\n return []\n\n def get_pids(self, all_pids: bool = False) -> list[tuple[str, str]]:\n \"\"\"\n Gets list of processes running inside docker container\n\n Args:\n all_pids: whether to return all pids, or whether to exclude ps\n and parent PIDs\n\n Returns:\n list of PIDs\n \"\"\"\n assert self.container_obj is not None\n pids = self.container_obj.exec_run(\"ps -eo pid,comm,ppid --no-headers\").output.decode().split(\"\\n\")\n pids = [x.split() for x in pids if x]\n if not all_pids:\n # Get just the PIDs of processes that are descendants of parent_pids and not others\n pids = [\n (x[0], x[1])\n for x in pids\n if x[1] != \"ps\"\n and x[0] not in self.parent_pids\n and x[1] != getattr(self.interactive_session, \"name\", None)\n and x[2] in self.parent_pids\n ]\n return pids\n\n # ctf\n def validate_submission(self, submission: str) -> bool:\n \"\"\"\n Function for validating submission for CTF challenges.\n\n Args:\n submission: extracted submission\n\n Returns:\n validation of the submission found against known flag\n \"\"\"\n submission = submission.strip()\n\n def wrap(s: str) -> str:\n return f\"flag{{{s}}}\"\n\n if self.challenge is not None:\n assert \"flag\" in self.challenge\n solution = self.challenge[\"flag\"]\n return any(\n (\n submission == solution,\n wrap(submission) == solution,\n submission == wrap(solution),\n )\n )\n\n return True\n\n def get_submission(self, output: str) -> str | None:\n \"\"\"\n Function for extracting diff patch submission at the end of an episode.\n\n Args:\n output: `submit` observation\n\n Returns:\n submission: diff patch submission\n \"\"\"\n pattern = r\"\\<\\<SUBMISSION\\|\\|(.*)\\|\\|SUBMISSION\\>\\>\"\n match = re.search(pattern, output, re.DOTALL)\n if match is None:\n return None\n return match.group(1)\n\n def run_shell_script(self, script_path: Path, *, location: str) -> None:\n \"\"\"Run custom script supplied by user at `script_path`\n\n Args:\n script_path: path to script file\n location: location of script file 'host' or 'container'\n \"\"\"\n if location == \"host\":\n return self._run_shell_script_host(script_path)\n elif location == \"container\":\n raise NotImplementedError\n msg = f\"Invalid 'location': {location}\"\n raise ValueError(msg)\n\n def _run_shell_script_host(self, script_path: Path) -> None:\n \"\"\"Run shell script file (located on host) in container\"\"\"\n if not script_path.is_file():\n msg = f\"Script not found at {script_path}\"\n raise FileNotFoundError(msg)\n shell_commands = Path(script_path).read_text().splitlines(keepends=True)\n for i, cmd in enumerate(shell_commands):\n self.communicate_with_handling(\n cmd,\n error_msg=f\"Failed to execute line {i}.\",\n timeout_duration=LONG_TIMEOUT,\n )\n\n def _get_install_configs(self) -> dict | None:\n \"\"\"Return config for environment setup\"\"\"\n assert self.record is not None # mypy\n if (\n self.record[\"problem_statement_source\"] != \"swe-bench\" or self.record[\"repo_type\"] == \"local\"\n ) and self.args.environment_setup is None:\n self.logger.warning(\n \"install_environment is set to True, but the data path is a GitHub URL \"\n \"without an environment config file (environment_config key/flag). \"\n \"Skipping conda environment installation.\",\n )\n return None\n if self.args.environment_setup is not None:\n assert isinstance(self.args.environment_setup, (str, os.PathLike))\n if Path(self.args.environment_setup).suffix in [\".yml\", \".yaml\"]:\n try:\n return yaml.safe_load(Path(self.args.environment_setup).read_text())\n except Exception as e:\n msg = \"Environment config file needs to be a yaml file\"\n raise ValueError(msg) from e\n elif Path(self.args.environment_setup).suffix == \".sh\":\n return {\n \"shell_script_path\": self.args.environment_setup,\n }\n else:\n msg = \"Environment config file needs to be a yaml file or shell script\"\n raise ValueError(msg)\n else:\n try:\n return MAP_REPO_VERSION_TO_SPECS[self.record[\"repo\"]][str(self.record[\"version\"])]\n except KeyError as e:\n msg = (\n \"Tried to look up install configs in swe-bench, but failed. \"\n \"You can set a custom environment config with the environment_config key/flag.\"\n )\n raise ValueError(msg) from e\n\n def _conda_environment_exists(self, env_name: str) -> bool:\n env_check = self.communicate(f\"conda env list | grep {env_name}\", timeout_duration=LONG_TIMEOUT)\n return env_check.strip() != \"\"\n\n def install_env(self) -> None:\n \"\"\"\n Creates conda environment and installs third party dependencies to allow code execution\n \"\"\"\n t0 = time.perf_counter()\n for hook in self.hooks:\n hook.on_install_env_started()\n install_configs = self._get_install_configs()\n if not install_configs:\n return\n if \"shell_script_path\" in install_configs:\n assert len(install_configs) == 1\n self.run_shell_script(Path(install_configs[\"shell_script_path\"]), location=\"host\")\n return\n assert self.record is not None # mypy\n # Create environment if does not exist yet\n env_name = f\"{self._repo_name}__{self.record['version']}\"\n if not self._conda_environment_exists(env_name):\n self.logger.info(f\"{env_name} conda env not found, creating...\")\n packages = install_configs.get(\"packages\", \"\")\n if packages == \"requirements.txt\":\n # Create conda environment\n self.communicate_with_handling(\n f\"conda create -n {env_name} python={install_configs['python']} -y\",\n error_msg=\"Failed to create conda environment\",\n timeout_duration=LONG_TIMEOUT,\n )\n self.logger.debug(\"Created conda environment\")\n # Write reqs to requirements.txt in docker container\n content_reqs = get_requirements(self.record)\n copy_file_to_container(self.container_obj, content_reqs, PATH_TO_REQS)\n # Create conda environment + install reqs\n self.communicate_with_handling(\n f\"conda activate {env_name}\",\n error_msg=\"Failed to activate conda environment\",\n )\n self.communicate_with_handling(\n f\"pip install -r {PATH_TO_REQS}\",\n error_msg=\"Failed to install requirements.txt\",\n timeout_duration=LONG_TIMEOUT,\n )\n self.logger.debug(\"Installed requirements from requirements.txt\")\n self.communicate(f\"rm {PATH_TO_REQS}\")\n elif packages == \"environment.yml\":\n # Write environment.yml to file\n content_env_yml = get_environment_yml(self.record, env_name)\n # Hotfix for\n if not install_configs.get(\"no_use_env\"):\n content_env_yml += f'\\n - python={install_configs[\"python\"]}\\n'\n copy_file_to_container(self.container_obj, content_env_yml, PATH_TO_ENV_YML)\n if install_configs.get(\"no_use_env\"):\n # Create conda environment\n self.communicate_with_handling(\n f\"conda create -c conda-forge -n {env_name} python={install_configs['python']} -y\",\n error_msg=\"Failed to create conda environment\",\n timeout_duration=LONG_TIMEOUT,\n )\n self.logger.debug(\"Created conda environment\")\n # Install packages\n self.communicate_with_handling(\n f\"conda env update -f {PATH_TO_ENV_YML}\",\n error_msg=\"Failed to install environment.yml\",\n timeout_duration=LONG_TIMEOUT,\n )\n self.logger.debug(\"Installed packages from environment.yml\")\n else:\n # Create environment + install packages\n self.communicate_with_handling(\n f\"conda env create --file {PATH_TO_ENV_YML}\",\n error_msg=\"Failed to create conda environment with environment.yml\",\n timeout_duration=LONG_TIMEOUT,\n )\n self.logger.debug(\"Created conda environment with environment.yml\")\n self.communicate(f\"rm {PATH_TO_ENV_YML}\")\n else:\n python_env = f\"python{install_configs['python']}\"\n if self._conda_environment_exists(python_env):\n self.communicate_with_handling(\n f\"conda create --name {env_name} --clone {python_env}\",\n error_msg=\"Failed to clone conda environment\",\n timeout_duration=LONG_TIMEOUT,\n )\n self.logger.debug(\"Cloned python conda environment\")\n else:\n self.logger.debug(f\"Could not find {python_env}, creating new environment\")\n self.communicate_with_handling(\n f\"conda create -n {env_name} python={install_configs['python']} -y\",\n error_msg=\"Failed to create conda environment\",\n timeout_duration=LONG_TIMEOUT,\n )\n self.communicate_with_handling(\n f\"conda activate {env_name}\",\n error_msg=\"Failed to activate conda environment\",\n )\n if packages.strip():\n self.communicate_with_handling(\n f\"conda install {packages} -y\",\n error_msg=\"Failed to install packages\",\n timeout_duration=LONG_TIMEOUT,\n )\n self.logger.debug(\"Installed conda packages\")\n # Install extra pip packages if specified\n if install_configs.get(\"pip_packages\"):\n self.communicate_with_handling(\n f\"source activate {env_name} && pip install {' '.join(install_configs['pip_packages'])}\",\n error_msg=\"Failed to install pip packages\",\n timeout_duration=LONG_TIMEOUT,\n )\n self.logger.debug(\"Installed extra pip dependencies\")\n\n # Activate environment\n self.communicate_with_handling(f\"conda activate {env_name}\", error_msg=\"Failed to activate conda environment\")\n\n # Install repo at base commit\n if install_configs.get(\"pre_install\"):\n self.logger.info(\"Running pre-install commands...\")\n for pre_install_cmd in install_configs[\"pre_install\"]:\n self.communicate_with_handling(\n pre_install_cmd,\n error_msg=\"Pre-install commands failed to execute successfully\",\n timeout_duration=LONG_TIMEOUT,\n )\n self.logger.debug(\"Ran pre-install commands\")\n self.logger.info(f\"Installing {self._repo_name} at base commit...\")\n if install_configs.get(\"install\"):\n install_cmd = install_configs[\"install\"]\n self.communicate_with_handling(\n install_cmd,\n error_msg=\"Install command failed to execute successfully\",\n timeout_duration=LONG_TIMEOUT,\n )\n self.logger.debug(\"Ran install command\")\n if install_configs.get(\"post_install\"):\n self.logger.info(\"Running post-install commands...\")\n for post_install_cmd in install_configs[\"post_install\"]:\n self.communicate_with_handling(\n post_install_cmd,\n error_msg=\"Post-install commands failed to execute successfully\",\n )\n self.logger.debug(\"Ran post-install commands\")\n\n self.logger.info(\"Installation step took %.2f seconds\", time.perf_counter() - t0)\n\n def add_commands(self, commands: list[dict]) -> None:\n \"\"\"\n Adds custom commands to container\n \"\"\"\n for command in commands:\n name = command[\"name\"]\n contents = command[\"contents\"]\n copy_file_to_container(self.container_obj, contents, f\"/root/commands/{name}\")\n if command[\"type\"] == \"source_file\":\n self.communicate_with_handling(\n f\"source /root/commands/{name}\",\n error_msg=(\n f\"Failed to source {name}. If you meant to make a script,\"\n \" start the file with a shebang (e.g. #!/usr/bin/env python).\"\n ),\n )\n elif command[\"type\"] == \"script\":\n self.communicate_with_handling(\n f\"chmod +x /root/commands/{name}\",\n error_msg=f\"Failed to chmod {name}\",\n )\n elif command[\"type\"] == \"utility\":\n # nothing to do for utility scripts\n pass\n else:\n msg = f\"Invalid command type: {command['type']}\"\n raise ValueError(msg)\n\n def interrupt(self) -> str:\n \"\"\"\n Send interrupt signal to container and exhaust stdout buffer with a communicate call\n \"\"\"\n assert self.container is not None\n assert self.container_obj is not None\n pids = self.get_pids()\n for pid, _ in pids:\n # Sending signal several times ensures that the process is dead\n for _ in range(3):\n self.container_obj.exec_run(f\"kill -9 {pid}\")\n observation = \"\"\n try:\n observation += read_with_timeout(self.container, self.get_pids, 20)\n except TimeoutError:\n pass\n try:\n # This is a workaround because of bash behaviour\n # when sometimes we get the prints of Killed after we press some \"Enter\" in stdin\n self.communicate(input=\"echo 'interrupted'\", timeout_duration=5)\n output = self.communicate(input=\"echo 'interrupted'\", timeout_duration=5)\n assert output.strip().endswith(\"interrupted\"), \"container health check failed\"\n except TimeoutError:\n msg = \"Failed to interrupt container\"\n raise RuntimeError(msg)\n return observation\n\n def open_pr(self, *, trajectory, _dry_run: bool = False) -> None:\n \"\"\"Create PR to repository\n\n Args:\n trajectory: Trajectory of actions taken by the agent\n _dry_run: Whether to actually push anything or just simulate it\n \"\"\"\n self.logger.info(\"Opening PR\")\n # TODO: have better way of handling this\n # Adding random string suffix to avoid name conflicts if we had a previously failed run\n issue_url = self.args.data_path\n try:\n issue = get_gh_issue_data(issue_url, token=self._github_token)\n except InvalidGithubURL as e:\n msg = \"Data path must be a github issue URL if --open_pr is set.\"\n raise ValueError(msg) from e\n branch_name = f\"swe-agent-fix-#{issue.number}-\" + str(random.random())[2:10]\n\n self.communicate_with_handling(\n input=\"rm -f model.patch\",\n error_msg=\"Failed to remove model patch\",\n timeout_duration=10,\n )\n self.communicate_with_handling(\n input=f\"git checkout -b {branch_name}\",\n error_msg=\"Failed to switch to new branch\",\n timeout_duration=10,\n )\n self.communicate_with_handling(\n input=\"git add .\",\n error_msg=\"Failed to add commits\",\n timeout_duration=10,\n )\n dry_run_flag = \"--allow-empty\" if _dry_run else \"\"\n commit_msg = [\n shlex.quote(\"Fix: {issue.title}\"),\n shlex.quote(\"Closes #{issue.number}\"),\n ]\n self.communicate_with_handling(\n input=f\"git commit -m {commit_msg[0]} -m {commit_msg[1]} {dry_run_flag}\",\n error_msg=\"Failed to commit changes\",\n timeout_duration=10,\n )\n\n owner, repo, _ = parse_gh_issue_url(issue_url)\n # If `--repo_path` was specified with a different github URL, then the record will contain\n # the forking user\n assert self.record is not None\n if self.record[\"repo_type\"] != \"github\":\n # We already validated that `--data_path` is a github issue URL\n # so this is the only case where we can reach here\n msg = \"--repo_path must point to a github URL if --open_pr is set\"\n raise ValueError(msg)\n forker, _ = self.record[\"repo\"].split(\"/\")\n head = branch_name\n remote = \"origin\"\n if forker != owner:\n head = f\"{forker}:{branch_name}\"\n token_prefix = \"\"\n if self._github_token:\n token_prefix = f\"{self._github_token}@\"\n fork_url = f\"https://{token_prefix}github.com/{forker}/{repo}.git\"\n self.logger.debug(f\"Using fork: {fork_url}\")\n self.communicate_with_handling(\n input=f\"git remote add fork {fork_url}\",\n error_msg=\"Failed to create new git remote\",\n timeout_duration=10,\n )\n remote = \"fork\"\n dry_run_prefix = \"echo \" if _dry_run else \"\"\n self.communicate_with_handling(\n input=f\"{dry_run_prefix} git push {remote} {branch_name}\",\n error_msg=(\n \"Failed to push branch to remote. Please check your token and permissions. \"\n \"You might want to push to a fork with the push_gh_repo_url option.\"\n ),\n timeout_duration=10,\n )\n body = (\n f\"This is a PR opened by AI tool [SWE Agent](https://github.com/princeton-nlp/SWE-agent/) \"\n f\"to close [#{issue.number}]({issue_url}) ({issue.title}).\\n\\nCloses #{issue.number}.\"\n )\n body += \"\\n\\n\" + format_trajectory_markdown(trajectory)\n api = GhApi(token=self._github_token)\n if not _dry_run:\n pr_info = api.pulls.create( # type: ignore\n owner=owner,\n repo=repo,\n title=f\"SWE-agent[bot] PR to fix: {issue.title}\",\n head=head,\n base=\"main\",\n body=body,\n draft=True,\n )\n self.logger.info(\n f\"\ud83c\udf89 PR created as a draft at {pr_info.html_url}. Please review it carefully, push \"\n \"any required changes onto the branch and then click \"\n \"'Ready for Review' to bring it to the attention of the maintainers.\",\n )\n\n def read_file(self, path: str | PurePath) -> str:\n \"\"\"Read file contents from container\n\n Args:\n path: Path to file relative to repository root\n\n Returns:\n file_contents: Contents of file as string\n \"\"\"\n path_in_container = f\"/{self._repo_name}/{path}\"\n return self.communicate(f\"cat {str(path_in_container)}\")\n</code></pre>"},{"location":"reference/env/#sweagent.environment.swe_env.SWEEnv.add_commands","title":"<code>add_commands(commands)</code>","text":"<p>Adds custom commands to container</p> Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>def add_commands(self, commands: list[dict]) -> None:\n \"\"\"\n Adds custom commands to container\n \"\"\"\n for command in commands:\n name = command[\"name\"]\n contents = command[\"contents\"]\n copy_file_to_container(self.container_obj, contents, f\"/root/commands/{name}\")\n if command[\"type\"] == \"source_file\":\n self.communicate_with_handling(\n f\"source /root/commands/{name}\",\n error_msg=(\n f\"Failed to source {name}. If you meant to make a script,\"\n \" start the file with a shebang (e.g. #!/usr/bin/env python).\"\n ),\n )\n elif command[\"type\"] == \"script\":\n self.communicate_with_handling(\n f\"chmod +x /root/commands/{name}\",\n error_msg=f\"Failed to chmod {name}\",\n )\n elif command[\"type\"] == \"utility\":\n # nothing to do for utility scripts\n pass\n else:\n msg = f\"Invalid command type: {command['type']}\"\n raise ValueError(msg)\n</code></pre>"},{"location":"reference/env/#sweagent.environment.swe_env.SWEEnv.add_hook","title":"<code>add_hook(hook)</code>","text":"<p>Add <code>EnvHook</code> to the environment.</p> <p>This allows to inject custom functionality at different stages of the environment lifecycle, in particular to connect SWE-agent to a new interface (like a GUI).</p> Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>def add_hook(self, hook: EnvHook):\n \"\"\"Add `EnvHook` to the environment.\n\n This allows to inject custom functionality at different stages of the environment\n lifecycle, in particular to connect SWE-agent to a new interface (like a GUI).\n \"\"\"\n hook.on_init()\n self.hooks.append(hook)\n</code></pre>"},{"location":"reference/env/#sweagent.environment.swe_env.SWEEnv.close","title":"<code>close()</code>","text":"<p>Handle environment shutdown</p> Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>def close(self) -> None:\n \"\"\"\n Handle environment shutdown\n \"\"\"\n self.logger.info(\"Beginning environment shutdown...\")\n try:\n self.communicate(input=\"exit\")\n except KeyboardInterrupt:\n raise\n except:\n self.logger.warning(\"Errors when exiting container\", exc_info=True)\n assert self.container is not None # mypy\n self.container.terminate()\n if self.docker_compose is not None:\n terminate_docker_compose(self.docker_compose)\n if self.interactive_session is not None:\n try:\n self.interactive_session.session_process.terminate()\n except KeyboardInterrupt:\n raise\n except Exception:\n self.logger.warning(\"Failed to stop interactive session: %s\", traceback.format_exc())\n self.interactive_session = None\n else:\n self.logger.info(\"Interactive session stopped\")\n self.interactive_session = None\n if self.container_obj is None:\n pass\n elif self.persistent:\n # stopping is Podman specific, but doesn't hurt to include\n # https://stackoverflow.com/a/32428199/\n # Want to avoid https://github.com/princeton-nlp/SWE-agent/issues/496\n # Note that container_obj.status might not be updated throughout the container\n # lifecycle, so let's get the container_obj again\n assert self.container_name\n try:\n self.container_obj = docker.from_env().containers.get(self.container_name)\n except Exception:\n self.logger.warning(f\"Failed to get fresh container object: {traceback.format_exc()}\", exc_info=True)\n if self.container_obj.status not in {\"paused\", \"exited\", \"dead\", \"stopping\"}:\n try:\n self.container_obj.pause()\n except Exception:\n self.logger.warning(\"Failed to pause container.\", exc_info=True)\n except KeyboardInterrupt:\n raise\n else:\n self.logger.info(\"Agent container paused\")\n else:\n self.logger.info(f\"Agent container status: {self.container_obj.status}\")\n else:\n try:\n self.container_obj.remove(force=True)\n except KeyboardInterrupt:\n raise\n except docker.errors.NotFound:\n # We already tried to exit the container, so it's actually good if\n # it's not found\n pass\n except Exception:\n self.logger.warning(\"Failed to remove container\", exc_info=True)\n else:\n self.logger.info(\"Agent container stopped\")\n for hook in self.hooks:\n hook.on_close()\n</code></pre>"},{"location":"reference/env/#sweagent.environment.swe_env.SWEEnv.communicate","title":"<code>communicate(input, timeout_duration=25, no_output_timeout_duration=None, *, set_last_action=False)</code>","text":"<p>Sends input to container and returns output</p> <p>Parameters:</p> Name Type Description Default <code>input</code> <code>str</code> <p>input to send to container</p> required <code>timeout_duration</code> <code>int | float</code> <p>duration to wait for output</p> <code>25</code> <code>set_last_action</code> <code>bool</code> <p>whether to set the LAST_ACTION environment variable</p> <code>False</code> <p>Returns:</p> Name Type Description <code>output</code> <code>str</code> <p>output from container</p> Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>def communicate(\n self,\n input: str,\n timeout_duration: int | float = 25,\n no_output_timeout_duration: int | float | None = None,\n *,\n set_last_action: bool = False,\n) -> str:\n \"\"\"\n Sends input to container and returns output\n\n Args:\n input: input to send to container\n timeout_duration: duration to wait for output\n set_last_action: whether to set the LAST_ACTION environment variable\n\n Returns:\n output: output from container\n \"\"\"\n assert self.container is not None\n if no_output_timeout_duration is None:\n no_output_timeout_duration = timeout_duration\n if input.strip() != \"exit\":\n self.logger.log(logging.TRACE, \"Input:\\n%s\", input) # type: ignore\n output, valid = self._check_syntax(input)\n if not valid:\n return output # shows syntax errors\n output = self._communicate(\n input,\n timeout_duration=timeout_duration,\n no_output_timeout_duration=no_output_timeout_duration,\n )\n self.logger.log(logging.TRACE, \"Output:\\n%s\", output) # type: ignore\n self.communicate_output = output\n if set_last_action:\n # Cannot merge this with last command, because of multiline command\n # handling.\n last_action_string = shlex.quote(input.strip())\n input = f\"export LAST_ACTION={last_action_string}\"\n self._communicate(input, timeout_duration=5, no_output_timeout_duration=5)\n return output\n else:\n self.container.terminate()\n self.returncode = 0\n self.communicate_output = \"\"\n return \"\"\n</code></pre>"},{"location":"reference/env/#sweagent.environment.swe_env.SWEEnv.communicate_with_handling","title":"<code>communicate_with_handling(input, error_msg, timeout_duration=25)</code>","text":"<p>Wrapper for communicate function that raises error if return code is non-zero</p> <p>Parameters:</p> Name Type Description Default <code>input</code> <code>str</code> <p>input to send to container</p> required <code>error_msg</code> <code>str</code> <p>error message to raise if return code is non-zero</p> required <code>timeout_duration</code> <code>int | float</code> <p>duration to wait for output</p> <code>25</code> <p>Returns:</p> Name Type Description <code>output</code> <code>str</code> <p>output from container</p> Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>def communicate_with_handling(self, input: str, error_msg: str, timeout_duration: int | float = 25) -> str:\n \"\"\"\n Wrapper for communicate function that raises error if return code is non-zero\n\n Args:\n input: input to send to container\n error_msg: error message to raise if return code is non-zero\n timeout_duration: duration to wait for output\n\n Returns:\n output: output from container\n \"\"\"\n logs = self.communicate(input, timeout_duration=timeout_duration)\n if self.returncode != 0:\n self.logger.error(f\"{error_msg}: {logs}\")\n self.close()\n msg = f\"{error_msg}: {logs}\"\n raise RuntimeError(msg)\n return logs\n</code></pre>"},{"location":"reference/env/#sweagent.environment.swe_env.SWEEnv.get_available_actions","title":"<code>get_available_actions()</code>","text":"<p>Returns list of available actions in current environment state</p> <p>Currently not in use.</p> Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>def get_available_actions(self) -> list[str]:\n \"\"\"\n Returns list of available actions in current environment state\n\n Currently not in use.\n \"\"\"\n return []\n</code></pre>"},{"location":"reference/env/#sweagent.environment.swe_env.SWEEnv.get_pids","title":"<code>get_pids(all_pids=False)</code>","text":"<p>Gets list of processes running inside docker container</p> <p>Parameters:</p> Name Type Description Default <code>all_pids</code> <code>bool</code> <p>whether to return all pids, or whether to exclude ps and parent PIDs</p> <code>False</code> <p>Returns:</p> Type Description <code>list[tuple[str, str]]</code> <p>list of PIDs</p> Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>def get_pids(self, all_pids: bool = False) -> list[tuple[str, str]]:\n \"\"\"\n Gets list of processes running inside docker container\n\n Args:\n all_pids: whether to return all pids, or whether to exclude ps\n and parent PIDs\n\n Returns:\n list of PIDs\n \"\"\"\n assert self.container_obj is not None\n pids = self.container_obj.exec_run(\"ps -eo pid,comm,ppid --no-headers\").output.decode().split(\"\\n\")\n pids = [x.split() for x in pids if x]\n if not all_pids:\n # Get just the PIDs of processes that are descendants of parent_pids and not others\n pids = [\n (x[0], x[1])\n for x in pids\n if x[1] != \"ps\"\n and x[0] not in self.parent_pids\n and x[1] != getattr(self.interactive_session, \"name\", None)\n and x[2] in self.parent_pids\n ]\n return pids\n</code></pre>"},{"location":"reference/env/#sweagent.environment.swe_env.SWEEnv.get_submission","title":"<code>get_submission(output)</code>","text":"<p>Function for extracting diff patch submission at the end of an episode.</p> <p>Parameters:</p> Name Type Description Default <code>output</code> <code>str</code> <p><code>submit</code> observation</p> required <p>Returns:</p> Name Type Description <code>submission</code> <code>str | None</code> <p>diff patch submission</p> Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>def get_submission(self, output: str) -> str | None:\n \"\"\"\n Function for extracting diff patch submission at the end of an episode.\n\n Args:\n output: `submit` observation\n\n Returns:\n submission: diff patch submission\n \"\"\"\n pattern = r\"\\<\\<SUBMISSION\\|\\|(.*)\\|\\|SUBMISSION\\>\\>\"\n match = re.search(pattern, output, re.DOTALL)\n if match is None:\n return None\n return match.group(1)\n</code></pre>"},{"location":"reference/env/#sweagent.environment.swe_env.SWEEnv.install_env","title":"<code>install_env()</code>","text":"<p>Creates conda environment and installs third party dependencies to allow code execution</p> Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>def install_env(self) -> None:\n \"\"\"\n Creates conda environment and installs third party dependencies to allow code execution\n \"\"\"\n t0 = time.perf_counter()\n for hook in self.hooks:\n hook.on_install_env_started()\n install_configs = self._get_install_configs()\n if not install_configs:\n return\n if \"shell_script_path\" in install_configs:\n assert len(install_configs) == 1\n self.run_shell_script(Path(install_configs[\"shell_script_path\"]), location=\"host\")\n return\n assert self.record is not None # mypy\n # Create environment if does not exist yet\n env_name = f\"{self._repo_name}__{self.record['version']}\"\n if not self._conda_environment_exists(env_name):\n self.logger.info(f\"{env_name} conda env not found, creating...\")\n packages = install_configs.get(\"packages\", \"\")\n if packages == \"requirements.txt\":\n # Create conda environment\n self.communicate_with_handling(\n f\"conda create -n {env_name} python={install_configs['python']} -y\",\n error_msg=\"Failed to create conda environment\",\n timeout_duration=LONG_TIMEOUT,\n )\n self.logger.debug(\"Created conda environment\")\n # Write reqs to requirements.txt in docker container\n content_reqs = get_requirements(self.record)\n copy_file_to_container(self.container_obj, content_reqs, PATH_TO_REQS)\n # Create conda environment + install reqs\n self.communicate_with_handling(\n f\"conda activate {env_name}\",\n error_msg=\"Failed to activate conda environment\",\n )\n self.communicate_with_handling(\n f\"pip install -r {PATH_TO_REQS}\",\n error_msg=\"Failed to install requirements.txt\",\n timeout_duration=LONG_TIMEOUT,\n )\n self.logger.debug(\"Installed requirements from requirements.txt\")\n self.communicate(f\"rm {PATH_TO_REQS}\")\n elif packages == \"environment.yml\":\n # Write environment.yml to file\n content_env_yml = get_environment_yml(self.record, env_name)\n # Hotfix for\n if not install_configs.get(\"no_use_env\"):\n content_env_yml += f'\\n - python={install_configs[\"python\"]}\\n'\n copy_file_to_container(self.container_obj, content_env_yml, PATH_TO_ENV_YML)\n if install_configs.get(\"no_use_env\"):\n # Create conda environment\n self.communicate_with_handling(\n f\"conda create -c conda-forge -n {env_name} python={install_configs['python']} -y\",\n error_msg=\"Failed to create conda environment\",\n timeout_duration=LONG_TIMEOUT,\n )\n self.logger.debug(\"Created conda environment\")\n # Install packages\n self.communicate_with_handling(\n f\"conda env update -f {PATH_TO_ENV_YML}\",\n error_msg=\"Failed to install environment.yml\",\n timeout_duration=LONG_TIMEOUT,\n )\n self.logger.debug(\"Installed packages from environment.yml\")\n else:\n # Create environment + install packages\n self.communicate_with_handling(\n f\"conda env create --file {PATH_TO_ENV_YML}\",\n error_msg=\"Failed to create conda environment with environment.yml\",\n timeout_duration=LONG_TIMEOUT,\n )\n self.logger.debug(\"Created conda environment with environment.yml\")\n self.communicate(f\"rm {PATH_TO_ENV_YML}\")\n else:\n python_env = f\"python{install_configs['python']}\"\n if self._conda_environment_exists(python_env):\n self.communicate_with_handling(\n f\"conda create --name {env_name} --clone {python_env}\",\n error_msg=\"Failed to clone conda environment\",\n timeout_duration=LONG_TIMEOUT,\n )\n self.logger.debug(\"Cloned python conda environment\")\n else:\n self.logger.debug(f\"Could not find {python_env}, creating new environment\")\n self.communicate_with_handling(\n f\"conda create -n {env_name} python={install_configs['python']} -y\",\n error_msg=\"Failed to create conda environment\",\n timeout_duration=LONG_TIMEOUT,\n )\n self.communicate_with_handling(\n f\"conda activate {env_name}\",\n error_msg=\"Failed to activate conda environment\",\n )\n if packages.strip():\n self.communicate_with_handling(\n f\"conda install {packages} -y\",\n error_msg=\"Failed to install packages\",\n timeout_duration=LONG_TIMEOUT,\n )\n self.logger.debug(\"Installed conda packages\")\n # Install extra pip packages if specified\n if install_configs.get(\"pip_packages\"):\n self.communicate_with_handling(\n f\"source activate {env_name} && pip install {' '.join(install_configs['pip_packages'])}\",\n error_msg=\"Failed to install pip packages\",\n timeout_duration=LONG_TIMEOUT,\n )\n self.logger.debug(\"Installed extra pip dependencies\")\n\n # Activate environment\n self.communicate_with_handling(f\"conda activate {env_name}\", error_msg=\"Failed to activate conda environment\")\n\n # Install repo at base commit\n if install_configs.get(\"pre_install\"):\n self.logger.info(\"Running pre-install commands...\")\n for pre_install_cmd in install_configs[\"pre_install\"]:\n self.communicate_with_handling(\n pre_install_cmd,\n error_msg=\"Pre-install commands failed to execute successfully\",\n timeout_duration=LONG_TIMEOUT,\n )\n self.logger.debug(\"Ran pre-install commands\")\n self.logger.info(f\"Installing {self._repo_name} at base commit...\")\n if install_configs.get(\"install\"):\n install_cmd = install_configs[\"install\"]\n self.communicate_with_handling(\n install_cmd,\n error_msg=\"Install command failed to execute successfully\",\n timeout_duration=LONG_TIMEOUT,\n )\n self.logger.debug(\"Ran install command\")\n if install_configs.get(\"post_install\"):\n self.logger.info(\"Running post-install commands...\")\n for post_install_cmd in install_configs[\"post_install\"]:\n self.communicate_with_handling(\n post_install_cmd,\n error_msg=\"Post-install commands failed to execute successfully\",\n )\n self.logger.debug(\"Ran post-install commands\")\n\n self.logger.info(\"Installation step took %.2f seconds\", time.perf_counter() - t0)\n</code></pre>"},{"location":"reference/env/#sweagent.environment.swe_env.SWEEnv.interrupt","title":"<code>interrupt()</code>","text":"<p>Send interrupt signal to container and exhaust stdout buffer with a communicate call</p> Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>def interrupt(self) -> str:\n \"\"\"\n Send interrupt signal to container and exhaust stdout buffer with a communicate call\n \"\"\"\n assert self.container is not None\n assert self.container_obj is not None\n pids = self.get_pids()\n for pid, _ in pids:\n # Sending signal several times ensures that the process is dead\n for _ in range(3):\n self.container_obj.exec_run(f\"kill -9 {pid}\")\n observation = \"\"\n try:\n observation += read_with_timeout(self.container, self.get_pids, 20)\n except TimeoutError:\n pass\n try:\n # This is a workaround because of bash behaviour\n # when sometimes we get the prints of Killed after we press some \"Enter\" in stdin\n self.communicate(input=\"echo 'interrupted'\", timeout_duration=5)\n output = self.communicate(input=\"echo 'interrupted'\", timeout_duration=5)\n assert output.strip().endswith(\"interrupted\"), \"container health check failed\"\n except TimeoutError:\n msg = \"Failed to interrupt container\"\n raise RuntimeError(msg)\n return observation\n</code></pre>"},{"location":"reference/env/#sweagent.environment.swe_env.SWEEnv.open_pr","title":"<code>open_pr(*, trajectory, _dry_run=False)</code>","text":"<p>Create PR to repository</p> <p>Parameters:</p> Name Type Description Default <code>trajectory</code> <p>Trajectory of actions taken by the agent</p> required <code>_dry_run</code> <code>bool</code> <p>Whether to actually push anything or just simulate it</p> <code>False</code> Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>def open_pr(self, *, trajectory, _dry_run: bool = False) -> None:\n \"\"\"Create PR to repository\n\n Args:\n trajectory: Trajectory of actions taken by the agent\n _dry_run: Whether to actually push anything or just simulate it\n \"\"\"\n self.logger.info(\"Opening PR\")\n # TODO: have better way of handling this\n # Adding random string suffix to avoid name conflicts if we had a previously failed run\n issue_url = self.args.data_path\n try:\n issue = get_gh_issue_data(issue_url, token=self._github_token)\n except InvalidGithubURL as e:\n msg = \"Data path must be a github issue URL if --open_pr is set.\"\n raise ValueError(msg) from e\n branch_name = f\"swe-agent-fix-#{issue.number}-\" + str(random.random())[2:10]\n\n self.communicate_with_handling(\n input=\"rm -f model.patch\",\n error_msg=\"Failed to remove model patch\",\n timeout_duration=10,\n )\n self.communicate_with_handling(\n input=f\"git checkout -b {branch_name}\",\n error_msg=\"Failed to switch to new branch\",\n timeout_duration=10,\n )\n self.communicate_with_handling(\n input=\"git add .\",\n error_msg=\"Failed to add commits\",\n timeout_duration=10,\n )\n dry_run_flag = \"--allow-empty\" if _dry_run else \"\"\n commit_msg = [\n shlex.quote(\"Fix: {issue.title}\"),\n shlex.quote(\"Closes #{issue.number}\"),\n ]\n self.communicate_with_handling(\n input=f\"git commit -m {commit_msg[0]} -m {commit_msg[1]} {dry_run_flag}\",\n error_msg=\"Failed to commit changes\",\n timeout_duration=10,\n )\n\n owner, repo, _ = parse_gh_issue_url(issue_url)\n # If `--repo_path` was specified with a different github URL, then the record will contain\n # the forking user\n assert self.record is not None\n if self.record[\"repo_type\"] != \"github\":\n # We already validated that `--data_path` is a github issue URL\n # so this is the only case where we can reach here\n msg = \"--repo_path must point to a github URL if --open_pr is set\"\n raise ValueError(msg)\n forker, _ = self.record[\"repo\"].split(\"/\")\n head = branch_name\n remote = \"origin\"\n if forker != owner:\n head = f\"{forker}:{branch_name}\"\n token_prefix = \"\"\n if self._github_token:\n token_prefix = f\"{self._github_token}@\"\n fork_url = f\"https://{token_prefix}github.com/{forker}/{repo}.git\"\n self.logger.debug(f\"Using fork: {fork_url}\")\n self.communicate_with_handling(\n input=f\"git remote add fork {fork_url}\",\n error_msg=\"Failed to create new git remote\",\n timeout_duration=10,\n )\n remote = \"fork\"\n dry_run_prefix = \"echo \" if _dry_run else \"\"\n self.communicate_with_handling(\n input=f\"{dry_run_prefix} git push {remote} {branch_name}\",\n error_msg=(\n \"Failed to push branch to remote. Please check your token and permissions. \"\n \"You might want to push to a fork with the push_gh_repo_url option.\"\n ),\n timeout_duration=10,\n )\n body = (\n f\"This is a PR opened by AI tool [SWE Agent](https://github.com/princeton-nlp/SWE-agent/) \"\n f\"to close [#{issue.number}]({issue_url}) ({issue.title}).\\n\\nCloses #{issue.number}.\"\n )\n body += \"\\n\\n\" + format_trajectory_markdown(trajectory)\n api = GhApi(token=self._github_token)\n if not _dry_run:\n pr_info = api.pulls.create( # type: ignore\n owner=owner,\n repo=repo,\n title=f\"SWE-agent[bot] PR to fix: {issue.title}\",\n head=head,\n base=\"main\",\n body=body,\n draft=True,\n )\n self.logger.info(\n f\"\ud83c\udf89 PR created as a draft at {pr_info.html_url}. Please review it carefully, push \"\n \"any required changes onto the branch and then click \"\n \"'Ready for Review' to bring it to the attention of the maintainers.\",\n )\n</code></pre>"},{"location":"reference/env/#sweagent.environment.swe_env.SWEEnv.read_file","title":"<code>read_file(path)</code>","text":"<p>Read file contents from container</p> <p>Parameters:</p> Name Type Description Default <code>path</code> <code>str | PurePath</code> <p>Path to file relative to repository root</p> required <p>Returns:</p> Name Type Description <code>file_contents</code> <code>str</code> <p>Contents of file as string</p> Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>def read_file(self, path: str | PurePath) -> str:\n \"\"\"Read file contents from container\n\n Args:\n path: Path to file relative to repository root\n\n Returns:\n file_contents: Contents of file as string\n \"\"\"\n path_in_container = f\"/{self._repo_name}/{path}\"\n return self.communicate(f\"cat {str(path_in_container)}\")\n</code></pre>"},{"location":"reference/env/#sweagent.environment.swe_env.SWEEnv.reset","title":"<code>reset(index=None, apply_test_patch=False)</code>","text":"<p>Function to reset container between each task instance.</p> <ul> <li>Clones instance's repository</li> <li>Cleans repository of prior modifications</li> <li>Resets environment variables</li> <li>Check out base commit</li> </ul> <p>Parameters:</p> Name Type Description Default <code>index</code> <code>int | None</code> <p>index of task instance to reset to</p> <code>None</code> <p>Returns:</p> Name Type Description <code>observation</code> <code>str | None</code> <p>output from container</p> <code>info</code> <code>dict</code> <p>additional information (e.g. debugging information)</p> Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>def reset(self, index: int | None = None, apply_test_patch: bool = False) -> tuple[str | None, dict]:\n \"\"\"\n Function to reset container between each task instance.\n\n * Clones instance's repository\n * Cleans repository of prior modifications\n * Resets environment variables\n * Check out base commit\n\n Args:\n index: index of task instance to reset to\n\n Returns:\n observation: output from container\n info: additional information (e.g. debugging information)\n \"\"\"\n info = {}\n info[\"commit_sha\"] = self.commit_sha\n\n # Get task instance\n self.idx = index if index is not None else self.idx\n self.record = self.data[self.idx]\n self.idx += 1\n\n # Set query, gold command\n self.base_commit = self.record[\"base_commit\"]\n self.query = self.record[\"problem_statement\"]\n self.challenge = self.record.get(\"challenge\")\n self.reward = None\n\n ### Reset Container ###\n self._init_docker_compose()\n\n if self.args.cache_task_images:\n cached_image = self._get_cached_task_image_name()\n if image_exists(cached_image):\n self.logger.info(f\"Restore environment from cached image {cached_image}\")\n self.close() # stop current container\n self._init_container(cached_image=cached_image)\n self.communicate(\"export $(xargs </.env)\")\n envs = self.communicate(\"env\")\n self.logger.debug(f\"Environment variables restored from the image:\\n{envs}\\n\")\n if apply_test_patch:\n self._apply_test_patch()\n return None, info\n else:\n self.logger.info(f\"Cached image {cached_image} not found, rebuilding task environment...\")\n\n # Init docker network\n self._init_docker_network()\n\n # Clone repository if not already cloned\n self.communicate(input=\"cd /\")\n folders = self.communicate(input=\"ls\").split(\"\\n\")\n if self._repo_name not in folders:\n self._copy_repo()\n\n self._reset_repository()\n self._reset_environment_variables()\n\n # Set up environment\n self.communicate_with_handling(\n \"source /root/miniconda3/etc/profile.d/conda.sh\",\n error_msg=\"Failed to source conda\",\n )\n\n system = self.communicate(\"uname -s\").strip().lower()\n arch = self.communicate(\"uname -m\").strip().lower()\n if system == \"linux\" and arch == \"x86_64\":\n self.communicate_with_handling(\n \"apt update; apt install build-essential -y\",\n error_msg=\"Failed to install build-essential\",\n timeout_duration=LONG_TIMEOUT,\n )\n\n # Call install environment helper function if specified\n if self.install_environment:\n self.install_env()\n # Install mypy for linting purposes\n self.communicate_with_handling(\"pip install flake8\", error_msg=\"Failed to install flake8 (lint library)\")\n\n if self.args.cache_task_images:\n envs = self.communicate(\"env\")\n self.logger.debug(f\"Environment variables to save:\\n{envs}\\n\")\n self.communicate(\"env >> /.env\")\n assert self.container_obj is not None # mypy\n self.container_obj.commit(cached_image)\n self.logger.info(f\"Container with environment {self.container_obj.id} cached as image {cached_image}\")\n\n if apply_test_patch:\n self._apply_test_patch()\n # Write any metadata to info if necessary\n return None, info\n</code></pre>"},{"location":"reference/env/#sweagent.environment.swe_env.SWEEnv.reset_for_new_attempt","title":"<code>reset_for_new_attempt()</code>","text":"<p>Compared to <code>reset</code>, which prepares the container for a new instance, this prepares the container for taking another shot at the same instance.</p> Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>def reset_for_new_attempt(\n self,\n) -> None:\n \"\"\"Compared to `reset`, which prepares the container for a new instance,\n this prepares the container for taking another shot at the same instance.\n \"\"\"\n self._reset_repository()\n self._reset_environment_variables()\n</code></pre>"},{"location":"reference/env/#sweagent.environment.swe_env.SWEEnv.run_shell_script","title":"<code>run_shell_script(script_path, *, location)</code>","text":"<p>Run custom script supplied by user at <code>script_path</code></p> <p>Parameters:</p> Name Type Description Default <code>script_path</code> <code>Path</code> <p>path to script file</p> required <code>location</code> <code>str</code> <p>location of script file 'host' or 'container'</p> required Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>def run_shell_script(self, script_path: Path, *, location: str) -> None:\n \"\"\"Run custom script supplied by user at `script_path`\n\n Args:\n script_path: path to script file\n location: location of script file 'host' or 'container'\n \"\"\"\n if location == \"host\":\n return self._run_shell_script_host(script_path)\n elif location == \"container\":\n raise NotImplementedError\n msg = f\"Invalid 'location': {location}\"\n raise ValueError(msg)\n</code></pre>"},{"location":"reference/env/#sweagent.environment.swe_env.SWEEnv.step","title":"<code>step(action)</code>","text":"<p>Runs an action proposed by the agent in the environment and returns the corresponding output.</p> <p>Parameters:</p> Name Type Description Default <code>action</code> <code>str</code> <p>command to run in bash shell</p> required <p>Returns:</p> Name Type Description <code>observation</code> <code>str | None</code> <p>output from container</p> <code>reward</code> <code>int</code> <p>Always set to 0</p> <code>done</code> <code>bool</code> <p>whether task is over</p> <code>info</code> <code>AgentInfo</code> <p>additional information (e.g. debugging information)</p> Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>def step(self, action: str) -> tuple[str | None, int, bool, AgentInfo]:\n \"\"\"\n Runs an action proposed by the agent in the environment and returns the corresponding output.\n\n Args:\n action: command to run in bash shell\n\n Returns:\n observation: output from container\n reward: Always set to 0\n done: whether task is over\n info: additional information (e.g. debugging information)\n \"\"\"\n info: AgentInfo = {}\n # Make sure to have the right keys even if the submission is missing/empty\n info.update(self._get_edited_files_with_context(patch=\"\")) # type: ignore\n\n observation = \"\"\n # Handle special actions\n action = action.strip()\n if action == \"skip\":\n observation = \"Skipped\"\n info[\"exit_status\"] = \"skipped\"\n return observation, 0, True, info\n if action == \"exit_forfeit\":\n observation = \"Exited\"\n info[\"exit_status\"] = action\n return observation, 0, True, info\n if action in {\"exit_context\", \"exit_cost\", \"exit_error\", \"exit_format\", \"exit_api\"}:\n try:\n observation = self.communicate(input=\"submit\")\n submission = self.get_submission(observation)\n assert submission is not None and submission.strip() != \"\", AssertionError(\"No submission found.\")\n self.logger.info(f\"Found submission: {submission}\")\n info[\"exit_status\"] = f\"submitted ({action})\"\n info[\"submission\"] = submission\n info.update(self._get_edited_files_with_context(patch=submission)) # type: ignore\n observation = \"Exited (autosubmitted)\"\n self.logger.info(\"Exiting with autosubmission\")\n return observation, 0, True, info\n except KeyboardInterrupt:\n raise\n except:\n observation = \"Exited\"\n info[\"exit_status\"] = action\n return observation, 0, True, info\n\n # Attempt to run action in container\n observation = \"\"\n try:\n observation = self.communicate(\n input=action,\n timeout_duration=AGENT_ACTION_TIMEOUT,\n no_output_timeout_duration=AGENT_ACTION_NO_OUTPUT_TIMEOUT,\n set_last_action=True,\n )\n except TimeoutError as e:\n try:\n observation += e.args[1] if len(e.args) > 1 else \"\"\n observation += self.interrupt()\n observation += \"\\nEXECUTION TIMED OUT\"\n observation += (\n f\" BECAUSE NO OUTPUT WAS PRODUCED FOR MORE THAN {AGENT_ACTION_NO_OUTPUT_TIMEOUT} SECONDS.\\nPLEASE REFINE YOUR RUNNING COMMAND SO IT WILL PRODUCE OUTPUT IN THE SPECIFIED TIME FRAME.\"\n if isinstance(e, NoOutputTimeoutError)\n else f\" BECAUSE THE COMMAND WAS RUNNING FOR MORE THAN {AGENT_ACTION_TIMEOUT} SECONDS.\"\n )\n except RuntimeError as e:\n observation += e.args[1] if len(e.args) > 1 else \"\"\n observation += \"\\nEXECUTION TIMED OUT AND INTERRUPT FAILED. RESTARTING PROCESS.\"\n info[\"exit_status\"] = \"early_exit\"\n self.logger.warning(f\"Failed to interrupt container: {e}\\nRESTARTING PROCESS.\")\n self.reset_container()\n return observation, 0, True, info\n except RuntimeError as e:\n observation += e.args[1] if len(e.args) > 1 else \"\"\n observation += \"\\nCOMMAND FAILED TO EXECUTE. RESTARTING PROCESS.\"\n info[\"exit_status\"] = \"early_exit\"\n self.logger.warning(f\"Failed to execute command: {e}\\nRESTARTING PROCESS.\")\n self.reset_container()\n return observation, 0, True, info\n except BrokenPipeError as e:\n observation += \"\\nBROKEN PIPE ERROR. RESTARTING PROCESS.\"\n info[\"exit_status\"] = \"early_exit\"\n self.logger.error(f\"Broken pipe error: {e}\\nRESTARTING PROCESS.\")\n self.reset_container()\n return observation, 0, True, info\n except UnicodeError as e:\n observation += \"\\nCOMMAND PRODUCED TOO MANY NON-UNICODE CHARACTERS. PLEASE TRY ANOTHER COMMAND.\\nIF YOU WANT TO VIEW BINARY FILES, PLEASE USE `xxd` OR `hexdump` INSTEAD.\\n\"\n self.logger.error(f\"Unicode error: {e}\")\n except Exception:\n observation += \"\\nEXECUTION FAILED OR COMMAND MALFORMED\"\n self.logger.exception(\"Unknown exception\")\n\n # Record submission and end episode if `submit` keyword found\n submission = self.get_submission(observation)\n if submission is not None:\n if self.validate_submission(submission):\n self.logger.info(f\"Found submission: {submission}\")\n info[\"exit_status\"] = \"submitted\"\n info[\"submission\"] = submission if submission.strip() != \"\" else None\n info.update(self._get_edited_files_with_context(patch=submission)) # type: ignore\n observation = submission if submission.strip() != \"\" else None\n return observation, 0, True, info\n else:\n # Currently only validating CTF challenges\n assert self.challenge is not None\n self.logger.warning(f\"Wrong submission found: {submission} (real flag is {self.challenge['flag']})\")\n observation = \"Wrong flag!\"\n return observation, 0, False, info\n\n observation = self._handle_interactive_commands(observation)\n\n return observation, 0, False, info\n</code></pre>"},{"location":"reference/env/#sweagent.environment.swe_env.SWEEnv.validate_submission","title":"<code>validate_submission(submission)</code>","text":"<p>Function for validating submission for CTF challenges.</p> <p>Parameters:</p> Name Type Description Default <code>submission</code> <code>str</code> <p>extracted submission</p> required <p>Returns:</p> Type Description <code>bool</code> <p>validation of the submission found against known flag</p> Source code in <code>sweagent/environment/swe_env.py</code> <pre><code>def validate_submission(self, submission: str) -> bool:\n \"\"\"\n Function for validating submission for CTF challenges.\n\n Args:\n submission: extracted submission\n\n Returns:\n validation of the submission found against known flag\n \"\"\"\n submission = submission.strip()\n\n def wrap(s: str) -> str:\n return f\"flag{{{s}}}\"\n\n if self.challenge is not None:\n assert \"flag\" in self.challenge\n solution = self.challenge[\"flag\"]\n return any(\n (\n submission == solution,\n wrap(submission) == solution,\n submission == wrap(solution),\n )\n )\n\n return True\n</code></pre>"},{"location":"reference/env_utils/","title":"Environment utils","text":""},{"location":"reference/env_utils/#sweagent.environment.utils.InstanceBuilder","title":"<code>InstanceBuilder</code>","text":"Source code in <code>sweagent/environment/utils.py</code> <pre><code>class InstanceBuilder:\n def __init__(self, token: str | None = None):\n \"\"\"This helper class is used to build the data for an instance object,\n retrieving problem statements from github issues or local files and setting\n repo paths from github urls or local paths.\n \"\"\"\n # Args that will be passed to the Instance constructor\n self.args = {}\n self.token = token\n self._instance_id_problem_suffix = \"\"\n\n def set_problem_statement_from_gh_issue(self, issue_url: str):\n owner, repo, issue_number = parse_gh_issue_url(issue_url)\n self.args[\"problem_statement\"] = get_problem_statement_from_github_issue(\n owner,\n repo,\n issue_number,\n token=self.token,\n )\n self.args[\"instance_id\"] = f\"{owner}__{repo}-i{issue_number}\"\n self.args[\"problem_statement_source\"] = \"online\"\n\n def set_server_description(self, server_name: str | None, port: int | None) -> None:\n \"\"\"For CTF challenges\"\"\"\n if server_name is None or port is None:\n self.args[\"challenge\"][\"server_description\"] = \"\"\n return\n if self.args[\"challenge\"][\"category\"] in {\"web\", \"misc\"} and self.args[\"challenge\"].get(\"proto\") != \"nc\":\n self.args[\"challenge\"][\"server_description\"] = (\n f\"The challenge web server is running on `{server_name}` port `{port}` and you can access it from within the container environment using `curl http://{server_name}:{port}`.\"\n )\n else:\n self.args[\"challenge\"][\"server_description\"] = (\n f\"The challenge web server is running on `{server_name}` port `{port}` and you can access it from within the container environment using `connect_start {server_name} {port}`.\"\n )\n\n def set_problem_statement_from_challenge_json(self, file_path: str) -> None:\n \"\"\"For CTF challenges\"\"\"\n challenge = json.loads(Path(file_path).read_text())\n self.args[\"challenge\"] = challenge\n self.args[\"challenge\"][\"files\"] = challenge.get(\"files\", [])\n self.args[\"challenge\"][\"points\"] = challenge.get(\"points\", 10)\n self.args[\"challenge\"][\"category_friendly\"] = CTF_CHALLENGES_CATEGORIES.get(challenge[\"category\"])\n if (Path(file_path).parent / \"docker-compose.yml\").is_file():\n logger.debug(f\"Found docker_compose file in {Path(file_path).parent}\")\n self.args[\"challenge\"][\"docker_compose\"] = Path(file_path).parent / \"docker-compose.yml\"\n self.args[\"challenge\"][\"port\"] = challenge.get(\"internal_port\") or challenge.get(\"port\")\n if \"box\" in challenge:\n self.args[\"challenge\"][\"server_name\"] = challenge[\"box\"] or \"127.0.0.1\"\n else:\n self.args[\"challenge\"][\"server_name\"] = \"\"\n self.args[\"challenge\"][\"file_path\"] = file_path\n self.set_server_description(self.args[\"challenge\"][\"server_name\"], self.args[\"challenge\"][\"port\"])\n self.set_problem_statement_from_text(f\"{challenge['name']} {challenge['description']}\")\n self.args[\"instance_id\"] = (\n # sanitize 'name' to only alphanumeric characters\n challenge.get(\"category\", \"misc\") + \"_\" + \"\".join(a for a in self.args[\"challenge\"][\"name\"] if a.isalnum())\n )\n\n def set_problem_statement_from_file(self, file_path: str):\n if Path(file_path).name == \"challenge.json\":\n self.set_problem_statement_from_challenge_json(file_path)\n else:\n self.set_problem_statement_from_text(Path(file_path).read_text())\n\n def set_problem_statement_from_text(self, text: str):\n self.args[\"problem_statement\"] = text\n self.args[\"instance_id\"] = hashlib.sha256(self.args[\"problem_statement\"].encode()).hexdigest()[:6]\n self.args[\"problem_statement_source\"] = \"local\"\n\n def set_problem_statement(self, data_path: str):\n \"\"\"Get problem statement for a single instance from a github issue url or a\n path to a markdown or text file.\n \"\"\"\n if data_path.startswith(\"text://\"):\n return self.set_problem_statement_from_text(data_path.removeprefix(\"text://\"))\n if is_github_issue_url(data_path):\n return self.set_problem_statement_from_gh_issue(data_path)\n if Path(data_path).is_file():\n return self.set_problem_statement_from_file(data_path)\n msg = f\"Not sure how to get problem statement from {data_path=}.\"\n raise ValueError(msg)\n\n def set_repo_info_from_gh_url(self, url: str, base_commit: str | None = None):\n owner, repo = parse_gh_repo_url(url)\n self.args[\"repo\"] = f\"{owner}/{repo}\"\n self.args[\"repo_type\"] = \"github\"\n # Always get commit hash, because base_commit can also be branch or tag\n api = GhApi(token=self.token)\n self.args[\"base_commit\"] = get_commit(api, owner, repo, ref=base_commit).sha\n if base_commit != self.args[\"base_commit\"]:\n logger.info(f\"Base commit reference {base_commit} resolved to commit hash {self.args['base_commit']}\")\n self.args[\"version\"] = self.args[\"base_commit\"][:7]\n\n def set_repo_info_from_local_path(self, path: str, base_commit: str | None = None):\n self.args[\"repo\"] = str(Path(path).resolve())\n self.args[\"repo_type\"] = \"local\"\n if base_commit:\n self.args[\"base_commit\"] = base_commit\n else:\n try:\n repo = Repo(path, search_parent_directories=True)\n except InvalidGitRepositoryError as e:\n msg = f\"Could not find git repository at {path=}.\"\n raise ValueError(msg) from e\n if repo.is_dirty() and \"PYTEST_CURRENT_TEST\" not in os.environ:\n msg = f\"Local git repository {path} is dirty. Please commit or stash changes.\"\n raise ValueError(msg)\n self.args[\"base_commit\"] = repo.head.object.hexsha\n self.args[\"version\"] = self.args[\"base_commit\"][:7]\n\n def set_repo_info(self, repo: str, base_commit: str | None = None):\n if is_github_repo_url(repo):\n self.set_repo_info_from_gh_url(repo, base_commit=base_commit)\n elif Path(repo).is_dir():\n self.set_repo_info_from_local_path(repo, base_commit=base_commit)\n else:\n msg = f\"Could not determine repo path from {repo=}.\"\n raise ValueError(msg)\n\n def set_from_dict(self, instance_dict: dict[str, Any]):\n self.args |= instance_dict\n\n def set_missing_fields(self):\n # TODO: This field is only needed while swe_env is using some questionable logic\n # to determine whether to clone from a mirror or not. This should be removed in the future.\n # Values: 'swe-bench' (loaded from json/jsonl for swe-bench style inference),\n # 'online' (loaded from github issue or similar) or 'local' (loaded from local file)\n if \"problem_statement_source\" not in self.args:\n self.args[\"problem_statement_source\"] = \"swe-bench\"\n if \"repo_type\" not in self.args:\n self.args[\"repo_type\"] = \"github\"\n\n def validate(self):\n required_fields = [\n \"problem_statement\",\n \"instance_id\",\n \"repo\",\n \"repo_type\",\n \"base_commit\",\n \"version\",\n \"problem_statement_source\",\n ]\n if not all(x in self.args for x in required_fields):\n missing = set(required_fields) - set(self.args.keys())\n msg = f\"Missing required fields: {missing=}\"\n raise ValueError(msg)\n if self.args[\"repo_type\"] not in {\"github\", \"local\"}:\n msg = f\"Invalid repo type: {self.args['repo_type']=}\"\n raise ValueError(msg)\n if self.args[\"repo_type\"] == \"github\" and self.args[\"repo\"].count(\"/\") != 1:\n msg = f\"Invalid repo format for {self.args['repo_type']=}: {self.args['repo']=}\"\n raise ValueError(msg)\n\n def build(self) -> dict[str, Any]:\n self.set_missing_fields()\n self.validate()\n return self.args\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.InstanceBuilder.__init__","title":"<code>__init__(token=None)</code>","text":"<p>This helper class is used to build the data for an instance object, retrieving problem statements from github issues or local files and setting repo paths from github urls or local paths.</p> Source code in <code>sweagent/environment/utils.py</code> <pre><code>def __init__(self, token: str | None = None):\n \"\"\"This helper class is used to build the data for an instance object,\n retrieving problem statements from github issues or local files and setting\n repo paths from github urls or local paths.\n \"\"\"\n # Args that will be passed to the Instance constructor\n self.args = {}\n self.token = token\n self._instance_id_problem_suffix = \"\"\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.InstanceBuilder.set_problem_statement","title":"<code>set_problem_statement(data_path)</code>","text":"<p>Get problem statement for a single instance from a github issue url or a path to a markdown or text file.</p> Source code in <code>sweagent/environment/utils.py</code> <pre><code>def set_problem_statement(self, data_path: str):\n \"\"\"Get problem statement for a single instance from a github issue url or a\n path to a markdown or text file.\n \"\"\"\n if data_path.startswith(\"text://\"):\n return self.set_problem_statement_from_text(data_path.removeprefix(\"text://\"))\n if is_github_issue_url(data_path):\n return self.set_problem_statement_from_gh_issue(data_path)\n if Path(data_path).is_file():\n return self.set_problem_statement_from_file(data_path)\n msg = f\"Not sure how to get problem statement from {data_path=}.\"\n raise ValueError(msg)\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.InstanceBuilder.set_problem_statement_from_challenge_json","title":"<code>set_problem_statement_from_challenge_json(file_path)</code>","text":"<p>For CTF challenges</p> Source code in <code>sweagent/environment/utils.py</code> <pre><code>def set_problem_statement_from_challenge_json(self, file_path: str) -> None:\n \"\"\"For CTF challenges\"\"\"\n challenge = json.loads(Path(file_path).read_text())\n self.args[\"challenge\"] = challenge\n self.args[\"challenge\"][\"files\"] = challenge.get(\"files\", [])\n self.args[\"challenge\"][\"points\"] = challenge.get(\"points\", 10)\n self.args[\"challenge\"][\"category_friendly\"] = CTF_CHALLENGES_CATEGORIES.get(challenge[\"category\"])\n if (Path(file_path).parent / \"docker-compose.yml\").is_file():\n logger.debug(f\"Found docker_compose file in {Path(file_path).parent}\")\n self.args[\"challenge\"][\"docker_compose\"] = Path(file_path).parent / \"docker-compose.yml\"\n self.args[\"challenge\"][\"port\"] = challenge.get(\"internal_port\") or challenge.get(\"port\")\n if \"box\" in challenge:\n self.args[\"challenge\"][\"server_name\"] = challenge[\"box\"] or \"127.0.0.1\"\n else:\n self.args[\"challenge\"][\"server_name\"] = \"\"\n self.args[\"challenge\"][\"file_path\"] = file_path\n self.set_server_description(self.args[\"challenge\"][\"server_name\"], self.args[\"challenge\"][\"port\"])\n self.set_problem_statement_from_text(f\"{challenge['name']} {challenge['description']}\")\n self.args[\"instance_id\"] = (\n # sanitize 'name' to only alphanumeric characters\n challenge.get(\"category\", \"misc\") + \"_\" + \"\".join(a for a in self.args[\"challenge\"][\"name\"] if a.isalnum())\n )\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.InstanceBuilder.set_server_description","title":"<code>set_server_description(server_name, port)</code>","text":"<p>For CTF challenges</p> Source code in <code>sweagent/environment/utils.py</code> <pre><code>def set_server_description(self, server_name: str | None, port: int | None) -> None:\n \"\"\"For CTF challenges\"\"\"\n if server_name is None or port is None:\n self.args[\"challenge\"][\"server_description\"] = \"\"\n return\n if self.args[\"challenge\"][\"category\"] in {\"web\", \"misc\"} and self.args[\"challenge\"].get(\"proto\") != \"nc\":\n self.args[\"challenge\"][\"server_description\"] = (\n f\"The challenge web server is running on `{server_name}` port `{port}` and you can access it from within the container environment using `curl http://{server_name}:{port}`.\"\n )\n else:\n self.args[\"challenge\"][\"server_description\"] = (\n f\"The challenge web server is running on `{server_name}` port `{port}` and you can access it from within the container environment using `connect_start {server_name} {port}`.\"\n )\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.PatchFormatter","title":"<code>PatchFormatter</code>","text":"Source code in <code>sweagent/environment/utils.py</code> <pre><code>class PatchFormatter:\n def __init__(\n self,\n patch: str,\n read_method: Callable[[str], str],\n ):\n \"\"\"Given the final patch and access to the container that contains the repository,\n extract relevant lines from the modified file.\n\n Args:\n patch: The patch as a string.\n read_method: Callable with path to file (relative to repository root) as argument\n that returns the file content as a string.\n \"\"\"\n self._patch = PatchSet(patch)\n self._patched_files: dict[str, str] = {}\n self._original_files: dict[str, str] = {}\n self._patch_applied = True\n self._read_file = read_method\n self._read_files(original=False)\n\n @staticmethod\n def _merge_intervals(starts: list[int], stops: list[int]) -> tuple[list[int], list[int]]:\n \"\"\"Given two lists of integers, starts and stops, merges all overlapping intervals.\n\n For example `starts=[1, 5, 18]`, `stops=[10, 13, 20]`\n should return `starts=[1, 18]`, `stops=[13, 20]`\n \"\"\"\n\n intervals = sorted(zip(starts, stops))\n merged = []\n for start, stop in intervals:\n if not merged or merged[-1][1] < start:\n # No overlap\n merged.append([start, stop])\n else:\n # Overlap\n merged[-1][1] = max(merged[-1][1], stop)\n # Unzip again\n merged_starts, merged_stops = zip(*merged)\n return list(merged_starts), list(merged_stops)\n\n def format_file(self, text: str, starts: list[int], stops: list[int], *, linenos: bool = True) -> str:\n \"\"\"Reads file and returns string representation of the relevant lines.\n\n Args:\n path: The path to the file within the repo location\n starts: The starting line numbers of the relevant lines. The first line is line 1.\n stops: The stopping line numbers of the relevant lines. The stop is not inclusive.\n The first line is line 1.\n linenos: Whether to include line numbers\n \"\"\"\n assert len(starts) == len(stops)\n assert all(start >= 1 for start in starts)\n assert all(start < stop for start, stop in zip(starts, stops))\n starts, stops = self._merge_intervals(starts, stops)\n assert all(hunk1_start < hunk2_start for hunk1_start, hunk2_start in zip(starts, starts[1:]))\n out: list[str] = []\n if starts[0] > 1:\n # Count from 1\n out.append(f\"[{starts[0]-1} lines above omitted]\")\n last_stop: int | None = None\n lines = text.splitlines()\n for start, stop in zip(starts, stops):\n assert start >= 1\n if last_stop is not None:\n n_omitted = start - last_stop\n # Check that we have non-overlapping hunks\n assert n_omitted >= 0\n if n_omitted:\n out.append(f\"\\n[{n_omitted} lines omitted]\\n\")\n # Count from 1\n these_lines = lines[start - 1 : stop - 1]\n if linenos:\n out.append(\"\\n\".join([f\"{i:6d}: {l}\" for i, l in enumerate(these_lines, start=start)]))\n else:\n out.append(\"\\n\".join(these_lines))\n last_stop = stop\n if last_stop < len(lines):\n # Stop is not inclusive\n omitted = len(lines) - last_stop\n assert omitted > 0\n out.append(f\"[{omitted} lines below omitted]\")\n return \"\\n\".join(out)\n\n def _get_hunk_lines(self, original: bool, *, context_length: int) -> dict[str, tuple[list[int], list[int]]]:\n \"\"\"Get the starts and stops for all files in the patch.\n\n Args:\n original: Whether to read the original file or the patched file\n context_length: The number of lines to include above and below the hunk\n\n Returns:\n A dictionary with the file path as key and a tuple of lists of starts and stops as value.\n \"\"\"\n out: dict[str, tuple[list[int], list[int]]] = {}\n for patch in self._patch:\n if not patch.is_modified_file:\n continue\n starts: list[int] = []\n stops: list[int] = []\n for hunk in patch:\n if original:\n # 1 is the lowest line number\n start = max(1, hunk.source_start - context_length)\n stop = hunk.source_start + hunk.source_length + context_length\n else:\n start = max(1, hunk.target_start - context_length)\n stop = hunk.target_start + hunk.target_length + context_length\n starts.append(start)\n stops.append(stop)\n out[patch.path] = (starts, stops)\n return out\n\n def _read_files(self, original: bool) -> None:\n for patch in self._patch:\n path = patch.path\n if not patch.is_modified_file:\n continue\n if original:\n msg = \"Original file reading not implemented\"\n raise NotImplementedError(msg)\n else:\n assert self._patch_applied\n self._patched_files[path] = self._read_file(path)\n\n @staticmethod\n def concat_files_strings(files: dict[str, str]) -> str:\n \"\"\"Concatenate multiple `read_files` outputs into a single string.\"\"\"\n out = []\n for path, content in files.items():\n out.append(f\"[File: {path}]\\n{content}\")\n return \"\\n\\n\".join(out)\n\n def get_files_str(self, *, original: bool, context_length: int | None = 50, linenos: bool = True) -> str:\n hunk_lines = self._get_hunk_lines(original=original, context_length=context_length)\n sources = self._original_files if original else self._patched_files\n return self.concat_files_strings(\n {path: self.format_file(text, *hunk_lines[path], linenos=linenos) for path, text in sources.items()}\n )\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.PatchFormatter.__init__","title":"<code>__init__(patch, read_method)</code>","text":"<p>Given the final patch and access to the container that contains the repository, extract relevant lines from the modified file.</p> <p>Parameters:</p> Name Type Description Default <code>patch</code> <code>str</code> <p>The patch as a string.</p> required <code>read_method</code> <code>Callable[[str], str]</code> <p>Callable with path to file (relative to repository root) as argument that returns the file content as a string.</p> required Source code in <code>sweagent/environment/utils.py</code> <pre><code>def __init__(\n self,\n patch: str,\n read_method: Callable[[str], str],\n):\n \"\"\"Given the final patch and access to the container that contains the repository,\n extract relevant lines from the modified file.\n\n Args:\n patch: The patch as a string.\n read_method: Callable with path to file (relative to repository root) as argument\n that returns the file content as a string.\n \"\"\"\n self._patch = PatchSet(patch)\n self._patched_files: dict[str, str] = {}\n self._original_files: dict[str, str] = {}\n self._patch_applied = True\n self._read_file = read_method\n self._read_files(original=False)\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.PatchFormatter.concat_files_strings","title":"<code>concat_files_strings(files)</code> <code>staticmethod</code>","text":"<p>Concatenate multiple <code>read_files</code> outputs into a single string.</p> Source code in <code>sweagent/environment/utils.py</code> <pre><code>@staticmethod\ndef concat_files_strings(files: dict[str, str]) -> str:\n \"\"\"Concatenate multiple `read_files` outputs into a single string.\"\"\"\n out = []\n for path, content in files.items():\n out.append(f\"[File: {path}]\\n{content}\")\n return \"\\n\\n\".join(out)\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.PatchFormatter.format_file","title":"<code>format_file(text, starts, stops, *, linenos=True)</code>","text":"<p>Reads file and returns string representation of the relevant lines.</p> <p>Parameters:</p> Name Type Description Default <code>path</code> <p>The path to the file within the repo location</p> required <code>starts</code> <code>list[int]</code> <p>The starting line numbers of the relevant lines. The first line is line 1.</p> required <code>stops</code> <code>list[int]</code> <p>The stopping line numbers of the relevant lines. The stop is not inclusive. The first line is line 1.</p> required <code>linenos</code> <code>bool</code> <p>Whether to include line numbers</p> <code>True</code> Source code in <code>sweagent/environment/utils.py</code> <pre><code>def format_file(self, text: str, starts: list[int], stops: list[int], *, linenos: bool = True) -> str:\n \"\"\"Reads file and returns string representation of the relevant lines.\n\n Args:\n path: The path to the file within the repo location\n starts: The starting line numbers of the relevant lines. The first line is line 1.\n stops: The stopping line numbers of the relevant lines. The stop is not inclusive.\n The first line is line 1.\n linenos: Whether to include line numbers\n \"\"\"\n assert len(starts) == len(stops)\n assert all(start >= 1 for start in starts)\n assert all(start < stop for start, stop in zip(starts, stops))\n starts, stops = self._merge_intervals(starts, stops)\n assert all(hunk1_start < hunk2_start for hunk1_start, hunk2_start in zip(starts, starts[1:]))\n out: list[str] = []\n if starts[0] > 1:\n # Count from 1\n out.append(f\"[{starts[0]-1} lines above omitted]\")\n last_stop: int | None = None\n lines = text.splitlines()\n for start, stop in zip(starts, stops):\n assert start >= 1\n if last_stop is not None:\n n_omitted = start - last_stop\n # Check that we have non-overlapping hunks\n assert n_omitted >= 0\n if n_omitted:\n out.append(f\"\\n[{n_omitted} lines omitted]\\n\")\n # Count from 1\n these_lines = lines[start - 1 : stop - 1]\n if linenos:\n out.append(\"\\n\".join([f\"{i:6d}: {l}\" for i, l in enumerate(these_lines, start=start)]))\n else:\n out.append(\"\\n\".join(these_lines))\n last_stop = stop\n if last_stop < len(lines):\n # Stop is not inclusive\n omitted = len(lines) - last_stop\n assert omitted > 0\n out.append(f\"[{omitted} lines below omitted]\")\n return \"\\n\".join(out)\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.copy_anything_to_container","title":"<code>copy_anything_to_container(container, host_path, container_path)</code>","text":"<p>Copy files or directories from host to container</p> <p>Note: Will need to set ownership on the copied files in the container.</p> Source code in <code>sweagent/environment/utils.py</code> <pre><code>def copy_anything_to_container(container: Container, host_path: str, container_path: str) -> None:\n \"\"\"Copy files or directories from host to container\n\n Note: Will need to set ownership on the copied files in the container.\n \"\"\"\n if not Path(host_path).exists():\n msg = f\"Path {host_path} does not exist, cannot copy it to container.\"\n raise FileNotFoundError(msg)\n cmd = [\"docker\", \"cp\", host_path, f\"{container.id}:{container_path}\"]\n logger.debug(f\"Copying {host_path} to container at {container_path} with command: {shlex.join(cmd)}\")\n try:\n subprocess.run(cmd, check=True)\n except subprocess.CalledProcessError as e:\n msg = f\"Error copying {host_path} to container at {container_path}: {e}\"\n raise RuntimeError(msg) from e\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.copy_file_to_container","title":"<code>copy_file_to_container(container, contents, container_path)</code>","text":"<p>Copies a given string into a Docker container at a specified path.</p> <p>Parameters:</p> Name Type Description Default <code>container</code> <code>Container</code> <p>Docker SDK container object.</p> required <code>contents</code> <code>str</code> <p>The string to copy into the container.</p> required <code>container_path</code> <code>str</code> <p>The path inside the container where the string should be copied to.</p> required <p>Returns:</p> Type Description <code>None</code> <p>None</p> Source code in <code>sweagent/environment/utils.py</code> <pre><code>def copy_file_to_container(container: Container, contents: str, container_path: str) -> None:\n \"\"\"\n Copies a given string into a Docker container at a specified path.\n\n Args:\n container: Docker SDK container object.\n contents: The string to copy into the container.\n container_path: The path inside the container where the string should be copied to.\n\n Returns:\n None\n \"\"\"\n temp_file_name = None\n\n try:\n # Create a temporary file\n with tempfile.NamedTemporaryFile(delete=False) as temp_file:\n temp_file_name = temp_file.name\n # Write the string to the temporary file and ensure it's written to disk\n temp_file.write(contents.encode(\"utf-8\"))\n temp_file.flush()\n os.fsync(temp_file.fileno())\n\n # Create a TAR archive in memory containing the temporary file\n with tempfile.NamedTemporaryFile():\n with open(temp_file_name, \"rb\") as temp_file:\n # Prepare the TAR archive\n with BytesIO() as tar_stream:\n with tarfile.open(fileobj=tar_stream, mode=\"w\") as tar:\n tar_info = tarfile.TarInfo(name=Path(container_path).name)\n tar_info.size = Path(temp_file_name).stat().st_size\n tar.addfile(tarinfo=tar_info, fileobj=temp_file)\n tar_stream.seek(0)\n # Copy the TAR stream to the container\n container.put_archive(path=Path(container_path).parent, data=tar_stream.read())\n\n except Exception as e:\n logger.error(f\"An error occurred: {e}\")\n logger.error(traceback.format_exc())\n finally:\n # Cleanup: Remove the temporary file if it was created\n if temp_file_name and Path(temp_file_name).exists():\n os.remove(temp_file_name)\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.format_trajectory_markdown","title":"<code>format_trajectory_markdown(trajectory)</code>","text":"<p>Format a trajectory as a markdown string for use in gh PR description.</p> Source code in <code>sweagent/environment/utils.py</code> <pre><code>def format_trajectory_markdown(trajectory: list[dict[str, str]]):\n \"\"\"Format a trajectory as a markdown string for use in gh PR description.\"\"\"\n prefix = [\n \"<details>\",\n \"<summary>Thought process ('trajectory') of SWE-agent (click to expand)</summary>\",\n \"\",\n \"\",\n ]\n steps = []\n for i, step in enumerate(trajectory):\n step_strs = [\n f\"**\ud83e\uddd1\u200d\ud83d\ude92 Response ({i})**: \",\n f\"{step['response'].strip()}\",\n f\"**\ud83d\udc40\u200d Observation ({i})**:\",\n \"```\",\n f\"{remove_triple_backticks(step['observation']).strip()}\",\n \"```\",\n ]\n steps.append(\"\\n\".join(step_strs))\n suffix = [\n \"\",\n \"</details>\",\n ]\n return \"\\n\".join(prefix) + \"\\n\\n---\\n\\n\".join(steps) + \"\\n\".join(suffix)\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.get_associated_commit_urls","title":"<code>get_associated_commit_urls(org, repo, issue_number, *, token='')</code>","text":"<p>Return the URLs of commits that would close an issue.</p> Source code in <code>sweagent/environment/utils.py</code> <pre><code>def get_associated_commit_urls(org: str, repo: str, issue_number: str, *, token: str = \"\") -> list[str]:\n \"\"\"Return the URLs of commits that would close an issue.\"\"\"\n api = GhApi(token=token)\n # Strangely the \"pull_request\" field of api.issues.get is often not set\n # so we have to go through the events to check if there's a commit\n events = api.issues.list_events(org, repo, issue_number)\n commit_urls = []\n for event in events:\n if event.event != \"referenced\":\n continue\n if not event.commit_id:\n continue\n commit = api.repos.get_commit(org, repo, event.commit_id)\n message = commit.commit.message\n if f\"fixes #{issue_number}\" in message.lower() or f\"closes #{issue_number}\" in message.lower():\n commit_urls.append(commit.html_url)\n return commit_urls\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.get_commit","title":"<code>get_commit(api, owner, repo, ref=None)</code>","text":"<p>Get commit object from github api</p> <p>Parameters:</p> Name Type Description Default <code>api</code> <code>GhApi</code> required <code>owner</code> <code>str</code> <p>Repo owner, e.g., \"princeton-nlp\"</p> required <code>repo</code> <code>str</code> <p>Repo, e.g., \"SWE-agent\"</p> required <code>ref</code> <code>str</code> <p>Branch, tag or commit hash</p> <code>None</code> <p>Returns:</p> Name Type Description <code>_type_</code> <p>description</p> Source code in <code>sweagent/environment/utils.py</code> <pre><code>def get_commit(api: GhApi, owner: str, repo: str, ref: str | None = None):\n \"\"\"Get commit object from github api\n\n Args:\n api (GhApi):\n owner (str): Repo owner, e.g., \"princeton-nlp\"\n repo (str): Repo, e.g., \"SWE-agent\"\n ref (str, optional): Branch, tag or commit hash\n\n Returns:\n _type_: _description_\n \"\"\"\n if ref:\n return api.repos.get_commit(owner, repo, ref)\n return api.repos.list_commits(owner, repo)[0]\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.get_container","title":"<code>get_container(ctr_name, image_name, container_mounts, persistent=False)</code>","text":"<p>Get a container object for a given container name and image name</p> <p>Parameters:</p> Name Type Description Default <code>ctr_name</code> <code>str</code> <p>Name of container</p> required <code>image_name</code> <code>str</code> <p>Name of image</p> required <code>persistent</code> <code>bool</code> <p>Whether to use a persistent container or not</p> <code>False</code> <p>Returns: Container object</p> Source code in <code>sweagent/environment/utils.py</code> <pre><code>def get_container(\n ctr_name: str, image_name: str, container_mounts: list[str], persistent: bool = False\n) -> tuple[subprocess.Popen, set]:\n \"\"\"\n Get a container object for a given container name and image name\n\n Arguments:\n ctr_name (str): Name of container\n image_name (str): Name of image\n persistent (bool): Whether to use a persistent container or not\n Returns:\n Container object\n \"\"\"\n if not image_exists(image_name):\n msg = (\n f\"Image {image_name} not found. Please ensure it is built and available. \"\n \"Please double-check that you followed all installation/setup instructions from the \"\n \"readme.\"\n )\n raise RuntimeError(msg)\n\n if persistent:\n return _get_persistent_container(ctr_name, image_name, container_mounts=container_mounts)\n else:\n return _get_non_persistent_container(ctr_name, image_name, container_mounts=container_mounts)\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.get_data_path_name","title":"<code>get_data_path_name(data_path)</code>","text":"<p>if data_path is a file, return the file stem elif it's a github url, return the owner__repo_name</p> Source code in <code>sweagent/environment/utils.py</code> <pre><code>def get_data_path_name(data_path: str) -> str:\n \"\"\"if data_path is a file, return the file stem\n elif it's a github url, return the owner__repo_name\n \"\"\"\n if data_path.startswith(\"text://\"):\n return hashlib.sha256(data_path.removeprefix(\"text://\").encode()).hexdigest()[:6]\n match = GITHUB_ISSUE_URL_PATTERN.search(data_path)\n if match:\n owner, repo, _ = match.groups()\n return f\"{owner}__{repo}\"\n return Path(data_path).stem\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.get_gh_issue_data","title":"<code>get_gh_issue_data(issue_url, *, token='')</code>","text":"<p>Returns github issue data in the form of a dictionary. See https://docs.github.com/en/rest/issues/issues?apiVersion=2022-11-28#get-an-issue for return format</p> Source code in <code>sweagent/environment/utils.py</code> <pre><code>def get_gh_issue_data(issue_url: str, *, token: str = \"\"):\n \"\"\"Returns github issue data in the form of a dictionary.\n See https://docs.github.com/en/rest/issues/issues?apiVersion=2022-11-28#get-an-issue\n for return format\n \"\"\"\n owner, repo, issue_number = parse_gh_issue_url(issue_url)\n api = GhApi(token=token)\n return api.issues.get(owner, repo, issue_number)\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.get_instances","title":"<code>get_instances(file_path, base_commit=None, split=None, token=None, *, repo_path='')</code>","text":"<p>Getter function for handling json, jsonl files</p> <p>Parameters:</p> Name Type Description Default <code>file_path</code> <code>str</code> <p>Path to file</p> required <p>Returns:</p> Type Description <code>list[dict[str, Any]]</code> <p>List of instances as dictionaries</p> Source code in <code>sweagent/environment/utils.py</code> <pre><code>def get_instances(\n file_path: str,\n base_commit: str | None = None,\n split: str | None = None,\n token: str | None = None,\n *,\n repo_path: str = \"\",\n) -> list[dict[str, Any]]:\n \"\"\"\n Getter function for handling json, jsonl files\n\n Args:\n file_path (str): Path to file\n\n Returns:\n List of instances as dictionaries\n \"\"\"\n\n def instance_from_dict(instances):\n ib = InstanceBuilder(token=token)\n ib.set_from_dict(instances)\n return ib.build()\n\n def postproc_instance_list(instances):\n if isinstance(instances, dict):\n msg = \"Expected a list of instances, got a dictionary.\"\n raise ValueError(msg)\n return [instance_from_dict(x) for x in instances]\n\n # The next if statement is very brittle logic to determine if we're processing a single instance\n if (\n file_path.startswith(\"text://\")\n or (\n Path(file_path).is_file()\n and (Path(file_path).suffix in [\".md\", \".txt\"] or Path(file_path).name == \"challenge.json\")\n )\n or is_github_issue_url(file_path)\n ):\n ib = InstanceBuilder(token=token)\n ib.set_problem_statement(file_path)\n if repo_path:\n ib.set_repo_info(repo_path, base_commit=base_commit)\n elif is_github_repo_url(file_path):\n ib.set_repo_info_from_gh_url(file_path, base_commit=base_commit)\n else:\n msg = f\"Could not determine repo path from {file_path=}, {repo_path=}\"\n raise ValueError(msg)\n\n return [ib.build()]\n\n if base_commit:\n msg = \"base_commit must be empty if running over multiple problem statements\"\n raise ValueError(msg)\n\n if repo_path:\n if not Path(repo_path).exists():\n msg = f\"Specified repository path {repo_path} does not exist\"\n raise FileNotFoundError(msg)\n msg = \"repo_path must be empty if running over multiple problem statements\"\n raise ValueError(msg)\n\n # If file_path is a directory, attempt load from disk\n if Path(file_path).is_dir():\n try:\n dataset_or_dict = load_from_disk(file_path)\n if isinstance(dataset_or_dict, dict):\n return postproc_instance_list(dataset_or_dict[split])\n return postproc_instance_list(dataset_or_dict)\n except FileNotFoundError:\n # Raised by load_from_disk if the directory is not a dataset directory\n pass\n\n if base_commit is not None:\n msg = \"base_commit must be None if data_path is not a github issue url\"\n raise ValueError(msg)\n\n # If file_path is a file, load the file\n if file_path.endswith(\".json\"):\n with open(file_path) as file:\n return postproc_instance_list(json.load(file))\n if file_path.endswith(\".jsonl\"):\n return postproc_instance_list([json.loads(x) for x in Path(file_path).read_text().splitlines(keepends=True)])\n\n # Attempt load from HF datasets as a last resort\n try:\n return postproc_instance_list(load_dataset(file_path, split=split))\n except Exception as e:\n msg = (\n f\"Could not load instances from {file_path}. \"\n \"Please ensure --data_path is a GitHub URL, a SWE-bench HuggingFace dataset, or a JSON/JSONL file.\"\n )\n raise ValueError(msg) from e\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.get_problem_statement_from_github_issue","title":"<code>get_problem_statement_from_github_issue(owner, repo, issue_number, *, token='')</code>","text":"<p>Return problem statement from github issue</p> Source code in <code>sweagent/environment/utils.py</code> <pre><code>def get_problem_statement_from_github_issue(owner: str, repo: str, issue_number: str, *, token: str | None = \"\") -> str:\n \"\"\"Return problem statement from github issue\"\"\"\n api = GhApi(token=token)\n issue = api.issues.get(owner, repo, issue_number)\n title = issue.title if issue.title else \"\"\n body = issue.body if issue.body else \"\"\n return f\"{title}\\n{body}\\n\"\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.image_exists","title":"<code>image_exists(image_name)</code>","text":"<p>Check that the image exists and give some better error messages.</p> <p>Parameters:</p> Name Type Description Default <code>image_name</code> <code>str</code> <p>Name of image</p> required <p>Returns: bool: True if image exists</p> Source code in <code>sweagent/environment/utils.py</code> <pre><code>def image_exists(image_name: str) -> bool:\n \"\"\"\n Check that the image exists and give some better error messages.\n\n Arguments:\n image_name: Name of image\n Returns:\n bool: True if image exists\n \"\"\"\n try:\n client = docker.from_env()\n except docker.errors.DockerException as e:\n docker_not_running = any(\n (\n \"connection aborted\" in str(e).lower(),\n \"connection refused\" in str(e).lower(),\n \"error while fetching server api version\" in str(e).lower(),\n ),\n )\n if docker_not_running:\n msg = (\n \"Probably the Docker daemon is not running. Please start the Docker daemon and try again. \"\n \"If Docker issues persist, please check out https://princeton-nlp.github.io/SWE-agent/installation/tips/\"\n )\n raise RuntimeError(msg) from e\n raise\n filterred_images = client.images.list(filters={\"reference\": image_name})\n if len(filterred_images) == 0:\n return False\n elif len(filterred_images) > 1:\n RuntimeError(f\"Multiple images found for {image_name}, that's weird.\")\n attrs = filterred_images[0].attrs\n if attrs is not None:\n logger.info(\n f\"Found image {image_name} with tags: {attrs['RepoTags']}, created: {attrs['Created']} \"\n f\"for {attrs['Os']} {attrs['Architecture']}.\",\n )\n return True\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.is_github_issue_url","title":"<code>is_github_issue_url(data_path)</code>","text":"<p>Check if data_path is an URL pointing to a github issue</p> Source code in <code>sweagent/environment/utils.py</code> <pre><code>def is_github_issue_url(data_path: str) -> bool:\n \"\"\"Check if data_path is an URL pointing to a github issue\"\"\"\n return GITHUB_ISSUE_URL_PATTERN.search(data_path) is not None\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.is_github_repo_url","title":"<code>is_github_repo_url(data_path)</code>","text":"<p>Check if data_path is an URL pointing to a github repository. Paths to issues or PRs will also match this pattern.</p> Source code in <code>sweagent/environment/utils.py</code> <pre><code>def is_github_repo_url(data_path: str) -> bool:\n \"\"\"Check if data_path is an URL pointing to a github repository.\n Paths to issues or PRs will also match this pattern.\n \"\"\"\n return GITHUB_REPO_URL_PATTERN.search(data_path) is not None\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.parse_gh_issue_url","title":"<code>parse_gh_issue_url(issue_url)</code>","text":"<p>Returns:</p> Name Type Description <code>owner</code> <code>str</code> <p>Repo owner</p> <code>repo</code> <code>str</code> <p>Repo name</p> <code>str</code> <p>issue number: Issue number as str</p> <p>Raises:</p> Type Description <code>InvalidGithubURL</code> <p>If the URL is not a valid github issue URL</p> Source code in <code>sweagent/environment/utils.py</code> <pre><code>def parse_gh_issue_url(issue_url: str) -> tuple[str, str, str]:\n \"\"\"\n Returns:\n owner: Repo owner\n repo: Repo name\n issue number: Issue number as str\n\n Raises:\n InvalidGithubURL: If the URL is not a valid github issue URL\n \"\"\"\n match = GITHUB_ISSUE_URL_PATTERN.search(issue_url)\n if not match:\n msg = f\"Invalid GitHub issue URL: {issue_url}\"\n raise InvalidGithubURL(msg)\n res = match.groups()\n assert len(res) == 3\n return tuple(res) # type: ignore\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.parse_gh_repo_url","title":"<code>parse_gh_repo_url(repo_url)</code>","text":"<p>Returns:</p> Name Type Description <code>owner</code> <code>str</code> <p>Repo owner/org</p> <code>repo</code> <code>str</code> <p>Repo name</p> <p>Raises:</p> Type Description <code>InvalidGithubURL</code> <p>If the URL is not a valid github repo URL</p> Source code in <code>sweagent/environment/utils.py</code> <pre><code>def parse_gh_repo_url(repo_url: str) -> tuple[str, str]:\n \"\"\"\n Returns:\n owner: Repo owner/org\n repo: Repo name\n\n Raises:\n InvalidGithubURL: If the URL is not a valid github repo URL\n \"\"\"\n match = GITHUB_REPO_URL_PATTERN.search(repo_url)\n if not match:\n msg = f\"Invalid GitHub issue URL: {repo_url}\"\n raise InvalidGithubURL(msg)\n res = match.groups()\n assert len(res) == 2\n return tuple(res) # type: ignore\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.read_session_with_timeout","title":"<code>read_session_with_timeout(session, terminal_pattern, timeout_duration, no_output_timeout_duration)</code>","text":"<p>Read data from a subprocess with a timeout. This function uses a file descriptor to read data from the subprocess in a non-blocking way.</p> <p>Parameters:</p> Name Type Description Default <code>session</code> <code>Popen</code> <p>The session subprocess.</p> required <code>terminal_pattern</code> <code>str</code> <p>the terminal pattern to indicate end of output.</p> required <code>timeout_duration</code> <code>int | float</code> <p>The timeout duration in seconds.</p> required <p>Returns:</p> Type Description <code>str</code> <p>Output</p> <p>Raises:</p> Type Description <code>TimeoutError</code> <p>If the timeout duration is reached while reading from the subprocess.</p> Source code in <code>sweagent/environment/utils.py</code> <pre><code>def read_session_with_timeout(\n session: subprocess.Popen,\n terminal_pattern: str,\n timeout_duration: int | float,\n no_output_timeout_duration: int | float,\n) -> str:\n \"\"\"\n Read data from a subprocess with a timeout.\n This function uses a file descriptor to read data from the subprocess in a non-blocking way.\n\n Args:\n session: The session subprocess.\n terminal_pattern: the terminal pattern to indicate end of output.\n timeout_duration: The timeout duration in seconds.\n\n Returns:\n Output\n\n Raises:\n TimeoutError: If the timeout duration is reached while reading from the subprocess.\n \"\"\"\n buffer = b\"\"\n fd = session.stdout.fileno()\n start_time = time.time()\n end_time = start_time + timeout_duration\n end_time_no_output = start_time + no_output_timeout_duration\n\n # Select is not available on windows\n import select\n\n def ready_to_read(fd) -> bool:\n return bool(select.select([fd], [], [], 0.01)[0])\n\n command_done = False\n while time.time() < min(end_time, end_time_no_output) and session.poll() is None:\n if ready_to_read(fd):\n try:\n data = os.read(fd, 4096)\n except BlockingIOError:\n logger.error(\"BlockingIOError while reading from subprocess.\", exc_info=True)\n break\n if data:\n end_time_no_output = time.time() + no_output_timeout_duration\n buffer += data\n if terminal_pattern in buffer.decode(\"utf-8\", errors=\"backslashreplace\").replace(\"\\r\\n\", \"\\n\"):\n command_done = True\n break\n time.sleep(0.01) # Prevents CPU hogging\n\n decoded = buffer.decode(\"utf-8\", errors=\"backslashreplace\").replace(\"\\r\\n\", \"\\n\")\n body = \"\\n\".join(line for line in decoded.splitlines() if not line.startswith(terminal_pattern))\n\n if session.poll() is not None:\n msg = f\"Subprocess exited unexpectedly.\\nCurrent buffer: {decoded}\"\n raise RuntimeError(msg, body)\n current_time = time.time()\n if not command_done and current_time >= min(end_time, end_time_no_output):\n if current_time >= end_time:\n msg = f\"Timeout reached while reading from subprocess.\\nCurrent buffer: {decoded}\"\n raise TimeoutError(msg, body)\n else:\n msg = f\"No output timeout reached while reading from subprocess.\\nCurrent buffer: {decoded}\"\n raise NoOutputTimeoutError(msg, body)\n\n return body\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.read_with_timeout","title":"<code>read_with_timeout(container, pid_func, timeout_duration)</code>","text":"<p>Read data from a subprocess with a timeout. This function uses a file descriptor to read data from the subprocess in a non-blocking way.</p> <p>Parameters:</p> Name Type Description Default <code>container</code> <code>Popen</code> <p>The subprocess container.</p> required <code>pid_func</code> <code>Callable</code> <p>A function that returns a list of process IDs (except the PID of the main process).</p> required <code>timeout_duration</code> <code>int | float</code> <p>The timeout duration in seconds.</p> required <p>Returns:</p> Name Type Description <code>output</code> <code>str</code> <p>The data read from the subprocess, stripped of trailing newline characters.</p> <p>Raises:</p> Type Description <code>TimeoutError</code> <p>If the timeout duration is reached while reading from the subprocess.</p> Source code in <code>sweagent/environment/utils.py</code> <pre><code>def read_with_timeout(container: subprocess.Popen, pid_func: Callable, timeout_duration: int | float) -> str:\n \"\"\"\n Read data from a subprocess with a timeout.\n This function uses a file descriptor to read data from the subprocess in a non-blocking way.\n\n Args:\n container: The subprocess container.\n pid_func: A function that returns a list of process IDs (except the PID of the main process).\n timeout_duration: The timeout duration in seconds.\n\n Returns:\n output: The data read from the subprocess, stripped of trailing newline characters.\n\n Raises:\n TimeoutError: If the timeout duration is reached while reading from the subprocess.\n \"\"\"\n buffer = b\"\"\n fd = container.stdout.fileno()\n end_time = time.time() + timeout_duration\n\n # Select is not available on windows\n is_windows = platform.system() == \"Windows\"\n if not is_windows:\n import select\n else:\n os.set_blocking(fd, False)\n\n def ready_to_read(fd) -> bool:\n if is_windows:\n # We can't do the extra check\n return True\n return bool(select.select([fd], [], [], 0.01)[0])\n\n while time.time() < end_time:\n pids = pid_func()\n if len(pids) > 0:\n # There are still PIDs running\n time.sleep(0.05)\n continue\n if ready_to_read(fd):\n data = os.read(fd, 4096)\n if data:\n buffer += data\n else:\n # No more data to read\n break\n time.sleep(0.05) # Prevents CPU hogging\n\n if container.poll() is not None:\n msg = f\"Subprocess exited unexpectedly.\\nCurrent buffer: {buffer.decode()}\"\n raise RuntimeError(msg)\n if time.time() >= end_time:\n msg = f\"Timeout reached while reading from subprocess.\\nCurrent buffer: {buffer.decode()}\\nRunning PIDs: {pids}\"\n raise TimeoutError(msg)\n\n decoded = buffer.decode(\"utf-8\", errors=\"backslashreplace\").replace(\"\\r\\n\", \"\\n\")\n return \"\\n\".join(line for line in decoded.splitlines())\n</code></pre>"},{"location":"reference/env_utils/#sweagent.environment.utils.read_with_timeout_experimental","title":"<code>read_with_timeout_experimental(container, timeout_duration, no_output_timeout_duration)</code>","text":"<p>Read data from a subprocess with a timeout. This function uses a file descriptor to read data from the subprocess in a non-blocking way.</p> <p>NOTE: This is an experimental implementation that is faster than <code>read_with_timeout</code>, but has not been thoroughly tested.</p> <p>Parameters:</p> Name Type Description Default <code>container</code> <code>Popen</code> <p>The subprocess container.</p> required <code>timeout_duration</code> <code>int | float</code> <p>The timeout duration in seconds.</p> required <code>no_output_timeout_duration</code> <code>int | float</code> <p>The timeout duration to wait if no output is produced, in seconds.</p> required <p>Returns:</p> Type Description <code>tuple[str, str]</code> <p>Output and exit code, both as strings (!)</p> <p>Raises:</p> Type Description <code>TimeoutError</code> <p>If the timeout duration is reached while reading from the subprocess.</p> Source code in <code>sweagent/environment/utils.py</code> <pre><code>def read_with_timeout_experimental(\n container: subprocess.Popen, timeout_duration: int | float, no_output_timeout_duration: int | float\n) -> tuple[str, str]:\n \"\"\"\n Read data from a subprocess with a timeout.\n This function uses a file descriptor to read data from the subprocess in a non-blocking way.\n\n NOTE: This is an experimental implementation that is faster than `read_with_timeout`, but\n has not been thoroughly tested.\n\n Args:\n container: The subprocess container.\n timeout_duration: The timeout duration in seconds.\n no_output_timeout_duration: The timeout duration to wait if no output is produced, in seconds.\n\n Returns:\n Output and exit code, both as strings (!)\n\n Raises:\n TimeoutError: If the timeout duration is reached while reading from the subprocess.\n \"\"\"\n buffer = b\"\"\n fd = container.stdout.fileno()\n start_time = time.time()\n end_time = start_time + timeout_duration\n end_time_no_output = start_time + no_output_timeout_duration\n\n # Select is not available on windows\n is_windows = platform.system() == \"Windows\"\n if not is_windows:\n import select\n else:\n os.set_blocking(fd, False)\n\n def ready_to_read(fd) -> bool:\n if is_windows:\n # We can't do the extra check\n return True\n return bool(select.select([fd], [], [], 0.01)[0])\n\n process_done = False\n\n while time.time() < min(end_time, end_time_no_output):\n if ready_to_read(fd):\n try:\n data = os.read(fd, 4096)\n except BlockingIOError:\n logger.error(\"BlockingIOError while reading from subprocess.\", exc_info=True)\n break\n if data:\n end_time_no_output = time.time() + no_output_timeout_duration\n buffer += data\n if PROCESS_DONE_MARKER_START in buffer.decode(\"utf-8\", errors=\"backslashreplace\").replace(\"\\r\\n\", \"\\n\"):\n process_done = True\n break\n time.sleep(0.01) # Prevents CPU hogging\n\n decoded = buffer.decode(\"utf-8\", errors=\"backslashreplace\").replace(\"\\r\\n\", \"\\n\")\n body = \"\\n\".join(line for line in decoded.splitlines() if not line.startswith(PROCESS_DONE_MARKER_START))\n\n if container.poll() is not None:\n msg = f\"Subprocess exited unexpectedly.\\nCurrent buffer: {decoded}\"\n raise RuntimeError(msg, body)\n\n current_time = time.time()\n if not process_done and current_time >= min(end_time, end_time_no_output):\n if current_time >= end_time:\n msg = f\"Timeout reached while reading from subprocess.\\nCurrent buffer: {decoded}\"\n raise TimeoutError(msg, body)\n else:\n msg = f\"No output timeout reached while reading from subprocess.\\nCurrent buffer: {decoded}\"\n raise NoOutputTimeoutError(msg, body)\n\n _check_for_too_many_non_unicode_bytes(buffer=buffer)\n _results = PROCESS_DONE_REGEX.search(decoded)\n if _results is None:\n msg = f\"Could not find process done marker in last line: {decoded=}, {body=}\"\n raise ValueError(msg)\n exit_code = _results.group(1)\n return body.replace(f\"{PROCESS_DONE_MARKER_START}{exit_code}{PROCESS_DONE_MARKER_END}\", \"\"), exit_code\n</code></pre>"},{"location":"reference/models/","title":"Models","text":""},{"location":"reference/models/#sweagent.agent.models.AnthropicModel","title":"<code>AnthropicModel</code>","text":"<p> Bases: <code>BaseModel</code></p> Source code in <code>sweagent/agent/models.py</code> <pre><code>class AnthropicModel(BaseModel):\n MODELS = {\n \"claude-instant\": {\n \"max_context\": 100_000,\n \"cost_per_input_token\": 1.63e-06,\n \"cost_per_output_token\": 5.51e-06,\n },\n \"claude-2.0\": {\n \"max_context\": 100_000,\n \"cost_per_input_token\": 1.102e-05,\n \"cost_per_output_token\": 3.268e-05,\n },\n \"claude-2.1\": {\n \"max_context\": 100_000,\n \"cost_per_input_token\": 1.102e-05,\n \"cost_per_output_token\": 3.268e-05,\n },\n \"claude-3-opus-20240229\": {\n \"max_context\": 200_000,\n \"max_tokens\": 4096, # Max tokens to generate for Claude 3 models\n \"cost_per_input_token\": 1.5e-05,\n \"cost_per_output_token\": 7.5e-05,\n },\n \"claude-3-sonnet-20240229\": {\n \"max_context\": 200_000,\n \"max_tokens\": 4096,\n \"cost_per_input_token\": 3e-06,\n \"cost_per_output_token\": 1.5e-05,\n },\n \"claude-3-5-sonnet-20240620\": {\n \"max_context\": 200_000,\n \"max_tokens\": 4096,\n \"cost_per_input_token\": 3e-06,\n \"cost_per_output_token\": 1.5e-05,\n },\n \"claude-3-haiku-20240307\": {\n \"max_context\": 200_000,\n \"max_tokens\": 4096,\n \"cost_per_input_token\": 2.5e-07,\n \"cost_per_output_token\": 1.25e-06,\n },\n }\n\n SHORTCUTS = {\n \"claude-2\": \"claude-2.1\",\n \"claude-opus\": \"claude-3-opus-20240229\",\n \"claude-sonnet\": \"claude-3-sonnet-20240229\",\n \"claude-haiku\": \"claude-3-haiku-20240307\",\n \"claude-sonnet-3.5\": \"claude-3-5-sonnet-20240620\",\n }\n\n def __init__(self, args: ModelArguments, commands: list[Command]):\n super().__init__(args, commands)\n\n # Set Anthropic key\n self.api = Anthropic(api_key=keys_config[\"ANTHROPIC_API_KEY\"])\n\n def history_to_messages(\n self,\n history: list[dict[str, str]],\n is_demonstration: bool = False,\n ) -> str | list[dict[str, str]]:\n \"\"\"\n Create `prompt` by filtering out all keys except for role/content per `history` turn\n Reference: https://docs.anthropic.com/claude/reference/complete_post\n \"\"\"\n return anthropic_history_to_messages(self, history, is_demonstration)\n\n @retry(\n wait=wait_random_exponential(min=1, max=15),\n reraise=True,\n stop=stop_after_attempt(_MAX_RETRIES),\n retry=retry_if_not_exception_type((CostLimitExceededError, RuntimeError)),\n )\n def query(self, history: list[dict[str, str]]) -> str:\n \"\"\"\n Query the Anthropic API with the given `history` and return the response.\n \"\"\"\n return anthropic_query(self, history)\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.AnthropicModel.history_to_messages","title":"<code>history_to_messages(history, is_demonstration=False)</code>","text":"<p>Create <code>prompt</code> by filtering out all keys except for role/content per <code>history</code> turn Reference: https://docs.anthropic.com/claude/reference/complete_post</p> Source code in <code>sweagent/agent/models.py</code> <pre><code>def history_to_messages(\n self,\n history: list[dict[str, str]],\n is_demonstration: bool = False,\n) -> str | list[dict[str, str]]:\n \"\"\"\n Create `prompt` by filtering out all keys except for role/content per `history` turn\n Reference: https://docs.anthropic.com/claude/reference/complete_post\n \"\"\"\n return anthropic_history_to_messages(self, history, is_demonstration)\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.AnthropicModel.query","title":"<code>query(history)</code>","text":"<p>Query the Anthropic API with the given <code>history</code> and return the response.</p> Source code in <code>sweagent/agent/models.py</code> <pre><code>@retry(\n wait=wait_random_exponential(min=1, max=15),\n reraise=True,\n stop=stop_after_attempt(_MAX_RETRIES),\n retry=retry_if_not_exception_type((CostLimitExceededError, RuntimeError)),\n)\ndef query(self, history: list[dict[str, str]]) -> str:\n \"\"\"\n Query the Anthropic API with the given `history` and return the response.\n \"\"\"\n return anthropic_query(self, history)\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.BaseModel","title":"<code>BaseModel</code>","text":"Source code in <code>sweagent/agent/models.py</code> <pre><code>class BaseModel:\n MODELS = {}\n SHORTCUTS = {}\n\n def __init__(self, args: ModelArguments, commands: list[Command]):\n self.args = args\n self.commands = commands\n self.model_metadata = {}\n self.stats = APIStats()\n\n # Map `model_name` to API-compatible name `api_model`\n self.api_model = (\n self.SHORTCUTS[self.args.model_name] if self.args.model_name in self.SHORTCUTS else self.args.model_name\n )\n\n # Map model name to metadata (cost, context info)\n MODELS = {\n **{dest: self.MODELS[src] for dest, src in self.SHORTCUTS.items()},\n **self.MODELS,\n }\n if args.model_name in MODELS:\n self.model_metadata = MODELS[args.model_name]\n elif args.model_name.startswith(\"ft:\"):\n ft_model = args.model_name.split(\":\")[1]\n self.model_metadata = MODELS[ft_model]\n elif args.model_name.startswith(\"ollama:\"):\n self.api_model = args.model_name.split(\"ollama:\", 1)[1]\n self.model_metadata = self.MODELS[self.api_model]\n elif args.model_name.startswith(\"azure:\"):\n azure_model = args.model_name.split(\"azure:\", 1)[1]\n self.model_metadata = MODELS[azure_model]\n elif args.model_name.startswith(\"bedrock:\"):\n self.api_model = args.model_name.split(\"bedrock:\", 1)[1]\n self.model_metadata = MODELS[self.api_model]\n elif args.model_name.startswith(\"groq:\"):\n self.api_model = args.model_name.split(\"groq:\", 1)[1]\n self.model_metadata = MODELS[self.api_model]\n else:\n msg = f\"Unregistered model ({args.model_name}). Add model name to MODELS metadata to {self.__class__}\"\n raise ValueError(msg)\n\n def reset_stats(self, other: APIStats | None = None):\n if other is None:\n self.stats = APIStats(total_cost=self.stats.total_cost)\n logger.info(\"Resetting model stats\")\n else:\n # Make sure to copy the stats to avoid modifying the original\n self.stats = copy.deepcopy(other)\n\n def update_stats(self, input_tokens: int, output_tokens: int) -> float:\n \"\"\"\n Calculates the cost of a response from the openai API.\n\n Args:\n input_tokens (int): The number of tokens in the prompt.\n output_tokens (int): The number of tokens in the response.\n\n Returns:\n float: The cost of the response.\n \"\"\"\n # Calculate cost and update cost related fields\n cost = (\n self.model_metadata[\"cost_per_input_token\"] * input_tokens\n + self.model_metadata[\"cost_per_output_token\"] * output_tokens\n )\n self.stats.total_cost += cost\n self.stats.instance_cost += cost\n self.stats.tokens_sent += input_tokens\n self.stats.tokens_received += output_tokens\n self.stats.api_calls += 1\n\n # Log updated cost values to std. err\n logger.debug(\n f\"input_tokens={input_tokens:,}, \"\n f\"output_tokens={output_tokens:,}, \"\n f\"instance_cost={self.stats.instance_cost:.2f}, \"\n f\"cost={cost:.2f}\",\n )\n logger.debug(\n f\"total_tokens_sent={self.stats.tokens_sent:,}, \"\n f\"total_tokens_received={self.stats.tokens_received:,}, \"\n f\"total_cost={self.stats.total_cost:.2f}, \"\n f\"total_api_calls={self.stats.api_calls:,}\",\n )\n\n # Check whether total cost or instance cost limits have been exceeded\n if 0 < self.args.total_cost_limit <= self.stats.total_cost:\n logger.warning(f\"Cost {self.stats.total_cost:.2f} exceeds limit {self.args.total_cost_limit:.2f}\")\n msg = \"Total cost limit exceeded\"\n raise CostLimitExceededError(msg)\n\n if 0 < self.args.per_instance_cost_limit <= self.stats.instance_cost:\n logger.warning(f\"Cost {self.stats.instance_cost:.2f} exceeds limit {self.args.per_instance_cost_limit:.2f}\")\n msg = \"Instance cost limit exceeded\"\n raise CostLimitExceededError(msg)\n return cost\n\n def query(self, history: list[dict[str, str]]) -> str:\n msg = \"Use a subclass of BaseModel\"\n raise NotImplementedError(msg)\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.BaseModel.update_stats","title":"<code>update_stats(input_tokens, output_tokens)</code>","text":"<p>Calculates the cost of a response from the openai API.</p> <p>Args: input_tokens (int): The number of tokens in the prompt. output_tokens (int): The number of tokens in the response.</p> <p>Returns: float: The cost of the response.</p> Source code in <code>sweagent/agent/models.py</code> <pre><code>def update_stats(self, input_tokens: int, output_tokens: int) -> float:\n \"\"\"\n Calculates the cost of a response from the openai API.\n\n Args:\n input_tokens (int): The number of tokens in the prompt.\n output_tokens (int): The number of tokens in the response.\n\n Returns:\n float: The cost of the response.\n \"\"\"\n # Calculate cost and update cost related fields\n cost = (\n self.model_metadata[\"cost_per_input_token\"] * input_tokens\n + self.model_metadata[\"cost_per_output_token\"] * output_tokens\n )\n self.stats.total_cost += cost\n self.stats.instance_cost += cost\n self.stats.tokens_sent += input_tokens\n self.stats.tokens_received += output_tokens\n self.stats.api_calls += 1\n\n # Log updated cost values to std. err\n logger.debug(\n f\"input_tokens={input_tokens:,}, \"\n f\"output_tokens={output_tokens:,}, \"\n f\"instance_cost={self.stats.instance_cost:.2f}, \"\n f\"cost={cost:.2f}\",\n )\n logger.debug(\n f\"total_tokens_sent={self.stats.tokens_sent:,}, \"\n f\"total_tokens_received={self.stats.tokens_received:,}, \"\n f\"total_cost={self.stats.total_cost:.2f}, \"\n f\"total_api_calls={self.stats.api_calls:,}\",\n )\n\n # Check whether total cost or instance cost limits have been exceeded\n if 0 < self.args.total_cost_limit <= self.stats.total_cost:\n logger.warning(f\"Cost {self.stats.total_cost:.2f} exceeds limit {self.args.total_cost_limit:.2f}\")\n msg = \"Total cost limit exceeded\"\n raise CostLimitExceededError(msg)\n\n if 0 < self.args.per_instance_cost_limit <= self.stats.instance_cost:\n logger.warning(f\"Cost {self.stats.instance_cost:.2f} exceeds limit {self.args.per_instance_cost_limit:.2f}\")\n msg = \"Instance cost limit exceeded\"\n raise CostLimitExceededError(msg)\n return cost\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.BedrockModel","title":"<code>BedrockModel</code>","text":"<p> Bases: <code>BaseModel</code></p> Source code in <code>sweagent/agent/models.py</code> <pre><code>class BedrockModel(BaseModel):\n MODELS = {\n \"anthropic.claude-instant-v1\": {\n \"max_context\": 100_000,\n \"max_tokens_to_sample\": 4096,\n \"cost_per_input_token\": 8e-07,\n \"cost_per_output_token\": 2.4e-06,\n },\n \"anthropic.claude-v2\": {\n \"max_context\": 100_000,\n \"max_tokens_to_sample\": 4096,\n \"cost_per_input_token\": 8e-06,\n \"cost_per_output_token\": 2.4e-05,\n },\n \"anthropic.claude-v2:1\": {\n \"max_context\": 100_000,\n \"max_tokens\": 4096,\n \"cost_per_input_token\": 8e-06,\n \"cost_per_output_token\": 2.4e-05,\n },\n \"anthropic.claude-3-opus-20240229-v1:0\": {\n \"max_context\": 200_000,\n \"max_tokens\": 4096,\n \"cost_per_input_token\": 1.5e-05,\n \"cost_per_output_token\": 7.5e-05,\n },\n \"anthropic.claude-3-sonnet-20240229-v1:0\": {\n \"max_context\": 200_000,\n \"max_tokens\": 4096,\n \"cost_per_input_token\": 3e-06,\n \"cost_per_output_token\": 1.5e-05,\n },\n \"anthropic.claude-3-haiku-20240307-v1:0\": {\n \"max_context\": 200_000,\n \"max_tokens\": 4096,\n \"cost_per_input_token\": 2.5e-07,\n \"cost_per_output_token\": 1.25e-06,\n },\n }\n\n def __init__(self, args: ModelArguments, commands: list[Command]):\n super().__init__(args, commands)\n\n # Extract provider from model ID\n # https://docs.aws.amazon.com/bedrock/latest/userguide/model-ids.html\n self.model_provider = self.api_model.split(\".\")[0]\n if self.model_provider == \"anthropic\":\n # Note: this assumes AWS credentials are already configured.\n # https://boto3.amazonaws.com/v1/documentation/api/latest/guide/credentials.html\n self.api = AnthropicBedrock()\n elif self.model_provider in [\"ai21\", \"amazon\", \"cohere\", \"meta\", \"mistral\"]:\n msg = f\"{self.api_model} is not supported!\"\n raise NotImplementedError(msg)\n else:\n msg = f\"Provider {self.model_provider} is not supported by Amazon Bedrock!\"\n raise ValueError(msg)\n\n def history_to_messages(\n self,\n history: list[dict[str, str]],\n is_demonstration: bool = False,\n ) -> str | list[dict[str, str]]:\n \"\"\"\n Create `prompt` from the history of messages\n \"\"\"\n if self.model_provider == \"anthropic\":\n return anthropic_history_to_messages(self, history, is_demonstration)\n else:\n msg = f\"{self.api_model} is not supported!\"\n raise NotImplementedError(msg)\n\n @retry(\n wait=wait_random_exponential(min=1, max=15),\n reraise=True,\n stop=stop_after_attempt(_MAX_RETRIES),\n retry=retry_if_not_exception_type((CostLimitExceededError, RuntimeError)),\n )\n def query(self, history: list[dict[str, str]]) -> str:\n \"\"\"\n Query Amazon Bedrock with the given `history` and return the response.\n \"\"\"\n if self.model_provider == \"anthropic\":\n return anthropic_query(self, history)\n else:\n msg = f\"{self.api_model} is not supported!\"\n raise NotImplementedError(msg)\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.BedrockModel.history_to_messages","title":"<code>history_to_messages(history, is_demonstration=False)</code>","text":"<p>Create <code>prompt</code> from the history of messages</p> Source code in <code>sweagent/agent/models.py</code> <pre><code>def history_to_messages(\n self,\n history: list[dict[str, str]],\n is_demonstration: bool = False,\n) -> str | list[dict[str, str]]:\n \"\"\"\n Create `prompt` from the history of messages\n \"\"\"\n if self.model_provider == \"anthropic\":\n return anthropic_history_to_messages(self, history, is_demonstration)\n else:\n msg = f\"{self.api_model} is not supported!\"\n raise NotImplementedError(msg)\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.BedrockModel.query","title":"<code>query(history)</code>","text":"<p>Query Amazon Bedrock with the given <code>history</code> and return the response.</p> Source code in <code>sweagent/agent/models.py</code> <pre><code>@retry(\n wait=wait_random_exponential(min=1, max=15),\n reraise=True,\n stop=stop_after_attempt(_MAX_RETRIES),\n retry=retry_if_not_exception_type((CostLimitExceededError, RuntimeError)),\n)\ndef query(self, history: list[dict[str, str]]) -> str:\n \"\"\"\n Query Amazon Bedrock with the given `history` and return the response.\n \"\"\"\n if self.model_provider == \"anthropic\":\n return anthropic_query(self, history)\n else:\n msg = f\"{self.api_model} is not supported!\"\n raise NotImplementedError(msg)\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.HumanModel","title":"<code>HumanModel</code>","text":"<p> Bases: <code>BaseModel</code></p> Source code in <code>sweagent/agent/models.py</code> <pre><code>class HumanModel(BaseModel):\n MODELS = {\"human\": {}}\n\n def __init__(self, args: ModelArguments, commands: list[Command]):\n super().__init__(args, commands)\n\n # Determine which commands require multi-line input\n self.multi_line_command_endings = {\n command.name: command.end_name for command in commands if command.end_name is not None\n }\n\n def history_to_messages(\n self,\n history: list[dict[str, str]],\n is_demonstration: bool = False,\n ) -> str | list[dict[str, str]]:\n \"\"\"\n Create `messages` by filtering out all keys except for role/content per `history` turn\n \"\"\"\n # Remove system messages if it is a demonstration\n if is_demonstration:\n history = [entry for entry in history if entry[\"role\"] != \"system\"]\n return \"\\n\".join([entry[\"content\"] for entry in history])\n # Return history components with just role, content fields\n return [{k: v for k, v in entry.items() if k in [\"role\", \"content\"]} for entry in history]\n\n def query(self, history: list[dict[str, str]], action_prompt: str = \"> \") -> str:\n \"\"\"\n Logic for handling user input to pass to SWEEnv\n \"\"\"\n action = input(action_prompt)\n command_name = action.split()[0] if action.strip() else \"\"\n\n # Special handling for multi-line input actions (i.e. edit)\n if command_name in self.multi_line_command_endings:\n buffer = [action]\n end_keyword = self.multi_line_command_endings[command_name]\n while True:\n action = input(\"... \")\n buffer.append(action)\n if action.rstrip() == end_keyword:\n # Continue reading input until terminating keyword inputted\n break\n action = \"\\n\".join(buffer)\n elif action.strip() == \"start_multiline_command\": # do arbitrary multi-line input\n buffer = []\n while True:\n action = input(\"... \")\n if action.rstrip() == \"end_multiline_command\":\n break\n buffer.append(action)\n action = \"\\n\".join(buffer)\n return action\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.HumanModel.history_to_messages","title":"<code>history_to_messages(history, is_demonstration=False)</code>","text":"<p>Create <code>messages</code> by filtering out all keys except for role/content per <code>history</code> turn</p> Source code in <code>sweagent/agent/models.py</code> <pre><code>def history_to_messages(\n self,\n history: list[dict[str, str]],\n is_demonstration: bool = False,\n) -> str | list[dict[str, str]]:\n \"\"\"\n Create `messages` by filtering out all keys except for role/content per `history` turn\n \"\"\"\n # Remove system messages if it is a demonstration\n if is_demonstration:\n history = [entry for entry in history if entry[\"role\"] != \"system\"]\n return \"\\n\".join([entry[\"content\"] for entry in history])\n # Return history components with just role, content fields\n return [{k: v for k, v in entry.items() if k in [\"role\", \"content\"]} for entry in history]\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.HumanModel.query","title":"<code>query(history, action_prompt='> ')</code>","text":"<p>Logic for handling user input to pass to SWEEnv</p> Source code in <code>sweagent/agent/models.py</code> <pre><code>def query(self, history: list[dict[str, str]], action_prompt: str = \"> \") -> str:\n \"\"\"\n Logic for handling user input to pass to SWEEnv\n \"\"\"\n action = input(action_prompt)\n command_name = action.split()[0] if action.strip() else \"\"\n\n # Special handling for multi-line input actions (i.e. edit)\n if command_name in self.multi_line_command_endings:\n buffer = [action]\n end_keyword = self.multi_line_command_endings[command_name]\n while True:\n action = input(\"... \")\n buffer.append(action)\n if action.rstrip() == end_keyword:\n # Continue reading input until terminating keyword inputted\n break\n action = \"\\n\".join(buffer)\n elif action.strip() == \"start_multiline_command\": # do arbitrary multi-line input\n buffer = []\n while True:\n action = input(\"... \")\n if action.rstrip() == \"end_multiline_command\":\n break\n buffer.append(action)\n action = \"\\n\".join(buffer)\n return action\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.HumanThoughtModel","title":"<code>HumanThoughtModel</code>","text":"<p> Bases: <code>HumanModel</code></p> Source code in <code>sweagent/agent/models.py</code> <pre><code>class HumanThoughtModel(HumanModel):\n MODELS = {\"human_thought\": {}}\n\n def query(self, history: list[dict[str, str]]) -> str:\n \"\"\"\n Logic for handling user input (both thought + action) to pass to SWEEnv\n \"\"\"\n thought_all = \"\"\n thought = input(\"Thought (end w/ END_THOUGHT): \")\n while True:\n if \"END_THOUGHT\" in thought:\n thought = thought.split(\"END_THOUGHT\")[0]\n thought_all += thought\n break\n thought_all += thought\n thought = input(\"... \")\n\n action = super().query(history, action_prompt=\"Action: \")\n\n return f\"{thought_all}\\n```\\n{action}\\n```\"\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.HumanThoughtModel.query","title":"<code>query(history)</code>","text":"<p>Logic for handling user input (both thought + action) to pass to SWEEnv</p> Source code in <code>sweagent/agent/models.py</code> <pre><code>def query(self, history: list[dict[str, str]]) -> str:\n \"\"\"\n Logic for handling user input (both thought + action) to pass to SWEEnv\n \"\"\"\n thought_all = \"\"\n thought = input(\"Thought (end w/ END_THOUGHT): \")\n while True:\n if \"END_THOUGHT\" in thought:\n thought = thought.split(\"END_THOUGHT\")[0]\n thought_all += thought\n break\n thought_all += thought\n thought = input(\"... \")\n\n action = super().query(history, action_prompt=\"Action: \")\n\n return f\"{thought_all}\\n```\\n{action}\\n```\"\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.InstantEmptySubmitTestModel","title":"<code>InstantEmptySubmitTestModel</code>","text":"<p> Bases: <code>BaseModel</code></p> Source code in <code>sweagent/agent/models.py</code> <pre><code>class InstantEmptySubmitTestModel(BaseModel):\n MODELS = {\n \"instant_empty_submit\": {\n \"max_context\": 100_000,\n \"max_tokens_to_sample\": 4096,\n \"cost_per_input_token\": 0,\n \"cost_per_output_token\": 0,\n }\n }\n\n def __init__(self, args: ModelArguments, commands: list[Command]):\n \"\"\"This model immediately submits. Useful for testing purposes\"\"\"\n super().__init__(args, commands)\n self._action_idx = 0\n\n def query(self, history: list[dict[str, str]]) -> str:\n # Need to at least do _something_ to submit\n if self._action_idx == 0:\n self._action_idx = 1\n action = \"DISCUSSION\\nLet's reproduce the bug by creating a `reproduce.py` file.\\n\\n```\\ncreate reproduce.py\\n```\\n\"\n elif self._action_idx == 1:\n self._action_idx = 0\n action = \"DISCUSSION\\nThe task should be resolved, so let's submit the patch.\\n\\n```\\nsubmit\\n```\\n\"\n self.update_stats(0, 0)\n return action\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.InstantEmptySubmitTestModel.__init__","title":"<code>__init__(args, commands)</code>","text":"<p>This model immediately submits. Useful for testing purposes</p> Source code in <code>sweagent/agent/models.py</code> <pre><code>def __init__(self, args: ModelArguments, commands: list[Command]):\n \"\"\"This model immediately submits. Useful for testing purposes\"\"\"\n super().__init__(args, commands)\n self._action_idx = 0\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.ModelArguments","title":"<code>ModelArguments</code> <code>dataclass</code>","text":"<p> Bases: <code>FrozenSerializable</code></p> <p>Arguments configuring the model and its behavior.</p> Source code in <code>sweagent/agent/models.py</code> <pre><code>@dataclass(frozen=True)\nclass ModelArguments(FrozenSerializable):\n \"\"\"Arguments configuring the model and its behavior.\"\"\"\n\n # Name of the model to use\n model_name: str\n # Cost limit for every instance (task)\n per_instance_cost_limit: float = 0.0\n # Total cost limit\n total_cost_limit: float = 0.0\n # Sampling temperature\n temperature: float = 1.0\n # Sampling top-p\n top_p: float = 1.0\n # Path to replay file when using the replay model\n replay_path: str | None = None\n # Host URL when using Ollama model\n host_url: str = \"localhost:11434\"\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.OllamaModel","title":"<code>OllamaModel</code>","text":"<p> Bases: <code>BaseModel</code></p> Source code in <code>sweagent/agent/models.py</code> <pre><code>class OllamaModel(BaseModel):\n MODELS = defaultdict(\n lambda: {\n \"max_context\": 128_000,\n \"cost_per_input_token\": 0,\n \"cost_per_output_token\": 0,\n },\n )\n\n def __init__(self, args: ModelArguments, commands: list[Command]):\n super().__init__(args, commands)\n from ollama import Client\n\n self.client = Client(host=args.host_url)\n\n def history_to_messages(\n self,\n history: list[dict[str, str]],\n is_demonstration: bool = False,\n ) -> str | list[dict[str, str]]:\n \"\"\"\n Create `messages` by filtering out all keys except for role/content per `history` turn\n \"\"\"\n # Remove system messages if it is a demonstration\n if is_demonstration:\n history = [entry for entry in history if entry[\"role\"] != \"system\"]\n return \"\\n\".join([entry[\"content\"] for entry in history])\n # Return history components with just role, content fields\n return [{k: v for k, v in entry.items() if k in [\"role\", \"content\"]} for entry in history]\n\n @retry(\n wait=wait_random_exponential(min=1, max=15),\n reraise=True,\n stop=stop_after_attempt(_MAX_RETRIES),\n retry=retry_if_not_exception_type((CostLimitExceededError, RuntimeError)),\n )\n def query(self, history: list[dict[str, str]]) -> str:\n \"\"\"\n Query the Ollama API with the given `history` and return the response.\n \"\"\"\n response = self.client.chat(\n model=self.api_model,\n messages=self.history_to_messages(history),\n options={\n \"temperature\": self.args.temperature,\n \"top_p\": self.args.top_p,\n },\n )\n # Calculate + update costs, return response\n if \"prompt_eval_count\" in response:\n input_tokens = response[\"prompt_eval_count\"]\n else:\n logger.warning(\n \"Prompt eval count not found in response. Using 0. \"\n \"This might be because the prompt has been cached. \"\n \"See https://github.com/princeton-nlp/SWE-agent/issues/44 \"\n \"and https://github.com/ollama/ollama/issues/3427.\",\n )\n input_tokens = 0\n output_tokens = response[\"eval_count\"]\n self.update_stats(input_tokens, output_tokens)\n return response[\"message\"][\"content\"]\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.OllamaModel.history_to_messages","title":"<code>history_to_messages(history, is_demonstration=False)</code>","text":"<p>Create <code>messages</code> by filtering out all keys except for role/content per <code>history</code> turn</p> Source code in <code>sweagent/agent/models.py</code> <pre><code>def history_to_messages(\n self,\n history: list[dict[str, str]],\n is_demonstration: bool = False,\n) -> str | list[dict[str, str]]:\n \"\"\"\n Create `messages` by filtering out all keys except for role/content per `history` turn\n \"\"\"\n # Remove system messages if it is a demonstration\n if is_demonstration:\n history = [entry for entry in history if entry[\"role\"] != \"system\"]\n return \"\\n\".join([entry[\"content\"] for entry in history])\n # Return history components with just role, content fields\n return [{k: v for k, v in entry.items() if k in [\"role\", \"content\"]} for entry in history]\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.OllamaModel.query","title":"<code>query(history)</code>","text":"<p>Query the Ollama API with the given <code>history</code> and return the response.</p> Source code in <code>sweagent/agent/models.py</code> <pre><code>@retry(\n wait=wait_random_exponential(min=1, max=15),\n reraise=True,\n stop=stop_after_attempt(_MAX_RETRIES),\n retry=retry_if_not_exception_type((CostLimitExceededError, RuntimeError)),\n)\ndef query(self, history: list[dict[str, str]]) -> str:\n \"\"\"\n Query the Ollama API with the given `history` and return the response.\n \"\"\"\n response = self.client.chat(\n model=self.api_model,\n messages=self.history_to_messages(history),\n options={\n \"temperature\": self.args.temperature,\n \"top_p\": self.args.top_p,\n },\n )\n # Calculate + update costs, return response\n if \"prompt_eval_count\" in response:\n input_tokens = response[\"prompt_eval_count\"]\n else:\n logger.warning(\n \"Prompt eval count not found in response. Using 0. \"\n \"This might be because the prompt has been cached. \"\n \"See https://github.com/princeton-nlp/SWE-agent/issues/44 \"\n \"and https://github.com/ollama/ollama/issues/3427.\",\n )\n input_tokens = 0\n output_tokens = response[\"eval_count\"]\n self.update_stats(input_tokens, output_tokens)\n return response[\"message\"][\"content\"]\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.OpenAIModel","title":"<code>OpenAIModel</code>","text":"<p> Bases: <code>BaseModel</code></p> Source code in <code>sweagent/agent/models.py</code> <pre><code>class OpenAIModel(BaseModel):\n MODELS = {\n \"gpt-3.5-turbo-0125\": {\n \"max_context\": 16_385,\n \"cost_per_input_token\": 5e-07,\n \"cost_per_output_token\": 1.5e-06,\n },\n \"gpt-3.5-turbo-1106\": {\n \"max_context\": 16_385,\n \"cost_per_input_token\": 1.5e-06,\n \"cost_per_output_token\": 2e-06,\n },\n \"gpt-3.5-turbo-16k-0613\": {\n \"max_context\": 16_385,\n \"cost_per_input_token\": 1.5e-06,\n \"cost_per_output_token\": 2e-06,\n },\n \"gpt-4-32k-0613\": {\n \"max_context\": 32_768,\n \"cost_per_input_token\": 6e-05,\n \"cost_per_output_token\": 0.00012,\n },\n \"gpt-4-0613\": {\n \"max_context\": 8_192,\n \"cost_per_input_token\": 3e-05,\n \"cost_per_output_token\": 6e-05,\n },\n \"gpt-4-1106-preview\": {\n \"max_context\": 128_000,\n \"cost_per_input_token\": 1e-05,\n \"cost_per_output_token\": 3e-05,\n },\n \"gpt-4-0125-preview\": {\n \"max_context\": 128_000,\n \"cost_per_input_token\": 1e-05,\n \"cost_per_output_token\": 3e-05,\n },\n \"gpt-4-turbo-2024-04-09\": {\n \"max_context\": 128_000,\n \"cost_per_input_token\": 1e-05,\n \"cost_per_output_token\": 3e-05,\n },\n \"gpt-4o-2024-05-13\": {\n \"max_context\": 128_000,\n \"cost_per_input_token\": 5e-06,\n \"cost_per_output_token\": 15e-06,\n },\n \"gpt-4o-mini-2024-07-18\": {\n \"max_context\": 128_000,\n \"cost_per_input_token\": 1.5e-07,\n \"cost_per_output_token\": 6e-07,\n },\n \"o1-preview-2024-09-12\": {\n \"max_context\": 128_000,\n \"cost_per_input_token\": 15e-06,\n \"cost_per_output_token\": 60e-06,\n },\n \"o1-mini-2024-09-12\": {\n \"max_context\": 128_000,\n \"cost_per_input_token\": 3e-6,\n \"cost_per_output_token\": 12e-6,\n },\n }\n\n SHORTCUTS = {\n \"gpt3\": \"gpt-3.5-turbo-1106\",\n \"gpt3-legacy\": \"gpt-3.5-turbo-16k-0613\",\n \"gpt4\": \"gpt-4-1106-preview\",\n \"gpt4-legacy\": \"gpt-4-0613\",\n \"gpt4-0125\": \"gpt-4-0125-preview\",\n \"gpt3-0125\": \"gpt-3.5-turbo-0125\",\n \"gpt4-turbo\": \"gpt-4-turbo-2024-04-09\",\n \"gpt4o\": \"gpt-4o-2024-05-13\",\n \"gpt-4o-mini\": \"gpt-4o-mini-2024-07-18\",\n \"gpt4omini\": \"gpt-4o-mini-2024-07-18\",\n \"o1\": \"o1-preview-2024-09-12\",\n \"o1-mini\": \"o1-mini-2024-09-12\",\n }\n\n def __init__(self, args: ModelArguments, commands: list[Command]):\n super().__init__(args, commands)\n\n logging.getLogger(\"openai\").setLevel(logging.WARNING)\n logging.getLogger(\"httpx\").setLevel(logging.WARNING)\n\n self._setup_client()\n\n def _setup_client(self):\n if self.args.model_name.startswith(\"azure\"):\n logger.warning(\n \"The --model CLI argument is ignored when using the Azure GPT endpoint. \"\n \"The model is determined by the AZURE_OPENAI_DEPLOYMENT key/\"\n \"environment variable (this might change in the future).\",\n )\n self.api_model = keys_config[\"AZURE_OPENAI_DEPLOYMENT\"]\n self.client = AzureOpenAI(\n api_key=keys_config[\"AZURE_OPENAI_API_KEY\"],\n azure_endpoint=keys_config[\"AZURE_OPENAI_ENDPOINT\"],\n api_version=keys_config.get(\"AZURE_OPENAI_API_VERSION\", \"2024-02-01\"),\n )\n else:\n api_base_url: str | None = keys_config.get(\"OPENAI_API_BASE_URL\", None)\n self.client = OpenAI(api_key=keys_config[\"OPENAI_API_KEY\"], base_url=api_base_url)\n\n def history_to_messages(\n self,\n history: list[dict[str, str]],\n is_demonstration: bool = False,\n ) -> str | list[dict[str, str]]:\n \"\"\"\n Create `messages` by filtering out all keys except for role/content per `history` turn\n \"\"\"\n # Remove system messages if it is a demonstration\n if is_demonstration:\n history = [entry for entry in history if entry[\"role\"] != \"system\"]\n return \"\\n\".join([entry[\"content\"] for entry in history])\n # Return history components with just role, content fields\n return [{k: v for k, v in entry.items() if k in [\"role\", \"content\"]} for entry in history]\n\n @retry(\n wait=wait_random_exponential(min=1, max=15),\n reraise=True,\n stop=stop_after_attempt(_MAX_RETRIES),\n retry=retry_if_not_exception_type((CostLimitExceededError, RuntimeError)),\n )\n def query(self, history: list[dict[str, str]]) -> str:\n \"\"\"\n Query the OpenAI API with the given `history` and return the response.\n \"\"\"\n try:\n # Perform OpenAI API call\n response = self.client.chat.completions.create(\n messages=self.history_to_messages(history),\n model=self.api_model,\n temperature=self.args.temperature,\n top_p=self.args.top_p,\n )\n except BadRequestError:\n msg = f\"Context window ({self.model_metadata['max_context']} tokens) exceeded\"\n raise ContextWindowExceededError(msg)\n # Calculate + update costs, return response\n input_tokens = response.usage.prompt_tokens\n output_tokens = response.usage.completion_tokens\n self.update_stats(input_tokens, output_tokens)\n return response.choices[0].message.content\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.OpenAIModel.history_to_messages","title":"<code>history_to_messages(history, is_demonstration=False)</code>","text":"<p>Create <code>messages</code> by filtering out all keys except for role/content per <code>history</code> turn</p> Source code in <code>sweagent/agent/models.py</code> <pre><code>def history_to_messages(\n self,\n history: list[dict[str, str]],\n is_demonstration: bool = False,\n) -> str | list[dict[str, str]]:\n \"\"\"\n Create `messages` by filtering out all keys except for role/content per `history` turn\n \"\"\"\n # Remove system messages if it is a demonstration\n if is_demonstration:\n history = [entry for entry in history if entry[\"role\"] != \"system\"]\n return \"\\n\".join([entry[\"content\"] for entry in history])\n # Return history components with just role, content fields\n return [{k: v for k, v in entry.items() if k in [\"role\", \"content\"]} for entry in history]\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.OpenAIModel.query","title":"<code>query(history)</code>","text":"<p>Query the OpenAI API with the given <code>history</code> and return the response.</p> Source code in <code>sweagent/agent/models.py</code> <pre><code>@retry(\n wait=wait_random_exponential(min=1, max=15),\n reraise=True,\n stop=stop_after_attempt(_MAX_RETRIES),\n retry=retry_if_not_exception_type((CostLimitExceededError, RuntimeError)),\n)\ndef query(self, history: list[dict[str, str]]) -> str:\n \"\"\"\n Query the OpenAI API with the given `history` and return the response.\n \"\"\"\n try:\n # Perform OpenAI API call\n response = self.client.chat.completions.create(\n messages=self.history_to_messages(history),\n model=self.api_model,\n temperature=self.args.temperature,\n top_p=self.args.top_p,\n )\n except BadRequestError:\n msg = f\"Context window ({self.model_metadata['max_context']} tokens) exceeded\"\n raise ContextWindowExceededError(msg)\n # Calculate + update costs, return response\n input_tokens = response.usage.prompt_tokens\n output_tokens = response.usage.completion_tokens\n self.update_stats(input_tokens, output_tokens)\n return response.choices[0].message.content\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.ReplayModel","title":"<code>ReplayModel</code>","text":"<p> Bases: <code>BaseModel</code></p> Source code in <code>sweagent/agent/models.py</code> <pre><code>class ReplayModel(BaseModel):\n MODELS = {\"replay\": {}}\n\n def __init__(self, args: ModelArguments, commands: list[Command]):\n super().__init__(args, commands)\n\n if self.args.replay_path is None or not Path(self.args.replay_path).exists():\n msg = \"--replay_path must point to a file that exists to run a replay policy\"\n raise ValueError(msg)\n\n self.replays = [\n list(json.loads(x).values())[0] for x in Path(self.args.replay_path).read_text().splitlines(keepends=True)\n ]\n self.replay_idx = 0\n self.action_idx = 0\n\n def _next_replay(self) -> None:\n \"\"\"Called after last action\"\"\"\n self.replay_idx += 1\n self.action_idx = 0\n\n def query(self, history: list[dict[str, str]]) -> str:\n \"\"\"\n Logic for tracking which replay action to pass to SWEEnv\n \"\"\"\n actions = self.replays[self.replay_idx]\n try:\n action = actions[self.action_idx]\n except IndexError:\n msg = (\n \"This seems to be an incomplete trajectory. \"\n \"We reached the end of it, but `submit` was not called. \"\n \"Calling it now.\"\n )\n logger.warning(msg)\n action = \"```\\nsubmit\\n```\"\n\n self.action_idx += 1\n\n # Assuming `submit` is always last action of replay trajectory\n if action == \"submit\":\n self._next_replay()\n\n return action\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.ReplayModel.query","title":"<code>query(history)</code>","text":"<p>Logic for tracking which replay action to pass to SWEEnv</p> Source code in <code>sweagent/agent/models.py</code> <pre><code>def query(self, history: list[dict[str, str]]) -> str:\n \"\"\"\n Logic for tracking which replay action to pass to SWEEnv\n \"\"\"\n actions = self.replays[self.replay_idx]\n try:\n action = actions[self.action_idx]\n except IndexError:\n msg = (\n \"This seems to be an incomplete trajectory. \"\n \"We reached the end of it, but `submit` was not called. \"\n \"Calling it now.\"\n )\n logger.warning(msg)\n action = \"```\\nsubmit\\n```\"\n\n self.action_idx += 1\n\n # Assuming `submit` is always last action of replay trajectory\n if action == \"submit\":\n self._next_replay()\n\n return action\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.TogetherModel","title":"<code>TogetherModel</code>","text":"<p> Bases: <code>BaseModel</code></p> Source code in <code>sweagent/agent/models.py</code> <pre><code>class TogetherModel(BaseModel):\n # Check https://docs.together.ai/docs/inference-models for model names, context\n # Check https://www.together.ai/pricing for pricing\n MODELS = {\n \"meta-llama/Llama-2-13b-chat-hf\": {\n \"max_context\": 4096,\n \"cost_per_input_token\": 2.25e-07,\n \"cost_per_output_token\": 2.25e-07,\n },\n \"meta-llama/Llama-2-70b-chat-hf\": {\n \"max_context\": 4096,\n \"cost_per_input_token\": 9e-07,\n \"cost_per_output_token\": 9e-07,\n },\n \"mistralai/Mistral-7B-Instruct-v0.2\": {\n \"max_context\": 32768,\n \"cost_per_input_token\": 2e-07,\n \"cost_per_output_token\": 2e-07,\n },\n \"togethercomputer/RedPajama-INCITE-7B-Chat\": {\n \"max_context\": 2048,\n \"cost_per_input_token\": 2e-07,\n \"cost_per_output_token\": 2e-07,\n },\n \"mistralai/Mixtral-8x7B-Instruct-v0.1\": {\n \"max_context\": 32768,\n \"cost_per_input_token\": 6e-07,\n \"cost_per_output_token\": 6e-07,\n },\n }\n\n SHORTCUTS = {\n \"llama13b\": \"meta-llama/Llama-2-13b-chat-hf\",\n \"llama70b\": \"meta-llama/Llama-2-70b-chat-hf\",\n \"mistral7b\": \"mistralai/Mistral-7B-Instruct-v0.2\",\n \"mixtral8x7b\": \"mistralai/Mixtral-8x7B-Instruct-v0.1\",\n \"redpajama7b\": \"togethercomputer/RedPajama-INCITE-7B-Chat\",\n }\n\n def __init__(self, args: ModelArguments, commands: list[Command]):\n super().__init__(args, commands)\n assert together.version >= \"1.1.0\", \"Please upgrade to Together SDK v1.1.0 or later.\"\n\n # Set Together key\n together.api_key = keys_config[\"TOGETHER_API_KEY\"]\n\n def history_to_messages(self, history: list[dict[str, str]], is_demonstration: bool = False) -> str:\n \"\"\"\n Create `prompt` by filtering out all keys except for role/content per `history` turn\n \"\"\"\n # Remove system messages if it is a demonstration\n if is_demonstration:\n history = [entry for entry in history if entry[\"role\"] != \"system\"]\n # Map history to TogetherAI format\n mapping = {\"user\": \"human\", \"assistant\": \"bot\", \"system\": \"bot\"}\n prompt = [f'<{mapping[d[\"role\"]]}>: {d[\"content\"]}' for d in history]\n prompt = \"\\n\".join(prompt)\n return f\"{prompt}\\n<bot>:\"\n\n @retry(\n wait=wait_random_exponential(min=1, max=15),\n reraise=True,\n stop=stop_after_attempt(_MAX_RETRIES),\n retry=retry_if_not_exception_type((CostLimitExceededError, RuntimeError)),\n )\n def query(self, history: list[dict[str, str]]) -> str:\n \"\"\"\n Query the Together API with the given `history` and return the response.\n \"\"\"\n # Perform Together API call\n prompt = self.history_to_messages(history)\n # Anthropic's count_tokens is convenient because it caches and utilizes huggingface/tokenizers, so we will use.\n max_tokens_to_sample = self.model_metadata[\"max_context\"] - Anthropic().count_tokens(prompt)\n completion = together.Complete.create(\n model=self.api_model,\n prompt=prompt,\n max_tokens=max_tokens_to_sample,\n stop=[\"<human>\"],\n temperature=self.args.temperature,\n top_p=self.args.top_p,\n )\n # Calculate + update costs, return response\n response = completion[\"choices\"][0][\"text\"].split(\"<human>\")[0]\n input_tokens = completion[\"usage\"][\"prompt_tokens\"]\n output_tokens = completion[\"usage\"][\"completion_tokens\"]\n self.update_stats(input_tokens, output_tokens)\n return response\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.TogetherModel.history_to_messages","title":"<code>history_to_messages(history, is_demonstration=False)</code>","text":"<p>Create <code>prompt</code> by filtering out all keys except for role/content per <code>history</code> turn</p> Source code in <code>sweagent/agent/models.py</code> <pre><code>def history_to_messages(self, history: list[dict[str, str]], is_demonstration: bool = False) -> str:\n \"\"\"\n Create `prompt` by filtering out all keys except for role/content per `history` turn\n \"\"\"\n # Remove system messages if it is a demonstration\n if is_demonstration:\n history = [entry for entry in history if entry[\"role\"] != \"system\"]\n # Map history to TogetherAI format\n mapping = {\"user\": \"human\", \"assistant\": \"bot\", \"system\": \"bot\"}\n prompt = [f'<{mapping[d[\"role\"]]}>: {d[\"content\"]}' for d in history]\n prompt = \"\\n\".join(prompt)\n return f\"{prompt}\\n<bot>:\"\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.TogetherModel.query","title":"<code>query(history)</code>","text":"<p>Query the Together API with the given <code>history</code> and return the response.</p> Source code in <code>sweagent/agent/models.py</code> <pre><code>@retry(\n wait=wait_random_exponential(min=1, max=15),\n reraise=True,\n stop=stop_after_attempt(_MAX_RETRIES),\n retry=retry_if_not_exception_type((CostLimitExceededError, RuntimeError)),\n)\ndef query(self, history: list[dict[str, str]]) -> str:\n \"\"\"\n Query the Together API with the given `history` and return the response.\n \"\"\"\n # Perform Together API call\n prompt = self.history_to_messages(history)\n # Anthropic's count_tokens is convenient because it caches and utilizes huggingface/tokenizers, so we will use.\n max_tokens_to_sample = self.model_metadata[\"max_context\"] - Anthropic().count_tokens(prompt)\n completion = together.Complete.create(\n model=self.api_model,\n prompt=prompt,\n max_tokens=max_tokens_to_sample,\n stop=[\"<human>\"],\n temperature=self.args.temperature,\n top_p=self.args.top_p,\n )\n # Calculate + update costs, return response\n response = completion[\"choices\"][0][\"text\"].split(\"<human>\")[0]\n input_tokens = completion[\"usage\"][\"prompt_tokens\"]\n output_tokens = completion[\"usage\"][\"completion_tokens\"]\n self.update_stats(input_tokens, output_tokens)\n return response\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.anthropic_history_to_messages","title":"<code>anthropic_history_to_messages(model, history, is_demonstration=False)</code>","text":"<p>Create <code>prompt</code> by filtering out all keys except for role/content per <code>history</code> turn Reference: https://docs.anthropic.com/claude/reference/complete_post</p> Source code in <code>sweagent/agent/models.py</code> <pre><code>def anthropic_history_to_messages(\n model: AnthropicModel | BedrockModel,\n history: list[dict[str, str]],\n is_demonstration: bool = False,\n) -> str | list[dict[str, str]]:\n \"\"\"\n Create `prompt` by filtering out all keys except for role/content per `history` turn\n Reference: https://docs.anthropic.com/claude/reference/complete_post\n \"\"\"\n # Preserve behavior for older models\n if model.api_model in [\"claude-instant\", \"claude-2.0\"] or (\n isinstance(model, BedrockModel) and model.api_model in [\"anthropic.claude-instant-v1\", \"anthropic.claude-v2\"]\n ):\n # Remove system messages if it is a demonstration\n if is_demonstration:\n history = [entry for entry in history if entry[\"role\"] != \"system\"]\n # Map history to Claude format\n prompt = \"\\n\\n\"\n for entry in history:\n if entry[\"role\"] in {\"user\", \"system\"}:\n prompt += f'{HUMAN_PROMPT} {entry[\"content\"]}\\n\\n'\n elif entry[\"role\"] == \"assistant\":\n prompt += f'{AI_PROMPT} {entry[\"content\"]}\\n\\n'\n prompt += AI_PROMPT\n return prompt\n\n # Remove system messages if it is a demonstration\n if is_demonstration:\n history = [entry for entry in history if entry[\"role\"] != \"system\"]\n return \"\\n\".join([entry[\"content\"] for entry in history])\n\n # Return history components with just role, content fields (no system message)\n messages = [\n {k: v for k, v in entry.items() if k in [\"role\", \"content\"]} for entry in history if entry[\"role\"] != \"system\"\n ]\n compiled_messages = [] # Combine messages from the same role\n last_role = None\n for message in reversed(messages):\n if last_role == message[\"role\"]:\n compiled_messages[-1][\"content\"] = message[\"content\"] + \"\\n\" + compiled_messages[-1][\"content\"]\n else:\n compiled_messages.append(message)\n last_role = message[\"role\"]\n compiled_messages = list(reversed(compiled_messages))\n # Replace any empty content values with a \"(No output)\"\n for message in compiled_messages:\n if message[\"content\"].strip() == \"\":\n message[\"content\"] = \"(No output)\"\n return compiled_messages\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.anthropic_query","title":"<code>anthropic_query(model, history)</code>","text":"<p>Query the Anthropic API with the given <code>history</code> and return the response.</p> Source code in <code>sweagent/agent/models.py</code> <pre><code>def anthropic_query(model: AnthropicModel | BedrockModel, history: list[dict[str, str]]) -> str:\n \"\"\"\n Query the Anthropic API with the given `history` and return the response.\n \"\"\"\n # Preserve behavior for older models\n if model.api_model in [\"claude-instant\", \"claude-2.0\", \"claude-2.1\"] or (\n isinstance(model, BedrockModel) and model.api_model in [\"anthropic.claude-instant-v1\", \"anthropic.claude-v2\"]\n ):\n # Perform Anthropic API call\n prompt = anthropic_history_to_messages(model, history)\n if isinstance(model, BedrockModel):\n # Use a dummy Anthropic client since count_tokens\n # is not available in AnthropicBedrock\n # https://github.com/anthropics/anthropic-sdk-python/issues/353\n input_tokens = Anthropic().count_tokens(prompt)\n else:\n input_tokens = model.api.count_tokens(prompt)\n completion = model.api.completions.create(\n model=model.api_model,\n prompt=prompt,\n max_tokens_to_sample=model.model_metadata[\"max_context\"] - input_tokens\n if isinstance(model, Anthropic)\n else model.model_metadata[\"max_tokens_to_sample\"],\n temperature=model.args.temperature,\n top_p=model.args.top_p,\n )\n # Calculate + update costs, return response\n response = completion.completion\n if isinstance(model, BedrockModel):\n output_tokens = Anthropic().count_tokens(response)\n else:\n output_tokens = model.api.count_tokens(response)\n model.update_stats(input_tokens, output_tokens)\n return response\n\n # Get system message(s)\n system_message = \"\\n\".join([entry[\"content\"] for entry in history if entry[\"role\"] == \"system\"])\n messages = anthropic_history_to_messages(model, history)\n\n # Perform Anthropic API call\n response = model.api.messages.create(\n messages=messages,\n max_tokens=model.model_metadata[\"max_tokens\"],\n model=model.api_model,\n temperature=model.args.temperature,\n top_p=model.args.top_p,\n system=system_message,\n )\n\n # Calculate + update costs, return response\n model.update_stats(response.usage.input_tokens, response.usage.output_tokens)\n return \"\\n\".join([x.text for x in response.content])\n</code></pre>"},{"location":"reference/models/#sweagent.agent.models.get_model","title":"<code>get_model(args, commands=None)</code>","text":"<p>Returns correct model object given arguments and commands</p> Source code in <code>sweagent/agent/models.py</code> <pre><code>def get_model(args: ModelArguments, commands: list[Command] | None = None):\n \"\"\"\n Returns correct model object given arguments and commands\n \"\"\"\n if commands is None:\n commands = []\n if args.model_name == \"instant_empty_submit\":\n return InstantEmptySubmitTestModel(args, commands)\n if args.model_name == \"human\":\n return HumanModel(args, commands)\n if args.model_name == \"human_thought\":\n return HumanThoughtModel(args, commands)\n if args.model_name == \"replay\":\n return ReplayModel(args, commands)\n elif (\n args.model_name.startswith(\"gpt\")\n or args.model_name.startswith(\"ft:gpt\")\n or args.model_name.startswith(\"azure:gpt\")\n or args.model_name in OpenAIModel.SHORTCUTS\n ):\n return OpenAIModel(args, commands)\n elif args.model_name.startswith(\"claude\"):\n return AnthropicModel(args, commands)\n elif args.model_name.startswith(\"bedrock\"):\n return BedrockModel(args, commands)\n elif args.model_name.startswith(\"ollama\"):\n return OllamaModel(args, commands)\n elif args.model_name.startswith(\"deepseek\"):\n return DeepSeekModel(args, commands)\n elif args.model_name in TogetherModel.SHORTCUTS:\n return TogetherModel(args, commands)\n elif args.model_name in GroqModel.SHORTCUTS:\n return GroqModel(args, commands)\n elif args.model_name == \"instant_empty_submit\":\n return InstantEmptySubmitTestModel(args, commands)\n else:\n msg = f\"Invalid model name: {args.model_name}\"\n raise ValueError(msg)\n</code></pre>"},{"location":"usage/","title":"Usage","text":"<p>We currently provide two interfaces to SWE-agent:</p> <ul> <li> <p> Command line interface (CLI)</p> <p>The default way of running SWE-agent with maximum options.</p> <p> Get started</p> </li> <li> <p> Graphical user interface</p> <p>We provide a browser-based graphical user interface particularly optimized for developers wanting to use SWE-agent as a tool.</p> <p> Get started</p> </li> </ul> <p>For EnIGMA usage, only the CLI is currently supported. Get started here.</p>"},{"location":"usage/benchmarking/","title":"Benchmarking","text":"<p>Scope</p> <p>This page talks about benchmarking on SWE-bench to measure the software engineering capabilities of SWE-agent. Benchmarking for the other modes (programming challenges, cybersecurity) are coming soon.</p> <p>There are two steps to the SWE-agent/SWE-bench pipeline. First SWE-agent takes an input GitHub issue and returns a pull request that attempts to fix it. We call that step inference. The second step (currently only available for issues in the SWE-bench benchmark) is to evaluate the pull request to verify that it has indeed fixed the issue.</p> <p>Architectures</p> <p>At this moment, there are known issues with a small number of repositories that don't install properly for <code>arm64</code> / <code>aarch64</code> architecture computers. We're working on a fix, but if you'd like to run and evaluate on the entirety of SWE-bench, the easiest way is by using an <code>x86</code> machine.</p>"},{"location":"usage/benchmarking/#inference","title":"\ud83d\udc69\u200d\ud83d\udcbb Inference","text":"<p>Run SWE-agent on SWE-bench Lite and generate patches.</p> <pre><code>python run.py --model_name gpt4 \\\n --per_instance_cost_limit 2.00 \\\n --config_file ./config/default.yaml\n</code></pre> <p>If you'd like to run on a single issue from SWE-bench, use the <code>--instance_filter</code> option as follows: <pre><code>python run.py --model_name gpt4 \\\n --instance_filter marshmallow-code__marshmallow-1359\n</code></pre></p> <p>The above examples use the default value of <code>--data_path</code> (<code>princeton-nlp/SWE-bench_Lite</code>, which will be looked up from huggingface). You can specify any other huggingface datasets as well, or supply the path to a pre-downloaded dataset. By default, SWE-agent evaluates on the <code>dev</code> split of that dataset. You can change that by supplying the <code>--split</code> argument to the above commands (obviously you shouldn't tune your model on the <code>test</code> dataset).</p>"},{"location":"usage/benchmarking/#evaluation","title":"\ud83e\uddea Evaluation","text":"<p>Previous <code>evaluation/</code> scripts</p> <p>We have removed the scripts that were previously in the <code>evaluation/</code> subfolder. They were relatively thin wrappers around <code>swe-bench</code>. After the <code>swe-bench</code> 2.0 update, we recommend to use <code>swe-bench</code> directly.</p> <p>You can directly run SWE-bench to evaluate the SWE-agent results.</p> <p>After installing SWE-bench, you can run <code>run_evaluation</code> as such:</p> <pre><code>python -m swebench.harness.run_evaluation \\\n --predictions_path /path/to/all_preds.jsonl \\\n --max_workers 1 \\\n --run_id test \\\n --split dev\n</code></pre> <p>Head over to SWE-bench for details.</p> <p>Default split</p> <p>When running <code>swe-agent</code> it uses the SWE-bench lite <code>dev</code> split by default (i.e., when not specifying <code>--data_path</code> or <code>--split</code>). However, <code>swe-bench</code> assumes the SWE-bench lite <code>test</code> split by default. When you get warnings from <code>swe-bench</code> about missing instances, make sure you specify <code>--split test</code> or <code>--split dev</code> appropriately.</p> <ul> <li> <p> Something broken? Report bug</p> </li> <li> <p> Something unclear? Ask question</p> </li> </ul>"},{"location":"usage/cl_tutorial/","title":"Command line usage tutorial","text":"<p>This tutorial walks you through running SWE-agent from the command line. Beginners might also be interested in the our web-based GUI (see here). This tutorial focuses on using SWE-agent as a tool to solve individual issues. Benchmarking SWE-agent is covered separately. Finally, we have a different tutorial for using SWE-agent for coding challenges.</p>"},{"location":"usage/cl_tutorial/#getting-started","title":"Getting started","text":"<p>For the CLI, use the <code>run.py</code> script. Let's start with an absolutely trivial example and solve an issue about a simple syntax error (<code>swe-agent/test-repo #1</code>)</p> <pre><code>python run.py \\\n --model_name gpt4 \\\n --data_path https://github.com/SWE-agent/test-repo/issues/1 \\\n --config_file config/default_from_url.yaml \\\n --per_instance_cost_limit 2.00\n</code></pre> Output <pre><code>INFO \ud83d\udcd9 Arguments: actions:\n apply_patch_locally: false\n open_pr: false\n push_gh_repo_url: ''\n skip_if_commits_reference_issue: true\n agent:\n config:\n _commands:\n - arguments:\n line_number:\n description: the line number to move the window to (if not provided, the\n window will start at the top of the file)\n required: false\n type: integer\n path:\n description: the path to the file to open\n required: true\n type: string\n code: 'open() { if [ -z \"$1\" ] then echo \"Usage: open <file>\" return fi #\n Check if the second argument is provided if [ -n \"$2\" ]; then #\n Check if the provided argument is a valid number if ! [[ $2 =~ ^[0-9]+$\n ]]; then echo \"Usage: open <file> [<line_number>]\" echo\n \"Error: <line_number> must be a number\" return # Exit if the line\n number is not valid fi local max_line=$(awk ''END {print NR}''\n $1) if [ $2 -gt $max_line ]; then echo \"Warning: <line_number>\n ($2) is greater than the number of lines in the file ($max_line)\" echo\n \"Warning: Setting <line_number> to $max_line\" local line_number=$(jq\n -n \"$max_line\") # Set line number to max if greater than max elif\n [ $2 -lt 1 ]; then echo \"Warning: <line_number> ($2) is less than\n 1\" echo \"Warning: Setting <line_number> to 1\" local\n line_number=$(jq -n \"1\") # Set line number to 1 if less than 1 else local\n OFFSET=$(jq -n \"$WINDOW/6\" | jq ''floor'') local line_number=$(jq\n -n \"[$2 + $WINDOW/2 - $OFFSET, 1] | max | floor\") fi else local\n line_number=$(jq -n \"$WINDOW/2\") # Set default line number if not provided fi if\n [ -f \"$1\" ]; then export CURRENT_FILE=$(realpath $1) export\n CURRENT_LINE=$line_number _constrain_line _print elif [ -d\n \"$1\" ]; then echo \"Error: $1 is a directory. You can only open files.\n Use cd or ls to navigate directories.\" else echo \"File $1 not found\" fi}'\n docstring: opens the file at the given path in the editor. If line_number is\n provided, the window will be move to include that line\n end_name: null\n name: open\n signature: open <path> [<line_number>]\n - arguments:\n line_number:\n description: the line number to move the window to\n required: true\n type: integer\n code: 'goto() { if [ $# -gt 1 ]; then echo \"goto allows only one line\n number at a time.\" return fi if [ -z \"$CURRENT_FILE\" ] then echo\n \"No file open. Use the open command first.\" return fi if [ -z\n \"$1\" ] then echo \"Usage: goto <line>\" return fi if\n ! [[ $1 =~ ^[0-9]+$ ]] then echo \"Usage: goto <line>\" echo\n \"Error: <line> must be a number\" return fi local max_line=$(awk\n ''END {print NR}'' $CURRENT_FILE) if [ $1 -gt $max_line ] then echo\n \"Error: <line> must be less than or equal to $max_line\" return fi local\n OFFSET=$(jq -n \"$WINDOW/6\" | jq ''floor'') export CURRENT_LINE=$(jq -n\n \"[$1 + $WINDOW/2 - $OFFSET, 1] | max | floor\") _constrain_line _print}'\n docstring: moves the window to show <line_number>\n end_name: null\n name: goto\n signature: goto <line_number>\n - arguments: null\n code: scroll_down() { if [ -z \"$CURRENT_FILE\" ] then echo \"No file\n open. Use the open command first.\" return fi export CURRENT_LINE=$(jq\n -n \"$CURRENT_LINE + $WINDOW - $OVERLAP\") _constrain_line _print}\n docstring: moves the window down {WINDOW} lines\n end_name: null\n name: scroll_down\n signature: scroll_down\n - arguments: null\n code: scroll_up() { if [ -z \"$CURRENT_FILE\" ] then echo \"No file\n open. Use the open command first.\" return fi export CURRENT_LINE=$(jq\n -n \"$CURRENT_LINE - $WINDOW + $OVERLAP\") _constrain_line _print}\n docstring: moves the window down {WINDOW} lines\n end_name: null\n name: scroll_up\n signature: scroll_up\n - arguments:\n filename:\n description: the name of the file to create\n required: true\n type: string\n code: \"create() { if [ -z \\\"$1\\\" ]; then echo \\\"Usage: create <filename>\\\"\\\n \\ return fi # Check if the file already exists if [ -e \\\"\\\n $1\\\" ]; then echo \\\"Error: File '$1' already exists.\\\"\\t\\topen \\\"$1\\\"\\\n \\ return fi # Create the file an empty new line printf \\\"\\\\\\\n n\\\" > \\\"$1\\\" # Use the existing open command to open the created file \\\n \\ open \\\"$1\\\"}\"\n docstring: creates and opens a new file with the given name\n end_name: null\n name: create\n signature: create <filename>\n - arguments: null\n code: 'submit() { cd $ROOT # Check if the patch file exists and is non-empty if\n [ -s \"/root/test.patch\" ]; then # Apply the patch in reverse git\n apply -R < \"/root/test.patch\" fi git add -A git diff --cached > model.patch echo\n \"<<SUBMISSION||\" cat model.patch echo \"||SUBMISSION>>\"}'\n docstring: submits your current code and terminates the session\n end_name: null\n name: submit\n signature: submit\n - arguments:\n dir:\n description: the directory to search in (if not provided, searches in the\n current directory)\n required: false\n type: string\n search_term:\n description: the term to search for\n required: true\n type: string\n code: 'search_dir() { if [ $# -eq 1 ]; then local search_term=\"$1\" local\n dir=\"./\" elif [ $# -eq 2 ]; then local search_term=\"$1\" if\n [ -d \"$2\" ]; then local dir=\"$2\" else echo \"Directory\n $2 not found\" return fi else echo \"Usage: search_dir\n <search_term> [<dir>]\" return fi dir=$(realpath \"$dir\") local\n matches=$(find \"$dir\" -type f ! -path ''*/.*'' -exec grep -nIH -- \"$search_term\"\n {} + | cut -d: -f1 | sort | uniq -c) # if no matches, return if [ -z\n \"$matches\" ]; then echo \"No matches found for \\\"$search_term\\\" in $dir\" return fi #\n Calculate total number of matches local num_matches=$(echo \"$matches\" |\n awk ''{sum+=$1} END {print sum}'') # calculate total number of files matched local\n num_files=$(echo \"$matches\" | wc -l | awk ''{$1=$1; print $0}'') # if num_files\n is > 100, print an error if [ $num_files -gt 100 ]; then echo \"More\n than $num_files files matched for \\\"$search_term\\\" in $dir. Please narrow\n your search.\" return fi echo \"Found $num_matches matches for\n \\\"$search_term\\\" in $dir:\" echo \"$matches\" | awk ''{$2=$2; gsub(/^\\.+\\/+/,\n \"./\", $2); print $2 \" (\"$1\" matches)\"}'' echo \"End of matches for \\\"$search_term\\\"\n in $dir\"}'\n docstring: searches for search_term in all files in dir. If dir is not provided,\n searches in the current directory\n end_name: null\n name: search_dir\n signature: search_dir <search_term> [<dir>]\n - arguments:\n file:\n description: the file to search in (if not provided, searches in the current\n open file)\n required: false\n type: string\n search_term:\n description: the term to search for\n required: true\n type: string\n code: 'search_file() { # Check if the first argument is provided if [\n -z \"$1\" ]; then echo \"Usage: search_file <search_term> [<file>]\" return fi #\n Check if the second argument is provided if [ -n \"$2\" ]; then #\n Check if the provided argument is a valid file if [ -f \"$2\" ]; then local\n file=\"$2\" # Set file if valid else echo \"Usage: search_file\n <search_term> [<file>]\" echo \"Error: File name $2 not found. Please\n provide a valid file name.\" return # Exit if the file is not valid fi else #\n Check if a file is open if [ -z \"$CURRENT_FILE\" ]; then echo\n \"No file open. Use the open command first.\" return # Exit if no\n file is open fi local file=\"$CURRENT_FILE\" # Set file to the\n current open file fi local search_term=\"$1\" file=$(realpath \"$file\") #\n Use grep to directly get the desired formatted output local matches=$(grep\n -nH -- \"$search_term\" \"$file\") # Check if no matches were found if [\n -z \"$matches\" ]; then echo \"No matches found for \\\"$search_term\\\" in\n $file\" return fi # Calculate total number of matches local\n num_matches=$(echo \"$matches\" | wc -l | awk ''{$1=$1; print $0}'') # calculate\n total number of lines matched local num_lines=$(echo \"$matches\" | cut -d:\n -f1 | sort | uniq | wc -l | awk ''{$1=$1; print $0}'') # if num_lines is\n > 100, print an error if [ $num_lines -gt 100 ]; then echo \"More\n than $num_lines lines matched for \\\"$search_term\\\" in $file. Please narrow\n your search.\" return fi # Print the total number of matches and\n the matches themselves echo \"Found $num_matches matches for \\\"$search_term\\\"\n in $file:\" echo \"$matches\" | cut -d: -f1-2 | sort -u -t: -k2,2n | while\n IFS=: read -r filename line_number; do echo \"Line $line_number:$(sed\n -n \"${line_number}p\" \"$file\")\" done echo \"End of matches for \\\"$search_term\\\"\n in $file\"}'\n docstring: searches for search_term in file. If file is not provided, searches\n in the current open file\n end_name: null\n name: search_file\n signature: search_file <search_term> [<file>]\n - arguments:\n dir:\n description: the directory to search in (if not provided, searches in the\n current directory)\n required: false\n type: string\n file_name:\n description: the name of the file to search for\n required: true\n type: string\n code: 'find_file() { if [ $# -eq 1 ]; then local file_name=\"$1\" local\n dir=\"./\" elif [ $# -eq 2 ]; then local file_name=\"$1\" if\n [ -d \"$2\" ]; then local dir=\"$2\" else echo \"Directory\n $2 not found\" return fi else echo \"Usage: find_file\n <file_name> [<dir>]\" return fi dir=$(realpath \"$dir\") local\n matches=$(find \"$dir\" -type f -name \"$file_name\") # if no matches, return if\n [ -z \"$matches\" ]; then echo \"No matches found for \\\"$file_name\\\" in\n $dir\" return fi # Calculate total number of matches local\n num_matches=$(echo \"$matches\" | wc -l | awk ''{$1=$1; print $0}'') echo\n \"Found $num_matches matches for \\\"$file_name\\\" in $dir:\" echo \"$matches\"\n | awk ''{print $0}''}'\n docstring: finds all files with the given name in dir. If dir is not provided,\n searches in the current directory\n end_name: null\n name: find_file\n signature: find_file <file_name> [<dir>]\n - arguments:\n end_line:\n description: the line number to end the edit at (inclusive)\n required: true\n type: integer\n replacement_text:\n description: the text to replace the current selection with\n required: true\n type: string\n start_line:\n description: the line number to start the edit at\n required: true\n type: integer\n code: 'edit() { if [ -z \"$CURRENT_FILE\" ] then echo ''No file open.\n Use the `open` command first.'' return fi local start_line=\"$(echo\n $1: | cut -d: -f1)\" local end_line=\"$(echo $1: | cut -d: -f2)\" if [\n -z \"$start_line\" ] || [ -z \"$end_line\" ] then echo \"Usage: edit\n <start_line>:<end_line>\" return fi local re=''^[0-9]+$'' if\n ! [[ $start_line =~ $re ]]; then echo \"Usage: edit <start_line>:<end_line>\" echo\n \"Error: start_line must be a number\" return fi if ! [[ $end_line\n =~ $re ]]; then echo \"Usage: edit <start_line>:<end_line>\" echo\n \"Error: end_line must be a number\" return fi # Bash array starts\n at 0, so let''s adjust local start_line=$((start_line - 1)) local end_line=$((end_line)) local\n line_count=0 local replacement=() while IFS= read -r line do replacement+=(\"$line\") ((line_count++)) done #\n Create a backup of the current file cp \"$CURRENT_FILE\" \"/root/$(basename\n \"$CURRENT_FILE\")_backup\" # Read the file line by line into an array mapfile\n -t lines < \"$CURRENT_FILE\" local new_lines=(\"${lines[@]:0:$start_line}\"\n \"${replacement[@]}\" \"${lines[@]:$((end_line))}\") # Write the new stuff\n directly back into the original file printf \"%s\\n\" \"${new_lines[@]}\" >|\n \"$CURRENT_FILE\" # Run linter if [[ $CURRENT_FILE == *.py ]]; then lint_output=$(flake8\n --isolated --select=F821,F822,F831,E111,E112,E113,E999,E902 \"$CURRENT_FILE\"\n 2>&1) else # do nothing lint_output=\"\" fi # if there\n is no output, then the file is good if [ -z \"$lint_output\" ]; then export\n CURRENT_LINE=$start_line _constrain_line _print echo\n \"File updated. Please review the changes and make sure they are correct (correct\n indentation, no duplicate lines, etc). Edit the file again if necessary.\" else echo\n \"Your proposed edit has introduced new syntax error(s). Please read this error\n message carefully and then retry editing the file.\" echo \"\" echo\n \"ERRORS:\" _split_string \"$lint_output\" echo \"\" # Save\n original values original_current_line=$CURRENT_LINE original_window=$WINDOW #\n Update values export CURRENT_LINE=$(( (line_count / 2) + start_line\n )) # Set to \"center\" of edit export WINDOW=$((line_count + 10)) # Show\n +/- 5 lines around edit echo \"This is how your edit would have looked\n if applied\" echo \"-------------------------------------------------\" _constrain_line _print echo\n \"-------------------------------------------------\" echo \"\" #\n Restoring CURRENT_FILE to original contents. cp \"/root/$(basename \"$CURRENT_FILE\")_backup\"\n \"$CURRENT_FILE\" export CURRENT_LINE=$(( ((end_line - start_line + 1)\n / 2) + start_line )) export WINDOW=$((end_line - start_line + 10)) echo\n \"This is the original code before your edit\" echo \"-------------------------------------------------\" _constrain_line _print echo\n \"-------------------------------------------------\" # Restore original\n values export CURRENT_LINE=$original_current_line export WINDOW=$original_window echo\n \"Your changes have NOT been applied. Please fix your edit command and try\n again.\" echo \"You either need to 1) Specify the correct start/end line\n arguments or 2) Correct your edit code.\" echo \"DO NOT re-run the same\n failed edit command. Running it again will lead to the same error.\" fi #\n Remove backup file rm -f \"/root/$(basename \"$CURRENT_FILE\")_backup\"}'\n docstring: replaces lines <start_line> through <end_line> (inclusive) with the\n given text in the open file. The replacement text is terminated by a line\n with only end_of_edit on it. All of the <replacement text> will be entered,\n so make sure your indentation is formatted properly. Python files will be\n checked for syntax errors after the edit. If the system detects a syntax error,\n the edit will not be executed. Simply try to edit the file again, but make\n sure to read the error message and modify the edit command you issue accordingly.\n Issuing the same command a second time will just lead to the same error message\n again.\n end_name: end_of_edit\n name: edit\n signature: |-\n edit <start_line>:<end_line>\n <replacement_text>\n end_of_edit\n _subroutines: {}\n blocklist:\n - vim\n - vi\n - emacs\n - nano\n - nohup\n - git\n blocklist_error_template: Interactive operation '{name}' is not supported by this\n environment\n blocklist_standalone:\n - python\n - python3\n - ipython\n - bash\n - sh\n - exit\n - /bin/bash\n - /bin/sh\n - nohup\n - vi\n - vim\n - emacs\n - nano\n command_docs: |+\n open:\n docstring: opens the file at the given path in the editor. If line_number is provided, the window will be move to include that line\n signature: open <path> [<line_number>]\n arguments:\n - path (string) [required]: the path to the file to open\n - line_number (integer) [optional]: the line number to move the window to (if not provided, the window will start at the top of the file)\n\n goto:\n docstring: moves the window to show <line_number>\n signature: goto <line_number>\n arguments:\n - line_number (integer) [required]: the line number to move the window to\n\n scroll_down:\n docstring: moves the window down {WINDOW} lines\n signature: scroll_down\n\n scroll_up:\n docstring: moves the window down {WINDOW} lines\n signature: scroll_up\n\n create:\n docstring: creates and opens a new file with the given name\n signature: create <filename>\n arguments:\n - filename (string) [required]: the name of the file to create\n\n submit:\n docstring: submits your current code and terminates the session\n signature: submit\n\n search_dir:\n docstring: searches for search_term in all files in dir. If dir is not provided, searches in the current directory\n signature: search_dir <search_term> [<dir>]\n arguments:\n - search_term (string) [required]: the term to search for\n - dir (string) [optional]: the directory to search in (if not provided, searches in the current directory)\n\n search_file:\n docstring: searches for search_term in file. If file is not provided, searches in the current open file\n signature: search_file <search_term> [<file>]\n arguments:\n - search_term (string) [required]: the term to search for\n - file (string) [optional]: the file to search in (if not provided, searches in the current open file)\n\n find_file:\n docstring: finds all files with the given name in dir. If dir is not provided, searches in the current directory\n signature: find_file <file_name> [<dir>]\n arguments:\n - file_name (string) [required]: the name of the file to search for\n - dir (string) [optional]: the directory to search in (if not provided, searches in the current directory)\n\n edit:\n docstring: replaces lines <start_line> through <end_line> (inclusive) with the given text in the open file. The replacement text is terminated by a line with only end_of_edit on it. All of the\n <replacement text> will be entered, so make sure your indentation is formatted properly. Python files will be checked for syntax errors after the edit. If the system detects a syntax error, the edit will\n not be executed. Simply try to edit the file again, but make sure to read the error message and modify the edit command you issue accordingly. Issuing the same command a second time will just lead to the\n same error message again.\n signature: edit <start_line>:<end_line>\n <replacement_text>\n end_of_edit\n arguments:\n - start_line (integer) [required]: the line number to start the edit at\n - end_line (integer) [required]: the line number to end the edit at (inclusive)\n - replacement_text (string) [required]: the text to replace the current selection with\n\n command_files:\n - /Users/fuchur/Documents/24/git_sync/SWE-agent/config/commands/defaults.sh\n - /Users/fuchur/Documents/24/git_sync/SWE-agent/config/commands/search.sh\n - /Users/fuchur/Documents/24/git_sync/SWE-agent/config/commands/edit_linting.sh\n - /Users/fuchur/Documents/24/git_sync/SWE-agent/config/commands/_split_string.py\n demonstration_template: |\n Here is a demonstration of how to correctly accomplish this task.\n It is included to show you how to correctly use the interface.\n You do not need to follow exactly what is done in the demonstration.\n --- DEMONSTRATION ---\n {demonstration}\n --- END OF DEMONSTRATION ---\n demonstrations:\n -\n /Users/fuchur/Documents/24/git_sync/SWE-agent/trajectories/demonstrations/replay__marshmallow-code__marshmallow-1867__default__t-0.20__p-0.95__c-2.00__install-1___install_from_source/marshmallow-code__ma\n rshmallow-1867.traj\n env_variables:\n CURRENT_FILE: ''\n CURRENT_LINE: '0'\n OVERLAP: '2'\n SEARCH_FILES: ()\n SEARCH_INDEX: '0'\n SEARCH_RESULTS: ()\n WINDOW: '100'\n format_error_template: |\n Your output was not formatted correctly. You must always include one discussion and one command as part of your response. Make sure you do not have multiple discussion/command tags.\n Please make sure your output precisely matches the following format:\n DISCUSSION\n Discuss here with yourself about what your planning and what you're going to do in this step.\n\n ```\n command(s) that you're going to run\n ```\n history_processor: {}\n history_processor_args: {}\n instance_template: |-\n We're currently solving the following issue within our repository. Here's the issue text:\n ISSUE:\n {issue}\n\n INSTRUCTIONS:\n Now, you're going to solve this issue on your own. Your terminal session has started and you're in the repository's root directory. You can use any bash commands or the special interface to help\n you. Edit all the files you need to and run any checks or tests that you want.\n Remember, YOU CAN ONLY ENTER ONE COMMAND AT A TIME. You should always wait for feedback after every command.\n When you're satisfied with all of the changes you've made, you can submit your changes to the code base by simply running the submit command.\n Note however that you cannot use any interactive session commands (e.g. python, vim) in this environment, but you can write scripts and run them. E.g. you can write a python script and then run it\n with `python <script_name>.py`.\n\n NOTE ABOUT THE EDIT COMMAND: Indentation really matters! When editing a file, make sure to insert appropriate indentation before each line!\n\n IMPORTANT TIPS:\n 1. Always start by trying to replicate the bug that the issues discusses.\n If the issue includes code for reproducing the bug, we recommend that you re-implement that in your environment, and run it to make sure you can reproduce the bug.\n Then start trying to fix it.\n When you think you've fixed the bug, re-run the bug reproduction script to make sure that the bug has indeed been fixed.\n\n If the bug reproduction script does not print anything when it successfully runs, we recommend adding a print(\"Script completed successfully, no errors.\") command at the end of the file,\n so that you can be sure that the script indeed ran fine all the way through.\n\n 2. If you run a command and it doesn't work, try running a different command. A command that did not work once will not work the second time unless you modify it!\n\n 3. If you open a file and need to get to an area around a specific line that is not in the first 100 lines, say line 583, don't just use the scroll_down command multiple times. Instead, use the\n goto 583 command. It's much quicker.\n\n 4. If the bug reproduction script requires inputting/reading a specific file, such as buggy-input.png, and you'd like to understand how to input that file, conduct a search in the existing repo\n code, to see whether someone else has already done that. Do this by running the command: find_file \"buggy-input.png\" If that doesn't work, use the linux 'find' command.\n\n 5. Always make sure to look at the currently open file and the current working directory (which appears right after the currently open file). The currently open file might be in a different\n directory than the working directory! Note that some commands, such as 'create', open files, so they might change the current open file.\n\n 6. When editing files, it is easy to accidentally specify a wrong line number or to write code with incorrect indentation. Always check the code after you issue an edit to make sure that it\n reflects what you wanted to accomplish. If it didn't, issue another command to fix it.\n\n 7. It may be necessary to install the repository from source before you can run code. Please think about how to install the environment from the repository directory if you need to do so.\n\n\n (Open file: {open_file})\n (Current directory: {working_dir})\n bash-$\n next_step_no_output_template: |-\n Your command ran successfully and did not produce any output.\n (Open file: {open_file})\n (Current directory: {working_dir})\n bash-$\n next_step_template: |-\n {observation}\n (Open file: {open_file})\n (Current directory: {working_dir})\n bash-$\n parse_command: {}\n parse_function: {}\n put_demos_in_history: false\n state_command:\n arguments: null\n code: |\n state() {\n local working_dir=\"$PWD\";\n if [ -z $CURRENT_FILE ]; then\n echo '{\"open_file\": \"n/a\", \"working_dir\": \"'$working_dir'\"}';\n else\n echo '{\"open_file\": \"'$(realpath $CURRENT_FILE)'\", \"working_dir\": \"'$working_dir'\"}';\n fi\n };\n docstring: null\n end_name: null\n name: state\n signature: null\n strategy_template: null\n submit_command: submit\n subroutine_types: []\n system_template: |-\n SETTING: You are an autonomous programmer, and you're working directly in the command line with a special interface.\n\n The special interface consists of a file editor that shows you {WINDOW} lines of a file at a time.\n In addition to typical bash commands, you can also use the following commands to help you navigate and edit files.\n\n COMMANDS:\n {command_docs}\n\n Please note that THE EDIT COMMAND REQUIRES PROPER INDENTATION.\n If you'd like to add the line ' print(x)' you must fully write that out, with all those spaces before the code! Indentation is important and code that is not indented correctly will fail and\n require fixing before it can be run.\n\n RESPONSE FORMAT:\n Your shell prompt is formatted as follows:\n (Open file: <path>) <cwd> $\n\n You need to format your output using two fields; discussion and command.\n Your output should always include _one_ discussion and _one_ command field EXACTLY as in the following example:\n DISCUSSION\n First I'll start by using ls to see what files are in the current directory. Then maybe we can look at some relevant files to see what they look like.\n ```\n ls -a\n ```\n\n You should only include a *SINGLE* command in the command section and then wait for a response from the shell before continuing with more discussion and commands. Everything you include in the\n DISCUSSION section will be saved for future reference.\n If you'd like to issue two commands at once, PLEASE DO NOT DO THAT! Please instead first submit just the first command, and then after receiving a response you'll be able to issue the second\n command.\n You're free to use any other bash commands you want (e.g. find, grep, cat, ls, cd) in addition to the special commands listed above.\n However, the environment does NOT support interactive session commands (e.g. python, vim), so please do not invoke them.\n util_functions:\n - arguments: null\n code: '_print() { local total_lines=$(awk ''END {print NR}'' $CURRENT_FILE) echo\n \"[File: $(realpath $CURRENT_FILE) ($total_lines lines total)]\" lines_above=$(jq\n -n \"$CURRENT_LINE - $WINDOW/2\" | jq ''[0, .] | max | floor'') lines_below=$(jq\n -n \"$total_lines - $CURRENT_LINE - $WINDOW/2\" | jq ''[0, .] | max | round'') if\n [ $lines_above -gt 0 ]; then echo \"($lines_above more lines above)\" fi cat\n $CURRENT_FILE | grep -n $ | head -n $(jq -n \"[$CURRENT_LINE + $WINDOW/2, $WINDOW/2]\n | max | floor\") | tail -n $(jq -n \"$WINDOW\") if [ $lines_below -gt 0 ];\n then echo \"($lines_below more lines below)\" fi}'\n docstring: null\n end_name: null\n name: _print\n signature: _print\n - arguments: null\n code: _constrain_line() { if [ -z \"$CURRENT_FILE\" ] then echo \"No\n file open. Use the open command first.\" return fi local max_line=$(awk\n 'END {print NR}' $CURRENT_FILE) local half_window=$(jq -n \"$WINDOW/2\" |\n jq 'floor') export CURRENT_LINE=$(jq -n \"[$CURRENT_LINE, $max_line - $half_window]\n | min\") export CURRENT_LINE=$(jq -n \"[$CURRENT_LINE, $half_window] | max\")}\n docstring: null\n end_name: null\n name: _constrain_line\n signature: _constrain_line\n config_file: config/default_from_url.yaml\n model:\n host_url: localhost:11434\n model_name: azure:gpt4\n per_instance_cost_limit: 2.0\n replay_path: null\n temperature: 0.0\n top_p: 0.95\n total_cost_limit: 0.0\n environment:\n base_commit: null\n cache_task_images: false\n container_name: null\n data_path: https://github.com/SWE-agent/test-repo/issues/1\n environment_setup: null\n image_name: sweagent/swe-agent:latest\n install_environment: true\n no_mirror: false\n repo_path: ''\n split: dev\n timeout: null\n verbose: true\n instance_filter: .*\n print_config: true\n raise_exceptions: false\n skip_existing: true\n suffix: ''\n\nINFO Base commit reference None resolved to commit hash 8c179cd2be750cd9f2bb91b21adb39934311e9b8\nINFO \ud83d\udcbd Loaded dataset from https://github.com/SWE-agent/test-repo/issues/1\nINFO Found image sweagent/swe-agent:latest with tags: ['sweagent/swe-agent:latest'], created: 2024-06-05T01:13:45.176471384Z for linux arm64.\nDEBUG Starting container with command: docker run -i --rm --name sweagent-swe-agent-latest-01edf87adc sweagent/swe-agent:latest /bin/bash -l\nINFO \ud83c\udf31 Environment Initialized\nDEBUG Environment initialization took 2.09 seconds\nINFO \u25b6\ufe0f Beginning task 0\nINFO Trying to clone from non-mirror...\nWARNING install_environment is set to True, but the data path is a GitHub URL without an environment config file (environment_config key/flag). Skipping conda environment installation.\nINFO Initializing agent settings for container 26cd13d1f31252475cde7e1ae1981d11f43e88d2066c2532611f5f2182d42737\nINFO Resetting model stats\nINFO SYSTEM (primary)\n SETTING: You are an autonomous programmer, and you're working directly in the command line with a special interface.\n\n The special interface consists of a file editor that shows you 100 lines of a file at a time.\n In addition to typical bash commands, you can also use the following commands to help you navigate and edit files.\n\n COMMANDS:\n open:\n docstring: opens the file at the given path in the editor. If line_number is provided, the window will be move to include that line\n signature: open <path> [<line_number>]\n arguments:\n - path (string) [required]: the path to the file to open\n - line_number (integer) [optional]: the line number to move the window to (if not provided, the window will start at the top of the file)\n\n goto:\n docstring: moves the window to show <line_number>\n signature: goto <line_number>\n arguments:\n - line_number (integer) [required]: the line number to move the window to\n\n scroll_down:\n docstring: moves the window down {WINDOW} lines\n signature: scroll_down\n\n scroll_up:\n docstring: moves the window down {WINDOW} lines\n signature: scroll_up\n\n create:\n docstring: creates and opens a new file with the given name\n signature: create <filename>\n arguments:\n - filename (string) [required]: the name of the file to create\n\n submit:\n docstring: submits your current code and terminates the session\n signature: submit\n\n search_dir:\n docstring: searches for search_term in all files in dir. If dir is not provided, searches in the current directory\n signature: search_dir <search_term> [<dir>]\n arguments:\n - search_term (string) [required]: the term to search for\n - dir (string) [optional]: the directory to search in (if not provided, searches in the current directory)\n\n search_file:\n docstring: searches for search_term in file. If file is not provided, searches in the current open file\n signature: search_file <search_term> [<file>]\n arguments:\n - search_term (string) [required]: the term to search for\n - file (string) [optional]: the file to search in (if not provided, searches in the current open file)\n\n find_file:\n docstring: finds all files with the given name in dir. If dir is not provided, searches in the current directory\n signature: find_file <file_name> [<dir>]\n arguments:\n - file_name (string) [required]: the name of the file to search for\n - dir (string) [optional]: the directory to search in (if not provided, searches in the current directory)\n\n edit:\n docstring: replaces lines <start_line> through <end_line> (inclusive) with the given text in the open file. The replacement text is terminated by a line with only end_of_edit on it. All of the\n <replacement text> will be entered, so make sure your indentation is formatted properly. Python files will be checked for syntax errors after the edit. If the system detects a syntax error, the edit will\n not be executed. Simply try to edit the file again, but make sure to read the error message and modify the edit command you issue accordingly. Issuing the same command a second time will just lead to the\n same error message again.\n signature: edit <start_line>:<end_line>\n <replacement_text>\n end_of_edit\n arguments:\n - start_line (integer) [required]: the line number to start the edit at\n - end_line (integer) [required]: the line number to end the edit at (inclusive)\n - replacement_text (string) [required]: the text to replace the current selection with\n\n\n\n Please note that THE EDIT COMMAND REQUIRES PROPER INDENTATION.\n If you'd like to add the line ' print(x)' you must fully write that out, with all those spaces before the code! Indentation is important and code that is not indented correctly will fail and\n require fixing before it can be run.\n\n RESPONSE FORMAT:\n Your shell prompt is formatted as follows:\n (Open file: <path>) <cwd> $\n\n You need to format your output using two fields; discussion and command.\n Your output should always include _one_ discussion and _one_ command field EXACTLY as in the following example:\n DISCUSSION\n First I'll start by using ls to see what files are in the current directory. Then maybe we can look at some relevant files to see what they look like.\n ```\n ls -a\n ```\n\n You should only include a *SINGLE* command in the command section and then wait for a response from the shell before continuing with more discussion and commands. Everything you include in the DISCUSSION\n section will be saved for future reference.\n If you'd like to issue two commands at once, PLEASE DO NOT DO THAT! Please instead first submit just the first command, and then after receiving a response you'll be able to issue the second command.\n You're free to use any other bash commands you want (e.g. find, grep, cat, ls, cd) in addition to the special commands listed above.\n However, the environment does NOT support interactive session commands (e.g. python, vim), so please do not invoke them.\nINFO DEMONSTRATION:\n /Users/fuchur/Documents/24/git_sync/SWE-agent/trajectories/demonstrations/replay__marshmallow-code__marshmallow-1867__default__t-0.20__p-0.95__c-2.00__install-1___install_from_source/marshmallow-code__ma\n rshmallow-1867.traj\nINFO Trajectory will be saved to trajectories/fuchur/azure-gpt4__SWE-agent__test-repo__default_from_url__t-0.00__p-0.95__c-2.00__install-1/SWE-agent__test-repo-i1.traj\nINFO \ud83e\udd16 MODEL INPUT\n We're currently solving the following issue within our repository. Here's the issue text:\n ISSUE:\n SyntaxError: invalid syntax\n I'm running `missing_colon.py` as follows:\n\n ```python\n division(23, 0)\n ```\n\n but I get the following error:\n\n ```\n File \"/Users/fuchur/Documents/24/git_sync/swe-agent-test-repo/tests/./missing_colon.py\", line 4\n def division(a: float, b: float) -> float\n ^\n SyntaxError: invalid syntax\n ```\n\n\n INSTRUCTIONS:\n Now, you're going to solve this issue on your own. Your terminal session has started and you're in the repository's root directory. You can use any bash commands or the special interface to help you.\n Edit all the files you need to and run any checks or tests that you want.\n Remember, YOU CAN ONLY ENTER ONE COMMAND AT A TIME. You should always wait for feedback after every command.\n When you're satisfied with all of the changes you've made, you can submit your changes to the code base by simply running the submit command.\n Note however that you cannot use any interactive session commands (e.g. python, vim) in this environment, but you can write scripts and run them. E.g. you can write a python script and then run it with\n `python <script_name>.py`.\n\n NOTE ABOUT THE EDIT COMMAND: Indentation really matters! When editing a file, make sure to insert appropriate indentation before each line!\n\n IMPORTANT TIPS:\n 1. Always start by trying to replicate the bug that the issues discusses.\n If the issue includes code for reproducing the bug, we recommend that you re-implement that in your environment, and run it to make sure you can reproduce the bug.\n Then start trying to fix it.\n When you think you've fixed the bug, re-run the bug reproduction script to make sure that the bug has indeed been fixed.\n\n If the bug reproduction script does not print anything when it successfully runs, we recommend adding a print(\"Script completed successfully, no errors.\") command at the end of the file,\n so that you can be sure that the script indeed ran fine all the way through.\n\n 2. If you run a command and it doesn't work, try running a different command. A command that did not work once will not work the second time unless you modify it!\n\n 3. If you open a file and need to get to an area around a specific line that is not in the first 100 lines, say line 583, don't just use the scroll_down command multiple times. Instead, use the goto 583\n command. It's much quicker.\n\n 4. If the bug reproduction script requires inputting/reading a specific file, such as buggy-input.png, and you'd like to understand how to input that file, conduct a search in the existing repo code, to\n see whether someone else has already done that. Do this by running the command: find_file \"buggy-input.png\" If that doesn't work, use the linux 'find' command.\n\n 5. Always make sure to look at the currently open file and the current working directory (which appears right after the currently open file). The currently open file might be in a different directory\n than the working directory! Note that some commands, such as 'create', open files, so they might change the current open file.\n\n 6. When editing files, it is easy to accidentally specify a wrong line number or to write code with incorrect indentation. Always check the code after you issue an edit to make sure that it reflects what\n you wanted to accomplish. If it didn't, issue another command to fix it.\n\n 7. It may be necessary to install the repository from source before you can run code. Please think about how to install the environment from the repository directory if you need to do so.\n\n\n (Open file: n/a)\n (Current directory: /SWE-agent__test-repo)\n bash-$\nINFO input_tokens=10,194, output_tokens=56, instance_cost=0.10, cost=0.10\nINFO total_tokens_sent=10,194, total_tokens_received=56, total_cost=0.10, total_api_calls=1\nINFO \ud83d\udcad THOUGHT (primary)\n First, let's locate the file `missing_colon.py` mentioned in the issue to understand the syntax error and fix it. We'll use the `find_file` command to search for this file in the current directory.\n\nINFO \ud83c\udfac ACTION (primary)\n find_file \"missing_colon.py\"\n\nINFO \ud83e\udd16 MODEL INPUT\n Found 1 matches for \"missing_colon.py\" in /SWE-agent__test-repo:\n /SWE-agent__test-repo/tests/missing_colon.py\n (Open file: n/a)\n (Current directory: /SWE-agent__test-repo)\n bash-$\nINFO input_tokens=10,314, output_tokens=45, instance_cost=0.21, cost=0.10\nINFO total_tokens_sent=20,508, total_tokens_received=101, total_cost=0.21, total_api_calls=2\nINFO \ud83d\udcad THOUGHT (primary)\n Now that we've located the file `missing_colon.py` in the `tests` directory, let's open it to inspect the syntax error mentioned in the issue.\n\nINFO \ud83c\udfac ACTION (primary)\n open tests/missing_colon.py\n\nINFO \ud83e\udd16 MODEL INPUT\n [File: /SWE-agent__test-repo/tests/missing_colon.py (10 lines total)]\n 1:#!/usr/bin/env python3\n 2:\n 3:\n 4:def division(a: float, b: float) -> float\n 5: return a/b\n 6:\n 7:\n 8:if __name__ == \"__main__\":\n 9: print(division(123, 15))\n 10:\n (Open file: /SWE-agent__test-repo/tests/missing_colon.py)\n (Current directory: /SWE-agent__test-repo)\n bash-$\nINFO input_tokens=10,484, output_tokens=59, instance_cost=0.31, cost=0.11\nINFO total_tokens_sent=30,992, total_tokens_received=160, total_cost=0.31, total_api_calls=3\nINFO \ud83d\udcad THOUGHT (primary)\n The syntax error is due to the missing colon at the end of the function definition on line 4. Let's add the missing colon to correct the syntax error.\n\nINFO \ud83c\udfac ACTION (primary)\n edit 4:4\n def division(a: float, b: float) -> float:\n end_of_edit\n\nINFO \ud83e\udd16 MODEL INPUT\n [File: /SWE-agent__test-repo/tests/missing_colon.py (10 lines total)]\n 1:#!/usr/bin/env python3\n 2:\n 3:\n 4:def division(a: float, b: float) -> float:\n 5: return a/b\n 6:\n 7:\n 8:if __name__ == \"__main__\":\n 9: print(division(123, 15))\n 10:\n File updated. Please review the changes and make sure they are correct (correct indentation, no duplicate lines, etc). Edit the file again if necessary.\n (Open file: /SWE-agent__test-repo/tests/missing_colon.py)\n (Current directory: /SWE-agent__test-repo)\n bash-$\nINFO input_tokens=10,698, output_tokens=47, instance_cost=0.42, cost=0.11\nINFO total_tokens_sent=41,690, total_tokens_received=207, total_cost=0.42, total_api_calls=4\nINFO \ud83d\udcad THOUGHT (primary)\n The syntax error has been corrected by adding the missing colon at the end of the function definition. Now, let's run the script to ensure that it executes without any syntax errors.\n\nINFO \ud83c\udfac ACTION (primary)\n python tests/missing_colon.py\n\nINFO \ud83e\udd16 MODEL INPUT\n 8.2\n (Open file: /SWE-agent__test-repo/tests/missing_colon.py)\n (Current directory: /SWE-agent__test-repo)\n bash-$\nINFO input_tokens=10,790, output_tokens=45, instance_cost=0.53, cost=0.11\nINFO total_tokens_sent=52,480, total_tokens_received=252, total_cost=0.53, total_api_calls=5\nINFO \ud83d\udcad THOUGHT (primary)\n The script executed successfully and printed the result of the division function, which confirms that the syntax error has been resolved. Now that the issue is fixed, we can submit the changes to the\n code base.\n\nINFO \ud83c\udfac ACTION (primary)\n submit\n\nINFO Found submission:\n diff --git a/tests/missing_colon.py b/tests/missing_colon.py\n index 20edef5..5857437 100755\n --- a/tests/missing_colon.py\n +++ b/tests/missing_colon.py\n @@ -1,7 +1,7 @@\n #!/usr/bin/env python3\n\n\n -def division(a: float, b: float) -> float\n +def division(a: float, b: float) -> float:\n return a/b\n\n\n\nINFO Trajectory saved to trajectories/fuchur/azure-gpt4__SWE-agent__test-repo__default_from_url__t-0.00__p-0.95__c-2.00__install-1/SWE-agent__test-repo-i1.traj\nINFO Saved predictions to trajectories/fuchur/azure-gpt4__SWE-agent__test-repo__default_from_url__t-0.00__p-0.95__c-2.00__install-1/all_preds.jsonl\n\u256d\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500 \ud83c\udf89 Submission successful \ud83c\udf89 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u256e\n\u2502 SWE-agent has produced a patch that it believes will solve the issue you submitted! \u2502\n\u2502 Use the code snippet below to inspect or apply it! \u2502\n\u2570\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u256f\n\n # The patch has been saved to your local filesystem at:\n PATCH_FILE_PATH='/Users/fuchur/Documents/24/git_sync/SWE-agent/trajectories/fuchur/azure-gpt4__SWE-agent__test-repo__default_from_url__t-0.00__p-0.95__c-2.00__install-1/patches/SWE-agent__test-repo-i1.patch'\n # Inspect it:\n cat \"${PATCH_FILE_PATH}\"\n # Apply it to a local repository:\n cd <your local repo root>\n git apply \"${PATCH_FILE_PATH}\"\n</code></pre> <p>Here,</p> <ul> <li><code>--model_name</code> sets the language model that is used by SWE-agent (with <code>gpt4</code> being the default). More information on the available models in our FAQ</li> <li><code>--data_path</code> points to the source of the problem statement (for example, the GitHub issue that you want to solve). You can also point it to local files (see below)</li> <li><code>--config_file</code> includes settings such as the prompts. Changing the config file is the easiest way to get started with modifying SWE-agent (more advanced options are discussed here).</li> <li><code>--per_instance_cost_limit</code> limits the total inference cost to $2 (default is $3).</li> </ul> <p>All options</p> <p>Run <code>python run.py --help</code> to see all available options for <code>run.py</code>. This tutorial will only cover a subset of options.</p> <p>Running more than once</p> <ul> <li>The complete details of the run are saved as a \"trajectory\" file (more about them here). They can also be turned into new demonstrations.</li> <li>If you run the same command more than once, you will find that SWE-agent aborts with <code>Skipping existing trajectory</code>. You can either remove the trajectory from the warning message, or add the <code>--skip_existing=False</code> flag.</li> <li>If you solve multiple issues from the same repository/in the same environment, you can specify the <code>--cache_task_images</code> flag. This will create a persistent docker image with the initialized environment required for the problem.</li> </ul>"},{"location":"usage/cl_tutorial/#specifying-the-repository","title":"Specifying the repository","text":"<p>Operating in batch mode: Running on SWE-bench and other benchmark sets</p> <p>If you want to run SWE-agent in batch mode on SWE-bench or another whole evaluation set, see benchmarking. This tutorial focuses on using SWE-agent on individual issues.</p> <p>In the above example, the repository/codebase is inferred from the <code>--data_path</code>. This options is currently only available for GitHub issues. For all other use cases, you can specify <code>--repo_path</code>, which accepts either GitHub URLs or paths to local repositories.</p> <p>To try it out, let's clone the test repository from the previous section.</p> <pre><code>git clone git@github.com:SWE-agent/test-repo.git\n</code></pre> <p>and then run</p> <pre><code>python run.py \\\n --data_path /path/to/test-repo/problem_statements/1.md \\\n --repo_path /path/to/test-repo \\\n --config_file config/default_from_url.yaml \\\n --apply_patch_locally\n</code></pre> <p>where you replaced paths with the prefix <code>/path/to/.../</code> with the actual paths to the corresponding file/directory.</p> <p>We have also added a new flag, <code>--apply_patch_locally</code>, which will make SWE-agent apply the changes to the local repository (if it believes that it has successfully solved the issue).</p> <p>You can mix and match the different ways of specifying problem statements and repositories. For example, any of the following combination of options also works</p> <ul> <li>Local problem statement with GitHub repository (<code>--data_path /path/to/problem.md --repo_path https://github.com/...</code>): Let SWE-agent work on something that wasn't reported yet</li> <li>GitHub issue with local repository (<code>--data_path https://github.com/.../issues/.. --repo_path /path/to/... --apply_patch_locally</code>): Let SWE-agent solve a GitHub issue locally (for example to edit the solution afterwards)</li> <li>GitHub issue with different GitHub repository: Useful with the <code>--open_pr</code> flag (see below) when working from a fork.</li> </ul> <p>In addition, if <code>--repo_path</code> points to a GitHub repository, you can use <code>--base_commit</code> to specify</p> <ul> <li>A branch name (e.g., <code>dev</code>),</li> <li>A tag (e.g., <code>v1.0.0</code>),</li> <li>A commit hash (e.g., <code>a4464baca1f28d7733337df6e4daa6c1ed920336</code>).</li> </ul> <p>SWE-agent will then start from this commit when trying to solve the problem.</p> <p>Uncommitted changes</p> <p>When running with a local <code>--repo_path</code>, SWE-agent will use the last commit, i.e., all local, uncommitted changes will not be seen by SWE-agent.</p>"},{"location":"usage/cl_tutorial/#installing-dependencies-and-setting-up-the-environment","title":"Installing dependencies and setting up the environment","text":"<p>Now let's move on to a slightly more complicated issue (<code>swe-agent/test-repo #22</code>).</p> <p>What makes it more complicated? This time the problematic code is part of a library <code>testpkg</code>, so SWE-agent first has to install the package in order to reproduce the issue before searching for the problematic code.</p> <p>In most circumstances, GPT4 will attempt to install the package and requirements (usually with some form of <code>pip install .</code> or <code>pip install pkg</code>). However, this wastes valuable queries to the LM. In addition, you might need to run your software for a specific python version or have other specific environment settings. The <code>--environment_setup</code> flag is used to fix this problem.</p> <p>Let's try it:</p> <pre><code>python run.py \\\n --data_path https://github.com/SWE-agent/test-repo/issues/22 \\\n --config_file config/default_from_url.yaml \\\n --environment_setup config/environment_setup/py310_default.yaml\n</code></pre> <p>This time, <code>pip install -e .</code> is called before SWE-agent gets to work, installing the package defined in the repository.</p> <p>Let's take a look at the <code>py310_default.yaml</code> config file</p> <pre><code>python: '3.10'\n# Use uv pip for speedup, but fallback to classic pip if we fail\n# Upgrade pip to avoid https://stackoverflow.com/a/73779542/\ninstall: 'uv pip install -e . || (python -m pip install --upgrade pip && python -m pip install -e .)'\n</code></pre> <p>Here, <code>install</code> is an arbitrary command that is run, while <code>python</code> will be the required python version. The default install command will create an editable install of the python package. We first try to use <code>uv pip</code> (a much faster implementation of <code>pip</code> in rust), but fall back to \"normal\" pip if it fails.</p> <p>Editable installs</p> <p>Using editable installs is crucial for SWE-agent so that changes to the package code take effect without having to reinstall the package.</p> <p>The config file can have the following keys:</p> <ul> <li><code>python</code>: Python version (will be set up via conda)</li> <li><code>packages</code>: Either <code>requirements.txt</code>, <code>environment.yml</code> (finds the corresponding file and installs from there) or a whitespace separated list of conda packages</li> <li><code>pip_packages</code>: A list of additional python packages that are installed with <code>pip install PACKAGE</code></li> <li><code>pre_install</code>: A list of custom commands</li> <li><code>install</code>: A custom command</li> <li><code>post_install</code>: A list of custom commands</li> </ul> <p>Instead of the <code>setup.yaml</code> file, you can also directly specify the path to a shell script, e.g., <code>--environment_setup /path/to/setup.sh</code>. If you have very specific requirements, that can not be installed via conda, you can create a custom Docker image.</p>"},{"location":"usage/cl_tutorial/#installing-non-python-dependencies","title":"Installing non-python dependencies","text":"<p>While SWE-agent has so far only been benchmarked and optimized for python project, you can still use it on repositories of any language.</p> <p>In most cases, this means creating a custom Docker image. However, you can for example install node dependencies with <code>--environment_setup setup.sh</code>, where <code>setup.sh</code> looks as follows:</p> <pre><code>apt-get update\nyes|apt-get install curl\nyes|curl -L https://bit.ly/n-install | bash\n/root/n/bin/n latest\nnpm install\n</code></pre> <p>However, this will take some time, so make sure to cache the environment (see the next section) or create a custom Docker image.</p>"},{"location":"usage/cl_tutorial/#speeding-up-swe-agent","title":"Speeding up SWE-agent","text":"<p>Speed up in v0.6</p> <p>SWE-agent v0.6 (June 4th, 2024) introduced major speedups. Please upgrade to the latest version. To make use of <code>uv pip</code>, make sure that you have the latest <code>sweagent/swe-agent:latest</code> image.</p> <p>After the Docker container has been started, the target repository cloned/copied, and the dependencies installed, almost all of the remaining runtime can be attributed to waiting for your LM provider to answer your API calls.</p> <p>Therefore, speeding SWE-agent is mostly about speeding up the setup stages. We currently offer three ways to cache the setup stages:</p> <ul> <li>By specifying <code>--container_name</code>, you run SWE-agent with a persistent Docker container: Rather than being deleted after every task, the Docker container will only be paused and can be resumed. Cloned repositories from previous runs with the same container name, as well as any installed conda environments (versioned by the version of the package you are installing) will be already available.</li> <li>Alternatively, you can specify <code>--cache_task_images</code>. For every repository/base commit/environment setup, we commit the changes from the installation stage to the Docker image. The corresponding containers are temporary as usual. Unlike the persistent containers, there will be a new image for almost every base commit (that is, probably for every task when evaluating on a benchmark), which makes this only relevant when running over the same tasks more than once (for example when testing different agent configurations or LMs).</li> <li>You can also build your own Docker image and ensure that all relevant conda environments and repositories are available (check the logs from the previous runs to get the names for repositories and environments).</li> </ul> <p>Confused about the two options?</p> <p>Probably <code>--container_name my_container_name</code> will do what you want.</p> <p>What's the difference between Docker images and containers?</p> <p>Docker containers are running instances of Docker images (that you can think of as snapshots of what happens after you build the <code>Dockerfile</code>). More information.</p>"},{"location":"usage/cl_tutorial/#taking-actions","title":"Taking actions","text":"<ul> <li>As mentioned above, you can use <code>--apply_patch_locally</code> to have SWE-agent apply successful solution attempts to local files.</li> <li>Alternatively, when running on a GitHub issue, you can have the agent automatically open a PR if the issue has been solved by supplying the <code>--open_pr</code> flag. Please use this feature responsibly (on your own repositories or after careful consideration).</li> </ul> <p>Alternatively, you can always retrieve the patch that was generated by SWE-agent. Watch out for the following message in the log:</p> <pre><code>\u256d\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500 \ud83c\udf89 Submission successful \ud83c\udf89 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u256e\n\u2502 SWE-agent has produced a patch that it believes will solve the issue you submitted! \u2502\n\u2502 Use the code snippet below to inspect or apply it! \u2502\n\u2570\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u256f\n</code></pre> <p>And follow the instructions below it:</p> <pre><code> # The patch has been saved to your local filesystem at:\n PATCH_FILE_PATH='/Users/.../patches/05917d.patch'\n # Inspect it:\n cat \"${PATCH_FILE_PATH}\"\n # Apply it to a local repository:\n cd <your local repo root>\n git apply \"${PATCH_FILE_PATH}\"\n</code></pre> <ul> <li> <p> Something broken? Report bug</p> </li> <li> <p> Something unclear? Ask question</p> </li> </ul>"},{"location":"usage/coding_challenges/","title":"Using SWE-agent for coding challenges","text":"<p>Command line tutorial</p> <p>We also provide a more general command line tutorial, which covers many more advanced features of SWE-agent, but focuses on its use for software engineering problems.</p> <p>It is easy to use SWE-agent to do more than just software engineering. For example, you can tell SWE-agent to work on leetcode or humaneval-style problems.</p> <p>For this, put the problem you want to solve in a markdown file <code>problem.md</code>, for example:</p> Example leetcode challenge This is the first missing positive challenge. <pre><code>Given an unsorted integer array nums.\nReturn the smallest positive integer that is not present in nums.\n\nYou must implement an algorithm that runs in O(n) time and uses O(1) auxiliary space.\n\n## Example 1:\n\n> Input: nums = [1,2,0]\n> Output: 3\n> Explanation: The numbers in the range [1,2] are all in the array.\n\n## Example 2:\n\n> Input: nums = [3,4,-1,1]\n> Output: 2\n> Explanation: 1 is in the array but 2 is missing.\n\n## Example 3:\n\n> Input: nums = [7,8,9,11,12]\n> Output: 1\n> Explanation: The smallest positive integer 1 is missing.\n\n## Constraints:\n\n1 <= nums.length <= 105\n-231 <= nums[i] <= 231 - 1\n</code></pre> <p>Second, we need to specify a repository wherein SWE-agent will work. Here, we can simply create an empty folder (outside of the SWE-agent repository), and add a <code>main.py</code> file</p> <pre><code>mkdir empty\ngit init\ntouch main.py\necho \"*.pyc\" > .gitignore # to avoid binary files in patches\n</code></pre> <p>and potentially populate it with the problem stub</p> <pre><code>from typing import List\n\n\nclass Solution:\n def firstMissingPositive(self, nums: List[int]) -> int:\n</code></pre> <p>Tip</p> <p>If some imports (like <code>List</code>) are missing in the problem stub (like they oftentimes do in leetcode) , SWE-agent will figure out how to add them. However, it might take an additional step, so it's best to directly specify them.</p> <p>Make sure to commit all changes to the repository:</p> <pre><code>git add . && git commit -m \"Add problem stub\"\n</code></pre> <p>Now, we can let SWE-agent solve the problem:</p> <pre><code>python run.py \\\n --data_path problem.md \\\n --repo_path /path/to/empty \\\n --config_file config/coding_challenge.yaml \\\n --model gpt4 \\\n --per_instance_cost_limit 3.0 \\\n --apply_patch_locally\n</code></pre> Output <pre><code>2024-07-12 17:57:39,876 INFO \ud83d\udcd9 Arguments: actions:\n apply_patch_locally: false\n open_pr: false\n push_gh_repo_url: ''\n skip_if_commits_reference_issue: true\nagent:\n config:\n _commands:\n - arguments:\n line_number:\n description: the line number to move the window to (if not provided, the\n window will start at the top of the file)\n required: false\n type: integer\n path:\n description: the path to the file to open\n required: true\n type: string\n code: 'open() { if [ -z \"$1\" ] then echo \"Usage: open <file>\" return fi #\n Check if the second argument is provided if [ -n \"$2\" ]; then #\n Check if the provided argument is a valid number if ! [[ $2 =~ ^[0-9]+$\n ]]; then echo \"Usage: open <file> [<line_number>]\" echo\n \"Error: <line_number> must be a number\" return # Exit if the line\n number is not valid fi local max_line=$(awk ''END {print NR}''\n $1) if [ $2 -gt $max_line ]; then echo \"Warning: <line_number>\n ($2) is greater than the number of lines in the file ($max_line)\" echo\n \"Warning: Setting <line_number> to $max_line\" local line_number=$(jq\n -n \"$max_line\") # Set line number to max if greater than max elif\n [ $2 -lt 1 ]; then echo \"Warning: <line_number> ($2) is less than\n 1\" echo \"Warning: Setting <line_number> to 1\" local\n line_number=$(jq -n \"1\") # Set line number to 1 if less than 1 else local\n OFFSET=$(jq -n \"$WINDOW/6\" | jq ''floor'') local line_number=$(jq\n -n \"[$2 + $WINDOW/2 - $OFFSET, 1] | max | floor\") fi else local\n line_number=$(jq -n \"$WINDOW/2\") # Set default line number if not provided fi if\n [ -f \"$1\" ]; then export CURRENT_FILE=$(realpath $1) export\n CURRENT_LINE=$line_number _constrain_line _print elif [ -d\n \"$1\" ]; then echo \"Error: $1 is a directory. You can only open files.\n Use cd or ls to navigate directories.\" else echo \"File $1 not found\" fi}'\n docstring: opens the file at the given path in the editor. If line_number is\n provided, the window will be move to include that line\n end_name: null\n name: open\n signature: open <path> [<line_number>]\n - arguments:\n line_number:\n description: the line number to move the window to\n required: true\n type: integer\n code: 'goto() { if [ $# -gt 1 ]; then echo \"goto allows only one line\n number at a time.\" return fi if [ -z \"$CURRENT_FILE\" ] then echo\n \"No file open. Use the open command first.\" return fi if [ -z\n \"$1\" ] then echo \"Usage: goto <line>\" return fi if\n ! [[ $1 =~ ^[0-9]+$ ]] then echo \"Usage: goto <line>\" echo\n \"Error: <line> must be a number\" return fi local max_line=$(awk\n ''END {print NR}'' $CURRENT_FILE) if [ $1 -gt $max_line ] then echo\n \"Error: <line> must be less than or equal to $max_line\" return fi local\n OFFSET=$(jq -n \"$WINDOW/6\" | jq ''floor'') export CURRENT_LINE=$(jq -n\n \"[$1 + $WINDOW/2 - $OFFSET, 1] | max | floor\") _constrain_line _print}'\n docstring: moves the window to show <line_number>\n end_name: null\n name: goto\n signature: goto <line_number>\n - arguments: null\n code: scroll_down() { if [ -z \"$CURRENT_FILE\" ] then echo \"No file\n open. Use the open command first.\" return fi export CURRENT_LINE=$(jq\n -n \"$CURRENT_LINE + $WINDOW - $OVERLAP\") _constrain_line _print _scroll_warning_message}\n docstring: moves the window down {WINDOW} lines\n end_name: null\n name: scroll_down\n signature: scroll_down\n - arguments: null\n code: scroll_up() { if [ -z \"$CURRENT_FILE\" ] then echo \"No file\n open. Use the open command first.\" return fi export CURRENT_LINE=$(jq\n -n \"$CURRENT_LINE - $WINDOW + $OVERLAP\") _constrain_line _print _scroll_warning_message}\n docstring: moves the window down {WINDOW} lines\n end_name: null\n name: scroll_up\n signature: scroll_up\n - arguments:\n filename:\n description: the name of the file to create\n required: true\n type: string\n code: \"create() { if [ -z \\\"$1\\\" ]; then echo \\\"Usage: create <filename>\\\"\\\n \\ return fi # Check if the file already exists if [ -e \\\"\\\n $1\\\" ]; then echo \\\"Error: File '$1' already exists.\\\"\\t\\topen \\\"$1\\\"\\\n \\ return fi # Create the file an empty new line printf \\\"\\\\\\\n n\\\" > \\\"$1\\\" # Use the existing open command to open the created file \\\n \\ open \\\"$1\\\"}\"\n docstring: creates and opens a new file with the given name\n end_name: null\n name: create\n signature: create <filename>\n - arguments: null\n code: 'submit() { cd $ROOT # Check if the patch file exists and is non-empty if\n [ -s \"/root/test.patch\" ]; then # Apply the patch in reverse git\n apply -R < \"/root/test.patch\" fi git add -A git diff --cached > model.patch echo\n \"<<SUBMISSION||\" cat model.patch echo \"||SUBMISSION>>\"}'\n docstring: submits your current code and terminates the session\n end_name: null\n name: submit\n signature: submit\n - arguments:\n dir:\n description: the directory to search in (if not provided, searches in the\n current directory)\n required: false\n type: string\n search_term:\n description: the term to search for\n required: true\n type: string\n code: 'search_dir() { if [ $# -eq 1 ]; then local search_term=\"$1\" local\n dir=\"./\" elif [ $# -eq 2 ]; then local search_term=\"$1\" if\n [ -d \"$2\" ]; then local dir=\"$2\" else echo \"Directory\n $2 not found\" return fi else echo \"Usage: search_dir\n <search_term> [<dir>]\" return fi dir=$(realpath \"$dir\") local\n matches=$(find \"$dir\" -type f ! -path ''*/.*'' -exec grep -nIH -- \"$search_term\"\n {} + | cut -d: -f1 | sort | uniq -c) # if no matches, return if [ -z\n \"$matches\" ]; then echo \"No matches found for \\\"$search_term\\\" in $dir\" return fi #\n Calculate total number of matches local num_matches=$(echo \"$matches\" |\n awk ''{sum+=$1} END {print sum}'') # calculate total number of files matched local\n num_files=$(echo \"$matches\" | wc -l | awk ''{$1=$1; print $0}'') # if num_files\n is > 100, print an error if [ $num_files -gt 100 ]; then echo \"More\n than $num_files files matched for \\\"$search_term\\\" in $dir. Please narrow\n your search.\" return fi echo \"Found $num_matches matches for\n \\\"$search_term\\\" in $dir:\" echo \"$matches\" | awk ''{$2=$2; gsub(/^\\.+\\/+/,\n \"./\", $2); print $2 \" (\"$1\" matches)\"}'' echo \"End of matches for \\\"$search_term\\\"\n in $dir\"}'\n docstring: searches for search_term in all files in dir. If dir is not provided,\n searches in the current directory\n end_name: null\n name: search_dir\n signature: search_dir <search_term> [<dir>]\n - arguments:\n file:\n description: the file to search in (if not provided, searches in the current\n open file)\n required: false\n type: string\n search_term:\n description: the term to search for\n required: true\n type: string\n code: 'search_file() { # Check if the first argument is provided if [\n -z \"$1\" ]; then echo \"Usage: search_file <search_term> [<file>]\" return fi #\n Check if the second argument is provided if [ -n \"$2\" ]; then #\n Check if the provided argument is a valid file if [ -f \"$2\" ]; then local\n file=\"$2\" # Set file if valid else echo \"Usage: search_file\n <search_term> [<file>]\" echo \"Error: File name $2 not found. Please\n provide a valid file name.\" return # Exit if the file is not valid fi else #\n Check if a file is open if [ -z \"$CURRENT_FILE\" ]; then echo\n \"No file open. Use the open command first.\" return # Exit if no\n file is open fi local file=\"$CURRENT_FILE\" # Set file to the\n current open file fi local search_term=\"$1\" file=$(realpath \"$file\") #\n Use grep to directly get the desired formatted output local matches=$(grep\n -nH -- \"$search_term\" \"$file\") # Check if no matches were found if [\n -z \"$matches\" ]; then echo \"No matches found for \\\"$search_term\\\" in\n $file\" return fi # Calculate total number of matches local\n num_matches=$(echo \"$matches\" | wc -l | awk ''{$1=$1; print $0}'') # calculate\n total number of lines matched local num_lines=$(echo \"$matches\" | cut -d:\n -f1 | sort | uniq | wc -l | awk ''{$1=$1; print $0}'') # if num_lines is\n > 100, print an error if [ $num_lines -gt 100 ]; then echo \"More\n than $num_lines lines matched for \\\"$search_term\\\" in $file. Please narrow\n your search.\" return fi # Print the total number of matches and\n the matches themselves echo \"Found $num_matches matches for \\\"$search_term\\\"\n in $file:\" echo \"$matches\" | cut -d: -f1-2 | sort -u -t: -k2,2n | while\n IFS=: read -r filename line_number; do echo \"Line $line_number:$(sed\n -n \"${line_number}p\" \"$file\")\" done echo \"End of matches for \\\"$search_term\\\"\n in $file\"}'\n docstring: searches for search_term in file. If file is not provided, searches\n in the current open file\n end_name: null\n name: search_file\n signature: search_file <search_term> [<file>]\n - arguments:\n dir:\n description: the directory to search in (if not provided, searches in the\n current directory)\n required: false\n type: string\n file_name:\n description: the name of the file to search for\n required: true\n type: string\n code: 'find_file() { if [ $# -eq 1 ]; then local file_name=\"$1\" local\n dir=\"./\" elif [ $# -eq 2 ]; then local file_name=\"$1\" if\n [ -d \"$2\" ]; then local dir=\"$2\" else echo \"Directory\n $2 not found\" return fi else echo \"Usage: find_file\n <file_name> [<dir>]\" return fi dir=$(realpath \"$dir\") local\n matches=$(find \"$dir\" -type f -name \"$file_name\") # if no matches, return if\n [ -z \"$matches\" ]; then echo \"No matches found for \\\"$file_name\\\" in\n $dir\" return fi # Calculate total number of matches local\n num_matches=$(echo \"$matches\" | wc -l | awk ''{$1=$1; print $0}'') echo\n \"Found $num_matches matches for \\\"$file_name\\\" in $dir:\" echo \"$matches\"\n | awk ''{print $0}''}'\n docstring: finds all files with the given name in dir. If dir is not provided,\n searches in the current directory\n end_name: null\n name: find_file\n signature: find_file <file_name> [<dir>]\n - arguments:\n end_line:\n description: the line number to end the edit at (inclusive)\n required: true\n type: integer\n replacement_text:\n description: the text to replace the current selection with\n required: true\n type: string\n start_line:\n description: the line number to start the edit at\n required: true\n type: integer\n code: 'edit() { if [ -z \"$CURRENT_FILE\" ] then echo ''No file open.\n Use the `open` command first.'' return fi local start_line=\"$(echo\n $1: | cut -d: -f1)\" local end_line=\"$(echo $1: | cut -d: -f2)\" if [\n -z \"$start_line\" ] || [ -z \"$end_line\" ] then echo \"Usage: edit\n <start_line>:<end_line>\" return fi local re=''^[0-9]+$'' if\n ! [[ $start_line =~ $re ]]; then echo \"Usage: edit <start_line>:<end_line>\" echo\n \"Error: start_line must be a number\" return fi if ! [[ $end_line\n =~ $re ]]; then echo \"Usage: edit <start_line>:<end_line>\" echo\n \"Error: end_line must be a number\" return fi local linter_cmd=\"flake8\n --isolated --select=F821,F822,F831,E111,E112,E113,E999,E902\" local linter_before_edit=$($linter_cmd\n \"$CURRENT_FILE\" 2>&1) # Bash array starts at 0, so let''s adjust local\n start_line=$((start_line - 1)) local end_line=$((end_line)) local line_count=0 local\n replacement=() while IFS= read -r line do replacement+=(\"$line\") ((line_count++)) done #\n Create a backup of the current file cp \"$CURRENT_FILE\" \"/root/$(basename\n \"$CURRENT_FILE\")_backup\" # Read the file line by line into an array mapfile\n -t lines < \"$CURRENT_FILE\" local new_lines=(\"${lines[@]:0:$start_line}\"\n \"${replacement[@]}\" \"${lines[@]:$((end_line))}\") # Write the new stuff\n directly back into the original file printf \"%s\\n\" \"${new_lines[@]}\" >|\n \"$CURRENT_FILE\" # Run linter if [[ $CURRENT_FILE == *.py ]]; then _lint_output=$($linter_cmd\n \"$CURRENT_FILE\" 2>&1) lint_output=$(_split_string \"$_lint_output\" \"$linter_before_edit\"\n \"$((start_line+1))\" \"$end_line\" \"$line_count\") else # do nothing lint_output=\"\" fi #\n if there is no output, then the file is good if [ -z \"$lint_output\" ];\n then export CURRENT_LINE=$start_line _constrain_line _print echo\n \"File updated. Please review the changes and make sure they are correct (correct\n indentation, no duplicate lines, etc). Edit the file again if necessary.\" else echo\n \"Your proposed edit has introduced new syntax error(s). Please read this error\n message carefully and then retry editing the file.\" echo \"\" echo\n \"ERRORS:\" echo \"$lint_output\" echo \"\" # Save original\n values original_current_line=$CURRENT_LINE original_window=$WINDOW #\n Update values export CURRENT_LINE=$(( (line_count / 2) + start_line\n )) # Set to \"center\" of edit export WINDOW=$((line_count + 10)) # Show\n +/- 5 lines around edit echo \"This is how your edit would have looked\n if applied\" echo \"-------------------------------------------------\" _constrain_line _print echo\n \"-------------------------------------------------\" echo \"\" #\n Restoring CURRENT_FILE to original contents. cp \"/root/$(basename \"$CURRENT_FILE\")_backup\"\n \"$CURRENT_FILE\" export CURRENT_LINE=$(( ((end_line - start_line + 1)\n / 2) + start_line )) export WINDOW=$((end_line - start_line + 10)) echo\n \"This is the original code before your edit\" echo \"-------------------------------------------------\" _constrain_line _print echo\n \"-------------------------------------------------\" # Restore original\n values export CURRENT_LINE=$original_current_line export WINDOW=$original_window echo\n \"Your changes have NOT been applied. Please fix your edit command and try\n again.\" echo \"You either need to 1) Specify the correct start/end line\n arguments or 2) Correct your edit code.\" echo \"DO NOT re-run the same\n failed edit command. Running it again will lead to the same error.\" fi #\n Remove backup file rm -f \"/root/$(basename \"$CURRENT_FILE\")_backup\"}'\n docstring: replaces lines <start_line> through <end_line> (inclusive) with the\n given text in the open file. The replacement text is terminated by a line\n with only end_of_edit on it. All of the <replacement text> will be entered,\n so make sure your indentation is formatted properly. Python files will be\n checked for syntax errors after the edit. If the system detects a syntax error,\n the edit will not be executed. Simply try to edit the file again, but make\n sure to read the error message and modify the edit command you issue accordingly.\n Issuing the same command a second time will just lead to the same error message\n again.\n end_name: end_of_edit\n name: edit\n signature: |-\n edit <start_line>:<end_line>\n <replacement_text>\n end_of_edit\n _subroutines: {}\n blocklist:\n - vim\n - vi\n - emacs\n - nano\n - nohup\n - git\n blocklist_error_template: Interactive operation '{name}' is not supported by this\n environment\n blocklist_standalone:\n - python\n - python3\n - ipython\n - bash\n - sh\n - exit\n - /bin/bash\n - /bin/sh\n - nohup\n - vi\n - vim\n - emacs\n - nano\n command_docs: |+\n open:\n docstring: opens the file at the given path in the editor. If line_number is provided, the window will be move to include that line\n signature: open <path> [<line_number>]\n arguments:\n - path (string) [required]: the path to the file to open\n - line_number (integer) [optional]: the line number to move the window to (if not provided, the window will start at the top of the file)\n\n goto:\n docstring: moves the window to show <line_number>\n signature: goto <line_number>\n arguments:\n - line_number (integer) [required]: the line number to move the window to\n\n scroll_down:\n docstring: moves the window down 100 lines\n signature: scroll_down\n\n scroll_up:\n docstring: moves the window down 100 lines\n signature: scroll_up\n\n create:\n docstring: creates and opens a new file with the given name\n signature: create <filename>\n arguments:\n - filename (string) [required]: the name of the file to create\n\n submit:\n docstring: submits your current code and terminates the session\n signature: submit\n\n search_dir:\n docstring: searches for search_term in all files in dir. If dir is not provided, searches in the current directory\n signature: search_dir <search_term> [<dir>]\n arguments:\n - search_term (string) [required]: the term to search for\n - dir (string) [optional]: the directory to search in (if not provided, searches in the current directory)\n\n search_file:\n docstring: searches for search_term in file. If file is not provided, searches in the current open file\n signature: search_file <search_term> [<file>]\n arguments:\n - search_term (string) [required]: the term to search for\n - file (string) [optional]: the file to search in (if not provided, searches in the current open file)\n\n find_file:\n docstring: finds all files with the given name in dir. If dir is not provided, searches in the current directory\n signature: find_file <file_name> [<dir>]\n arguments:\n - file_name (string) [required]: the name of the file to search for\n - dir (string) [optional]: the directory to search in (if not provided, searches in the current directory)\n\n edit:\n docstring: replaces lines <start_line> through <end_line> (inclusive) with the given text in the open file. The replacement text is terminated by a line with only end_of_edit on it. All of the <replacement text> will be entered, so make sure your indentation is formatted properly. Python files will be checked for syntax errors after the edit. If the system detects a syntax error, the edit will not be executed. Simply try to edit the file again, but make sure to read the error message and modify the edit command you issue accordingly. Issuing the same command a second time will just lead to the same error message again.\n signature: edit <start_line>:<end_line>\n <replacement_text>\n end_of_edit\n arguments:\n - start_line (integer) [required]: the line number to start the edit at\n - end_line (integer) [required]: the line number to end the edit at (inclusive)\n - replacement_text (string) [required]: the text to replace the current selection with\n\n command_files:\n - /Users/fuchur/Documents/24/git_sync/SWE-agent/config/commands/defaults.sh\n - /Users/fuchur/Documents/24/git_sync/SWE-agent/config/commands/search.sh\n - /Users/fuchur/Documents/24/git_sync/SWE-agent/config/commands/edit_linting.sh\n - /Users/fuchur/Documents/24/git_sync/SWE-agent/config/commands/_split_string.py\n demonstration_template: |\n Here is a demonstration of how to correctly accomplish this task.\n It is included to show you how to correctly use the interface.\n You do not need to follow exactly what is done in the demonstration.\n --- DEMONSTRATION ---\n {demonstration}\n --- END OF DEMONSTRATION ---\n demonstrations:\n - /Users/fuchur/Documents/24/git_sync/SWE-agent/trajectories/demonstrations/human_thought__swe-bench-HumanEvalFix-python__lcb__t-0.00__p-0.95__c-4.00__install-0/humanevalfix-python-0.traj\n env_variables:\n CURRENT_FILE: ''\n CURRENT_LINE: '0'\n OVERLAP: '2'\n SEARCH_FILES: ()\n SEARCH_INDEX: '0'\n SEARCH_RESULTS: ()\n WINDOW: '100'\n format_error_template: |\n Your output was not formatted correctly. You must always include one discussion and one command as part of your response. Make sure you do not have multiple discussion/command tags.\n Please make sure your output precisely matches the following format:\n DISCUSSION\n Discuss here with yourself about what your planning and what you're going to do in this step.\n\n ```\n command(s) that you're going to run\n ```\n history_processor: {}\n history_processor_args: {}\n instance_template: |-\n We're currently attempting to solve the following problem:\n ISSUE:\n {issue}\n\n INSTRUCTIONS:\n Now, you're going to solve this issue on your own. Your terminal session has started and you're in the repository's root directory. You can use any bash commands or the special interface to help you. Edit all the files you need to and run any checks or tests that you want.\n Remember, YOU CAN ONLY ENTER ONE COMMAND AT A TIME. You should always wait for feedback after every command.\n When you're satisfied with all of the changes you've made, you can submit your changes to the code base by simply running the submit command.\n Note however that you cannot use any interactive session commands (e.g. python, vim) in this environment, but you can write scripts and run them. E.g. you can write a python script and then run it with `python <script_name>.py`.\n\n NOTE ABOUT THE EDIT COMMAND: Indentation really matters! When editing a file, make sure to insert appropriate indentation before each line!\n\n IMPORTANT TIPS:\n 1. Write your solution in main.py. Always test your code thoroughly before submitting, and if any of the tests fail, try to fix the code before continuing.\n\n 2. If you run a command and it doesn't work, try running a different command. A command that did not work once will not work the second time unless you modify it!\n\n 3. If you open a file and need to get to an area around a specific line that is not in the first 100 lines, say line 583, don't just use the scroll_down command multiple times. Instead, use the goto 583 command. It's much quicker.\n\n 4. Always make sure to look at the currently open file and the current working directory (which appears right after the currently open file). The currently open file might be in a different directory than the working directory! Note that some commands, such as 'create', open files, so they might change the current open file.\n\n 5. When editing files, it is easy to accidentally specify a wrong line number or to write code with incorrect indentation. Always check the code after you issue an edit to make sure that it reflects what you wanted to accomplish. If it didn't, issue another command to fix it.\n\n (Open file: {open_file})\n (Current directory: {working_dir})\n bash-$\n next_step_no_output_template: |-\n Your command ran successfully and did not produce any output.\n (Open file: {open_file})\n (Current directory: {working_dir})\n bash-$\n next_step_template: |-\n {observation}\n (Open file: {open_file})\n (Current directory: {working_dir})\n bash-$\n parse_command: {}\n parse_function: {}\n put_demos_in_history: false\n state_command:\n arguments: null\n code: |\n state() {\n local working_dir=\"$PWD\";\n if [ -z $CURRENT_FILE ]; then\n echo '{\"open_file\": \"n/a\", \"working_dir\": \"'$working_dir'\"}';\n else\n echo '{\"open_file\": \"'$(realpath $CURRENT_FILE)'\", \"working_dir\": \"'$working_dir'\"}';\n fi\n };\n docstring: null\n end_name: null\n name: state\n signature: null\n strategy_template: null\n submit_command: submit\n subroutine_types: []\n system_template: |-\n SETTING: You are an autonomous programmer, and you're working directly in the command line with a special interface.\n\n The special interface consists of a file editor that shows you {WINDOW} lines of a file at a time.\n In addition to typical bash commands, you can also use the following commands to help you navigate and edit files.\n\n COMMANDS:\n {command_docs}\n\n Please note that THE EDIT COMMAND REQUIRES PROPER INDENTATION.\n If you'd like to add the line ' print(x)' you must fully write that out, with all those spaces before the code! Indentation is important and code that is not indented correctly will fail and require fixing before it can be run.\n\n RESPONSE FORMAT:\n Your shell prompt is formatted as follows:\n (Open file: <path>) <cwd> $\n\n You need to format your output using two fields; discussion and command.\n Your output should always include _one_ discussion and _one_ command field EXACTLY as in the following example:\n DISCUSSION\n First I'll start by using ls to see what files are in the current directory. Then maybe we can look at some relevant files to see what they look like.\n ```\n ls -a\n ```\n\n You should only include a *SINGLE* command in the command section and then wait for a response from the shell before continuing with more discussion and commands. Everything you include in the DISCUSSION section will be saved for future reference.\n If you'd like to issue two commands at once, PLEASE DO NOT DO THAT! Please instead first submit just the first command, and then after receiving a response you'll be able to issue the second command.\n You're free to use any other bash commands you want (e.g. find, grep, cat, ls, cd) in addition to the special commands listed above.\n However, the environment does NOT support interactive session commands (e.g. python, vim), so please do not invoke them.\n util_functions:\n - arguments: null\n code: '_print() { local total_lines=$(awk ''END {print NR}'' $CURRENT_FILE) echo\n \"[File: $(realpath $CURRENT_FILE) ($total_lines lines total)]\" lines_above=$(jq\n -n \"$CURRENT_LINE - $WINDOW/2\" | jq ''[0, .] | max | floor'') lines_below=$(jq\n -n \"$total_lines - $CURRENT_LINE - $WINDOW/2\" | jq ''[0, .] | max | round'') if\n [ $lines_above -gt 0 ]; then echo \"($lines_above more lines above)\" fi cat\n $CURRENT_FILE | grep -n $ | head -n $(jq -n \"[$CURRENT_LINE + $WINDOW/2, $WINDOW/2]\n | max | floor\") | tail -n $(jq -n \"$WINDOW\") if [ $lines_below -gt 0 ];\n then echo \"($lines_below more lines below)\" fi}'\n docstring: null\n end_name: null\n name: _print\n signature: _print\n - arguments: null\n code: _constrain_line() { if [ -z \"$CURRENT_FILE\" ] then echo \"No\n file open. Use the open command first.\" return fi local max_line=$(awk\n 'END {print NR}' $CURRENT_FILE) local half_window=$(jq -n \"$WINDOW/2\" |\n jq 'floor') export CURRENT_LINE=$(jq -n \"[$CURRENT_LINE, $max_line - $half_window]\n | min\") export CURRENT_LINE=$(jq -n \"[$CURRENT_LINE, $half_window] | max\")}\n docstring: null\n end_name: null\n name: _constrain_line\n signature: _constrain_line\n - arguments: null\n code: '_scroll_warning_message() { # Warn the agent if we scroll too many\n times # Message will be shown if scroll is called more than WARN_AFTER_SCROLLING_TIMES\n (default 3) times # Initialize variable if it''s not set export SCROLL_COUNT=${SCROLL_COUNT:-0} #\n Reset if the last command wasn''t about scrolling if [ \"$LAST_ACTION\" !=\n \"scroll_up\" ] && [ \"$LAST_ACTION\" != \"scroll_down\" ]; then export SCROLL_COUNT=0 fi #\n Increment because we''re definitely scrolling now export SCROLL_COUNT=$((SCROLL_COUNT\n + 1)) if [ $SCROLL_COUNT -ge ${WARN_AFTER_SCROLLING_TIMES:-3} ]; then echo\n \"\" echo \"WARNING: Scrolling many times in a row is very inefficient.\" echo\n \"If you know what you are looking for, use \\`search_file <pattern>\\` instead.\" echo\n \"\" fi}'\n docstring: null\n end_name: null\n name: _scroll_warning_message\n signature: _scroll_warning_message\n config_file: config/coding_challenge.yaml\n model:\n host_url: localhost:11434\n model_name: azure:gpt4\n per_instance_cost_limit: 3.0\n replay_path: null\n temperature: 0.0\n top_p: 0.95\n total_cost_limit: 0.0\nenvironment:\n base_commit: null\n cache_task_images: false\n container_name: null\n data_path: ../empty/problem.md\n environment_setup: null\n image_name: sweagent/swe-agent:latest\n install_environment: true\n no_mirror: false\n repo_path: ../empty\n split: dev\n timeout: null\n verbose: true\ninstance_filter: .*\nprint_config: true\nraise_exceptions: false\nskip_existing: false\nsuffix: ''\n\n2024-07-12 17:57:39,938 WARNING The --model CLI argument is ignored when using the Azure GPT endpoint. The model is determined by the AZURE_OPENAI_DEPLOYMENT key/environment variable (this might change in the future).\n2024-07-12 17:57:40,021 INFO \ud83d\udcbd Loaded dataset from ../empty/problem.md\n2024-07-12 17:57:40,059 INFO Found image sweagent/swe-agent:latest with tags: ['sweagent/swe-agent:latest'], created: 2024-07-01T19:58:23.043599678Z for linux arm64.\n2024-07-12 17:57:40,060 DEBUG Starting container with command: docker run -i --rm --name sweagent-swe-agent-latest-0abb363825 sweagent/swe-agent:latest /bin/bash -l\n2024-07-12 17:57:41,130 INFO \ud83c\udf31 Environment Initialized\n2024-07-12 17:57:41,433 DEBUG Environment initialization took 1.49 seconds\n2024-07-12 17:57:41,517 WARNING **************************************************\n2024-07-12 17:57:41,517 WARNING Found existing args.yaml with different arguments!\n2024-07-12 17:57:41,518 WARNING **************************************************\n2024-07-12 17:57:41,539 INFO \u25b6\ufe0f Beginning task 0\n2024-07-12 17:57:41,673 DEBUG Copying /Users/fuchur/Documents/24/git_sync/empty to container at /__Users__fuchur__Documents__24__git_sync__empty with command: docker cp /Users/fuchur/Documents/24/git_sync/empty eaeb81eeabf519b6f46f8daf8729b52c1c3846091c707c400d4083c948905888:/__Users__fuchur__Documents__24__git_sync__empty\n2024-07-12 17:57:42,876 WARNING install_environment is set to True, but the data path is a GitHub URL without an environment config file (environment_config key/flag). Skipping conda environment installation.\n2024-07-12 17:57:44,017 INFO Initializing agent settings for container eaeb81eeabf519b6f46f8daf8729b52c1c3846091c707c400d4083c948905888\n2024-07-12 17:57:44,667 INFO Resetting model stats\n2024-07-12 17:57:44,668 INFO SYSTEM (primary)\nSETTING: You are an autonomous programmer, and you're working directly in the command line with a special interface.\n\nThe special interface consists of a file editor that shows you 100 lines of a file at a time.\nIn addition to typical bash commands, you can also use the following commands to help you navigate and edit files.\n\nCOMMANDS:\nopen:\n docstring: opens the file at the given path in the editor. If line_number is provided, the window will be move to include that line\n signature: open <path> [<line_number>]\n arguments:\n - path (string) [required]: the path to the file to open\n - line_number (integer) [optional]: the line number to move the window to (if not provided, the window will start at the top of the file)\n\ngoto:\n docstring: moves the window to show <line_number>\n signature: goto <line_number>\n arguments:\n - line_number (integer) [required]: the line number to move the window to\n\nscroll_down:\n docstring: moves the window down 100 lines\n signature: scroll_down\n\nscroll_up:\n docstring: moves the window down 100 lines\n signature: scroll_up\n\ncreate:\n docstring: creates and opens a new file with the given name\n signature: create <filename>\n arguments:\n - filename (string) [required]: the name of the file to create\n\nsubmit:\n docstring: submits your current code and terminates the session\n signature: submit\n\nsearch_dir:\n docstring: searches for search_term in all files in dir. If dir is not provided, searches in the current directory\n signature: search_dir <search_term> [<dir>]\n arguments:\n - search_term (string) [required]: the term to search for\n - dir (string) [optional]: the directory to search in (if not provided, searches in the current directory)\n\nsearch_file:\n docstring: searches for search_term in file. If file is not provided, searches in the current open file\n signature: search_file <search_term> [<file>]\n arguments:\n - search_term (string) [required]: the term to search for\n - file (string) [optional]: the file to search in (if not provided, searches in the current open file)\n\nfind_file:\n docstring: finds all files with the given name in dir. If dir is not provided, searches in the current directory\n signature: find_file <file_name> [<dir>]\n arguments:\n - file_name (string) [required]: the name of the file to search for\n - dir (string) [optional]: the directory to search in (if not provided, searches in the current directory)\n\nedit:\n docstring: replaces lines <start_line> through <end_line> (inclusive) with the given text in the open file. The replacement text is terminated by a line with only end_of_edit on it. All of the <replacement text> will be entered, so make sure your indentation is formatted properly. Python files will be checked for syntax errors after the edit. If the system detects a syntax error, the edit will not be executed. Simply try to edit the file again, but make sure to read the error message and modify the edit command you issue accordingly. Issuing the same command a second time will just lead to the same error message again.\n signature: edit <start_line>:<end_line>\n<replacement_text>\nend_of_edit\n arguments:\n - start_line (integer) [required]: the line number to start the edit at\n - end_line (integer) [required]: the line number to end the edit at (inclusive)\n - replacement_text (string) [required]: the text to replace the current selection with\n\n\n\nPlease note that THE EDIT COMMAND REQUIRES PROPER INDENTATION.\nIf you'd like to add the line ' print(x)' you must fully write that out, with all those spaces before the code! Indentation is important and code that is not indented correctly will fail and require fixing before it can be run.\n\nRESPONSE FORMAT:\nYour shell prompt is formatted as follows:\n(Open file: <path>) <cwd> $\n\nYou need to format your output using two fields; discussion and command.\nYour output should always include _one_ discussion and _one_ command field EXACTLY as in the following example:\nDISCUSSION\nFirst I'll start by using ls to see what files are in the current directory. Then maybe we can look at some relevant files to see what they look like.\n ```\n ls -a\n ```\n\nYou should only include a *SINGLE* command in the command section and then wait for a response from the shell before continuing with more discussion and commands. Everything you include in the DISCUSSION section will be saved for future reference.\nIf you'd like to issue two commands at once, PLEASE DO NOT DO THAT! Please instead first submit just the first command, and then after receiving a response you'll be able to issue the second command.\nYou're free to use any other bash commands you want (e.g. find, grep, cat, ls, cd) in addition to the special commands listed above.\nHowever, the environment does NOT support interactive session commands (e.g. python, vim), so please do not invoke them.\n2024-07-12 17:57:44,679 INFO DEMONSTRATION: /Users/fuchur/Documents/24/git_sync/SWE-agent/trajectories/demonstrations/human_thought__swe-bench-HumanEvalFix-python__lcb__t-0.00__p-0.95__c-4.00__install-0/humanevalfix-python-0.traj\n2024-07-12 17:57:44,680 INFO Trajectory will be saved to trajectories/fuchur/azure-gpt4__problem__coding_challenge__t-0.00__p-0.95__c-3.00__install-1/26d111.traj\n2024-07-12 17:57:44,747 INFO \ud83e\udd16 MODEL INPUT\nWe're currently attempting to solve the following problem:\nISSUE:\nGiven an unsorted integer array nums.\nReturn the smallest positive integer that is not present in nums.\n\nYou must implement an algorithm that runs in O(n) time and uses O(1) auxiliary space.\n\n## Example 1:\n\n> Input: nums = [1,2,0]\n> Output: 3\n> Explanation: The numbers in the range [1,2] are all in the array.\n\n## Example 2:\n\n> Input: nums = [3,4,-1,1]\n> Output: 2\n> Explanation: 1 is in the array but 2 is missing.\n\n## Example 3:\n\n> Input: nums = [7,8,9,11,12]\n> Output: 1\n> Explanation: The smallest positive integer 1 is missing.\n\n## Constraints:\n\n1 <= nums.length <= 105\n-231 <= nums[i] <= 231 - 1\n\n\n\n\nINSTRUCTIONS:\nNow, you're going to solve this issue on your own. Your terminal session has started and you're in the repository's root directory. You can use any bash commands or the special interface to help you. Edit all the files you need to and run any checks or tests that you want.\nRemember, YOU CAN ONLY ENTER ONE COMMAND AT A TIME. You should always wait for feedback after every command.\nWhen you're satisfied with all of the changes you've made, you can submit your changes to the code base by simply running the submit command.\nNote however that you cannot use any interactive session commands (e.g. python, vim) in this environment, but you can write scripts and run them. E.g. you can write a python script and then run it with `python <script_name>.py`.\n\nNOTE ABOUT THE EDIT COMMAND: Indentation really matters! When editing a file, make sure to insert appropriate indentation before each line!\n\nIMPORTANT TIPS:\n1. Write your solution in main.py. Always test your code thoroughly before submitting, and if any of the tests fail, try to fix the code before continuing.\n\n2. If you run a command and it doesn't work, try running a different command. A command that did not work once will not work the second time unless you modify it!\n\n3. If you open a file and need to get to an area around a specific line that is not in the first 100 lines, say line 583, don't just use the scroll_down command multiple times. Instead, use the goto 583 command. It's much quicker.\n\n4. Always make sure to look at the currently open file and the current working directory (which appears right after the currently open file). The currently open file might be in a different directory than the working directory! Note that some commands, such as 'create', open files, so they might change the current open file.\n\n5. When editing files, it is easy to accidentally specify a wrong line number or to write code with incorrect indentation. Always check the code after you issue an edit to make sure that it reflects what you wanted to accomplish. If it didn't, issue another command to fix it.\n\n(Open file: n/a)\n(Current directory: /__Users__fuchur__Documents__24__git_sync__empty)\nbash-$\n2024-07-12 17:57:48,042 INFO input_tokens=3,705, output_tokens=48, instance_cost=0.04, cost=0.04\n2024-07-12 17:57:48,044 INFO total_tokens_sent=3,705, total_tokens_received=48, total_cost=0.04, total_api_calls=1\n2024-07-12 17:57:48,045 INFO \ud83d\udcad THOUGHT (primary)\nTo start addressing the issue, I need to access the main.py file where I will implement the solution for the given problem. First, I'll check if the main.py file exists in the current directory.\n\n2024-07-12 17:57:48,046 INFO \ud83c\udfac ACTION (primary)\nls -a\n\n2024-07-12 17:57:48,217 INFO \ud83e\udd16 MODEL INPUT\n.\n..\n.git\nmain.py\nproblem.md\n(Open file: n/a)\n(Current directory: /__Users__fuchur__Documents__24__git_sync__empty)\nbash-$\n2024-07-12 17:57:50,839 INFO input_tokens=3,800, output_tokens=36, instance_cost=0.08, cost=0.04\n2024-07-12 17:57:50,842 INFO total_tokens_sent=7,505, total_tokens_received=84, total_cost=0.08, total_api_calls=2\n2024-07-12 17:57:50,847 INFO \ud83d\udcad THOUGHT (primary)\nThe main.py file is present in the current directory. I will open this file to inspect its contents and then proceed to implement the required function.\n\n2024-07-12 17:57:50,856 INFO \ud83c\udfac ACTION (primary)\nopen main.py\n\n2024-07-12 17:57:51,196 INFO \ud83e\udd16 MODEL INPUT\n[File: /__Users__fuchur__Documents__24__git_sync__empty/main.py (5 lines total)]\n1:from typing import List\n2:\n3:\n4:class Solution:\n5: def firstMissingPositive(self, nums: List[int]) -> int:\n(Open file: /__Users__fuchur__Documents__24__git_sync__empty/main.py)\n(Current directory: /__Users__fuchur__Documents__24__git_sync__empty)\nbash-$\n2024-07-12 17:58:08,893 INFO input_tokens=3,947, output_tokens=334, instance_cost=0.13, cost=0.05\n2024-07-12 17:58:08,895 INFO total_tokens_sent=11,452, total_tokens_received=418, total_cost=0.13, total_api_calls=3\n2024-07-12 17:58:08,895 INFO \ud83d\udcad THOUGHT (primary)\nThe `firstMissingPositive` function is declared but not implemented. I will now edit the file to implement the function that finds the smallest positive integer not present in the array, adhering to the constraints of O(n) time complexity and O(1) space complexity.\n\n2024-07-12 17:58:08,896 INFO \ud83c\udfac ACTION (primary)\nedit 5:5\n # Implementation of the function to find the smallest missing positive integer\n def firstMissingPositive(self, nums: List[int]) -> int:\n n = len(nums)\n # Mark numbers (num < 0) and (num > n) with a special marker number (n+1)\n # We can ignore those because if missing number is in the range (1, n+1)\n # it must be in the array\n for i in range(n):\n if nums[i] <= 0 or nums[i] > n:\n nums[i] = n + 1\n\n # Use index as a hash key and number sign as a presence detector\n # For example, if nums[1] is negative that means that number `1` is present in the array\n for i in range(n):\n num = abs(nums[i])\n if num <= n:\n nums[num - 1] = -abs(nums[num - 1])\n\n # Now the index of the first positive number is equal to the smallest missing positive number\n for i in range(n):\n if nums[i] > 0:\n return i + 1\n\n # If no positive numbers were found, the smallest missing positive is n+1\n return n + 1\nend_of_edit\n\n2024-07-12 17:58:09,404 INFO \ud83e\udd16 MODEL INPUT\n[File: /__Users__fuchur__Documents__24__git_sync__empty/main.py (28 lines total)]\n1:from typing import List\n2:\n3:\n4:class Solution:\n5: # Implementation of the function to find the smallest missing positive integer\n6: def firstMissingPositive(self, nums: List[int]) -> int:\n7: n = len(nums)\n8: # Mark numbers (num < 0) and (num > n) with a special marker number (n+1)\n9: # We can ignore those because if missing number is in the range (1, n+1)\n10: # it must be in the array\n11: for i in range(n):\n12: if nums[i] <= 0 or nums[i] > n:\n13: nums[i] = n + 1\n14:\n15: # Use index as a hash key and number sign as a presence detector\n16: # For example, if nums[1] is negative that means that number `1` is present in the array\n17: for i in range(n):\n18: num = abs(nums[i])\n19: if num <= n:\n20: nums[num - 1] = -abs(nums[num - 1])\n21:\n22: # Now the index of the first positive number is equal to the smallest missing positive number\n23: for i in range(n):\n24: if nums[i] > 0:\n25: return i + 1\n26:\n27: # If no positive numbers were found, the smallest missing positive is n+1\n28: return n + 1\nFile updated. Please review the changes and make sure they are correct (correct indentation, no duplicate lines, etc). Edit the file again if necessary.\n(Open file: /__Users__fuchur__Documents__24__git_sync__empty/main.py)\n(Current directory: /__Users__fuchur__Documents__24__git_sync__empty)\nbash-$\n2024-07-12 17:58:20,954 INFO input_tokens=4,720, output_tokens=206, instance_cost=0.18, cost=0.05\n2024-07-12 17:58:20,958 INFO total_tokens_sent=16,172, total_tokens_received=624, total_cost=0.18, total_api_calls=4\n2024-07-12 17:58:20,962 INFO \ud83d\udcad THOUGHT (primary)\nThe implementation of the `firstMissingPositive` function has been successfully added to the `main.py` file. To ensure that the function works correctly, I will now write a script to test the function with the example inputs provided in the issue description.\n\n2024-07-12 17:58:20,973 INFO \ud83c\udfac ACTION (primary)\nedit 29:35\n# Adding test cases to verify the implementation of firstMissingPositive\nif __name__ == \"__main__\":\n sol = Solution()\n # Test case 1\n assert sol.firstMissingPositive([1, 2, 0]) == 3, \"Test case 1 failed\"\n # Test case 2\n assert sol.firstMissingPositive([3, 4, -1, 1]) == 2, \"Test case 2 failed\"\n # Test case 3\n assert sol.firstMissingPositive([7, 8, 9, 11, 12]) == 1, \"Test case 3 failed\"\n print(\"All test cases passed successfully.\")\nend_of_edit\n\n2024-07-12 17:58:21,472 INFO \ud83e\udd16 MODEL INPUT\n[File: /__Users__fuchur__Documents__24__git_sync__empty/main.py (38 lines total)]\n1:from typing import List\n2:\n3:\n4:class Solution:\n5: # Implementation of the function to find the smallest missing positive integer\n6: def firstMissingPositive(self, nums: List[int]) -> int:\n7: n = len(nums)\n8: # Mark numbers (num < 0) and (num > n) with a special marker number (n+1)\n9: # We can ignore those because if missing number is in the range (1, n+1)\n10: # it must be in the array\n11: for i in range(n):\n12: if nums[i] <= 0 or nums[i] > n:\n13: nums[i] = n + 1\n14:\n15: # Use index as a hash key and number sign as a presence detector\n16: # For example, if nums[1] is negative that means that number `1` is present in the array\n17: for i in range(n):\n18: num = abs(nums[i])\n19: if num <= n:\n20: nums[num - 1] = -abs(nums[num - 1])\n21:\n22: # Now the index of the first positive number is equal to the smallest missing positive number\n23: for i in range(n):\n24: if nums[i] > 0:\n25: return i + 1\n26:\n27: # If no positive numbers were found, the smallest missing positive is n+1\n28: return n + 1\n29:# Adding test cases to verify the implementation of firstMissingPositive\n30:if __name__ == \"__main__\":\n31: sol = Solution()\n32: # Test case 1\n33: assert sol.firstMissingPositive([1, 2, 0]) == 3, \"Test case 1 failed\"\n34: # Test case 2\n35: assert sol.firstMissingPositive([3, 4, -1, 1]) == 2, \"Test case 2 failed\"\n36: # Test case 3\n37: assert sol.firstMissingPositive([7, 8, 9, 11, 12]) == 1, \"Test case 3 failed\"\n38: print(\"All test cases passed successfully.\")\nFile updated. Please review the changes and make sure they are correct (correct indentation, no duplicate lines, etc). Edit the file again if necessary.\n(Open file: /__Users__fuchur__Documents__24__git_sync__empty/main.py)\n(Current directory: /__Users__fuchur__Documents__24__git_sync__empty)\nbash-$\n2024-07-12 17:58:24,886 INFO input_tokens=5,527, output_tokens=42, instance_cost=0.24, cost=0.06\n2024-07-12 17:58:24,891 INFO total_tokens_sent=21,699, total_tokens_received=666, total_cost=0.24, total_api_calls=5\n2024-07-12 17:58:24,894 INFO \ud83d\udcad THOUGHT (primary)\nThe test cases have been added to the `main.py` file. Now, I will run the script to ensure that all test cases pass and the implementation meets the requirements.\n\n2024-07-12 17:58:24,896 INFO \ud83c\udfac ACTION (primary)\npython main.py\n\n2024-07-12 17:58:25,084 INFO \ud83e\udd16 MODEL INPUT\nAll test cases passed successfully.\n(Open file: /__Users__fuchur__Documents__24__git_sync__empty/main.py)\n(Current directory: /__Users__fuchur__Documents__24__git_sync__empty)\nbash-$\n2024-07-12 17:58:28,523 INFO input_tokens=5,628, output_tokens=38, instance_cost=0.29, cost=0.06\n2024-07-12 17:58:28,524 INFO total_tokens_sent=27,327, total_tokens_received=704, total_cost=0.29, total_api_calls=6\n2024-07-12 17:58:28,524 INFO \ud83d\udcad THOUGHT (primary)\nThe test cases have successfully passed, indicating that the implementation of the `firstMissingPositive` function is correct. I will now submit the changes to the code base.\n\n2024-07-12 17:58:28,525 INFO \ud83c\udfac ACTION (primary)\nsubmit\n\n2024-07-12 17:58:28,666 INFO Found submission:\ndiff --git a/main.py b/main.py\nindex 1e7f08f..beabaa7 100644\n--- a/main.py\n+++ b/main.py\n@@ -2,4 +2,37 @@ from typing import List\n\n\n class Solution:\n- def firstMissingPositive(self, nums: List[int]) -> int:\n+ # Implementation of the function to find the smallest missing positive integer\n+ def firstMissingPositive(self, nums: List[int]) -> int:\n+ n = len(nums)\n+ # Mark numbers (num < 0) and (num > n) with a special marker number (n+1)\n+ # We can ignore those because if missing number is in the range (1, n+1)\n+ # it must be in the array\n+ for i in range(n):\n+ if nums[i] <= 0 or nums[i] > n:\n+ nums[i] = n + 1\n+\n+ # Use index as a hash key and number sign as a presence detector\n+ # For example, if nums[1] is negative that means that number `1` is present in the array\n+ for i in range(n):\n+ num = abs(nums[i])\n+ if num <= n:\n+ nums[num - 1] = -abs(nums[num - 1])\n+\n+ # Now the index of the first positive number is equal to the smallest missing positive number\n+ for i in range(n):\n+ if nums[i] > 0:\n+ return i + 1\n+\n+ # If no positive numbers were found, the smallest missing positive is n+1\n+ return n + 1\n+# Adding test cases to verify the implementation of firstMissingPositive\n+if __name__ == \"__main__\":\n+ sol = Solution()\n+ # Test case 1\n+ assert sol.firstMissingPositive([1, 2, 0]) == 3, \"Test case 1 failed\"\n+ # Test case 2\n+ assert sol.firstMissingPositive([3, 4, -1, 1]) == 2, \"Test case 2 failed\"\n+ # Test case 3\n+ assert sol.firstMissingPositive([7, 8, 9, 11, 12]) == 1, \"Test case 3 failed\"\n+ print(\"All test cases passed successfully.\")\n\n2024-07-12 17:58:28,684 INFO Trajectory saved to trajectories/fuchur/azure-gpt4__problem__coding_challenge__t-0.00__p-0.95__c-3.00__install-1/26d111.traj\n2024-07-12 17:58:28,685 INFO Saved predictions to trajectories/fuchur/azure-gpt4__problem__coding_challenge__t-0.00__p-0.95__c-3.00__install-1/all_preds.jsonl\n</code></pre> <p>SWE-agent will typically conclude with a message like</p> <pre><code>INFO Trajectory saved to trajectories/fuchur/azure-gpt4__problem__coding_challenge__t-0.00__p-0.95__c-3.00__install-1/26d111.traj\nINFO Saved predictions to\n trajectories/fuchur/azure-gpt4__problem__coding_challenge__t-0.00__p-0.95__c-3.00__install-1/all_preds.jsonl\n\u256d\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500 \ud83c\udf89 Submission successful \ud83c\udf89 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u256e\n\u2502 SWE-agent has produced a patch that it believes will solve the issue you submitted! \u2502\n\u2502 Use the code snippet below to inspect or apply it! \u2502\n\u2570\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u256f\n\n # The patch has been saved to your local filesystem at:\n PATCH_FILE_PATH='/Users/fuchur/Documents/24/git_sync/SWE-agent/trajectories/fuchur/azure-gpt4__problem__coding_challenge__t-0.00__p-0\n 5__c-3.00__install-1/patches/26d111.patch'\n # Inspect it:\n cat \"${PATCH_FILE_PATH}\"\n # Apply it to a local repository:\n cd <your local repo root>\n git apply \"${PATCH_FILE_PATH}\"\n</code></pre> <p>This informs you of the location of the trajectory and of the patch file that contains the solution.</p> <p>In our case, this looks like this:</p> Patch (solution) <pre><code>diff --git a/main.py b/main.py\nindex 1e7f08f..beabaa7 100644\n--- a/main.py\n+++ b/main.py\n@@ -2,4 +2,37 @@ from typing import List\n\n\n class Solution:\n- def firstMissingPositive(self, nums: List[int]) -> int:\n+ # Implementation of the function to find the smallest missing positive integer\n+ def firstMissingPositive(self, nums: List[int]) -> int:\n+ n = len(nums)\n+ # Mark numbers (num < 0) and (num > n) with a special marker number (n+1)\n+ # We can ignore those because if missing number is in the range (1, n+1)\n+ # it must be in the array\n+ for i in range(n):\n+ if nums[i] <= 0 or nums[i] > n:\n+ nums[i] = n + 1\n+\n+ # Use index as a hash key and number sign as a presence detector\n+ # For example, if nums[1] is negative that means that number `1` is present in the array\n+ for i in range(n):\n+ num = abs(nums[i])\n+ if num <= n:\n+ nums[num - 1] = -abs(nums[num - 1])\n+\n+ # Now the index of the first positive number is equal to the smallest missing positive number\n+ for i in range(n):\n+ if nums[i] > 0:\n+ return i + 1\n+\n+ # If no positive numbers were found, the smallest missing positive is n+1\n+ return n + 1\n+# Adding test cases to verify the implementation of firstMissingPositive\n+if __name__ == \"__main__\":\n+ sol = Solution()\n+ # Test case 1\n+ assert sol.firstMissingPositive([1, 2, 0]) == 3, \"Test case 1 failed\"\n+ # Test case 2\n+ assert sol.firstMissingPositive([3, 4, -1, 1]) == 2, \"Test case 2 failed\"\n+ # Test case 3\n+ assert sol.firstMissingPositive([7, 8, 9, 11, 12]) == 1, \"Test case 3 failed\"\n+ print(\"All test cases passed successfully.\")\n</code></pre> <p>Because of the <code>--apply_patch_locally</code> flag, the patch has also already been applied to the repository, so you can also retrieve the final solution from there.</p>"},{"location":"usage/coding_challenges/#improving-swe-agent-for-coding-challenges","title":"Improving SWE-agent for coding challenges","text":"<p>By default, the demonstration trajectory the agent uses while solving a coding challenge is one in which it needs to solve a small bug in a short piece of code (from the HumanEvalFix dataset). Since that process is not too similar to solving a coding challenge, performance would probably substantially improve if the agent was given a demonstration trajectory in which it has to solve an actual programming challenge. To learn how to do that, read this.</p>"},{"location":"usage/enigma/","title":"EnIGMA Tutorial","text":"<p>This tutorial walks you through running EnIGMA from the command-line. It is based on basic knowledge of the command-line of SWE-agent that is covered here. This tutorial focuses on using EnIGMA as a tool to solve individual CTF challenges.</p>"},{"location":"usage/enigma/#getting-started","title":"Getting started","text":"<p>For the CLI, use the <code>run.py</code> script. Let's start with an absolutely trivial example and solve a CTF challenge where the flag is leaked in its description. We will first need to clone NYU CTF benchmark.</p> <p>Then, assuming the following directory structure:</p> <pre><code>\u251c\u2500\u2500 SWE-agent\n\u2502 \u251c\u2500\u2500 run.py\n\u2502 \u2514\u2500\u2500 ...\n\u251c\u2500\u2500 LLM_CTF_Database\n\u2502 \u251c\u2500\u2500 2017\n\u2502 \u251c\u2500\u2500 2018\n\u2502 \u251c\u2500\u2500 2019\n\u2514 \u2514\u2500\u2500 ...\n</code></pre> <p>We will run the following command,</p> <pre><code>python run.py \\\n --model_name gpt4 \\\n --ctf \\\n --image_name sweagent/enigma:latest \\\n --data_path ../LLM_CTF_Database/2018/CSAW-Finals/misc/leaked_flag/challenge.json \\\n --repo_path ../LLM_CTF_Database/2018/CSAW-Finals/misc/leaked_flag \\\n --config_file config/default_ctf.yaml \\\n --per_instance_cost_limit 2.00\n</code></pre> Output <pre><code>2024-09-19 11:26:12,131 INFO \ud83d\udcd9 Arguments: actions:\n apply_patch_locally: false\n open_pr: false\n push_gh_repo_url: ''\n skip_if_commits_reference_issue: true\nagent:\n config:\n _commands:\n - arguments:\n line_number:\n description: the line number to move the window to (if not provided, the\n window will start at the top of the file)\n required: false\n type: integer\n path:\n description: the path to the file to open\n required: true\n type: string\n code: 'open() { if [ -z \"$1\" ] then echo \"Usage: open <file>\" return fi #\n Check if the second argument is provided if [ -n \"$2\" ]; then #\n Check if the provided argument is a valid number if ! [[ $2 =~ ^[0-9]+$\n ]]; then echo \"Usage: open <file> [<line_number>]\" echo\n \"Error: <line_number> must be a number\" return # Exit if the line\n number is not valid fi local max_line=$(awk ''END {print NR}''\n $1) if [ $2 -gt $max_line ]; then echo \"Warning: <line_number>\n ($2) is greater than the number of lines in the file ($max_line)\" echo\n \"Warning: Setting <line_number> to $max_line\" local line_number=$(jq\n -n \"$max_line\") # Set line number to max if greater than max elif\n [ $2 -lt 1 ]; then echo \"Warning: <line_number> ($2) is less than\n 1\" echo \"Warning: Setting <line_number> to 1\" local\n line_number=$(jq -n \"1\") # Set line number to 1 if less than 1 else local\n OFFSET=$(jq -n \"$WINDOW/6\" | jq ''floor'') local line_number=$(jq\n -n \"[$2 + $WINDOW/2 - $OFFSET, 1] | max | floor\") fi else local\n line_number=$(jq -n \"$WINDOW/2\") # Set default line number if not provided fi if\n [ -f \"$1\" ]; then export CURRENT_FILE=$(realpath $1) export\n CURRENT_LINE=$line_number _constrain_line _print elif [ -d\n \"$1\" ]; then echo \"Error: $1 is a directory. You can only open files.\n Use cd or ls to navigate directories.\" else echo \"File $1 not found\" fi}'\n docstring: opens the file at the given path in the editor. If line_number is\n provided, the window will be move to include that line\n end_name: null\n name: open\n signature: open <path> [<line_number>]\n - arguments:\n line_number:\n description: the line number to move the window to\n required: true\n type: integer\n code: 'goto() { if [ $# -gt 1 ]; then echo \"goto allows only one line\n number at a time.\" return fi if [ -z \"$CURRENT_FILE\" ] then echo\n \"No file open. Use the open command first.\" return fi if [ -z\n \"$1\" ] then echo \"Usage: goto <line>\" return fi if\n ! [[ $1 =~ ^[0-9]+$ ]] then echo \"Usage: goto <line>\" echo\n \"Error: <line> must be a number\" return fi local max_line=$(awk\n ''END {print NR}'' $CURRENT_FILE) if [ $1 -gt $max_line ] then echo\n \"Error: <line> must be less than or equal to $max_line\" return fi local\n OFFSET=$(jq -n \"$WINDOW/6\" | jq ''floor'') export CURRENT_LINE=$(jq -n\n \"[$1 + $WINDOW/2 - $OFFSET, 1] | max | floor\") _constrain_line _print}'\n docstring: moves the window to show <line_number>\n end_name: null\n name: goto\n signature: goto <line_number>\n - arguments: null\n code: scroll_down() { if [ -z \"$CURRENT_FILE\" ] then echo \"No file\n open. Use the open command first.\" return fi export CURRENT_LINE=$(jq\n -n \"$CURRENT_LINE + $WINDOW - $OVERLAP\") _constrain_line _print _scroll_warning_message}\n docstring: moves the window down {WINDOW} lines\n end_name: null\n name: scroll_down\n signature: scroll_down\n - arguments: null\n code: scroll_up() { if [ -z \"$CURRENT_FILE\" ] then echo \"No file\n open. Use the open command first.\" return fi export CURRENT_LINE=$(jq\n -n \"$CURRENT_LINE - $WINDOW + $OVERLAP\") _constrain_line _print _scroll_warning_message}\n docstring: moves the window down {WINDOW} lines\n end_name: null\n name: scroll_up\n signature: scroll_up\n - arguments:\n filename:\n description: the name of the file to create\n required: true\n type: string\n code: \"create() { if [ -z \\\"$1\\\" ]; then echo \\\"Usage: create <filename>\\\"\\\n \\ return fi # Check if the file already exists if [ -e \\\"\\\n $1\\\" ]; then echo \\\"Error: File '$1' already exists.\\\"\\t\\topen \\\"$1\\\"\\\n \\ return fi # Create the file an empty new line printf \\\"\\\\\\\n n\\\" > \\\"$1\\\" # Use the existing open command to open the created file \\\n \\ open \\\"$1\\\"}\"\n docstring: creates and opens a new file with the given name\n end_name: null\n name: create\n signature: create <filename>\n - arguments:\n dir:\n description: the directory to search in (if not provided, searches in the\n current directory)\n required: false\n type: string\n search_term:\n description: the term to search for\n required: true\n type: string\n code: 'search_dir() { if [ $# -eq 1 ]; then local search_term=\"$1\" local\n dir=\"./\" elif [ $# -eq 2 ]; then local search_term=\"$1\" if\n [ -d \"$2\" ]; then local dir=\"$2\" else echo \"Directory\n $2 not found\" return fi else echo \"Usage: search_dir\n <search_term> [<dir>]\" return fi dir=$(realpath \"$dir\") local\n matches=$(find \"$dir\" -type f ! -path ''*/.*'' -exec grep -nIH -- \"$search_term\"\n {} + | cut -d: -f1 | sort | uniq -c) # if no matches, return if [ -z\n \"$matches\" ]; then echo \"No matches found for \\\"$search_term\\\" in $dir\" return fi #\n Calculate total number of matches local num_matches=$(echo \"$matches\" |\n awk ''{sum+=$1} END {print sum}'') # calculate total number of files matched local\n num_files=$(echo \"$matches\" | wc -l | awk ''{$1=$1; print $0}'') # if num_files\n is > 100, print an error if [ $num_files -gt 100 ]; then echo \"More\n than $num_files files matched for \\\"$search_term\\\" in $dir. Please narrow\n your search.\" return fi echo \"Found $num_matches matches for\n \\\"$search_term\\\" in $dir:\" echo \"$matches\" | awk ''{$2=$2; gsub(/^\\.+\\/+/,\n \"./\", $2); print $2 \" (\"$1\" matches)\"}'' echo \"End of matches for \\\"$search_term\\\"\n in $dir\"}'\n docstring: searches for search_term in all files in dir. If dir is not provided,\n searches in the current directory\n end_name: null\n name: search_dir\n signature: search_dir <search_term> [<dir>]\n - arguments:\n file:\n description: the file to search in (if not provided, searches in the current\n open file)\n required: false\n type: string\n search_term:\n description: the term to search for\n required: true\n type: string\n code: 'search_file() { # Check if the first argument is provided if [\n -z \"$1\" ]; then echo \"Usage: search_file <search_term> [<file>]\" return fi #\n Check if the second argument is provided if [ -n \"$2\" ]; then #\n Check if the provided argument is a valid file if [ -f \"$2\" ]; then local\n file=\"$2\" # Set file if valid else echo \"Usage: search_file\n <search_term> [<file>]\" echo \"Error: File name $2 not found. Please\n provide a valid file name.\" return # Exit if the file is not valid fi else #\n Check if a file is open if [ -z \"$CURRENT_FILE\" ]; then echo\n \"No file open. Use the open command first.\" return # Exit if no\n file is open fi local file=\"$CURRENT_FILE\" # Set file to the\n current open file fi local search_term=\"$1\" file=$(realpath \"$file\") #\n Use grep to directly get the desired formatted output local matches=$(grep\n -nH -- \"$search_term\" \"$file\") # Check if no matches were found if [\n -z \"$matches\" ]; then echo \"No matches found for \\\"$search_term\\\" in\n $file\" return fi # Calculate total number of matches local\n num_matches=$(echo \"$matches\" | wc -l | awk ''{$1=$1; print $0}'') # calculate\n total number of lines matched local num_lines=$(echo \"$matches\" | cut -d:\n -f1 | sort | uniq | wc -l | awk ''{$1=$1; print $0}'') # if num_lines is\n > 100, print an error if [ $num_lines -gt 100 ]; then echo \"More\n than $num_lines lines matched for \\\"$search_term\\\" in $file. Please narrow\n your search.\" return fi # Print the total number of matches and\n the matches themselves echo \"Found $num_matches matches for \\\"$search_term\\\"\n in $file:\" echo \"$matches\" | cut -d: -f1-2 | sort -u -t: -k2,2n | while\n IFS=: read -r filename line_number; do echo \"Line $line_number:$(sed\n -n \"${line_number}p\" \"$file\")\" done echo \"End of matches for \\\"$search_term\\\"\n in $file\"}'\n docstring: searches for search_term in file. If file is not provided, searches\n in the current open file\n end_name: null\n name: search_file\n signature: search_file <search_term> [<file>]\n - arguments:\n dir:\n description: the directory to search in (if not provided, searches in the\n current directory)\n required: false\n type: string\n file_name:\n description: the name of the file to search for\n required: true\n type: string\n code: 'find_file() { if [ $# -eq 1 ]; then local file_name=\"$1\" local\n dir=\"./\" elif [ $# -eq 2 ]; then local file_name=\"$1\" if\n [ -d \"$2\" ]; then local dir=\"$2\" else echo \"Directory\n $2 not found\" return fi else echo \"Usage: find_file\n <file_name> [<dir>]\" return fi dir=$(realpath \"$dir\") local\n matches=$(find \"$dir\" -type f -name \"$file_name\") # if no matches, return if\n [ -z \"$matches\" ]; then echo \"No matches found for \\\"$file_name\\\" in\n $dir\" return fi # Calculate total number of matches local\n num_matches=$(echo \"$matches\" | wc -l | awk ''{$1=$1; print $0}'') echo\n \"Found $num_matches matches for \\\"$file_name\\\" in $dir:\" echo \"$matches\"\n | awk ''{print $0}''}'\n docstring: finds all files with the given name in dir. If dir is not provided,\n searches in the current directory\n end_name: null\n name: find_file\n signature: find_file <file_name> [<dir>]\n - arguments:\n end_line:\n description: the line number to end the edit at (inclusive)\n required: true\n type: integer\n replacement_text:\n description: the text to replace the current selection with\n required: true\n type: string\n start_line:\n description: the line number to start the edit at\n required: true\n type: integer\n code: 'edit() { if [ -z \"$CURRENT_FILE\" ] then echo ''No file open.\n Use the `open` command first.'' return fi local start_line=\"$(echo\n $1: | cut -d: -f1)\" local end_line=\"$(echo $1: | cut -d: -f2)\" if [\n -z \"$start_line\" ] || [ -z \"$end_line\" ] then echo \"Usage: edit\n <start_line>:<end_line>\" return fi local re=''^[0-9]+$'' if\n ! [[ $start_line =~ $re ]]; then echo \"Usage: edit <start_line>:<end_line>\" echo\n \"Error: start_line must be a number\" return fi if ! [[ $end_line\n =~ $re ]]; then echo \"Usage: edit <start_line>:<end_line>\" echo\n \"Error: end_line must be a number\" return fi local linter_cmd=\"flake8\n --isolated --select=F821,F822,F831,E111,E112,E113,E999,E902\" local linter_before_edit=$($linter_cmd\n \"$CURRENT_FILE\" 2>&1) # Bash array starts at 0, so let''s adjust local\n start_line=$((start_line - 1)) local end_line=$((end_line)) local line_count=0 local\n replacement=() while IFS= read -r line do replacement+=(\"$line\") ((line_count++)) done #\n Create a backup of the current file cp \"$CURRENT_FILE\" \"/root/$(basename\n \"$CURRENT_FILE\")_backup\" # Read the file line by line into an array mapfile\n -t lines < \"$CURRENT_FILE\" local new_lines=(\"${lines[@]:0:$start_line}\"\n \"${replacement[@]}\" \"${lines[@]:$((end_line))}\") # Write the new stuff\n directly back into the original file printf \"%s\\n\" \"${new_lines[@]}\" >|\n \"$CURRENT_FILE\" # Run linter if [[ $CURRENT_FILE == *.py ]]; then _lint_output=$($linter_cmd\n \"$CURRENT_FILE\" 2>&1) lint_output=$(_split_string \"$_lint_output\" \"$linter_before_edit\"\n \"$((start_line+1))\" \"$end_line\" \"$line_count\") else # do nothing lint_output=\"\" fi #\n if there is no output, then the file is good if [ -z \"$lint_output\" ];\n then export CURRENT_LINE=$start_line _constrain_line _print echo\n \"File updated. Please review the changes and make sure they are correct (correct\n indentation, no duplicate lines, etc). Edit the file again if necessary.\" else echo\n \"Your proposed edit has introduced new syntax error(s). Please read this error\n message carefully and then retry editing the file.\" echo \"\" echo\n \"ERRORS:\" echo \"$lint_output\" echo \"\" # Save original\n values original_current_line=$CURRENT_LINE original_window=$WINDOW #\n Update values export CURRENT_LINE=$(( (line_count / 2) + start_line\n )) # Set to \"center\" of edit export WINDOW=$((line_count + 10)) # Show\n +/- 5 lines around edit echo \"This is how your edit would have looked\n if applied\" echo \"-------------------------------------------------\" _constrain_line _print echo\n \"-------------------------------------------------\" echo \"\" #\n Restoring CURRENT_FILE to original contents. cp \"/root/$(basename \"$CURRENT_FILE\")_backup\"\n \"$CURRENT_FILE\" export CURRENT_LINE=$(( ((end_line - start_line + 1)\n / 2) + start_line )) export WINDOW=$((end_line - start_line + 10)) echo\n \"This is the original code before your edit\" echo \"-------------------------------------------------\" _constrain_line _print echo\n \"-------------------------------------------------\" # Restore original\n values export CURRENT_LINE=$original_current_line export WINDOW=$original_window echo\n \"Your changes have NOT been applied. Please fix your edit command and try\n again.\" echo \"You either need to 1) Specify the correct start/end line\n arguments or 2) Correct your edit code.\" echo \"DO NOT re-run the same\n failed edit command. Running it again will lead to the same error.\" fi #\n Remove backup file rm -f \"/root/$(basename \"$CURRENT_FILE\")_backup\"}'\n docstring: replaces lines <start_line> through <end_line> (inclusive) with the\n given text in the open file. The replacement text is terminated by a line\n with only end_of_edit on it. All of the <replacement text> will be entered,\n so make sure your indentation is formatted properly. Python files will be\n checked for syntax errors after the edit. If the system detects a syntax error,\n the edit will not be executed. Simply try to edit the file again, but make\n sure to read the error message and modify the edit command you issue accordingly.\n Issuing the same command a second time will just lead to the same error message\n again.\n end_name: end_of_edit\n name: edit\n signature: |-\n edit <start_line>:<end_line>\n <replacement_text>\n end_of_edit\n - arguments:\n flag:\n description: the flag found\n required: true\n type: string\n code: 'submit() { if [ -z \"$1\" ]; then echo \"Usage: submit <flag>\" return fi echo\n -n \"<<SUBMISSION||\" echo -n \"$@\" echo \"||SUBMISSION>>\"}'\n docstring: submits your current flag and terminates the session, you should\n be aware to properly escape the flag as this is a bash command and to put\n your flag under single quotes.\n end_name: null\n name: submit\n signature: submit '<flag>'\n - arguments:\n binary_path:\n description: The path to the binary to be decompiled\n required: true\n type: file path\n function_name:\n description: The function name to be decompiled, or main by default\n required: false\n type: string\n code: |\n #!/root/miniconda3/bin/python\n\n # @yaml\n # signature: decompile <binary_path> [--function_name <function_name>]\n # docstring: Decompile a binary and prints the decompilation of a given function name\n # arguments:\n # binary_path:\n # type: file path\n # description: The path to the binary to be decompiled\n # required: true\n # function_name:\n # type: string\n # description: The function name to be decompiled, or main by default\n # required: false\n\n import argparse\n import subprocess\n import re\n import json\n import tempfile\n from typing import Annotated\n from pathlib import Path\n\n GHIDRA_BINARY = \"analyzeHeadless\"\n\n class Decompile:\n def __init__(self):\n pass\n\n def __call__(self,\n path: Annotated[str,\"path to the binary to decompile\"],\n function: Annotated[str,\"the function to decompile\"] = 'main'):\n \"\"\"Decompile a function from a binary using Ghidra.\"\"\"\n if path is None or not Path(path).is_file():\n return f\"Error: Binary {path} does not exist! Please try again with a real binary file.\"\n if function is None:\n function = \"main\"\n return self.decompile(path, function)\n\n def find_function(self, dis, function):\n \"\"\"Returns the name of the function found in the dict\"\"\"\n if function in dis[\"functions\"]:\n return function\n # Looking for main entry point, so try other names also\n if function == \"main\":\n if \"main\" in dis:\n return dis[\"main\"]\n if \"_start\" in dis[\"functions\"]:\n return \"_start\"\n if \"invoke_main\" in dis[\"functions\"]:\n return \"invoke_main\"\n if \"entry\" in dis[\"functions\"]:\n return \"entry\"\n # Check if requesting radare2 unnamed function with address\n if re.match(r\"fcn\\.[0-9a-f]+$\", function):\n addr = function[4:]\n if addr in dis[\"addresses\"]:\n return dis[\"addresses\"][addr]\n # Nothing found\n return None\n\n def decompile(self, binary, function):\n # Look for the decompilation output in \"decomp\"\n basename = Path(binary).name\n decomp_output = Path(f\"/ghidra_out/{basename}.decomp.json\")\n if not decomp_output.exists():\n if not self.run_ghidra(basename, decomp_output):\n return f\"Error: Decompilation for {binary} not available\"\n self.ghidra_out = json.loads(decomp_output.read_text())\n\n if found := self.find_function(self.ghidra_out, function):\n ret = self.ghidra_out[\"functions\"][found]\n if found == function:\n # Exact name found\n return f\"Decompilation Found!\\n{ret}\"\n else:\n return f\"Function {function} not found! Instead, here is the decompilation of equivalent function {found}:\\n{ret}\"\n else:\n funclist = \", \".join(self.ghidra_out['functions'].keys())\n return f\"Error: Function {function} not found in {binary}.\\nThese are the available functions found: {funclist}\"\n\n def run_ghidra(self, binary, output):\n real_binary = Path(binary)\n if not real_binary or not real_binary.exists():\n return False\n output.parent.mkdir(parents=True, exist_ok=True)\n with tempfile.TemporaryDirectory() as tmpdir:\n tmpdir = Path(tmpdir)\n subprocess.run(\n [GHIDRA_BINARY, tmpdir, \"DummyProj\", \"-scriptpath\", '/ghidra_scripts',\n \"-import\", real_binary, \"-postscript\", \"DecompileToJson.java\", output],\n check=False, capture_output=True,\n )\n return output.exists()\n\n\n if __name__ == \"__main__\":\n parser = argparse.ArgumentParser(\n description=\"Print the contents of a Python file, skipping lengthy function and method definitions.\"\n )\n parser.add_argument(\"file_path\", type=str, help=\"The path to the binary to be decompiled\")\n parser.add_argument(\"--function_name\", type=str, help=\"The function name to be decompiled\", required=False, default=\"main\")\n args = parser.parse_args()\n print(Decompile()(args.file_path, args.function_name))\n docstring: Decompile a binary and prints the decompilation of a given function\n name\n end_name: null\n name: decompile\n signature: decompile <binary_path> [--function_name <function_name>]\n - arguments:\n binary_path:\n description: The path to the binary to be disassembled\n required: true\n type: file path\n function_name:\n description: The function name to be disassembled, or main by default\n required: false\n type: string\n code: |\n #!/root/miniconda3/bin/python\n\n # @yaml\n # signature: disassemble <binary_path> [--function_name <function_name>]\n # docstring: Disassembles a binary and prints the disassembly of a given function name\n # arguments:\n # binary_path:\n # type: file path\n # description: The path to the binary to be disassembled\n # required: true\n # function_name:\n # type: string\n # description: The function name to be disassembled, or main by default\n # required: false\n\n import argparse\n import subprocess\n import re\n import json\n import tempfile\n from typing import Annotated\n from pathlib import Path\n\n GHIDRA_BINARY = \"analyzeHeadless\"\n\n class Disassemble:\n def __init__(self):\n pass\n\n def __call__(self,\n path: Annotated[str,\"path to the binary to disassemble\"],\n function: Annotated[str,\"the function to disassemble\"] = 'main'):\n \"\"\"Disassemble a function from a binary using Ghidra.\"\"\"\n if path is None or not Path(path).is_file():\n return f\"Error: Binary {path} does not exist! Please try again with a real binary file.\"\n if function is None:\n function = \"main\"\n return self.disassemble(path, function)\n\n def find_function(self, dis, function):\n \"\"\"Returns the name of the function found in the dict\"\"\"\n if function in dis[\"functions\"]:\n return function\n # Looking for main entry point, so try other names also\n if function == \"main\":\n if \"main\" in dis:\n return dis[\"main\"]\n if \"_start\" in dis[\"functions\"]:\n return \"_start\"\n if \"invoke_main\" in dis[\"functions\"]:\n return \"invoke_main\"\n if \"entry\" in dis[\"functions\"]:\n return \"entry\"\n # Check if requesting radare2 unnamed function with address\n if re.match(r\"fcn\\.[0-9a-f]+$\", function):\n addr = function[4:]\n if addr in dis[\"addresses\"]:\n return dis[\"addresses\"][addr]\n # Nothing found\n return None\n\n def disassemble(self, binary, function):\n # Look for the disassembly output in \"disas\"\n basename = Path(binary).name\n disas_output = Path(f\"/ghidra_out/{basename}.disas.json\")\n if not disas_output.exists():\n if not self.run_ghidra(basename, disas_output):\n return f\"Error: Disassembly for {binary} not available\"\n self.ghidra_out = json.loads(disas_output.read_text())\n\n if found := self.find_function(self.ghidra_out, function):\n ret = self.ghidra_out[\"functions\"][found]\n if found == function:\n # Exact name found\n return f\"Disassembly Found!\\n{ret}\"\n else:\n return f\"Function {function} not found! Instead, here is the disassembly of equivalent function {found}:\\n{ret}\"\n else:\n funclist = \", \".join(self.ghidra_out[\"functions\"].keys())\n return f\"Error: Function {function} not found in {binary}.\\nThese are the available functions found: {funclist}\"\n\n def run_ghidra(self, binary, output):\n real_binary = Path(binary)\n if not real_binary or not real_binary.exists():\n return False\n output.parent.mkdir(parents=True, exist_ok=True)\n with tempfile.TemporaryDirectory() as tmpdir:\n tmpdir = Path(tmpdir)\n subprocess.run(\n [GHIDRA_BINARY, tmpdir, \"DummyProj\", \"-scriptpath\", '/ghidra_scripts',\n \"-import\", real_binary, \"-postscript\", \"DisassembleToJson.java\", output],\n check=False, capture_output=True,\n )\n return output.exists()\n\n\n if __name__ == \"__main__\":\n parser = argparse.ArgumentParser(\n description=\"Print the contents of a Python file, skipping lengthy function and method definitions.\"\n )\n parser.add_argument(\"file_path\", type=str, help=\"The path to the binary to be disassembled\")\n parser.add_argument(\"--function_name\", type=str, help=\"The function name to be disassembled\", required=False, default=\"main\")\n args = parser.parse_args()\n print(Disassemble()(args.file_path, args.function_name))\n docstring: Disassembles a binary and prints the disassembly of a given function\n name\n end_name: null\n name: disassemble\n signature: disassemble <binary_path> [--function_name <function_name>]\n - arguments:\n args:\n description: optional command-line arguments for the binary\n required: false\n type: string\n binary:\n description: the path to the binary to debug\n required: true\n type: string\n code: 'debug_start() { if [ -z \"$1\" ] then echo \"Usage: debug_start\n <binary>\" return fi if [ ! -x \"$1\" ] then echo \"Error:\n File $1 does not exist, or is not executable\" return fi fp=$(realpath\n $1) _debug_command \"SESSION=gdb\" _debug_command \"START\" _debug_command\n \"set confirm off\" _debug_command \"file $fp\" if [ ! -z \"$2\" ] then _debug_command\n \"set args ${@:2:$#}\" # Set arguments from 2 until the end fi _debug_command\n \"starti\" export INTERACTIVE_SESSION=\"gdb $@\"}'\n docstring: Starts a debug session with the given binary.\n end_name: null\n name: debug_start\n signature: debug_start <binary> [<args>]\n - arguments:\n breakpoint:\n description: The breakpoint location, which may be a function name, address,\n or filename and line number.\n required: true\n type: string\n code: 'debug_add_breakpoint() { if [ -z \"$1\" ] then echo \"Usage:\n debug_add_breakpoint <breakpoint>\" return fi _debug_command \"SESSION=gdb\" _debug_command\n ''break ''$1}'\n docstring: Adds a breakpoint in the debug session\n end_name: null\n name: debug_add_breakpoint\n signature: debug_add_breakpoint <breakpoint>\n - arguments: null\n code: debug_continue() { _debug_command \"SESSION=gdb\" _debug_command 'continue'}\n docstring: Continues the program execution in the debug session.\n end_name: null\n name: debug_continue\n signature: debug_continue\n - arguments:\n number:\n description: number of instructions to step (default is 1)\n required: false\n type: integer\n code: 'debug_step() { if [ -z \"$1\" ] then _debug_command \"SESSION=gdb\" _debug_command\n ''stepi'' elif [[ ((\"$1\" -eq \"$1\") && (\"$1\" -gt \"0\")) ]] # Check if integer\n and positive then _debug_command \"SESSION=gdb\" _debug_command\n ''stepi ''$1 else echo \"Please provide a positive integer for number\n of instructions.\" echo \"Usage: debug_step [number]\" fi}'\n docstring: Steps number of instructions in the debug session.\n end_name: null\n name: debug_step\n signature: debug_step [number]\n - arguments:\n command:\n description: command to execute (wrap in single quotes to avoid shell escaping\n and substitution)\n required: true\n type: string\n code: 'debug_exec() { if [ -z \"$1\" ] then echo \"Usage: debug_exec\n <command>\" return fi _debug_command \"SESSION=gdb\" _debug_command\n \"$1\"}'\n docstring: Executes arbitrary gdb command in debug session.\n end_name: null\n name: debug_exec\n signature: debug_exec <command>\n - arguments: null\n code: debug_stop() { _debug_command \"SESSION=gdb\" _debug_command \"quit\" _debug_command\n \"STOP\" unset INTERACTIVE_SESSION}\n docstring: Stops the current debug session.\n end_name: null\n name: debug_stop\n signature: debug_stop\n - arguments:\n port:\n description: desired port for connection\n required: true\n type: int\n server_address:\n description: the server address to initiate connection to\n required: true\n type: string\n code: 'connect_start() { if [ -z \"$1\" ] || [ -z \"$2\" ] then echo\n \"Usage: connect_start <server_address> <port>\" return fi _connect_command\n \"SESSION=connect\" _connect_command \"START\" _connect_command \"connect\n $1 $2\" export INTERACTIVE_SESSION=\"connect $@\"}'\n docstring: Starts a new interactive connection to the server address and port.\n end_name: null\n name: connect_start\n signature: connect_start <server_address> <port>\n - arguments:\n line:\n description: The line to send to the connection\n required: false\n type: string\n code: connect_sendline() { _connect_command \"SESSION=connect\" _connect_command\n 'sendline '$@}\n docstring: Sends a line with unicode/hexadecimal values to the connection. Sending\n hexadecimal bytes should be done using \\x<hh> where hh stands for the sepecific\n byte you want to send.\n end_name: null\n name: connect_sendline\n signature: connect_sendline [<line>]\n - arguments:\n command:\n description: command to execute (wrap in single quotes to avoid shell escaping\n and substitution)\n required: true\n type: string\n code: 'connect_exec() { if [ -z \"$1\" ] then echo \"Usage: connect_exec\n <command>\" return fi _connect_command \"SESSION=connect\" _connect_command\n \"$@\"}'\n docstring: Executes arbitrary connect command in connect session.\n end_name: null\n name: connect_exec\n signature: connect_exec <command>\n - arguments: null\n code: connect_stop() { _connect_command \"SESSION=connect\" _connect_command\n \"quit\" _connect_command \"STOP\" unset INTERACTIVE_SESSION}\n docstring: Stops the current connect session.\n end_name: null\n name: connect_stop\n signature: connect_stop\n - arguments: null\n code: \"exit_forfeit() { echo \\\"Forfeit \\U0001F625\\\" # Implementation note:\\\n \\ This is actually caught directly in swe_env.py # We only include this\\\n \\ stub so that the command is shown to the agent.}\"\n docstring: Give up on the current challenge and terminate the session.\n end_name: null\n name: exit_forfeit\n signature: exit_forfeit\n _subroutines: {}\n block_unless_regex:\n r2: \\b(?:radare2)\\b.*\\s+-c\\s+.*\n radare2: \\b(?:radare2)\\b.*\\s+-c\\s+.*\n blocklist:\n - vim\n - vi\n - emacs\n - nano\n - nohup\n - gdb\n blocklist_error_template: |\n Interactive operation '{name}' is not supported by this environment.\n Please consider using one of the interactive commands available to you in this environment.\n blocklist_standalone:\n - python\n - python3\n - ipython\n - bash\n - sh\n - exit\n - /bin/bash\n - /bin/sh\n - nohup\n - vi\n - vim\n - emacs\n - nano\n command_docs: |+\n open:\n docstring: opens the file at the given path in the editor. If line_number is provided, the window will be move to include that line\n signature: open <path> [<line_number>]\n arguments:\n - path (string) [required]: the path to the file to open\n - line_number (integer) [optional]: the line number to move the window to (if not provided, the window will start at the top of the file)\n\n goto:\n docstring: moves the window to show <line_number>\n signature: goto <line_number>\n arguments:\n - line_number (integer) [required]: the line number to move the window to\n\n scroll_down:\n docstring: moves the window down 100 lines\n signature: scroll_down\n\n scroll_up:\n docstring: moves the window down 100 lines\n signature: scroll_up\n\n create:\n docstring: creates and opens a new file with the given name\n signature: create <filename>\n arguments:\n - filename (string) [required]: the name of the file to create\n\n search_dir:\n docstring: searches for search_term in all files in dir. If dir is not provided, searches in the current directory\n signature: search_dir <search_term> [<dir>]\n arguments:\n - search_term (string) [required]: the term to search for\n - dir (string) [optional]: the directory to search in (if not provided, searches in the current directory)\n\n search_file:\n docstring: searches for search_term in file. If file is not provided, searches in the current open file\n signature: search_file <search_term> [<file>]\n arguments:\n - search_term (string) [required]: the term to search for\n - file (string) [optional]: the file to search in (if not provided, searches in the current open file)\n\n find_file:\n docstring: finds all files with the given name in dir. If dir is not provided, searches in the current directory\n signature: find_file <file_name> [<dir>]\n arguments:\n - file_name (string) [required]: the name of the file to search for\n - dir (string) [optional]: the directory to search in (if not provided, searches in the current directory)\n\n edit:\n docstring: replaces lines <start_line> through <end_line> (inclusive) with the given text in the open file. The replacement text is terminated by a line with only end_of_edit on it. All of the <replacement text> will be entered, so make sure your indentation is formatted properly. Python files will be checked for syntax errors after the edit. If the system detects a syntax error, the edit will not be executed. Simply try to edit the file again, but make sure to read the error message and modify the edit command you issue accordingly. Issuing the same command a second time will just lead to the same error message again.\n signature: edit <start_line>:<end_line>\n <replacement_text>\n end_of_edit\n arguments:\n - start_line (integer) [required]: the line number to start the edit at\n - end_line (integer) [required]: the line number to end the edit at (inclusive)\n - replacement_text (string) [required]: the text to replace the current selection with\n\n submit:\n docstring: submits your current flag and terminates the session, you should be aware to properly escape the flag as this is a bash command and to put your flag under single quotes.\n signature: submit '<flag>'\n arguments:\n - flag (string) [required]: the flag found\n\n decompile:\n docstring: Decompile a binary and prints the decompilation of a given function name\n signature: decompile <binary_path> [--function_name <function_name>]\n arguments:\n - binary_path (file path) [required]: The path to the binary to be decompiled\n - function_name (string) [optional]: The function name to be decompiled, or main by default\n\n disassemble:\n docstring: Disassembles a binary and prints the disassembly of a given function name\n signature: disassemble <binary_path> [--function_name <function_name>]\n arguments:\n - binary_path (file path) [required]: The path to the binary to be disassembled\n - function_name (string) [optional]: The function name to be disassembled, or main by default\n\n debug_start:\n docstring: Starts a debug session with the given binary.\n signature: debug_start <binary> [<args>]\n arguments:\n - binary (string) [required]: the path to the binary to debug\n - args (string) [optional]: optional command-line arguments for the binary\n\n debug_add_breakpoint:\n docstring: Adds a breakpoint in the debug session\n signature: debug_add_breakpoint <breakpoint>\n arguments:\n - breakpoint (string) [required]: The breakpoint location, which may be a function name, address, or filename and line number.\n\n debug_continue:\n docstring: Continues the program execution in the debug session.\n signature: debug_continue\n\n debug_step:\n docstring: Steps number of instructions in the debug session.\n signature: debug_step [number]\n arguments:\n - number (integer) [optional]: number of instructions to step (default is 1)\n\n debug_exec:\n docstring: Executes arbitrary gdb command in debug session.\n signature: debug_exec <command>\n arguments:\n - command (string) [required]: command to execute (wrap in single quotes to avoid shell escaping and substitution)\n\n debug_stop:\n docstring: Stops the current debug session.\n signature: debug_stop\n\n connect_start:\n docstring: Starts a new interactive connection to the server address and port.\n signature: connect_start <server_address> <port>\n arguments:\n - server_address (string) [required]: the server address to initiate connection to\n - port (int) [required]: desired port for connection\n\n connect_sendline:\n docstring: Sends a line with unicode/hexadecimal values to the connection. Sending hexadecimal bytes should be done using \\x<hh> where hh stands for the sepecific byte you want to send.\n signature: connect_sendline [<line>]\n arguments:\n - line (string) [optional]: The line to send to the connection\n\n connect_exec:\n docstring: Executes arbitrary connect command in connect session.\n signature: connect_exec <command>\n arguments:\n - command (string) [required]: command to execute (wrap in single quotes to avoid shell escaping and substitution)\n\n connect_stop:\n docstring: Stops the current connect session.\n signature: connect_stop\n\n exit_forfeit:\n docstring: Give up on the current challenge and terminate the session.\n signature: exit_forfeit\n\n command_files:\n - /home/talorabr/SWE-agent-ctf/config/commands/defaults.sh\n - /home/talorabr/SWE-agent-ctf/config/commands/search.sh\n - /home/talorabr/SWE-agent-ctf/config/commands/edit_linting.sh\n - /home/talorabr/SWE-agent-ctf/config/commands/_split_string.py\n - /home/talorabr/SWE-agent-ctf/config/commands/submit_ctf.sh\n - /home/talorabr/SWE-agent-ctf/config/commands/decompile.py\n - /home/talorabr/SWE-agent-ctf/config/commands/disassemble.py\n - /home/talorabr/SWE-agent-ctf/config/commands/debug.sh\n - /home/talorabr/SWE-agent-ctf/config/commands/_connect.py\n - /home/talorabr/SWE-agent-ctf/config/commands/server_connection.sh\n - /home/talorabr/SWE-agent-ctf/config/commands/forfeit.sh\n demonstration_template: null\n demonstrations: []\n env_variables:\n CURRENT_FILE: ''\n CURRENT_LINE: '0'\n OVERLAP: '2'\n SEARCH_FILES: ()\n SEARCH_INDEX: '0'\n SEARCH_RESULTS: ()\n WINDOW: '100'\n format_error_template: |\n Your output was not formatted correctly. You must always include one discussion and one command as part of your response. Make sure you do not have multiple discussion/command tags.\n Please make sure your output precisely matches the following format:\n DISCUSSION\n Discuss here with yourself about what your planning and what you're going to do in this step.\n\n ```\n command(s) that you're going to run\n ```\n history_processor: {}\n history_processor_args: {}\n instance_template: |-\n We're currently solving the following CTF challenge. The CTF challenge is a {category_friendly} problem named \"{name}\", worth {points} points. The description is:\n {description}\n Files included in the challenge: {files}\n\n {server_description}\n\n INSTRUCTIONS:\n Now, you're going to solve this challenge on your own. Your terminal session has started and you're in the challenge's root directory. You can use any bash commands or the special interface to help you.\n Solve the challenge, executing any commands you need and writing any code that may be helpful. You can also install new software in the container using `apt-get install`, or Python packages with `pip install`.\n Remember, YOU CAN ONLY ENTER ONE COMMAND AT A TIME. You should always wait for feedback after every command.\n When you think you have found the flag, you can submit your solution by simply running the submit command.\n Note however that you cannot use any interactive session commands (e.g. python, vim) in this environment, but you can write scripts and run them. E.g. you can write a python script and then run it with `python <script_name>.py`.\n\n NOTE ABOUT THE EDIT COMMAND: Indentation really matters! When editing a file, make sure to insert appropriate indentation before each line!\n\n IMPORTANT TIPS:\n 1. When you edit an existing file, try to minimize the changes you make to the file.\n\n 2. Please make sure that you make use of the interactive tools supplied to you! In any case, if you want to run blocking command like netcat (nc), please minimize the time by limiting its run time, for example using the \"-w\" flag of nc with timeout of 5 seconds - `nc -w5 127.0.0.1 80`.\n\n 3. Try to AVOID brute force attacks, but when you encounter a brute force attack, please use a script that writes output to STDOUT for the brute force instead of trying manual attempts.\n\n 4. Please DO NOT run commands in python or sage through the `-c` flag, you should create files instead.\n\n 5. Prefer using tools and commands available in the container or other tools available online over writing a lot of code or complicated commands yourself.\n In particular, prefer using `open` instead of `cat` and `search_file` instead of `grep`, and prefer using the interactive commands supplied to you!\n\n 6. DO NOT run any command that produces binary output! Instead try to use `hexdump` or `xxd`.\n\n 7. Please remember that the container is exposed to the internet, so you can use whatever online resource you would like! Use it wisely!\n\n 8. Do not use any interactive commands AT ALL! Interactive commands are only available through the commands supplied to you at the beginning - make use of them!\n\n 9. YOU CAN SOLVE THE CHALLENGES BY YOURSELF COMPLETELY - I TRUST IN YOU!\n\n (Open file: {open_file})\n (Current directory: {working_dir})\n (Interactive session: {interactive_session})\n bash-$\n next_step_no_output_template: |-\n Your command ran successfully and did not produce any output.\n (Open file: {open_file})\n (Current directory: {working_dir})\n (Interactive session: {interactive_session})\n bash-$\n next_step_template: |-\n {observation}\n (Open file: {open_file})\n (Current directory: {working_dir})\n (Interactive session: {interactive_session})\n bash-$\n parse_command: {}\n parse_function: {}\n put_demos_in_history: false\n review_loop_config: null\n state_command:\n arguments: null\n code: |\n state() {\n local working_dir=\"$PWD\";\n local open_file=\"n/a\";\n local interactive_session=\"${INTERACTIVE_SESSION:-n/a}\";\n if [ ! -z $CURRENT_FILE ]; then\n open_file=$(realpath $CURRENT_FILE);\n fi\n\n echo '{\"open_file\": \"'$open_file'\", \"working_dir\": \"'$working_dir'\", \"interactive_session\": \"'$interactive_session'\"}'\n };\n docstring: null\n end_name: null\n name: state\n signature: null\n strategy_template: null\n submit_command: submit\n subroutine_types: []\n summarizer_config:\n function: {}\n instance_template: null\n model: null\n system_template: null\n template: null\n window_length: 105\n system_template: |-\n SETTING: You are a skilled cybersecurity professional and top CTF player.\n Your goal is to solve the CTF challenge and find the flag. The flag format for this CTF is {flag_format}.\n\n You have access to a Linux container environment with a shell and a live network environment that has access to the CTF challenge server (if applicable), and you're working directly in the command line with a special interface.\n The container comes with pre-installed software and python packages that are useful for solving these challenges including: pwntools, radare2, wine, wine32, gmpy2, sagemath, pycryptodome, sympy, RsaCtfTool.py, tshark, sqlmap and nikto.\n\n The special interface consists of a file editor that shows you {WINDOW} lines of a file at a time.\n In addition to typical bash commands, you can also use the following commands to help you navigate and edit files.\n\n COMMANDS:\n {command_docs}\n\n Please note that THE EDIT COMMAND REQUIRES PROPER INDENTATION.\n If you'd like to add the line ' print(x)' you must fully write that out, with all those spaces before the code! Indentation is important and code that is not indented correctly will fail and require fixing before it can be run.\n\n RESPONSE FORMAT:\n Your shell prompt is formatted as follows:\n (Open file: <path>) <cwd> $\n\n You need to format your output using two fields; discussion and command.\n Your output should always include _one_ discussion and _one_ command field EXACTLY as in the following example:\n DISCUSSION\n First I'll start by using ls to see what files are in the current directory. Then maybe we can look at some relevant files to see what they look like.\n ```\n ls -a\n ```\n\n You should only include a *SINGLE* command in the command section and then wait for a response from the shell before continuing with more discussion and commands. Everything you include in the DISCUSSION section will be saved for future reference.\n If you'd like to issue two commands at once, PLEASE DO NOT DO THAT! Please instead first submit just the first command, and then after receiving a response you'll be able to issue the second command.\n You're free to use any other bash commands you want (e.g. find, grep, cat, ls, cd) in addition to the special commands listed above.\n However, the environment does NOT support interactive session commands (e.g. python, vim), so please do not invoke them.\n util_functions:\n - arguments: null\n code: '_print() { local total_lines=$(awk ''END {print NR}'' $CURRENT_FILE) echo\n \"[File: $(realpath $CURRENT_FILE) ($total_lines lines total)]\" lines_above=$(jq\n -n \"$CURRENT_LINE - $WINDOW/2\" | jq ''[0, .] | max | floor'') lines_below=$(jq\n -n \"$total_lines - $CURRENT_LINE - $WINDOW/2\" | jq ''[0, .] | max | round'') if\n [ $lines_above -gt 0 ]; then echo \"($lines_above more lines above)\" fi cat\n $CURRENT_FILE | grep -n $ | head -n $(jq -n \"[$CURRENT_LINE + $WINDOW/2, $WINDOW/2]\n | max | floor\") | tail -n $(jq -n \"$WINDOW\") if [ $lines_below -gt 0 ];\n then echo \"($lines_below more lines below)\" fi}'\n docstring: null\n end_name: null\n name: _print\n signature: _print\n - arguments: null\n code: _constrain_line() { if [ -z \"$CURRENT_FILE\" ] then echo \"No\n file open. Use the open command first.\" return fi local max_line=$(awk\n 'END {print NR}' $CURRENT_FILE) local half_window=$(jq -n \"$WINDOW/2\" |\n jq 'floor') export CURRENT_LINE=$(jq -n \"[$CURRENT_LINE, $max_line - $half_window]\n | min\") export CURRENT_LINE=$(jq -n \"[$CURRENT_LINE, $half_window] | max\")}\n docstring: null\n end_name: null\n name: _constrain_line\n signature: _constrain_line\n - arguments: null\n code: '_scroll_warning_message() { # Warn the agent if we scroll too many\n times # Message will be shown if scroll is called more than WARN_AFTER_SCROLLING_TIMES\n (default 3) times # Initialize variable if it''s not set export SCROLL_COUNT=${SCROLL_COUNT:-0} #\n Reset if the last command wasn''t about scrolling if [ \"$LAST_ACTION\" !=\n \"scroll_up\" ] && [ \"$LAST_ACTION\" != \"scroll_down\" ]; then export SCROLL_COUNT=0 fi #\n Increment because we''re definitely scrolling now export SCROLL_COUNT=$((SCROLL_COUNT\n + 1)) if [ $SCROLL_COUNT -ge ${WARN_AFTER_SCROLLING_TIMES:-3} ]; then echo\n \"\" echo \"WARNING: Scrolling many times in a row is very inefficient.\" echo\n \"If you know what you are looking for, use \\`search_file <pattern>\\` instead.\" echo\n \"\" fi}'\n docstring: null\n end_name: null\n name: _scroll_warning_message\n signature: _scroll_warning_message\n - arguments: null\n code: _debug_command() { echo \"<<INTERACTIVE||$@||INTERACTIVE>>\"}\n docstring: null\n end_name: null\n name: _debug_command\n signature: _debug_command\n - arguments: null\n code: _connect_command() { echo \"<<INTERACTIVE||$@||INTERACTIVE>>\"}\n docstring: null\n end_name: null\n name: _connect_command\n signature: _connect_command\n config_file: config/default_ctf.yaml\n model:\n host_url: localhost:11434\n model_name: gpt4\n per_instance_cost_limit: 2.0\n replay_path: null\n temperature: 0.0\n top_p: 0.95\n total_cost_limit: 0.0\nenvironment:\n base_commit: null\n cache_task_images: false\n container_mounts: []\n container_name: null\n data_path: ../LLM_CTF_Database/2018/CSAW-Finals/misc/leaked_flag/challenge.json\n environment_setup: null\n image_name: sweagent/enigma:latest\n install_environment: true\n interactive_sessions_config:\n connect:\n cmdline: /root/commands/_connect\n exit_command: connect_stop\n quit_commands_in_session:\n - quit\n signal_for_interrupt_limit: 3\n start_command: connect_start\n terminal_prompt_pattern: '(nc) '\n timeout_duration_on_interrupt: 5\n gdb:\n cmdline: gdb\n exit_command: debug_stop\n quit_commands_in_session:\n - quit\n signal_for_interrupt_limit: 3\n start_command: debug_start\n terminal_prompt_pattern: '(gdb) '\n timeout_duration_on_interrupt: 5\n no_mirror: false\n repo_path: ../LLM_CTF_Database/2018/CSAW-Finals/misc/leaked_flag\n split: dev\n timeout: null\n verbose: true\ninstance_filter: .*\nprint_config: true\nraise_exceptions: false\nskip_existing: true\nsuffix: ''\n\n2024-09-19 11:26:13,116 INFO \ud83d\udcbd Loaded dataset from ../LLM_CTF_Database/2018/CSAW-Finals/misc/leaked_flag/challenge.json\n2024-09-19 11:26:13,162 INFO Found image sweagent/enigma:latest with tags: ['sweagent/enigma:0.1.0', 'sweagent/enigma:latest', 'sweagent/swe-ctf:latest'], created: 2024-08-29T19:52:43.576514108Z for linux amd64.\n2024-09-19 11:26:13,163 DEBUG Starting container with command: docker run -i --rm --name sweagent-enigma-latest-480a0f32bf sweagent/enigma:latest /bin/bash -l\n2024-09-19 11:26:14,189 INFO \ud83c\udf31 Environment Initialized\n2024-09-19 11:26:14,191 TRACE Input:\nsource /root/.bashrc\n2024-09-19 11:26:14,251 TRACE Output:\n\n2024-09-19 11:26:14,251 TRACE Input:\nmkdir -p /root/commands\n2024-09-19 11:26:14,312 TRACE Output:\n\n2024-09-19 11:26:14,312 TRACE Input:\ntouch /root/commands/__init__.py\n2024-09-19 11:26:14,372 TRACE Output:\n\n2024-09-19 11:26:14,373 TRACE Input:\nexport PATH=$PATH:/root/commands\n2024-09-19 11:26:14,433 TRACE Output:\n\n2024-09-19 11:26:14,434 DEBUG Environment initialization took 1.36 seconds\n2024-09-19 11:26:14,470 INFO \u25b6\ufe0f Beginning task 0\n2024-09-19 11:26:14,471 DEBUG Attaching NIC to container with command: docker network connect ctfnet sweagent-enigma-latest-480a0f32bf\n2024-09-19 11:26:14,548 TRACE Input:\ncd /\n2024-09-19 11:26:14,609 TRACE Output:\n\n2024-09-19 11:26:14,609 TRACE Input:\nls\n2024-09-19 11:26:14,670 TRACE Output:\nbin\nboot\ndev\netc\nghidra_scripts\nhome\nlib\nlib32\nlib64\nlibx32\nmedia\nmnt\nopt\nproc\nroot\nrun\nsbin\nsnap\nsrv\nsys\ntmp\nusr\nvar\n2024-09-19 11:26:14,670 TRACE Input:\nmkdir __home__talorabr__LLM_CTF_Database__2018__CSAW-Finals__misc__leaked_flag\n2024-09-19 11:26:14,731 TRACE Output:\n\n2024-09-19 11:26:14,731 TRACE Input:\nchown -R root:root __home__talorabr__LLM_CTF_Database__2018__CSAW-Finals__misc__leaked_flag\n2024-09-19 11:26:14,791 TRACE Output:\n\n2024-09-19 11:26:14,791 TRACE Input:\necho -n > /root/files_to_edit.txt && cd /__home__talorabr__LLM_CTF_Database__2018__CSAW-Finals__misc__leaked_flag && export ROOT=$(pwd -P)\n2024-09-19 11:26:14,852 TRACE Output:\n\n2024-09-19 11:26:14,852 TRACE Input:\nexport CURRENT_FILE=\"\" && export CURRENT_LINE=0 && export SEARCH_RESULTS=() && export SEARCH_FILES=() && export SEARCH_INDEX=0\n2024-09-19 11:26:14,913 TRACE Output:\n\n2024-09-19 11:26:14,913 TRACE Input:\nsource /root/miniconda3/etc/profile.d/conda.sh\n2024-09-19 11:26:14,973 TRACE Output:\n\n2024-09-19 11:26:14,973 TRACE Input:\nuname -s\n2024-09-19 11:26:15,034 TRACE Output:\nLinux\n2024-09-19 11:26:15,034 TRACE Input:\nuname -m\n2024-09-19 11:26:15,095 TRACE Output:\nx86_64\n2024-09-19 11:26:15,095 TRACE Input:\napt update; apt install build-essential -y\n2024-09-19 11:26:24,204 TRACE Output:\n\nWARNING: apt does not have a stable CLI interface. Use with caution in scripts.\n\nGet:1 http://security.ubuntu.com/ubuntu jammy-security InRelease [129 kB]\nHit:2 http://archive.ubuntu.com/ubuntu jammy InRelease\nGet:3 http://archive.ubuntu.com/ubuntu jammy-updates InRelease [128 kB]\nGet:4 http://security.ubuntu.com/ubuntu jammy-security/main amd64 Packages [2308 kB]\nGet:5 http://security.ubuntu.com/ubuntu jammy-security/main i386 Packages [673 kB]\nGet:6 http://archive.ubuntu.com/ubuntu jammy-backports InRelease [127 kB]\nGet:7 http://security.ubuntu.com/ubuntu jammy-security/restricted amd64 Packages [3097 kB]\nGet:8 http://security.ubuntu.com/ubuntu jammy-security/restricted i386 Packages [45.6 kB]\nGet:9 http://security.ubuntu.com/ubuntu jammy-security/universe amd64 Packages [1150 kB]\nGet:10 http://security.ubuntu.com/ubuntu jammy-security/universe i386 Packages [782 kB]\nGet:11 http://security.ubuntu.com/ubuntu jammy-security/multiverse amd64 Packages [44.7 kB]\nGet:12 http://archive.ubuntu.com/ubuntu jammy-updates/main amd64 Packages [2585 kB]\nGet:13 http://archive.ubuntu.com/ubuntu jammy-updates/main i386 Packages [874 kB]\nGet:14 http://archive.ubuntu.com/ubuntu jammy-updates/restricted i386 Packages [47.6 kB]\nGet:15 http://archive.ubuntu.com/ubuntu jammy-updates/restricted amd64 Packages [3181 kB]\nGet:16 http://archive.ubuntu.com/ubuntu jammy-updates/universe amd64 Packages [1440 kB]\nGet:17 http://archive.ubuntu.com/ubuntu jammy-updates/universe i386 Packages [915 kB]\nGet:18 http://archive.ubuntu.com/ubuntu jammy-updates/multiverse amd64 Packages [51.8 kB]\nGet:19 http://archive.ubuntu.com/ubuntu jammy-backports/universe amd64 Packages [33.7 kB]\nGet:20 http://archive.ubuntu.com/ubuntu jammy-backports/universe i386 Packages [19.8 kB]\nFetched 17.6 MB in 2s (8383 kB/s)\nReading package lists...\nBuilding dependency tree...\nReading state information...\n63 packages can be upgraded. Run 'apt list --upgradable' to see them.\n\nWARNING: apt does not have a stable CLI interface. Use with caution in scripts.\n\nReading package lists...\nBuilding dependency tree...\nReading state information...\nbuild-essential is already the newest version (12.9ubuntu3).\n0 upgraded, 0 newly installed, 0 to remove and 63 not upgraded.\n2024-09-19 11:26:24,205 WARNING install_environment is set to True, but the data path is a GitHub URL without an environment config file (environment_config key/flag). Skipping conda environment installation.\n2024-09-19 11:26:24,206 TRACE Input:\npip install flake8\n2024-09-19 11:26:26,081 TRACE Output:\nCollecting flake8\n Downloading flake8-7.1.1-py2.py3-none-any.whl.metadata (3.8 kB)\nCollecting mccabe<0.8.0,>=0.7.0 (from flake8)\n Downloading mccabe-0.7.0-py2.py3-none-any.whl.metadata (5.0 kB)\nCollecting pycodestyle<2.13.0,>=2.12.0 (from flake8)\n Downloading pycodestyle-2.12.1-py2.py3-none-any.whl.metadata (4.5 kB)\nCollecting pyflakes<3.3.0,>=3.2.0 (from flake8)\n Downloading pyflakes-3.2.0-py2.py3-none-any.whl.metadata (3.5 kB)\nDownloading flake8-7.1.1-py2.py3-none-any.whl (57 kB)\nDownloading mccabe-0.7.0-py2.py3-none-any.whl (7.3 kB)\nDownloading pycodestyle-2.12.1-py2.py3-none-any.whl (31 kB)\nDownloading pyflakes-3.2.0-py2.py3-none-any.whl (62 kB)\nInstalling collected packages: pyflakes, pycodestyle, mccabe, flake8\nSuccessfully installed flake8-7.1.1 mccabe-0.7.0 pycodestyle-2.12.1 pyflakes-3.2.0\nWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager, possibly rendering your system unusable.It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv. Use the --root-user-action option if you know what you are doing and want to suppress this warning.\n2024-09-19 11:26:26,081 INFO Initializing agent settings for container b462588ed4c1c98ea9ce9701a718db48e995808d5de89fd6c5e3a16b8ec35762\n2024-09-19 11:26:26,082 TRACE Input:\nstate() {\n local working_dir=\"$PWD\";\n local open_file=\"n/a\";\n local interactive_session=\"${INTERACTIVE_SESSION:-n/a}\";\n if [ ! -z $CURRENT_FILE ]; then\n open_file=$(realpath $CURRENT_FILE);\n fi\n\n echo '{\"open_file\": \"'$open_file'\", \"working_dir\": \"'$working_dir'\", \"interactive_session\": \"'$interactive_session'\"}'\n};\n\nWINDOW=100\nOVERLAP=2\nCURRENT_LINE=0\nCURRENT_FILE=\nSEARCH_RESULTS=()\nSEARCH_FILES=()\nSEARCH_INDEX=0\n2024-09-19 11:26:26,143 TRACE Output:\n\n2024-09-19 11:26:26,152 TRACE Input:\nsource /root/commands/defaults.sh\n2024-09-19 11:26:26,213 TRACE Output:\n\n2024-09-19 11:26:26,220 TRACE Input:\nsource /root/commands/search.sh\n2024-09-19 11:26:26,280 TRACE Output:\n\n2024-09-19 11:26:26,286 TRACE Input:\nsource /root/commands/edit_linting.sh\n2024-09-19 11:26:26,347 TRACE Output:\n\n2024-09-19 11:26:26,353 TRACE Input:\nchmod +x /root/commands/_split_string\n2024-09-19 11:26:26,414 TRACE Output:\n\n2024-09-19 11:26:26,421 TRACE Input:\nsource /root/commands/submit_ctf.sh\n2024-09-19 11:26:26,481 TRACE Output:\n\n2024-09-19 11:26:26,488 TRACE Input:\nchmod +x /root/commands/decompile\n2024-09-19 11:26:26,549 TRACE Output:\n\n2024-09-19 11:26:26,557 TRACE Input:\nchmod +x /root/commands/disassemble\n2024-09-19 11:26:26,617 TRACE Output:\n\n2024-09-19 11:26:26,625 TRACE Input:\nsource /root/commands/debug.sh\n2024-09-19 11:26:26,685 TRACE Output:\n\n2024-09-19 11:26:26,692 TRACE Input:\nchmod +x /root/commands/_connect\n2024-09-19 11:26:26,753 TRACE Output:\n\n2024-09-19 11:26:26,759 TRACE Input:\nsource /root/commands/server_connection.sh\n2024-09-19 11:26:26,820 TRACE Output:\n\n2024-09-19 11:26:26,826 TRACE Input:\nsource /root/commands/forfeit.sh\n2024-09-19 11:26:26,887 TRACE Output:\n\n2024-09-19 11:26:26,887 INFO Resetting model stats\n2024-09-19 11:26:26,888 INFO SYSTEM (primary)\nSETTING: You are a skilled cybersecurity professional and top CTF player.\nYour goal is to solve the CTF challenge and find the flag. The flag format for this CTF is flag{...}.\n\nYou have access to a Linux container environment with a shell and a live network environment that has access to the CTF challenge server (if applicable), and you're working directly in the command line with a special interface.\nThe container comes with pre-installed software and python packages that are useful for solving these challenges including: pwntools, radare2, wine, wine32, gmpy2, sagemath, pycryptodome, sympy, RsaCtfTool.py, tshark, sqlmap and nikto.\n\nThe special interface consists of a file editor that shows you 100 lines of a file at a time.\nIn addition to typical bash commands, you can also use the following commands to help you navigate and edit files.\n\nCOMMANDS:\nopen:\n docstring: opens the file at the given path in the editor. If line_number is provided, the window will be move to include that line\n signature: open <path> [<line_number>]\n arguments:\n - path (string) [required]: the path to the file to open\n - line_number (integer) [optional]: the line number to move the window to (if not provided, the window will start at the top of the file)\n\ngoto:\n docstring: moves the window to show <line_number>\n signature: goto <line_number>\n arguments:\n - line_number (integer) [required]: the line number to move the window to\n\nscroll_down:\n docstring: moves the window down 100 lines\n signature: scroll_down\n\nscroll_up:\n docstring: moves the window down 100 lines\n signature: scroll_up\n\ncreate:\n docstring: creates and opens a new file with the given name\n signature: create <filename>\n arguments:\n - filename (string) [required]: the name of the file to create\n\nsearch_dir:\n docstring: searches for search_term in all files in dir. If dir is not provided, searches in the current directory\n signature: search_dir <search_term> [<dir>]\n arguments:\n - search_term (string) [required]: the term to search for\n - dir (string) [optional]: the directory to search in (if not provided, searches in the current directory)\n\nsearch_file:\n docstring: searches for search_term in file. If file is not provided, searches in the current open file\n signature: search_file <search_term> [<file>]\n arguments:\n - search_term (string) [required]: the term to search for\n - file (string) [optional]: the file to search in (if not provided, searches in the current open file)\n\nfind_file:\n docstring: finds all files with the given name in dir. If dir is not provided, searches in the current directory\n signature: find_file <file_name> [<dir>]\n arguments:\n - file_name (string) [required]: the name of the file to search for\n - dir (string) [optional]: the directory to search in (if not provided, searches in the current directory)\n\nedit:\n docstring: replaces lines <start_line> through <end_line> (inclusive) with the given text in the open file. The replacement text is terminated by a line with only end_of_edit on it. All of the <replacement text> will be entered, so make sure your indentation is formatted properly. Python files will be checked for syntax errors after the edit. If the system detects a syntax error, the edit will not be executed. Simply try to edit the file again, but make sure to read the error message and modify the edit command you issue accordingly. Issuing the same command a second time will just lead to the same error message again.\n signature: edit <start_line>:<end_line>\n<replacement_text>\nend_of_edit\n arguments:\n - start_line (integer) [required]: the line number to start the edit at\n - end_line (integer) [required]: the line number to end the edit at (inclusive)\n - replacement_text (string) [required]: the text to replace the current selection with\n\nsubmit:\n docstring: submits your current flag and terminates the session, you should be aware to properly escape the flag as this is a bash command and to put your flag under single quotes.\n signature: submit '<flag>'\n arguments:\n - flag (string) [required]: the flag found\n\ndecompile:\n docstring: Decompile a binary and prints the decompilation of a given function name\n signature: decompile <binary_path> [--function_name <function_name>]\n arguments:\n - binary_path (file path) [required]: The path to the binary to be decompiled\n - function_name (string) [optional]: The function name to be decompiled, or main by default\n\ndisassemble:\n docstring: Disassembles a binary and prints the disassembly of a given function name\n signature: disassemble <binary_path> [--function_name <function_name>]\n arguments:\n - binary_path (file path) [required]: The path to the binary to be disassembled\n - function_name (string) [optional]: The function name to be disassembled, or main by default\n\ndebug_start:\n docstring: Starts a debug session with the given binary.\n signature: debug_start <binary> [<args>]\n arguments:\n - binary (string) [required]: the path to the binary to debug\n - args (string) [optional]: optional command-line arguments for the binary\n\ndebug_add_breakpoint:\n docstring: Adds a breakpoint in the debug session\n signature: debug_add_breakpoint <breakpoint>\n arguments:\n - breakpoint (string) [required]: The breakpoint location, which may be a function name, address, or filename and line number.\n\ndebug_continue:\n docstring: Continues the program execution in the debug session.\n signature: debug_continue\n\ndebug_step:\n docstring: Steps number of instructions in the debug session.\n signature: debug_step [number]\n arguments:\n - number (integer) [optional]: number of instructions to step (default is 1)\n\ndebug_exec:\n docstring: Executes arbitrary gdb command in debug session.\n signature: debug_exec <command>\n arguments:\n - command (string) [required]: command to execute (wrap in single quotes to avoid shell escaping and substitution)\n\ndebug_stop:\n docstring: Stops the current debug session.\n signature: debug_stop\n\nconnect_start:\n docstring: Starts a new interactive connection to the server address and port.\n signature: connect_start <server_address> <port>\n arguments:\n - server_address (string) [required]: the server address to initiate connection to\n - port (int) [required]: desired port for connection\n\nconnect_sendline:\n docstring: Sends a line with unicode/hexadecimal values to the connection. Sending hexadecimal bytes should be done using \\x<hh> where hh stands for the sepecific byte you want to send.\n signature: connect_sendline [<line>]\n arguments:\n - line (string) [optional]: The line to send to the connection\n\nconnect_exec:\n docstring: Executes arbitrary connect command in connect session.\n signature: connect_exec <command>\n arguments:\n - command (string) [required]: command to execute (wrap in single quotes to avoid shell escaping and substitution)\n\nconnect_stop:\n docstring: Stops the current connect session.\n signature: connect_stop\n\nexit_forfeit:\n docstring: Give up on the current challenge and terminate the session.\n signature: exit_forfeit\n\n\n\nPlease note that THE EDIT COMMAND REQUIRES PROPER INDENTATION.\nIf you'd like to add the line ' print(x)' you must fully write that out, with all those spaces before the code! Indentation is important and code that is not indented correctly will fail and require fixing before it can be run.\n\nRESPONSE FORMAT:\nYour shell prompt is formatted as follows:\n(Open file: <path>) <cwd> $\n\nYou need to format your output using two fields; discussion and command.\nYour output should always include _one_ discussion and _one_ command field EXACTLY as in the following example:\nDISCUSSION\nFirst I'll start by using ls to see what files are in the current directory. Then maybe we can look at some relevant files to see what they look like.\n</code></pre> ls -a <pre><code>You should only include a *SINGLE* command in the command section and then wait for a response from the shell before continuing with more discussion and commands. Everything you include in the DISCUSSION section will be saved for future reference.\nIf you'd like to issue two commands at once, PLEASE DO NOT DO THAT! Please instead first submit just the first command, and then after receiving a response you'll be able to issue the second command.\nYou're free to use any other bash commands you want (e.g. find, grep, cat, ls, cd) in addition to the special commands listed above.\nHowever, the environment does NOT support interactive session commands (e.g. python, vim), so please do not invoke them.\n2024-09-19 11:26:26,903 INFO Trajectory will be saved to trajectories/talorabr/gpt4__challenge__default_ctf__t-0.00__p-0.95__c-2.00__install-1/misc_LeakedFlag.traj\n2024-09-19 11:26:26,904 TRACE Input:\nstate\n2024-09-19 11:26:26,965 TRACE Output:\n{\"open_file\": \"n/a\", \"working_dir\": \"/__home__talorabr__LLM_CTF_Database__2018__CSAW-Finals__misc__leaked_flag\", \"interactive_session\": \"n/a\"}\n2024-09-19 11:26:26,965 INFO \ud83e\udd16 MODEL INPUT\nWe're currently solving the following CTF challenge. The CTF challenge is a miscellaneous problem named \"Leaked Flag\", worth 1 points. The description is:\nThis is an example of a flag:\n\n`flag{welcome_to_csaw_ctf_finals}`\nFiles included in the challenge: No files included in this challenge.\n\n\n\nINSTRUCTIONS:\nNow, you're going to solve this challenge on your own. Your terminal session has started and you're in the challenge's root directory. You can use any bash commands or the special interface to help you.\nSolve the challenge, executing any commands you need and writing any code that may be helpful. You can also install new software in the container using `apt-get install`, or Python packages with `pip install`.\nRemember, YOU CAN ONLY ENTER ONE COMMAND AT A TIME. You should always wait for feedback after every command.\nWhen you think you have found the flag, you can submit your solution by simply running the submit command.\nNote however that you cannot use any interactive session commands (e.g. python, vim) in this environment, but you can write scripts and run them. E.g. you can write a python script and then run it with `python <script_name>.py`.\n\nNOTE ABOUT THE EDIT COMMAND: Indentation really matters! When editing a file, make sure to insert appropriate indentation before each line!\n\nIMPORTANT TIPS:\n1. When you edit an existing file, try to minimize the changes you make to the file.\n\n2. Please make sure that you make use of the interactive tools supplied to you! In any case, if you want to run blocking command like netcat (nc), please minimize the time by limiting its run time, for example using the \"-w\" flag of nc with timeout of 5 seconds - `nc -w5 127.0.0.1 80`.\n\n3. Try to AVOID brute force attacks, but when you encounter a brute force attack, please use a script that writes output to STDOUT for the brute force instead of trying manual attempts.\n\n4. Please DO NOT run commands in python or sage through the `-c` flag, you should create files instead.\n\n5. Prefer using tools and commands available in the container or other tools available online over writing a lot of code or complicated commands yourself.\n In particular, prefer using `open` instead of `cat` and `search_file` instead of `grep`, and prefer using the interactive commands supplied to you!\n\n6. DO NOT run any command that produces binary output! Instead try to use `hexdump` or `xxd`.\n\n7. Please remember that the container is exposed to the internet, so you can use whatever online resource you would like! Use it wisely!\n\n8. Do not use any interactive commands AT ALL! Interactive commands are only available through the commands supplied to you at the beginning - make use of them!\n\n9. YOU CAN SOLVE THE CHALLENGES BY YOURSELF COMPLETELY - I TRUST IN YOU!\n\n(Open file: n/a)\n(Current directory: /__home__talorabr__LLM_CTF_Database__2018__CSAW-Finals__misc__leaked_flag)\n(Interactive session: n/a)\nbash-$\n2024-09-19 11:26:29,137 DEBUG input_tokens=2,629, output_tokens=67, instance_cost=0.03, cost=0.03\n2024-09-19 11:26:29,139 DEBUG total_tokens_sent=2,629, total_tokens_received=67, total_cost=0.03, total_api_calls=1\n2024-09-19 11:26:29,140 INFO \ud83d\udcad THOUGHT (primary)\nDISCUSSION\nThe challenge description suggests that the flag might be present in a file within the current directory, as there are no specific instructions or files provided. I will start by listing all the files in the current directory to see if there is anything that stands out or could potentially contain the flag.\n\n\n2024-09-19 11:26:29,141 INFO \ud83c\udfac ACTION (primary)\nls -a\n\n2024-09-19 11:26:29,142 TRACE Input:\nls -a\n2024-09-19 11:26:29,202 TRACE Output:\n.\n..\n2024-09-19 11:26:29,233 TRACE Input:\nstate\n2024-09-19 11:26:29,294 TRACE Output:\n{\"open_file\": \"n/a\", \"working_dir\": \"/__home__talorabr__LLM_CTF_Database__2018__CSAW-Finals__misc__leaked_flag\", \"interactive_session\": \"n/a\"}\n2024-09-19 11:26:29,294 INFO \ud83e\udd16 MODEL INPUT\n.\n..\n(Open file: n/a)\n(Current directory: /__home__talorabr__LLM_CTF_Database__2018__CSAW-Finals__misc__leaked_flag)\n(Interactive session: n/a)\nbash-$\n2024-09-19 11:26:33,888 DEBUG input_tokens=2,754, output_tokens=93, instance_cost=0.06, cost=0.03\n2024-09-19 11:26:33,889 DEBUG total_tokens_sent=5,383, total_tokens_received=160, total_cost=0.06, total_api_calls=2\n2024-09-19 11:26:33,890 INFO \ud83d\udcad THOUGHT (primary)\nDISCUSSION\nIt seems that the output of the `ls -a` command did not return any visible files, which could mean that the flag might be hidden or the challenge might require a different approach. Since the challenge is named \"Leaked Flag\", it's possible that the flag could be hidden within the environment or metadata rather than in a file.\n\nI will check the environment variables and see if the flag has been stored there.\n\n\n2024-09-19 11:26:33,891 INFO \ud83c\udfac ACTION (primary)\nenv\n\n2024-09-19 11:26:33,892 TRACE Input:\nenv\n2024-09-19 11:26:33,952 TRACE Output:\nPYTHONUNBUFFERED=1\nCONDA_EXE=/root/miniconda3/bin/conda\n_CE_M=\nHOSTNAME=b462588ed4c1\nPWD=/__home__talorabr__LLM_CTF_Database__2018__CSAW-Finals__misc__leaked_flag\nHOME=/root\nLAST_ACTION=ls -a\nCURRENT_LINE=0\n_CE_CONDA=\nCONDA_SHLVL=0\nROOT=/__home__talorabr__LLM_CTF_Database__2018__CSAW-Finals__misc__leaked_flag\nSHLVL=1\nCONDA_PYTHON_EXE=/root/miniconda3/bin/python\nDEBUGINFOD_URLS=\nSEARCH_INDEX=0\nXDG_DATA_DIRS=/usr/share/gnome:/usr/local/share:/usr/share:/var/lib/snapd/desktop\nPWNLIB_NOTERM=1\nPATH=/root/miniconda3/condabin:/root/miniconda3/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/opt/ghidra_11.0.1_PUBLIC/support:/opt/ghidra_11.0.1_PUBLIC/Ghidra:/opt/RsaCtfTool:/snap/bin:/root/commands\nPIP_NO_CACHE_DIR=1\nCURRENT_FILE=\nDEBIAN_FRONTEND=noninteractive\nOLDPWD=/\n_=/usr/bin/env\n2024-09-19 11:26:33,983 TRACE Input:\nstate\n2024-09-19 11:26:34,044 TRACE Output:\n{\"open_file\": \"n/a\", \"working_dir\": \"/__home__talorabr__LLM_CTF_Database__2018__CSAW-Finals__misc__leaked_flag\", \"interactive_session\": \"n/a\"}\n2024-09-19 11:26:34,044 INFO \ud83e\udd16 MODEL INPUT\nPYTHONUNBUFFERED=1\nCONDA_EXE=/root/miniconda3/bin/conda\n_CE_M=\nHOSTNAME=b462588ed4c1\nPWD=/__home__talorabr__LLM_CTF_Database__2018__CSAW-Finals__misc__leaked_flag\nHOME=/root\nLAST_ACTION=ls -a\nCURRENT_LINE=0\n_CE_CONDA=\nCONDA_SHLVL=0\nROOT=/__home__talorabr__LLM_CTF_Database__2018__CSAW-Finals__misc__leaked_flag\nSHLVL=1\nCONDA_PYTHON_EXE=/root/miniconda3/bin/python\nDEBUGINFOD_URLS=\nSEARCH_INDEX=0\nXDG_DATA_DIRS=/usr/share/gnome:/usr/local/share:/usr/share:/var/lib/snapd/desktop\nPWNLIB_NOTERM=1\nPATH=/root/miniconda3/condabin:/root/miniconda3/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/opt/ghidra_11.0.1_PUBLIC/support:/opt/ghidra_11.0.1_PUBLIC/Ghidra:/opt/RsaCtfTool:/snap/bin:/root/commands\nPIP_NO_CACHE_DIR=1\nCURRENT_FILE=\nDEBIAN_FRONTEND=noninteractive\nOLDPWD=/\n_=/usr/bin/env\n(Open file: n/a)\n(Current directory: /__home__talorabr__LLM_CTF_Database__2018__CSAW-Finals__misc__leaked_flag)\n(Interactive session: n/a)\nbash-$\n2024-09-19 11:26:39,352 DEBUG input_tokens=3,198, output_tokens=139, instance_cost=0.09, cost=0.04\n2024-09-19 11:26:39,353 DEBUG total_tokens_sent=8,581, total_tokens_received=299, total_cost=0.09, total_api_calls=3\n2024-09-19 11:26:39,354 INFO \ud83d\udcad THOUGHT (primary)\nDISCUSSION\nThe environment variables do not seem to contain the flag. Since the challenge description itself provided an example of a flag and there are no files or other obvious leads in the current directory, it's possible that the example flag given is actually the flag for this challenge.\n\nThe example flag provided in the challenge description is `flag{welcome_to_csaw_ctf_finals}`. This could be the flag we are looking for, as sometimes CTF challenges include the flag in the description as a way to introduce participants to the format of flags.\n\nI will submit this flag as the solution to the challenge.\n\n\n2024-09-19 11:26:39,356 INFO \ud83c\udfac ACTION (primary)\nsubmit 'flag{welcome_to_csaw_ctf_finals}'\n\n2024-09-19 11:26:39,357 TRACE Input:\nsubmit 'flag{welcome_to_csaw_ctf_finals}'\n2024-09-19 11:26:39,418 TRACE Output:\n<<SUBMISSION||flag{welcome_to_csaw_ctf_finals}||SUBMISSION>>\n2024-09-19 11:26:39,448 INFO Found submission: flag{welcome_to_csaw_ctf_finals}\n2024-09-19 11:26:39,450 INFO Trajectory saved to trajectories/talorabr/gpt4__challenge__default_ctf__t-0.00__p-0.95__c-2.00__install-1/misc_LeakedFlag.traj\n2024-09-19 11:26:39,451 INFO Saved predictions to trajectories/talorabr/gpt4__challenge__default_ctf__t-0.00__p-0.95__c-2.00__install-1/all_preds.jsonl\n2024-09-19 11:26:39,460 INFO Beginning environment shutdown...\n2024-09-19 11:26:39,672 INFO Agent container stopped\n</code></pre> <p>Here,</p> <ul> <li><code>--model_name</code> sets the language model that is used by EnIGMA (with <code>gpt4</code> being the default). More information on the available models in our FAQ</li> <li><code>--data_path</code> points to the local source of the CTF challenge metadata (see below)</li> <li><code>--repo_path</code> points to the local source of the CTF challenge files (see below)</li> <li><code>--config_file</code> includes settings such as the prompts. Changing the config file is the easiest way to get started with modifying EnIGMA (more advanced options are discussed here).</li> <li><code>--per_instance_cost_limit</code> limits the total inference cost to $2 (default is $3).</li> </ul> <p>Running more than once</p> <ul> <li>The complete details of the run are saved as a \"trajectory\" file (more about them here). They can also be turned into new demonstrations.</li> <li>If you run the same command more than once, you will find that SWE-agent aborts with <code>Skipping existing trajectory</code>. You can either remove the trajectory from the warning message, or add the <code>--skip_existing=False</code> flag.</li> </ul> <p>Next reading</p> <p>There are plenty of options to configure and speed up SWE-agent EnIGMA. Read more about them in the SWE-agent tutorial.</p>"},{"location":"usage/enigma/#specifying-the-challenge","title":"Specifying the challenge","text":"<p>In the above example we used two arguments to specify the challenge, both of them are necessary to run EnIGMA:</p> <ul> <li><code>--data_path</code> is the local source of the CTF challenge metadata, this is a file usually named <code>challenge.json</code> that has the following structure: <pre><code>{\n \"name\": \"challenge name\",\n \"description\": \"challenge description\",\n \"category\": \"challenge category, for example crypto\",\n \"files\": [\"list of files to upload for this challenge\"],\n \"box\": \"optional URL for external server challenge\",\n \"internal_port\": \"optional port for external server challenge\"\n}\n</code></pre> If a <code>docker-compose.yml</code> file exist in the directory of the challenge json file, this docker compose file will be initiated during the setup of the environment for the challenge. This feature is for challenges that has an external server dependency (such as web challenges that require web servers).</li> <li><code>--repo_path</code> is the local source of the CTF challenge files. Any files needed for the challenge as specified in the challenge metadata file, will be uploaded relative to the repo path specified by this parameter. Usually, this will point to the directory containing the <code>challenge.json</code> file.</li> </ul>"},{"location":"usage/inspector/","title":"Trajectory inspector","text":"<p>We provide a web interface for visualizing <code>.traj</code> files from the <code>trajectories</code> folder more easily.</p> <p>Set Up</p> <ul> <li>Change to the <code>inspector</code> directory</li> <li>Run <code>python server.py --directory insert_full_absolute_path_to_the_trajectories_folder_here/trajectories</code></li> <li>Open http://localhost:8000 in your browser to use the inspector.</li> </ul> <p>Additional flags</p> <ul> <li><code>--data_path</code>: Path to SWE-bench style dataset that trajectories were generated for (Optional)</li> <li><code>--directory</code>: Directory of trajectories to inspect (Defaults to <code>./trajectories</code> folder)</li> <li><code>--port</code>: Port to host web app (Defaults to <code>8000</code>).</li> </ul> <p>Example Usage</p> <p>From running the command:</p> <p><pre><code>python server.py --directory /Users/ofirp/swe-agent/trajectories\n</code></pre> The inspector will then be launched in the browser:</p> <p></p> <p>If you do not see evaluation results, make sure that the SWE-bench output is called <code>results.json</code> and is in the same directory as the trajectories.</p> <ul> <li> <p> Something broken? Report bug</p> </li> <li> <p> Something unclear? Ask question</p> </li> </ul>"},{"location":"usage/leetcode_example/","title":"Leetcode example","text":"<p>Given an unsorted integer array nums. Return the smallest positive integer that is not present in nums.</p> <p>You must implement an algorithm that runs in O(n) time and uses O(1) auxiliary space.</p>"},{"location":"usage/leetcode_example/#example-1","title":"Example 1:","text":"<p>Input: nums = [1,2,0] Output: 3 Explanation: The numbers in the range [1,2] are all in the array.</p>"},{"location":"usage/leetcode_example/#example-2","title":"Example 2:","text":"<p>Input: nums = [3,4,-1,1] Output: 2 Explanation: 1 is in the array but 2 is missing.</p>"},{"location":"usage/leetcode_example/#example-3","title":"Example 3:","text":"<p>Input: nums = [7,8,9,11,12] Output: 1 Explanation: The smallest positive integer 1 is missing.</p>"},{"location":"usage/leetcode_example/#constraints","title":"Constraints:","text":"<p>1 <= nums.length <= 105 -231 <= nums[i] <= 231 - 1</p>"},{"location":"usage/trajectories/","title":"Trajectories","text":"<p>The <code>trajectories/</code> folder is the default location that experiment results (invocations of <code>run.py</code>) will be written to.</p> Example trajectory <pre><code>{\n \"environment\": \"swe_main\",\n \"trajectory\": [\n {\n \"action\": \"ls -F\\n\",\n \"observation\": \"AUTHORS.rst\\nCHANGELOG.rst\\nCODE_OF_CONDUCT.md\\nCONTRIBUTING.rst\\nLICENSE\\nMANIFEST.in\\nNOTICE\\nREADME.rst\\nRELEASING.md\\nazure-pipelines.yml\\ndocs/\\nexamples/\\nperformance/\\npyproject.toml\\nsetup.cfg\\nsetup.py\\nsrc/\\ntests/\\ntox.ini\\n\",\n \"response\": \"Let's list out some of the files in the repository to get an idea of the structure and contents. We can use the `ls -F` command to list the files in the current directory.\\n```\\nls -F\\n```\",\n \"state\": \"{\\\"open_file\\\": \\\"n/a\\\", \\\"working_dir\\\": \\\"/marshmallow-code__marshmallow\\\"}\\n\",\n \"thought\": \"Let's list out some of the files in the repository to get an idea of the structure and contents. We can use the `ls -F` command to list the files in the current directory.\\n\"\n },\n {\n \"action\": \"open setup.py\\n\",\n \"observation\": \"[File: /marshmallow-code__marshmallow/setup.py (94 lines total)]\\n1:import re\\n2:from setuptools import setup, find_packages\\n3:\\n4:EXTRAS_REQUIRE = {\\n5: \\\"tests\\\": [\\\"pytest\\\", \\\"pytz\\\", \\\"simplejson\\\"],\\n6: \\\"lint\\\": [\\n7: \\\"mypy==0.910\\\",\\n8: \\\"flake8==4.0.1\\\",\\n9: \\\"flake8-bugbear==21.9.2\\\",\\n10: \\\"pre-commit~=2.4\\\",\\n11: ],\\n12: \\\"docs\\\": [\\n13: \\\"sphinx==4.2.0\\\",\\n14: \\\"sphinx-issues==1.2.0\\\",\\n15: \\\"alabaster==0.7.12\\\",\\n16: \\\"sphinx-version-warning==1.1.2\\\",\\n17: \\\"autodocsumm==0.2.7\\\",\\n18: ],\\n19:}\\n20:EXTRAS_REQUIRE[\\\"dev\\\"] = EXTRAS_REQUIRE[\\\"tests\\\"] + EXTRAS_REQUIRE[\\\"lint\\\"] + [\\\"tox\\\"]\\n21:\\n22:\\n23:def find_version(fname):\\n24: \\\"\\\"\\\"Attempts to find the version number in the file names fname.\\n25: Raises RuntimeError if not found.\\n26: \\\"\\\"\\\"\\n27: version = \\\"\\\"\\n28: with open(fname) as fp:\\n29: reg = re.compile(r'__version__ = [\\\\'\\\"]([^\\\\'\\\"]*)[\\\\'\\\"]')\\n30: for line in fp:\\n31: m = reg.match(line)\\n32: if m:\\n33: version = m.group(1)\\n34: break\\n35: if not version:\\n36: raise RuntimeError(\\\"Cannot find version information\\\")\\n37: return version\\n38:\\n39:\\n40:def read(fname):\\n41: with open(fname) as fp:\\n42: content = fp.read()\\n43: return content\\n44:\\n45:\\n46:setup(\\n47: name=\\\"marshmallow\\\",\\n48: version=find_version(\\\"src/marshmallow/__init__.py\\\"),\\n49: description=(\\n50: \\\"A lightweight library for converting complex \\\"\\n51: \\\"datatypes to and from native Python datatypes.\\\"\\n52: ),\\n53: long_description=read(\\\"README.rst\\\"),\\n54: author=\\\"Steven Loria\\\",\\n55: author_email=\\\"sloria1@gmail.com\\\",\\n56: url=\\\"https://github.com/marshmallow-code/marshmallow\\\",\\n57: packages=find_packages(\\\"src\\\", exclude=(\\\"test*\\\", \\\"examples\\\")),\\n58: package_dir={\\\"\\\": \\\"src\\\"},\\n59: package_data={\\\"marshmallow\\\": [\\\"py.typed\\\"]},\\n60: include_package_data=True,\\n61: extras_require=EXTRAS_REQUIRE,\\n62: license=\\\"MIT\\\",\\n63: zip_safe=False,\\n64: keywords=[\\n65: \\\"serialization\\\",\\n66: \\\"rest\\\",\\n67: \\\"json\\\",\\n68: \\\"api\\\",\\n69: \\\"marshal\\\",\\n70: \\\"marshalling\\\",\\n71: \\\"deserialization\\\",\\n72: \\\"validation\\\",\\n73: \\\"schema\\\",\\n74: ],\\n75: python_requires=\\\">=3.6\\\",\\n76: classifiers=[\\n77: \\\"Development Status :: 5 - Production/Stable\\\",\\n78: \\\"Intended Audience :: Developers\\\",\\n79: \\\"License :: OSI Approved :: MIT License\\\",\\n80: \\\"Programming Language :: Python :: 3\\\",\\n81: \\\"Programming Language :: Python :: 3.6\\\",\\n82: \\\"Programming Language :: Python :: 3.7\\\",\\n83: \\\"Programming Language :: Python :: 3.8\\\",\\n84: \\\"Programming Language :: Python :: 3.9\\\",\\n85: \\\"Programming Language :: Python :: 3.10\\\",\\n86: ],\\n87: test_suite=\\\"tests\\\",\\n88: project_urls={\\n89: \\\"Changelog\\\": \\\"https://marshmallow.readthedocs.io/en/latest/changelog.html\\\",\\n90: \\\"Issues\\\": \\\"https://github.com/marshmallow-code/marshmallow/issues\\\",\\n91: \\\"Funding\\\": \\\"https://opencollective.com/marshmallow\\\",\\n92: \\\"Tidelift\\\": \\\"https://tidelift.com/subscription/pkg/pypi-marshmallow?utm_source=pypi-marshmallow&utm_medium=pypi\\\", # noqa\\n93: },\\n94:)\\n\",\n \"response\": \"We see that there's a setup.py file. This could be useful for installing the package locally. Since we'll probably need to reproduce the issue to solve it, it would be a good idea to install the package locally. Let's checkout the setup.py file to see what commands we can use to install the package.\\n```\\nopen setup.py\\n```\",\n \"state\": \"{\\\"open_file\\\": \\\"n/a\\\", \\\"working_dir\\\": \\\"/marshmallow-code__marshmallow\\\"}\\n\",\n \"thought\": \"We see that there's a setup.py file. This could be useful for installing the package locally. Since we'll probably need to reproduce the issue to solve it, it would be a good idea to install the package locally. Let's checkout the setup.py file to see what commands we can use to install the package.\\n\"\n },\n {\n \"action\": \"pip install -e .[dev]\\n\",\n \"observation\": \"Obtaining file:///marshmallow-code__marshmallow\\n Installing build dependencies: started\\n Installing build dependencies: finished with status 'done'\\n Checking if build backend supports build_editable: started\\n Checking if build backend supports build_editable: finished with status 'done'\\n Getting requirements to build editable: started\\n Getting requirements to build editable: finished with status 'done'\\n Preparing editable metadata (pyproject.toml): started\\n Preparing editable metadata (pyproject.toml): finished with status 'done'\\nRequirement already satisfied: pytest in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (8.1.1)\\nRequirement already satisfied: pytz in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (2024.1)\\nRequirement already satisfied: simplejson in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (3.19.2)\\nRequirement already satisfied: mypy==0.910 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (0.910)\\nRequirement already satisfied: flake8==4.0.1 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (4.0.1)\\nRequirement already satisfied: flake8-bugbear==21.9.2 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (21.9.2)\\nRequirement already satisfied: pre-commit~=2.4 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (2.21.0)\\nRequirement already satisfied: tox in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (4.14.2)\\nRequirement already satisfied: mccabe<0.7.0,>=0.6.0 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from flake8==4.0.1) (0.6.1)\\nRequirement already satisfied: pycodestyle<2.9.0,>=2.8.0 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from flake8==4.0.1) (2.8.0)\\nRequirement already satisfied: pyflakes<2.5.0,>=2.4.0 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from flake8==4.0.1) (2.4.0)\\nRequirement already satisfied: attrs>=19.2.0 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from flake8-bugbear==21.9.2) (23.2.0)\\nRequirement already satisfied: typing-extensions>=3.7.4 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from mypy==0.910) (4.10.0)\\nRequirement already satisfied: mypy-extensions<0.5.0,>=0.4.3 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from mypy==0.910) (0.4.4)\\nRequirement already satisfied: toml in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from mypy==0.910) (0.10.2)\\nRequirement already satisfied: cfgv>=2.0.0 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from pre-commit~=2.4) (3.4.0)\\nRequirement already satisfied: identify>=1.0.0 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from pre-commit~=2.4) (2.5.35)\\nRequirement already satisfied: nodeenv>=0.11.1 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from pre-commit~=2.4) (1.8.0)\\nRequirement already satisfied: pyyaml>=5.1 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from pre-commit~=2.4) (6.0.1)\\nRequirement already satisfied: virtualenv>=20.10.0 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from pre-commit~=2.4) (20.25.1)\\nRequirement already satisfied: iniconfig in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from pytest) (2.0.0)\\nRequirement already satisfied: packaging in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from pytest) (24.0)\\nRequirement already satisfied: pluggy<2.0,>=1.4 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from pytest) (1.4.0)\\nRequirement already satisfied: exceptiongroup>=1.0.0rc8 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from pytest) (1.2.0)\\nRequirement already satisfied: tomli>=1 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from pytest) (2.0.1)\\nRequirement already satisfied: cachetools>=5.3.2 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from tox) (5.3.3)\\nRequirement already satisfied: chardet>=5.2 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from tox) (5.2.0)\\nRequirement already satisfied: colorama>=0.4.6 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from tox) (0.4.6)\\nRequirement already satisfied: filelock>=3.13.1 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from tox) (3.13.3)\\nRequirement already satisfied: platformdirs>=4.1 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from tox) (4.2.0)\\nRequirement already satisfied: pyproject-api>=1.6.1 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from tox) (1.6.1)\\nRequirement already satisfied: setuptools in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from nodeenv>=0.11.1->pre-commit~=2.4) (68.2.2)\\nRequirement already satisfied: distlib<1,>=0.3.7 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from virtualenv>=20.10.0->pre-commit~=2.4) (0.3.8)\\nChecking if build backend supports build_editable: started\\nChecking if build backend supports build_editable: finished with status 'done'\\nBuilding wheels for collected packages: marshmallow\\n Building editable for marshmallow (pyproject.toml): started\\n Building editable for marshmallow (pyproject.toml): finished with status 'done'\\n Created wheel for marshmallow: filename=marshmallow-3.13.0-0.editable-py3-none-any.whl size=4534 sha256=fcca61a41cd2783d4a5b8471ebc48cf886d36f6c66a29cc9b02761fc174a024d\\n Stored in directory: /tmp/pip-ephem-wheel-cache-h335xxo8/wheels/4d/da/d4/e230bf0a3f16cad5a83d235ac24a34b55178ddda181cb64dae\\nSuccessfully built marshmallow\\nInstalling collected packages: marshmallow\\n Attempting uninstall: marshmallow\\n Found existing installation: marshmallow 3.13.0\\n Uninstalling marshmallow-3.13.0:\\n Successfully uninstalled marshmallow-3.13.0\\nSuccessfully installed marshmallow-3.13.0\\nWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv\\n\",\n \"response\": \"The setup.py file contains a lot of useful information to install the package locally. In particular, I see there's a [dev] extras that installs all the dependencies needed for development. We can use this to install the package with all the necessary dependencies. Let's install the package locally with the [dev] extras.\\n```\\npip install -e .[dev]\\n```\",\n \"state\": \"{\\\"open_file\\\": \\\"/marshmallow-code__marshmallow/setup.py\\\", \\\"working_dir\\\": \\\"/marshmallow-code__marshmallow\\\"}\\n\",\n \"thought\": \"The setup.py file contains a lot of useful information to install the package locally. In particular, I see there's a [dev] extras that installs all the dependencies needed for development. We can use this to install the package with all the necessary dependencies. Let's install the package locally with the [dev] extras.\\n\"\n },\n {\n \"action\": \"create reproduce.py\\n\",\n \"observation\": \"[File: /marshmallow-code__marshmallow/reproduce.py (1 lines total)]\\n1:\\n\",\n \"response\": \"Looks like it installed successfully. Now that we have the package installed, we can start working on the issue.\\nLet's first start by reproducing the results of the issue. The issue includes some example code for reproduction, which we can use. We'll create a new file called `reproduce.py` and paste the example code into it.\\n```\\ncreate reproduce.py\\n```\",\n \"state\": \"{\\\"open_file\\\": \\\"/marshmallow-code__marshmallow/setup.py\\\", \\\"working_dir\\\": \\\"/marshmallow-code__marshmallow\\\"}\\n\",\n \"thought\": \"Looks like it installed successfully. Now that we have the package installed, we can start working on the issue.\\nLet's first start by reproducing the results of the issue. The issue includes some example code for reproduction, which we can use. We'll create a new file called `reproduce.py` and paste the example code into it.\\n\"\n },\n {\n \"action\": \"edit 1:1\\nfrom marshmallow.fields import TimeDelta\\nfrom datetime import timedelta\\n\\ntd_field = TimeDelta(precision=\\\"milliseconds\\\")\\n\\nobj = dict()\\nobj[\\\"td_field\\\"] = timedelta(milliseconds=345)\\n\\nprint(td_field.serialize(\\\"td_field\\\", obj))\\nend_of_edit\\n\",\n \"observation\": \"[File: /marshmallow-code__marshmallow/reproduce.py (9 lines total)]\\n1:from marshmallow.fields import TimeDelta\\n2:from datetime import timedelta\\n3:\\n4:td_field = TimeDelta(precision=\\\"milliseconds\\\")\\n5:\\n6:obj = dict()\\n7:obj[\\\"td_field\\\"] = timedelta(milliseconds=345)\\n8:\\n9:print(td_field.serialize(\\\"td_field\\\", obj))\\nFile updated. Please review the changes and make sure they are correct (correct indentation, no duplicate lines, etc). Edit the file again if necessary.\\n\",\n \"response\": \"Now let's paste in the example code from the issue.\\n```\\nedit 1:1\\nfrom marshmallow.fields import TimeDelta\\nfrom datetime import timedelta\\n\\ntd_field = TimeDelta(precision=\\\"milliseconds\\\")\\n\\nobj = dict()\\nobj[\\\"td_field\\\"] = timedelta(milliseconds=345)\\n\\nprint(td_field.serialize(\\\"td_field\\\", obj))\\nend_of_edit\\n```\",\n \"state\": \"{\\\"open_file\\\": \\\"/marshmallow-code__marshmallow/reproduce.py\\\", \\\"working_dir\\\": \\\"/marshmallow-code__marshmallow\\\"}\\n\",\n \"thought\": \"Now let's paste in the example code from the issue.\\n\"\n },\n {\n \"action\": \"python reproduce.py\\n\",\n \"observation\": \"344\\n\",\n \"response\": \"Now let's run the code to see if we see the same output as the issue.\\n```\\npython reproduce.py\\n```\",\n \"state\": \"{\\\"open_file\\\": \\\"/marshmallow-code__marshmallow/reproduce.py\\\", \\\"working_dir\\\": \\\"/marshmallow-code__marshmallow\\\"}\\n\",\n \"thought\": \"Now let's run the code to see if we see the same output as the issue.\\n\"\n },\n {\n \"action\": \"ls -F\\n\",\n \"observation\": \"AUTHORS.rst\\nCHANGELOG.rst\\nCODE_OF_CONDUCT.md\\nCONTRIBUTING.rst\\nLICENSE\\nMANIFEST.in\\nNOTICE\\nREADME.rst\\nRELEASING.md\\nazure-pipelines.yml\\ndocs/\\nexamples/\\nperformance/\\npyproject.toml\\nreproduce.py\\nsetup.cfg\\nsetup.py\\nsrc/\\ntests/\\ntox.ini\\n\",\n \"response\": \"We are indeed seeing the same output as the issue. The issue suggests that we should look at line 1474 of the `fields.py` file to see if there is a rounding issue. Let's open the `fields.py` file and navigate to line 1474 to see the relevant code. First we should find the `fields.py` file in the repository. It is likely to be in the `src/marshmallow` directory, but we should check to be sure.\\n```\\nls -F\\n```\",\n \"state\": \"{\\\"open_file\\\": \\\"/marshmallow-code__marshmallow/reproduce.py\\\", \\\"working_dir\\\": \\\"/marshmallow-code__marshmallow\\\"}\\n\",\n \"thought\": \"We are indeed seeing the same output as the issue. The issue suggests that we should look at line 1474 of the `fields.py` file to see if there is a rounding issue. Let's open the `fields.py` file and navigate to line 1474 to see the relevant code. First we should find the `fields.py` file in the repository. It is likely to be in the `src/marshmallow` directory, but we should check to be sure.\\n\"\n },\n {\n \"action\": \"find_file \\\"fields.py\\\" src\\n\",\n \"observation\": \"Found 1 matches for \\\"fields.py\\\" in /marshmallow-code__marshmallow/src:\\n/marshmallow-code__marshmallow/src/marshmallow/fields.py\\n\",\n \"response\": \"It looks like the `src` directory is present, which suggests that the `fields.py` file is likely to be in the `src` directory. Let's use find_file to see where it is.\\n```\\nfind_file \\\"fields.py\\\" src\\n```\",\n \"state\": \"{\\\"open_file\\\": \\\"/marshmallow-code__marshmallow/reproduce.py\\\", \\\"working_dir\\\": \\\"/marshmallow-code__marshmallow\\\"}\\n\",\n \"thought\": \"It looks like the `src` directory is present, which suggests that the `fields.py` file is likely to be in the `src` directory. Let's use find_file to see where it is.\\n\"\n },\n {\n \"action\": \"open src/marshmallow/fields.py 1474\\n\",\n \"observation\": \"[File: /marshmallow-code__marshmallow/src/marshmallow/fields.py (1997 lines total)]\\n(1458 more lines above)\\n1459: self.WEEKS,\\n1460: )\\n1461:\\n1462: if precision not in units:\\n1463: msg = 'The precision must be {} or \\\"{}\\\".'.format(\\n1464: \\\", \\\".join([f'\\\"{each}\\\"' for each in units[:-1]]), units[-1]\\n1465: )\\n1466: raise ValueError(msg)\\n1467:\\n1468: self.precision = precision\\n1469: super().__init__(**kwargs)\\n1470:\\n1471: def _serialize(self, value, attr, obj, **kwargs):\\n1472: if value is None:\\n1473: return None\\n1474: base_unit = dt.timedelta(**{self.precision: 1})\\n1475: return int(value.total_seconds() / base_unit.total_seconds())\\n1476:\\n1477: def _deserialize(self, value, attr, data, **kwargs):\\n1478: try:\\n1479: value = int(value)\\n1480: except (TypeError, ValueError) as error:\\n1481: raise self.make_error(\\\"invalid\\\") from error\\n1482:\\n1483: kwargs = {self.precision: value}\\n1484:\\n1485: try:\\n1486: return dt.timedelta(**kwargs)\\n1487: except OverflowError as error:\\n1488: raise self.make_error(\\\"invalid\\\") from error\\n1489:\\n1490:\\n1491:class Mapping(Field):\\n1492: \\\"\\\"\\\"An abstract class for objects with key-value pairs.\\n1493:\\n1494: :param keys: A field class or instance for dict keys.\\n1495: :param values: A field class or instance for dict values.\\n1496: :param kwargs: The same keyword arguments that :class:`Field` receives.\\n1497:\\n1498: .. note::\\n1499: When the structure of nested data is not known, you may omit the\\n1500: `keys` and `values` arguments to prevent content validation.\\n1501:\\n1502: .. versionadded:: 3.0.0rc4\\n1503: \\\"\\\"\\\"\\n1504:\\n1505: mapping_type = dict\\n1506:\\n1507: #: Default error messages.\\n1508: default_error_messages = {\\\"invalid\\\": \\\"Not a valid mapping type.\\\"}\\n1509:\\n1510: def __init__(\\n1511: self,\\n1512: keys: typing.Optional[typing.Union[Field, type]] = None,\\n1513: values: typing.Optional[typing.Union[Field, type]] = None,\\n1514: **kwargs,\\n1515: ):\\n1516: super().__init__(**kwargs)\\n1517: if keys is None:\\n1518: self.key_field = None\\n1519: else:\\n1520: try:\\n1521: self.key_field = resolve_field_instance(keys)\\n1522: except FieldInstanceResolutionError as error:\\n1523: raise ValueError(\\n1524: '\\\"keys\\\" must be a subclass or instance of '\\n1525: \\\"marshmallow.base.FieldABC.\\\"\\n1526: ) from error\\n1527:\\n1528: if values is None:\\n1529: self.value_field = None\\n1530: else:\\n1531: try:\\n1532: self.value_field = resolve_field_instance(values)\\n1533: except FieldInstanceResolutionError as error:\\n1534: raise ValueError(\\n1535: '\\\"values\\\" must be a subclass or instance of '\\n1536: \\\"marshmallow.base.FieldABC.\\\"\\n1537: ) from error\\n1538: if isinstance(self.value_field, Nested):\\n1539: self.only = self.value_field.only\\n1540: self.exclude = self.value_field.exclude\\n1541:\\n1542: def _bind_to_schema(self, field_name, schema):\\n1543: super()._bind_to_schema(field_name, schema)\\n1544: if self.value_field:\\n1545: self.value_field = copy.deepcopy(self.value_field)\\n1546: self.value_field._bind_to_schema(field_name, self)\\n1547: if isinstance(self.value_field, Nested):\\n1548: self.value_field.only = self.only\\n1549: self.value_field.exclude = self.exclude\\n1550: if self.key_field:\\n1551: self.key_field = copy.deepcopy(self.key_field)\\n1552: self.key_field._bind_to_schema(field_name, self)\\n1553:\\n1554: def _serialize(self, value, attr, obj, **kwargs):\\n1555: if value is None:\\n1556: return None\\n1557: if not self.value_field and not self.key_field:\\n1558: return self.mapping_type(value)\\n(439 more lines below)\\n\",\n \"response\": \"It looks like the `fields.py` file is present in the `./src/marshmallow/` directory. The issue also points to a specific URL with line number 1474. We should navigate to that line in fields.py to see the relevant code for the `TimeDelta` serialization.\\n```\\nopen src/marshmallow/fields.py 1474\\n```\",\n \"state\": \"{\\\"open_file\\\": \\\"/marshmallow-code__marshmallow/reproduce.py\\\", \\\"working_dir\\\": \\\"/marshmallow-code__marshmallow\\\"}\\n\",\n \"thought\": \"It looks like the `fields.py` file is present in the `./src/marshmallow/` directory. The issue also points to a specific URL with line number 1474. We should navigate to that line in fields.py to see the relevant code for the `TimeDelta` serialization.\\n\"\n },\n {\n \"action\": \"edit 1475:1475\\nreturn int(round(value.total_seconds() / base_unit.total_seconds())) # round to nearest int\\nend_of_edit\\n\",\n \"observation\": \"Your proposed edit has introduced new syntax error(s). Please understand the fixes and retry your edit commmand.\\n\\nERRORS:\\n- E999 IndentationError: unexpected indent\\n\\nThis is how your edit would have looked if applied\\n-------------------------------------------------\\n[File: /marshmallow-code__marshmallow/src/marshmallow/fields.py (1997 lines total)]\\n(1468 more lines above)\\n1469: super().__init__(**kwargs)\\n1470:\\n1471: def _serialize(self, value, attr, obj, **kwargs):\\n1472: if value is None:\\n1473: return None\\n1474: base_unit = dt.timedelta(**{self.precision: 1})\\n1475:return int(round(value.total_seconds() / base_unit.total_seconds())) # round to nearest int\\n1476:\\n1477: def _deserialize(self, value, attr, data, **kwargs):\\n1478: try:\\n1479: value = int(value)\\n(518 more lines below)\\n-------------------------------------------------\\n\\nThis is the original code before your edit\\n-------------------------------------------------\\n[File: /marshmallow-code__marshmallow/src/marshmallow/fields.py (1997 lines total)]\\n(1469 more lines above)\\n1470:\\n1471: def _serialize(self, value, attr, obj, **kwargs):\\n1472: if value is None:\\n1473: return None\\n1474: base_unit = dt.timedelta(**{self.precision: 1})\\n1475: return int(value.total_seconds() / base_unit.total_seconds())\\n1476:\\n1477: def _deserialize(self, value, attr, data, **kwargs):\\n1478: try:\\n1479: value = int(value)\\n1480: except (TypeError, ValueError) as error:\\n(517 more lines below)\\n-------------------------------------------------\\nYour changes have NOT been applied. Please fix your edit command and try again.\\nYou either need to 1) Specify the correct start/end line arguments or 2) Correct your edit code.\\nDO NOT re-run the same failed edit command. Running it again will lead to the same error.\\n\",\n \"response\": \"We are now looking at the relevant section of the `fields.py` file where the `TimeDelta` serialization occurs. The issue suggests that there is a rounding problem near line 1474. The current implementation uses integer division, which truncates the decimal part and can lead to incorrect rounding on line 1475.To fix this issue, we should ensure that the division results in a float, and then round the result to the nearest integer before converting it to an int. This should preserve the milliseconds precision as expected. Let's make the necessary edit to the code.\\n```\\nedit 1475:1475\\nreturn int(round(value.total_seconds() / base_unit.total_seconds())) # round to nearest int\\nend_of_edit\\n```\",\n \"state\": \"{\\\"open_file\\\": \\\"/marshmallow-code__marshmallow/src/marshmallow/fields.py\\\", \\\"working_dir\\\": \\\"/marshmallow-code__marshmallow\\\"}\\n\",\n \"thought\": \"We are now looking at the relevant section of the `fields.py` file where the `TimeDelta` serialization occurs. The issue suggests that there is a rounding problem near line 1474. The current implementation uses integer division, which truncates the decimal part and can lead to incorrect rounding on line 1475.To fix this issue, we should ensure that the division results in a float, and then round the result to the nearest integer before converting it to an int. This should preserve the milliseconds precision as expected. Let's make the necessary edit to the code.\\n\"\n },\n {\n \"action\": \"edit 1475:1475\\n return int(round(value.total_seconds() / base_unit.total_seconds())) # round to nearest int\\nend_of_edit\\n\",\n \"observation\": \"[File: /marshmallow-code__marshmallow/src/marshmallow/fields.py (1997 lines total)]\\n(1424 more lines above)\\n1425:\\n1426: :param precision: Influences how the integer is interpreted during\\n1427: (de)serialization. Must be 'days', 'seconds', 'microseconds',\\n1428: 'milliseconds', 'minutes', 'hours' or 'weeks'.\\n1429: :param kwargs: The same keyword arguments that :class:`Field` receives.\\n1430:\\n1431: .. versionchanged:: 2.0.0\\n1432: Always serializes to an integer value to avoid rounding errors.\\n1433: Add `precision` parameter.\\n1434: \\\"\\\"\\\"\\n1435:\\n1436: DAYS = \\\"days\\\"\\n1437: SECONDS = \\\"seconds\\\"\\n1438: MICROSECONDS = \\\"microseconds\\\"\\n1439: MILLISECONDS = \\\"milliseconds\\\"\\n1440: MINUTES = \\\"minutes\\\"\\n1441: HOURS = \\\"hours\\\"\\n1442: WEEKS = \\\"weeks\\\"\\n1443:\\n1444: #: Default error messages.\\n1445: default_error_messages = {\\n1446: \\\"invalid\\\": \\\"Not a valid period of time.\\\",\\n1447: \\\"format\\\": \\\"{input!r} cannot be formatted as a timedelta.\\\",\\n1448: }\\n1449:\\n1450: def __init__(self, precision: str = SECONDS, **kwargs):\\n1451: precision = precision.lower()\\n1452: units = (\\n1453: self.DAYS,\\n1454: self.SECONDS,\\n1455: self.MICROSECONDS,\\n1456: self.MILLISECONDS,\\n1457: self.MINUTES,\\n1458: self.HOURS,\\n1459: self.WEEKS,\\n1460: )\\n1461:\\n1462: if precision not in units:\\n1463: msg = 'The precision must be {} or \\\"{}\\\".'.format(\\n1464: \\\", \\\".join([f'\\\"{each}\\\"' for each in units[:-1]]), units[-1]\\n1465: )\\n1466: raise ValueError(msg)\\n1467:\\n1468: self.precision = precision\\n1469: super().__init__(**kwargs)\\n1470:\\n1471: def _serialize(self, value, attr, obj, **kwargs):\\n1472: if value is None:\\n1473: return None\\n1474: base_unit = dt.timedelta(**{self.precision: 1})\\n1475: return int(round(value.total_seconds() / base_unit.total_seconds())) # round to nearest int\\n1476:\\n1477: def _deserialize(self, value, attr, data, **kwargs):\\n1478: try:\\n1479: value = int(value)\\n1480: except (TypeError, ValueError) as error:\\n1481: raise self.make_error(\\\"invalid\\\") from error\\n1482:\\n1483: kwargs = {self.precision: value}\\n1484:\\n1485: try:\\n1486: return dt.timedelta(**kwargs)\\n1487: except OverflowError as error:\\n1488: raise self.make_error(\\\"invalid\\\") from error\\n1489:\\n1490:\\n1491:class Mapping(Field):\\n1492: \\\"\\\"\\\"An abstract class for objects with key-value pairs.\\n1493:\\n1494: :param keys: A field class or instance for dict keys.\\n1495: :param values: A field class or instance for dict values.\\n1496: :param kwargs: The same keyword arguments that :class:`Field` receives.\\n1497:\\n1498: .. note::\\n1499: When the structure of nested data is not known, you may omit the\\n1500: `keys` and `values` arguments to prevent content validation.\\n1501:\\n1502: .. versionadded:: 3.0.0rc4\\n1503: \\\"\\\"\\\"\\n1504:\\n1505: mapping_type = dict\\n1506:\\n1507: #: Default error messages.\\n1508: default_error_messages = {\\\"invalid\\\": \\\"Not a valid mapping type.\\\"}\\n1509:\\n1510: def __init__(\\n1511: self,\\n1512: keys: typing.Optional[typing.Union[Field, type]] = None,\\n1513: values: typing.Optional[typing.Union[Field, type]] = None,\\n1514: **kwargs,\\n1515: ):\\n1516: super().__init__(**kwargs)\\n1517: if keys is None:\\n1518: self.key_field = None\\n1519: else:\\n1520: try:\\n1521: self.key_field = resolve_field_instance(keys)\\n1522: except FieldInstanceResolutionError as error:\\n1523: raise ValueError(\\n1524: '\\\"keys\\\" must be a subclass or instance of '\\n(473 more lines below)\\nFile updated. Please review the changes and make sure they are correct (correct indentation, no duplicate lines, etc). Edit the file again if necessary.\\n\",\n \"response\": \"My edit command did not use the proper indentation, I will fix my syntax in this follow up edit command.\\n```\\nedit 1475:1475\\n return int(round(value.total_seconds() / base_unit.total_seconds())) # round to nearest int\\nend_of_edit\\n```\",\n \"state\": \"{\\\"open_file\\\": \\\"/marshmallow-code__marshmallow/src/marshmallow/fields.py\\\", \\\"working_dir\\\": \\\"/marshmallow-code__marshmallow\\\"}\\n\",\n \"thought\": \"My edit command did not use the proper indentation, I will fix my syntax in this follow up edit command.\\n\"\n },\n {\n \"action\": \"python reproduce.py\\n\",\n \"observation\": \"345\\n\",\n \"response\": \"The code has been updated to use the `round` function, which should fix the rounding issue. Before submitting the changes, it would be prudent to run the reproduce.py code again to ensure that this change has the desired effect. I see that \\\"Current directory\\\" is still `/marshmallow-code__marshmallow`, so we can just run `reproduce.py` directly.\\n```\\npython reproduce.py\\n```\",\n \"state\": \"{\\\"open_file\\\": \\\"/marshmallow-code__marshmallow/src/marshmallow/fields.py\\\", \\\"working_dir\\\": \\\"/marshmallow-code__marshmallow\\\"}\\n\",\n \"thought\": \"The code has been updated to use the `round` function, which should fix the rounding issue. Before submitting the changes, it would be prudent to run the reproduce.py code again to ensure that this change has the desired effect. I see that \\\"Current directory\\\" is still `/marshmallow-code__marshmallow`, so we can just run `reproduce.py` directly.\\n\"\n },\n {\n \"action\": \"rm reproduce.py\\n\",\n \"observation\": \"\",\n \"response\": \"The output has changed from 344 to 345, which suggests that the rounding issue has been fixed. Let's remove the reproduce.py file since it is no longer needed.\\n```\\nrm reproduce.py\\n```\",\n \"state\": \"{\\\"open_file\\\": \\\"/marshmallow-code__marshmallow/src/marshmallow/fields.py\\\", \\\"working_dir\\\": \\\"/marshmallow-code__marshmallow\\\"}\\n\",\n \"thought\": \"The output has changed from 344 to 345, which suggests that the rounding issue has been fixed. Let's remove the reproduce.py file since it is no longer needed.\\n\"\n },\n {\n \"action\": \"submit\\n\",\n \"observation\": \"\\ndiff --git a/src/marshmallow/fields.py b/src/marshmallow/fields.py\\nindex ad388c7..20da768 100644\\n--- a/src/marshmallow/fields.py\\n+++ b/src/marshmallow/fields.py\\n@@ -1472,7 +1472,7 @@ class TimeDelta(Field):\\n if value is None:\\n return None\\n base_unit = dt.timedelta(**{self.precision: 1})\\n- return int(value.total_seconds() / base_unit.total_seconds())\\n+ return int(round(value.total_seconds() / base_unit.total_seconds())) # round to nearest int\\n \\n def _deserialize(self, value, attr, data, **kwargs):\\n try:\\n\",\n \"response\": \"rm doesn't have any output when it deletes successfully, so that must have worked. Now that we have fixed the issue, we should submit our changes to the code base. Let's submit the changes using the `submit` command.\\n```\\nsubmit\\n```\",\n \"state\": \"{\\\"open_file\\\": \\\"/marshmallow-code__marshmallow/src/marshmallow/fields.py\\\", \\\"working_dir\\\": \\\"/marshmallow-code__marshmallow\\\"}\\n\",\n \"thought\": \"rm doesn't have any output when it deletes successfully, so that must have worked. Now that we have fixed the issue, we should submit our changes to the code base. Let's submit the changes using the `submit` command.\\n\"\n }\n ],\n \"history\": [\n {\n \"role\": \"system\",\n \"content\": \"SETTING: You are an autonomous programmer, and you're working directly in the command line with a special interface.\\n\\nThe special interface consists of a file editor that shows you 100 lines of a file at a time.\\nIn addition to typical bash commands, you can also use the following commands to help you navigate and edit files.\\n\\nCOMMANDS:\\nopen:\\n docstring: opens the file at the given path in the editor. If line_number is provided, the window will be move to include that line\\n signature: open <path> [<line_number>]\\n arguments:\\n - path (string) [required]: the path to the file to open\\n - line_number (integer) [optional]: the line number to move the window to (if not provided, the window will start at the top of the file)\\n\\ngoto:\\n docstring: moves the window to show <line_number>\\n signature: goto <line_number>\\n arguments:\\n - line_number (integer) [required]: the line number to move the window to\\n\\nscroll_down:\\n docstring: moves the window down {WINDOW} lines\\n signature: scroll_down\\n\\nscroll_up:\\n docstring: moves the window down {WINDOW} lines\\n signature: scroll_down\\n\\ncreate:\\n docstring: creates and opens a new file with the given name\\n signature: create <filename>\\n arguments:\\n - filename (string) [required]: the name of the file to create\\n\\nsubmit:\\n docstring: submits your current code and terminates the session\\n signature: submit\\n\\nsearch_dir:\\n docstring: searches for search_term in all files in dir. If dir is not provided, searches in the current directory\\n signature: search_dir <search_term> [<dir>]\\n arguments:\\n - search_term (string) [required]: the term to search for\\n - dir (string) [optional]: the directory to search in (if not provided, searches in the current directory)\\n\\nsearch_file:\\n docstring: searches for search_term in file. If file is not provided, searches in the current open file\\n signature: search_file <search_term> [<file>]\\n arguments:\\n - search_term (string) [required]: the term to search for\\n - file (string) [optional]: the file to search in (if not provided, searches in the current open file)\\n\\nfind_file:\\n docstring: finds all files with the given name in dir. If dir is not provided, searches in the current directory\\n signature: find_file <file_name> [<dir>]\\n arguments:\\n - file_name (string) [required]: the name of the file to search for\\n - dir (string) [optional]: the directory to search in (if not provided, searches in the current directory)\\n\\nedit:\\n docstring: replaces lines <start_line> through <end_line> (inclusive) with the given text in the open file. The replacement text is terminated by a line with only end_of_edit on it. All of the <replacement text> will be entered, so make sure your indentation is formatted properly. Python files will be checked for syntax errors after the edit. If the system detects a syntax error, the edit will not be executed. Simply try to edit the file again, but make sure to read the error message and modify the edit command you issue accordingly. Issuing the same command a second time will just lead to the same error message again.\\n signature: edit <start_line>:<end_line>\\n<replacement_text>\\nend_of_edit\\n arguments:\\n - start_line (integer) [required]: the line number to start the edit at\\n - end_line (integer) [required]: the line number to end the edit at (inclusive)\\n - replacement_text (string) [required]: the text to replace the current selection with\\n\\n\\n\\nPlease note that THE EDIT COMMAND REQUIRES PROPER INDENTATION. \\nIf you'd like to add the line ' print(x)' you must fully write that out, with all those spaces before the code! Indentation is important and code that is not indented correctly will fail and require fixing before it can be run.\\n\\nRESPONSE FORMAT:\\nYour shell prompt is formatted as follows:\\n(Open file: <path>) <cwd> $\\n\\nYou need to format your output using two fields; discussion and command.\\nYour output should always include _one_ discussion and _one_ command field EXACTLY as in the following example:\\nDISCUSSION\\nFirst I'll start by using ls to see what files are in the current directory. Then maybe we can look at some relevant files to see what they look like.\\n```\\nls -a\\n```\\n\\nYou should only include a *SINGLE* command in the command section and then wait for a response from the shell before continuing with more discussion and commands. Everything you include in the DISCUSSION section will be saved for future reference.\\nIf you'd like to issue two commands at once, PLEASE DO NOT DO THAT! Please instead first submit just the first command, and then after receiving a response you'll be able to issue the second command. \\nYou're free to use any other bash commands you want (e.g. find, grep, cat, ls, cd) in addition to the special commands listed above.\\nHowever, the environment does NOT support interactive session commands (e.g. python, vim), so please do not invoke them.\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"user\",\n \"content\": \"We're currently solving the following issue within our repository. Here's the issue text:\\nISSUE:\\nTimeDelta serialization precision\\nHi there!\\n\\nI just found quite strange behaviour of `TimeDelta` field serialization\\n\\n```python3\\nfrom marshmallow.fields import TimeDelta\\nfrom datetime import timedelta\\n\\ntd_field = TimeDelta(precision=\\\"milliseconds\\\")\\n\\nobj = dict()\\nobj[\\\"td_field\\\"] = timedelta(milliseconds=345)\\n\\nprint(td_field.serialize(\\\"td_field\\\", obj))\\n```\\n\\nOutput of this snippet is `344`, but it seems that `345` is correct.\\n\\nLooks like a rounding issue here: https://github.com/marshmallow-code/marshmallow/blob/dev/src/marshmallow/fields.py#L1474\\n\\n\\nINSTRUCTIONS:\\nNow, you're going to solve this issue on your own. Your terminal session has started and you're in the repository's root directory. You can use any bash commands or the special interface to help you. Edit all the files you need to and run any checks or tests that you want. \\nRemember, YOU CAN ONLY ENTER ONE COMMAND AT A TIME. You should always wait for feedback after every command. \\nWhen you're satisfied with all of the changes you've made, you can submit your changes to the code base by simply running the submit command.\\nNote however that you cannot use any interactive session commands (e.g. python, vim) in this environment, but you can write scripts and run them. E.g. you can write a python script and then run it with `python <script_name>.py`.\\n\\nNOTE ABOUT THE EDIT COMMAND: Indentation really matters! When editing a file, make sure to insert appropriate indentation before each line! \\n\\nIMPORTANT TIPS:\\n1. Always start by trying to replicate the bug that the issues discusses. \\n If the issue includes code for reproducing the bug, we recommend that you re-implement that in your environment, and run it to make sure you can reproduce the bug.\\n Then start trying to fix it.\\n When you think you've fixed the bug, re-run the bug reproduction script to make sure that the bug has indeed been fixed.\\n \\n If the bug reproduction script does not print anything when it succesfully runs, we recommend adding a print(\\\"Script completed successfully, no errors.\\\") command at the end of the file,\\n so that you can be sure that the script indeed ran fine all the way through. \\n\\n2. If you run a command and it doesn't work, try running a different command. A command that did not work once will not work the second time unless you modify it!\\n\\n3. If you open a file and need to get to an area around a specific line that is not in the first 100 lines, say line 583, don't just use the scroll_down command multiple times. Instead, use the goto 583 command. It's much quicker. \\n \\n4. If the bug reproduction script requires inputting/reading a specific file, such as buggy-input.png, and you'd like to understand how to input that file, conduct a search in the existing repo code, to see whether someone else has already done that. Do this by running the command: find_file \\\"buggy-input.png\\\" If that doensn't work, use the linux 'find' command. \\n\\n5. Always make sure to look at the currently open file and the current working directory (which appears right after the currently open file). The currently open file might be in a different directory than the working directory! Note that some commands, such as 'create', open files, so they might change the current open file.\\n\\n6. When editing files, it is easy to accidentally specify a wrong line number or to write code with incorrect indentation. Always check the code after you issue an edit to make sure that it reflects what you wanted to accomplish. If it didn't, issue another command to fix it.\\n \\n\\n(Open file: n/a)\\n(Current directory: /marshmallow-code__marshmallow)\\nbash-$\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"assistant\",\n \"content\": \"Let's list out some of the files in the repository to get an idea of the structure and contents. We can use the `ls -F` command to list the files in the current directory.\\n```\\nls -F\\n```\",\n \"thought\": \"Let's list out some of the files in the repository to get an idea of the structure and contents. We can use the `ls -F` command to list the files in the current directory.\\n\",\n \"action\": \"ls -F\\n\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"user\",\n \"content\": \"AUTHORS.rst\\nCHANGELOG.rst\\nCODE_OF_CONDUCT.md\\nCONTRIBUTING.rst\\nLICENSE\\nMANIFEST.in\\nNOTICE\\nREADME.rst\\nRELEASING.md\\nazure-pipelines.yml\\ndocs/\\nexamples/\\nperformance/\\npyproject.toml\\nsetup.cfg\\nsetup.py\\nsrc/\\ntests/\\ntox.ini\\n\\n(Open file: n/a)\\n(Current directory: /marshmallow-code__marshmallow)\\nbash-$\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"assistant\",\n \"content\": \"We see that there's a setup.py file. This could be useful for installing the package locally. Since we'll probably need to reproduce the issue to solve it, it would be a good idea to install the package locally. Let's checkout the setup.py file to see what commands we can use to install the package.\\n```\\nopen setup.py\\n```\",\n \"thought\": \"We see that there's a setup.py file. This could be useful for installing the package locally. Since we'll probably need to reproduce the issue to solve it, it would be a good idea to install the package locally. Let's checkout the setup.py file to see what commands we can use to install the package.\\n\",\n \"action\": \"open setup.py\\n\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"user\",\n \"content\": \"[File: /marshmallow-code__marshmallow/setup.py (94 lines total)]\\n1:import re\\n2:from setuptools import setup, find_packages\\n3:\\n4:EXTRAS_REQUIRE = {\\n5: \\\"tests\\\": [\\\"pytest\\\", \\\"pytz\\\", \\\"simplejson\\\"],\\n6: \\\"lint\\\": [\\n7: \\\"mypy==0.910\\\",\\n8: \\\"flake8==4.0.1\\\",\\n9: \\\"flake8-bugbear==21.9.2\\\",\\n10: \\\"pre-commit~=2.4\\\",\\n11: ],\\n12: \\\"docs\\\": [\\n13: \\\"sphinx==4.2.0\\\",\\n14: \\\"sphinx-issues==1.2.0\\\",\\n15: \\\"alabaster==0.7.12\\\",\\n16: \\\"sphinx-version-warning==1.1.2\\\",\\n17: \\\"autodocsumm==0.2.7\\\",\\n18: ],\\n19:}\\n20:EXTRAS_REQUIRE[\\\"dev\\\"] = EXTRAS_REQUIRE[\\\"tests\\\"] + EXTRAS_REQUIRE[\\\"lint\\\"] + [\\\"tox\\\"]\\n21:\\n22:\\n23:def find_version(fname):\\n24: \\\"\\\"\\\"Attempts to find the version number in the file names fname.\\n25: Raises RuntimeError if not found.\\n26: \\\"\\\"\\\"\\n27: version = \\\"\\\"\\n28: with open(fname) as fp:\\n29: reg = re.compile(r'__version__ = [\\\\'\\\"]([^\\\\'\\\"]*)[\\\\'\\\"]')\\n30: for line in fp:\\n31: m = reg.match(line)\\n32: if m:\\n33: version = m.group(1)\\n34: break\\n35: if not version:\\n36: raise RuntimeError(\\\"Cannot find version information\\\")\\n37: return version\\n38:\\n39:\\n40:def read(fname):\\n41: with open(fname) as fp:\\n42: content = fp.read()\\n43: return content\\n44:\\n45:\\n46:setup(\\n47: name=\\\"marshmallow\\\",\\n48: version=find_version(\\\"src/marshmallow/__init__.py\\\"),\\n49: description=(\\n50: \\\"A lightweight library for converting complex \\\"\\n51: \\\"datatypes to and from native Python datatypes.\\\"\\n52: ),\\n53: long_description=read(\\\"README.rst\\\"),\\n54: author=\\\"Steven Loria\\\",\\n55: author_email=\\\"sloria1@gmail.com\\\",\\n56: url=\\\"https://github.com/marshmallow-code/marshmallow\\\",\\n57: packages=find_packages(\\\"src\\\", exclude=(\\\"test*\\\", \\\"examples\\\")),\\n58: package_dir={\\\"\\\": \\\"src\\\"},\\n59: package_data={\\\"marshmallow\\\": [\\\"py.typed\\\"]},\\n60: include_package_data=True,\\n61: extras_require=EXTRAS_REQUIRE,\\n62: license=\\\"MIT\\\",\\n63: zip_safe=False,\\n64: keywords=[\\n65: \\\"serialization\\\",\\n66: \\\"rest\\\",\\n67: \\\"json\\\",\\n68: \\\"api\\\",\\n69: \\\"marshal\\\",\\n70: \\\"marshalling\\\",\\n71: \\\"deserialization\\\",\\n72: \\\"validation\\\",\\n73: \\\"schema\\\",\\n74: ],\\n75: python_requires=\\\">=3.6\\\",\\n76: classifiers=[\\n77: \\\"Development Status :: 5 - Production/Stable\\\",\\n78: \\\"Intended Audience :: Developers\\\",\\n79: \\\"License :: OSI Approved :: MIT License\\\",\\n80: \\\"Programming Language :: Python :: 3\\\",\\n81: \\\"Programming Language :: Python :: 3.6\\\",\\n82: \\\"Programming Language :: Python :: 3.7\\\",\\n83: \\\"Programming Language :: Python :: 3.8\\\",\\n84: \\\"Programming Language :: Python :: 3.9\\\",\\n85: \\\"Programming Language :: Python :: 3.10\\\",\\n86: ],\\n87: test_suite=\\\"tests\\\",\\n88: project_urls={\\n89: \\\"Changelog\\\": \\\"https://marshmallow.readthedocs.io/en/latest/changelog.html\\\",\\n90: \\\"Issues\\\": \\\"https://github.com/marshmallow-code/marshmallow/issues\\\",\\n91: \\\"Funding\\\": \\\"https://opencollective.com/marshmallow\\\",\\n92: \\\"Tidelift\\\": \\\"https://tidelift.com/subscription/pkg/pypi-marshmallow?utm_source=pypi-marshmallow&utm_medium=pypi\\\", # noqa\\n93: },\\n94:)\\n\\n(Open file: /marshmallow-code__marshmallow/setup.py)\\n(Current directory: /marshmallow-code__marshmallow)\\nbash-$\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"assistant\",\n \"content\": \"The setup.py file contains a lot of useful information to install the package locally. In particular, I see there's a [dev] extras that installs all the dependencies needed for development. We can use this to install the package with all the necessary dependencies. Let's install the package locally with the [dev] extras.\\n```\\npip install -e .[dev]\\n```\",\n \"thought\": \"The setup.py file contains a lot of useful information to install the package locally. In particular, I see there's a [dev] extras that installs all the dependencies needed for development. We can use this to install the package with all the necessary dependencies. Let's install the package locally with the [dev] extras.\\n\",\n \"action\": \"pip install -e .[dev]\\n\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"user\",\n \"content\": \"Obtaining file:///marshmallow-code__marshmallow\\n Installing build dependencies: started\\n Installing build dependencies: finished with status 'done'\\n Checking if build backend supports build_editable: started\\n Checking if build backend supports build_editable: finished with status 'done'\\n Getting requirements to build editable: started\\n Getting requirements to build editable: finished with status 'done'\\n Preparing editable metadata (pyproject.toml): started\\n Preparing editable metadata (pyproject.toml): finished with status 'done'\\nRequirement already satisfied: pytest in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (8.1.1)\\nRequirement already satisfied: pytz in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (2024.1)\\nRequirement already satisfied: simplejson in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (3.19.2)\\nRequirement already satisfied: mypy==0.910 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (0.910)\\nRequirement already satisfied: flake8==4.0.1 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (4.0.1)\\nRequirement already satisfied: flake8-bugbear==21.9.2 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (21.9.2)\\nRequirement already satisfied: pre-commit~=2.4 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (2.21.0)\\nRequirement already satisfied: tox in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (4.14.2)\\nRequirement already satisfied: mccabe<0.7.0,>=0.6.0 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from flake8==4.0.1) (0.6.1)\\nRequirement already satisfied: pycodestyle<2.9.0,>=2.8.0 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from flake8==4.0.1) (2.8.0)\\nRequirement already satisfied: pyflakes<2.5.0,>=2.4.0 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from flake8==4.0.1) (2.4.0)\\nRequirement already satisfied: attrs>=19.2.0 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from flake8-bugbear==21.9.2) (23.2.0)\\nRequirement already satisfied: typing-extensions>=3.7.4 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from mypy==0.910) (4.10.0)\\nRequirement already satisfied: mypy-extensions<0.5.0,>=0.4.3 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from mypy==0.910) (0.4.4)\\nRequirement already satisfied: toml in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from mypy==0.910) (0.10.2)\\nRequirement already satisfied: cfgv>=2.0.0 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from pre-commit~=2.4) (3.4.0)\\nRequirement already satisfied: identify>=1.0.0 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from pre-commit~=2.4) (2.5.35)\\nRequirement already satisfied: nodeenv>=0.11.1 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from pre-commit~=2.4) (1.8.0)\\nRequirement already satisfied: pyyaml>=5.1 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from pre-commit~=2.4) (6.0.1)\\nRequirement already satisfied: virtualenv>=20.10.0 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from pre-commit~=2.4) (20.25.1)\\nRequirement already satisfied: iniconfig in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from pytest) (2.0.0)\\nRequirement already satisfied: packaging in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from pytest) (24.0)\\nRequirement already satisfied: pluggy<2.0,>=1.4 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from pytest) (1.4.0)\\nRequirement already satisfied: exceptiongroup>=1.0.0rc8 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from pytest) (1.2.0)\\nRequirement already satisfied: tomli>=1 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from pytest) (2.0.1)\\nRequirement already satisfied: cachetools>=5.3.2 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from tox) (5.3.3)\\nRequirement already satisfied: chardet>=5.2 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from tox) (5.2.0)\\nRequirement already satisfied: colorama>=0.4.6 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from tox) (0.4.6)\\nRequirement already satisfied: filelock>=3.13.1 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from tox) (3.13.3)\\nRequirement already satisfied: platformdirs>=4.1 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from tox) (4.2.0)\\nRequirement already satisfied: pyproject-api>=1.6.1 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from tox) (1.6.1)\\nRequirement already satisfied: setuptools in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from nodeenv>=0.11.1->pre-commit~=2.4) (68.2.2)\\nRequirement already satisfied: distlib<1,>=0.3.7 in /root/miniconda3/envs/marshmallow-code__marshmallow__3.13/lib/python3.9/site-packages (from virtualenv>=20.10.0->pre-commit~=2.4) (0.3.8)\\nChecking if build backend supports build_editable: started\\nChecking if build backend supports build_editable: finished with status 'done'\\nBuilding wheels for collected packages: marshmallow\\n Building editable for marshmallow (pyproject.toml): started\\n Building editable for marshmallow (pyproject.toml): finished with status 'done'\\n Created wheel for marshmallow: filename=marshmallow-3.13.0-0.editable-py3-none-any.whl size=4534 sha256=fcca61a41cd2783d4a5b8471ebc48cf886d36f6c66a29cc9b02761fc174a024d\\n Stored in directory: /tmp/pip-ephem-wheel-cache-h335xxo8/wheels/4d/da/d4/e230bf0a3f16cad5a83d235ac24a34b55178ddda181cb64dae\\nSuccessfully built marshmallow\\nInstalling collected packages: marshmallow\\n Attempting uninstall: marshmallow\\n Found existing installation: marshmallow 3.13.0\\n Uninstalling marshmallow-3.13.0:\\n Successfully uninstalled marshmallow-3.13.0\\nSuccessfully installed marshmallow-3.13.0\\nWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv\\n\\n(Open file: /marshmallow-code__marshmallow/setup.py)\\n(Current directory: /marshmallow-code__marshmallow)\\nbash-$\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"assistant\",\n \"content\": \"Looks like it installed successfully. Now that we have the package installed, we can start working on the issue.\\nLet's first start by reproducing the results of the issue. The issue includes some example code for reproduction, which we can use. We'll create a new file called `reproduce.py` and paste the example code into it.\\n```\\ncreate reproduce.py\\n```\",\n \"thought\": \"Looks like it installed successfully. Now that we have the package installed, we can start working on the issue.\\nLet's first start by reproducing the results of the issue. The issue includes some example code for reproduction, which we can use. We'll create a new file called `reproduce.py` and paste the example code into it.\\n\",\n \"action\": \"create reproduce.py\\n\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"user\",\n \"content\": \"[File: /marshmallow-code__marshmallow/reproduce.py (1 lines total)]\\n1:\\n\\n(Open file: /marshmallow-code__marshmallow/reproduce.py)\\n(Current directory: /marshmallow-code__marshmallow)\\nbash-$\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"assistant\",\n \"content\": \"Now let's paste in the example code from the issue.\\n```\\nedit 1:1\\nfrom marshmallow.fields import TimeDelta\\nfrom datetime import timedelta\\n\\ntd_field = TimeDelta(precision=\\\"milliseconds\\\")\\n\\nobj = dict()\\nobj[\\\"td_field\\\"] = timedelta(milliseconds=345)\\n\\nprint(td_field.serialize(\\\"td_field\\\", obj))\\nend_of_edit\\n```\",\n \"thought\": \"Now let's paste in the example code from the issue.\\n\",\n \"action\": \"edit 1:1\\nfrom marshmallow.fields import TimeDelta\\nfrom datetime import timedelta\\n\\ntd_field = TimeDelta(precision=\\\"milliseconds\\\")\\n\\nobj = dict()\\nobj[\\\"td_field\\\"] = timedelta(milliseconds=345)\\n\\nprint(td_field.serialize(\\\"td_field\\\", obj))\\nend_of_edit\\n\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"user\",\n \"content\": \"[File: /marshmallow-code__marshmallow/reproduce.py (9 lines total)]\\n1:from marshmallow.fields import TimeDelta\\n2:from datetime import timedelta\\n3:\\n4:td_field = TimeDelta(precision=\\\"milliseconds\\\")\\n5:\\n6:obj = dict()\\n7:obj[\\\"td_field\\\"] = timedelta(milliseconds=345)\\n8:\\n9:print(td_field.serialize(\\\"td_field\\\", obj))\\nFile updated. Please review the changes and make sure they are correct (correct indentation, no duplicate lines, etc). Edit the file again if necessary.\\n\\n(Open file: /marshmallow-code__marshmallow/reproduce.py)\\n(Current directory: /marshmallow-code__marshmallow)\\nbash-$\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"assistant\",\n \"content\": \"Now let's run the code to see if we see the same output as the issue.\\n```\\npython reproduce.py\\n```\",\n \"thought\": \"Now let's run the code to see if we see the same output as the issue.\\n\",\n \"action\": \"python reproduce.py\\n\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"user\",\n \"content\": \"344\\n\\n(Open file: /marshmallow-code__marshmallow/reproduce.py)\\n(Current directory: /marshmallow-code__marshmallow)\\nbash-$\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"assistant\",\n \"content\": \"We are indeed seeing the same output as the issue. The issue suggests that we should look at line 1474 of the `fields.py` file to see if there is a rounding issue. Let's open the `fields.py` file and navigate to line 1474 to see the relevant code. First we should find the `fields.py` file in the repository. It is likely to be in the `src/marshmallow` directory, but we should check to be sure.\\n```\\nls -F\\n```\",\n \"thought\": \"We are indeed seeing the same output as the issue. The issue suggests that we should look at line 1474 of the `fields.py` file to see if there is a rounding issue. Let's open the `fields.py` file and navigate to line 1474 to see the relevant code. First we should find the `fields.py` file in the repository. It is likely to be in the `src/marshmallow` directory, but we should check to be sure.\\n\",\n \"action\": \"ls -F\\n\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"user\",\n \"content\": \"AUTHORS.rst\\nCHANGELOG.rst\\nCODE_OF_CONDUCT.md\\nCONTRIBUTING.rst\\nLICENSE\\nMANIFEST.in\\nNOTICE\\nREADME.rst\\nRELEASING.md\\nazure-pipelines.yml\\ndocs/\\nexamples/\\nperformance/\\npyproject.toml\\nreproduce.py\\nsetup.cfg\\nsetup.py\\nsrc/\\ntests/\\ntox.ini\\n\\n(Open file: /marshmallow-code__marshmallow/reproduce.py)\\n(Current directory: /marshmallow-code__marshmallow)\\nbash-$\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"assistant\",\n \"content\": \"It looks like the `src` directory is present, which suggests that the `fields.py` file is likely to be in the `src` directory. Let's use find_file to see where it is.\\n```\\nfind_file \\\"fields.py\\\" src\\n```\",\n \"thought\": \"It looks like the `src` directory is present, which suggests that the `fields.py` file is likely to be in the `src` directory. Let's use find_file to see where it is.\\n\",\n \"action\": \"find_file \\\"fields.py\\\" src\\n\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"user\",\n \"content\": \"Found 1 matches for \\\"fields.py\\\" in /marshmallow-code__marshmallow/src:\\n/marshmallow-code__marshmallow/src/marshmallow/fields.py\\n\\n(Open file: /marshmallow-code__marshmallow/reproduce.py)\\n(Current directory: /marshmallow-code__marshmallow)\\nbash-$\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"assistant\",\n \"content\": \"It looks like the `fields.py` file is present in the `./src/marshmallow/` directory. The issue also points to a specific URL with line number 1474. We should navigate to that line in fields.py to see the relevant code for the `TimeDelta` serialization.\\n```\\nopen src/marshmallow/fields.py 1474\\n```\",\n \"thought\": \"It looks like the `fields.py` file is present in the `./src/marshmallow/` directory. The issue also points to a specific URL with line number 1474. We should navigate to that line in fields.py to see the relevant code for the `TimeDelta` serialization.\\n\",\n \"action\": \"open src/marshmallow/fields.py 1474\\n\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"user\",\n \"content\": \"[File: /marshmallow-code__marshmallow/src/marshmallow/fields.py (1997 lines total)]\\n(1458 more lines above)\\n1459: self.WEEKS,\\n1460: )\\n1461:\\n1462: if precision not in units:\\n1463: msg = 'The precision must be {} or \\\"{}\\\".'.format(\\n1464: \\\", \\\".join([f'\\\"{each}\\\"' for each in units[:-1]]), units[-1]\\n1465: )\\n1466: raise ValueError(msg)\\n1467:\\n1468: self.precision = precision\\n1469: super().__init__(**kwargs)\\n1470:\\n1471: def _serialize(self, value, attr, obj, **kwargs):\\n1472: if value is None:\\n1473: return None\\n1474: base_unit = dt.timedelta(**{self.precision: 1})\\n1475: return int(value.total_seconds() / base_unit.total_seconds())\\n1476:\\n1477: def _deserialize(self, value, attr, data, **kwargs):\\n1478: try:\\n1479: value = int(value)\\n1480: except (TypeError, ValueError) as error:\\n1481: raise self.make_error(\\\"invalid\\\") from error\\n1482:\\n1483: kwargs = {self.precision: value}\\n1484:\\n1485: try:\\n1486: return dt.timedelta(**kwargs)\\n1487: except OverflowError as error:\\n1488: raise self.make_error(\\\"invalid\\\") from error\\n1489:\\n1490:\\n1491:class Mapping(Field):\\n1492: \\\"\\\"\\\"An abstract class for objects with key-value pairs.\\n1493:\\n1494: :param keys: A field class or instance for dict keys.\\n1495: :param values: A field class or instance for dict values.\\n1496: :param kwargs: The same keyword arguments that :class:`Field` receives.\\n1497:\\n1498: .. note::\\n1499: When the structure of nested data is not known, you may omit the\\n1500: `keys` and `values` arguments to prevent content validation.\\n1501:\\n1502: .. versionadded:: 3.0.0rc4\\n1503: \\\"\\\"\\\"\\n1504:\\n1505: mapping_type = dict\\n1506:\\n1507: #: Default error messages.\\n1508: default_error_messages = {\\\"invalid\\\": \\\"Not a valid mapping type.\\\"}\\n1509:\\n1510: def __init__(\\n1511: self,\\n1512: keys: typing.Optional[typing.Union[Field, type]] = None,\\n1513: values: typing.Optional[typing.Union[Field, type]] = None,\\n1514: **kwargs,\\n1515: ):\\n1516: super().__init__(**kwargs)\\n1517: if keys is None:\\n1518: self.key_field = None\\n1519: else:\\n1520: try:\\n1521: self.key_field = resolve_field_instance(keys)\\n1522: except FieldInstanceResolutionError as error:\\n1523: raise ValueError(\\n1524: '\\\"keys\\\" must be a subclass or instance of '\\n1525: \\\"marshmallow.base.FieldABC.\\\"\\n1526: ) from error\\n1527:\\n1528: if values is None:\\n1529: self.value_field = None\\n1530: else:\\n1531: try:\\n1532: self.value_field = resolve_field_instance(values)\\n1533: except FieldInstanceResolutionError as error:\\n1534: raise ValueError(\\n1535: '\\\"values\\\" must be a subclass or instance of '\\n1536: \\\"marshmallow.base.FieldABC.\\\"\\n1537: ) from error\\n1538: if isinstance(self.value_field, Nested):\\n1539: self.only = self.value_field.only\\n1540: self.exclude = self.value_field.exclude\\n1541:\\n1542: def _bind_to_schema(self, field_name, schema):\\n1543: super()._bind_to_schema(field_name, schema)\\n1544: if self.value_field:\\n1545: self.value_field = copy.deepcopy(self.value_field)\\n1546: self.value_field._bind_to_schema(field_name, self)\\n1547: if isinstance(self.value_field, Nested):\\n1548: self.value_field.only = self.only\\n1549: self.value_field.exclude = self.exclude\\n1550: if self.key_field:\\n1551: self.key_field = copy.deepcopy(self.key_field)\\n1552: self.key_field._bind_to_schema(field_name, self)\\n1553:\\n1554: def _serialize(self, value, attr, obj, **kwargs):\\n1555: if value is None:\\n1556: return None\\n1557: if not self.value_field and not self.key_field:\\n1558: return self.mapping_type(value)\\n(439 more lines below)\\n\\n(Open file: /marshmallow-code__marshmallow/src/marshmallow/fields.py)\\n(Current directory: /marshmallow-code__marshmallow)\\nbash-$\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"assistant\",\n \"content\": \"We are now looking at the relevant section of the `fields.py` file where the `TimeDelta` serialization occurs. The issue suggests that there is a rounding problem near line 1474. The current implementation uses integer division, which truncates the decimal part and can lead to incorrect rounding on line 1475.To fix this issue, we should ensure that the division results in a float, and then round the result to the nearest integer before converting it to an int. This should preserve the milliseconds precision as expected. Let's make the necessary edit to the code.\\n```\\nedit 1475:1475\\nreturn int(round(value.total_seconds() / base_unit.total_seconds())) # round to nearest int\\nend_of_edit\\n```\",\n \"thought\": \"We are now looking at the relevant section of the `fields.py` file where the `TimeDelta` serialization occurs. The issue suggests that there is a rounding problem near line 1474. The current implementation uses integer division, which truncates the decimal part and can lead to incorrect rounding on line 1475.To fix this issue, we should ensure that the division results in a float, and then round the result to the nearest integer before converting it to an int. This should preserve the milliseconds precision as expected. Let's make the necessary edit to the code.\\n\",\n \"action\": \"edit 1475:1475\\nreturn int(round(value.total_seconds() / base_unit.total_seconds())) # round to nearest int\\nend_of_edit\\n\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"user\",\n \"content\": \"Your proposed edit has introduced new syntax error(s). Please understand the fixes and retry your edit commmand.\\n\\nERRORS:\\n- E999 IndentationError: unexpected indent\\n\\nThis is how your edit would have looked if applied\\n-------------------------------------------------\\n[File: /marshmallow-code__marshmallow/src/marshmallow/fields.py (1997 lines total)]\\n(1468 more lines above)\\n1469: super().__init__(**kwargs)\\n1470:\\n1471: def _serialize(self, value, attr, obj, **kwargs):\\n1472: if value is None:\\n1473: return None\\n1474: base_unit = dt.timedelta(**{self.precision: 1})\\n1475:return int(round(value.total_seconds() / base_unit.total_seconds())) # round to nearest int\\n1476:\\n1477: def _deserialize(self, value, attr, data, **kwargs):\\n1478: try:\\n1479: value = int(value)\\n(518 more lines below)\\n-------------------------------------------------\\n\\nThis is the original code before your edit\\n-------------------------------------------------\\n[File: /marshmallow-code__marshmallow/src/marshmallow/fields.py (1997 lines total)]\\n(1469 more lines above)\\n1470:\\n1471: def _serialize(self, value, attr, obj, **kwargs):\\n1472: if value is None:\\n1473: return None\\n1474: base_unit = dt.timedelta(**{self.precision: 1})\\n1475: return int(value.total_seconds() / base_unit.total_seconds())\\n1476:\\n1477: def _deserialize(self, value, attr, data, **kwargs):\\n1478: try:\\n1479: value = int(value)\\n1480: except (TypeError, ValueError) as error:\\n(517 more lines below)\\n-------------------------------------------------\\nYour changes have NOT been applied. Please fix your edit command and try again.\\nYou either need to 1) Specify the correct start/end line arguments or 2) Correct your edit code.\\nDO NOT re-run the same failed edit command. Running it again will lead to the same error.\\n\\n(Open file: /marshmallow-code__marshmallow/src/marshmallow/fields.py)\\n(Current directory: /marshmallow-code__marshmallow)\\nbash-$\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"assistant\",\n \"content\": \"My edit command did not use the proper indentation, I will fix my syntax in this follow up edit command.\\n```\\nedit 1475:1475\\n return int(round(value.total_seconds() / base_unit.total_seconds())) # round to nearest int\\nend_of_edit\\n```\",\n \"thought\": \"My edit command did not use the proper indentation, I will fix my syntax in this follow up edit command.\\n\",\n \"action\": \"edit 1475:1475\\n return int(round(value.total_seconds() / base_unit.total_seconds())) # round to nearest int\\nend_of_edit\\n\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"user\",\n \"content\": \"[File: /marshmallow-code__marshmallow/src/marshmallow/fields.py (1997 lines total)]\\n(1424 more lines above)\\n1425:\\n1426: :param precision: Influences how the integer is interpreted during\\n1427: (de)serialization. Must be 'days', 'seconds', 'microseconds',\\n1428: 'milliseconds', 'minutes', 'hours' or 'weeks'.\\n1429: :param kwargs: The same keyword arguments that :class:`Field` receives.\\n1430:\\n1431: .. versionchanged:: 2.0.0\\n1432: Always serializes to an integer value to avoid rounding errors.\\n1433: Add `precision` parameter.\\n1434: \\\"\\\"\\\"\\n1435:\\n1436: DAYS = \\\"days\\\"\\n1437: SECONDS = \\\"seconds\\\"\\n1438: MICROSECONDS = \\\"microseconds\\\"\\n1439: MILLISECONDS = \\\"milliseconds\\\"\\n1440: MINUTES = \\\"minutes\\\"\\n1441: HOURS = \\\"hours\\\"\\n1442: WEEKS = \\\"weeks\\\"\\n1443:\\n1444: #: Default error messages.\\n1445: default_error_messages = {\\n1446: \\\"invalid\\\": \\\"Not a valid period of time.\\\",\\n1447: \\\"format\\\": \\\"{input!r} cannot be formatted as a timedelta.\\\",\\n1448: }\\n1449:\\n1450: def __init__(self, precision: str = SECONDS, **kwargs):\\n1451: precision = precision.lower()\\n1452: units = (\\n1453: self.DAYS,\\n1454: self.SECONDS,\\n1455: self.MICROSECONDS,\\n1456: self.MILLISECONDS,\\n1457: self.MINUTES,\\n1458: self.HOURS,\\n1459: self.WEEKS,\\n1460: )\\n1461:\\n1462: if precision not in units:\\n1463: msg = 'The precision must be {} or \\\"{}\\\".'.format(\\n1464: \\\", \\\".join([f'\\\"{each}\\\"' for each in units[:-1]]), units[-1]\\n1465: )\\n1466: raise ValueError(msg)\\n1467:\\n1468: self.precision = precision\\n1469: super().__init__(**kwargs)\\n1470:\\n1471: def _serialize(self, value, attr, obj, **kwargs):\\n1472: if value is None:\\n1473: return None\\n1474: base_unit = dt.timedelta(**{self.precision: 1})\\n1475: return int(round(value.total_seconds() / base_unit.total_seconds())) # round to nearest int\\n1476:\\n1477: def _deserialize(self, value, attr, data, **kwargs):\\n1478: try:\\n1479: value = int(value)\\n1480: except (TypeError, ValueError) as error:\\n1481: raise self.make_error(\\\"invalid\\\") from error\\n1482:\\n1483: kwargs = {self.precision: value}\\n1484:\\n1485: try:\\n1486: return dt.timedelta(**kwargs)\\n1487: except OverflowError as error:\\n1488: raise self.make_error(\\\"invalid\\\") from error\\n1489:\\n1490:\\n1491:class Mapping(Field):\\n1492: \\\"\\\"\\\"An abstract class for objects with key-value pairs.\\n1493:\\n1494: :param keys: A field class or instance for dict keys.\\n1495: :param values: A field class or instance for dict values.\\n1496: :param kwargs: The same keyword arguments that :class:`Field` receives.\\n1497:\\n1498: .. note::\\n1499: When the structure of nested data is not known, you may omit the\\n1500: `keys` and `values` arguments to prevent content validation.\\n1501:\\n1502: .. versionadded:: 3.0.0rc4\\n1503: \\\"\\\"\\\"\\n1504:\\n1505: mapping_type = dict\\n1506:\\n1507: #: Default error messages.\\n1508: default_error_messages = {\\\"invalid\\\": \\\"Not a valid mapping type.\\\"}\\n1509:\\n1510: def __init__(\\n1511: self,\\n1512: keys: typing.Optional[typing.Union[Field, type]] = None,\\n1513: values: typing.Optional[typing.Union[Field, type]] = None,\\n1514: **kwargs,\\n1515: ):\\n1516: super().__init__(**kwargs)\\n1517: if keys is None:\\n1518: self.key_field = None\\n1519: else:\\n1520: try:\\n1521: self.key_field = resolve_field_instance(keys)\\n1522: except FieldInstanceResolutionError as error:\\n1523: raise ValueError(\\n1524: '\\\"keys\\\" must be a subclass or instance of '\\n(473 more lines below)\\nFile updated. Please review the changes and make sure they are correct (correct indentation, no duplicate lines, etc). Edit the file again if necessary.\\n\\n(Open file: /marshmallow-code__marshmallow/src/marshmallow/fields.py)\\n(Current directory: /marshmallow-code__marshmallow)\\nbash-$\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"assistant\",\n \"content\": \"The code has been updated to use the `round` function, which should fix the rounding issue. Before submitting the changes, it would be prudent to run the reproduce.py code again to ensure that this change has the desired effect. I see that \\\"Current directory\\\" is still `/marshmallow-code__marshmallow`, so we can just run `reproduce.py` directly.\\n```\\npython reproduce.py\\n```\",\n \"thought\": \"The code has been updated to use the `round` function, which should fix the rounding issue. Before submitting the changes, it would be prudent to run the reproduce.py code again to ensure that this change has the desired effect. I see that \\\"Current directory\\\" is still `/marshmallow-code__marshmallow`, so we can just run `reproduce.py` directly.\\n\",\n \"action\": \"python reproduce.py\\n\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"user\",\n \"content\": \"345\\n\\n(Open file: /marshmallow-code__marshmallow/src/marshmallow/fields.py)\\n(Current directory: /marshmallow-code__marshmallow)\\nbash-$\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"assistant\",\n \"content\": \"The output has changed from 344 to 345, which suggests that the rounding issue has been fixed. Let's remove the reproduce.py file since it is no longer needed.\\n```\\nrm reproduce.py\\n```\",\n \"thought\": \"The output has changed from 344 to 345, which suggests that the rounding issue has been fixed. Let's remove the reproduce.py file since it is no longer needed.\\n\",\n \"action\": \"rm reproduce.py\\n\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"user\",\n \"content\": \"Your command ran successfully and did not produce any output.\\n(Open file: /marshmallow-code__marshmallow/src/marshmallow/fields.py)\\n(Current directory: /marshmallow-code__marshmallow)\\nbash-$\",\n \"agent\": \"primary\"\n },\n {\n \"role\": \"assistant\",\n \"content\": \"rm doesn't have any output when it deletes successfully, so that must have worked. Now that we have fixed the issue, we should submit our changes to the code base. Let's submit the changes using the `submit` command.\\n```\\nsubmit\\n```\",\n \"thought\": \"rm doesn't have any output when it deletes successfully, so that must have worked. Now that we have fixed the issue, we should submit our changes to the code base. Let's submit the changes using the `submit` command.\\n\",\n \"action\": \"submit\\n\",\n \"agent\": \"primary\"\n }\n ],\n \"info\": {\n \"exit_status\": \"submitted\",\n \"submission\": \"\\ndiff --git a/src/marshmallow/fields.py b/src/marshmallow/fields.py\\nindex ad388c7..20da768 100644\\n--- a/src/marshmallow/fields.py\\n+++ b/src/marshmallow/fields.py\\n@@ -1472,7 +1472,7 @@ class TimeDelta(Field):\\n if value is None:\\n return None\\n base_unit = dt.timedelta(**{self.precision: 1})\\n- return int(value.total_seconds() / base_unit.total_seconds())\\n+ return int(round(value.total_seconds() / base_unit.total_seconds())) # round to nearest int\\n \\n def _deserialize(self, value, attr, data, **kwargs):\\n try:\\n\",\n \"model_stats\": {\n \"total_cost\": 0,\n \"instance_cost\": 0,\n \"tokens_sent\": 0,\n \"tokens_received\": 0,\n \"api_calls\": 0\n }\n }\n}\n</code></pre> <p>At a high level, the experiments folder is organized in the following manner: <pre><code>trajectories\n\u251c\u2500\u2500 <user 1> \ud83d\udc69\u200d\ud83d\udcbb\n\u2502 \u251c\u2500\u2500 <experiment 1> \ud83e\uddea\n\u2502 \u2502 \u251c\u2500\u2500 all_preds.jsonl\n\u2502 \u2502 \u251c\u2500\u2500 args.yaml\n\u2502 \u2502 \u251c\u2500\u2500 *.html (Webpage Files)\n\u2502 \u2502 \u2514\u2500\u2500 *.traj (Trajectories)\n\u2502 \u2514\u2500\u2500 <experiment 2> \ud83e\uddea\n\u2502 \u251c\u2500\u2500 all_preds.jsonl\n\u2502 \u251c\u2500\u2500 args.yaml\n\u2502 \u251c\u2500\u2500 *.html (Webpage Files)\n\u2502 \u2514\u2500\u2500 *.traj (Trajectories)\n\u251c\u2500\u2500 <user 2> \ud83d\udc68\u200d\ud83d\udcbb\n\u2502 \u251c\u2500\u2500 <experiment 1> \ud83e\uddea\n\u2502 \u2502 \u2514\u2500\u2500 ...\n\u2502 \u2514\u2500\u2500 <experiment 2> \ud83e\uddea\n\u2502 \u2514\u2500\u2500 ...\n...\n</code></pre> Where every experiment follows the pattern <code>trajectories/<user name>/<experiment name></code>. The <code><user name></code> is automatically inferred from your system, and the <code>experiment name</code> is inferred from the arguments of the <code>run.py</code>.</p> <p>Viewing trajectories</p> <p>We provide a trajectory viewer for an easy viewing of trajectories.</p>"},{"location":"usage/trajectories/#how-an-experiment-folder-is-generated","title":"How an Experiment Folder is Generated","text":"<p>Each call to <code>run.py</code> produces a single <code>trajectories/<user name>/<experiment name></code> folder containing the following assets:</p> <ul> <li><code>all_preds.jsonl</code>: A single file containing all of the predictions generated for the experiment (1 prediction per task instance), where each line is formatted as: <pre><code>{\n \"instance_id\": \"<Unique task instance ID>\",\n \"model_patch\": \"<.patch file content string>\",\n \"model_name_or_path\": \"<Model name here (Inferred from experiment configs)>\",\n}\n</code></pre></li> <li><code>args.yaml</code>: A summary of the configurations for the experiment run.</li> <li><code><instance_id>.traj</code>: A <code>.json</code> formatted file containing the (thought, action, observation) turns generated by SWE-agent towards solving <code><instance_id></code>.</li> <li><code><instance_id>.html</code>: An <code>.html</code> single webpage render of the trajectory, which can be directly opened in the browser for easier viewing of the trajectory.</li> </ul> <p>Tip</p> <ul> <li>Evaluation is not completed by <code>run.py</code>, it is a separate step (see benchmarking)</li> <li><code>all_preds.jsonl</code> can be referenced directly into SWE-bench to run evaluation (see benchmarking)</li> <li>Trajectories can be turned into custom demonstrations for SWE-agent (more information).</li> </ul> <ul> <li> <p> Something broken? Report bug</p> </li> <li> <p> Something unclear? Ask question</p> </li> </ul>"},{"location":"usage/usage_faq/","title":"Usage FAQ","text":""},{"location":"usage/usage_faq/#what-models-are-supported","title":"What models are supported?","text":"<p>Model support</p> <p>Note that SWE-agent is currently unlikely to perform well with small or local models.</p> <p>Models are configured in <code>models.py</code> (we're working on giving a complete list of model settings).</p> <p>Here are some few examples:</p> <pre><code>gpt4\ngpt4o\ngpt4-turbo\nclaude-2\nclaude-opus\nclaude-sonnet\nclaude-haiku\nclaude-sonnet-3.5\ndeepseek-coder\nazure:gpt4\n</code></pre>"},{"location":"usage/usage_faq/#ollama-support","title":"Ollama support","text":"<p>Models served with an ollama server can be used by specifying <code>--model</code> with <code>ollama:model_name</code> and <code>--host_url</code> to point to the url used to serve ollama (<code>http://localhost:11434</code> by default). See more details about using ollama here.</p> <pre><code>python run.py --model_name ollama:deepseek-coder:6.7b-instruct \\\n --host_url http://localhost:11434 \\\n --data_path https://github.com/pvlib/pvlib-python/issues/1603 \\\n --config_file config/default_from_url.yaml\n</code></pre>"},{"location":"usage/usage_faq/#models-for-testing","title":"Models for testing","text":"<p>We also provide models for testing SWE-agent without spending any credits</p> <ul> <li><code>HumanModel</code> and <code>HumandThoughtModel</code> will prompt for input from the user that stands in for the output of the LM. This can be used to create new demonstrations.</li> <li><code>ReplayModel</code> takes a trajectory as input and \"replays it\"</li> <li><code>InstantEmptySubmitTestModel</code> will create an empty <code>reproduce.py</code> and then submit</li> </ul>"},{"location":"usage/usage_faq/#debugging","title":"Debugging","text":"<ul> <li>If you get <code>Error code: 404</code>, please check your configured keys, in particular whether you set <code>OPENAI_API_BASE_URL</code> correctly (if you're not using it, the line should be deleted or commented out). Also see this issue for reference.</li> </ul> <ul> <li> <p> Something broken? Report bug</p> </li> <li> <p> Something unclear? Ask question</p> </li> </ul>"},{"location":"usage/web_ui/","title":"Using the web interface","text":"<p>Our graphical web interface is optimized for using SWE-agent as a developer tool, fixing single GitHub issues or working in local repositories. However, it is still missing some of the options of the command line interface.</p>"},{"location":"usage/web_ui/#quickstart","title":"Quickstart","text":"<p>To start our web UI, simply run</p> <pre><code>./start_web_ui.sh\n</code></pre> <p>from the root of the repository.</p> <p>Opening the webpage</p> <p>If the user interface doesn't automatically open in your browser, please open it at <code>http://localhost:3000</code>. Running from GitHub codespaces? More tips here.</p> <p>Running from Docker</p> <p>If you run SWE-agent from the <code>docker-run</code> Docker container, please see here for how to start the web server.</p>"},{"location":"usage/web_ui/#if-something-doesnt-work","title":"If something doesn't work","text":"<p>Please make sure that your port 8000 and 3000 are unoccupied before running the above script.</p> <p>Run</p> <pre><code>lsof -i :8000\nlsof -i :3000\n</code></pre> <p>to identify other programs serving to that port and kill them, then try again.</p> <p>If <code>./start_web_ui.sh</code> is running, but you see a warning message about the backend not being connected, either run</p> <pre><code># this should show some python processes\nlsof -i :8000\n</code></pre> <p>or head to <code>localhost:8000</code> (you should see a small dummy page).</p> <p>It might also make sense to start the backend frontend manually as explained in the next section.</p>"},{"location":"usage/web_ui/#manually-starting-frontend-and-backend","title":"Manually starting frontend and backend","text":"<p>The web UI consists of a frontend written in react (showing the pretty control elements) and a backend written with flask. The <code>./start_web_ui.sh</code> starts both of them in the background. However, this might not be best for development and debugging. This section explains how to start both parts separately.</p> <p>First, let's start the backend:</p> <pre><code>python sweagent/api/server.py\n</code></pre> <p>You should see output similar to the following:</p> <pre><code> * Serving Flask app 'server'\n * Debug mode: on\n2024-05-23 11:30:45,436 - werkzeug - INFO - WARNING: This is a development server. Do not use it in a production deployment. Use a production WSGI server instead.\n * Running on http://127.0.0.1:8000\n2024-05-23 11:30:45,437 - werkzeug - INFO - Press CTRL+C to quit\n2024-05-23 11:30:45,437 - werkzeug - INFO - * Restarting with watchdog (fsevents)\n2024-05-23 11:30:46,484 - werkzeug - WARNING - * Debugger is active!\n2024-05-23 11:30:46,492 - werkzeug - INFO - * Debugger PIN: 123-594-933\n</code></pre> <p>Port availability</p> <p>If see an error about port 8000 not being available, please first close any application that occupies it. The frontend currently expects the <code>flask</code> server on port 8000, so choosing a different port won't work.</p> <p>Now, open a new terminal tab and navigate to the <code>frontend</code> directory:</p> <pre><code>cd sweagent/frontend\n</code></pre> <p>First, let's install the react dependencies:</p> <pre><code>npm install\n</code></pre> <p>And start the server:</p> <pre><code>npm start\n</code></pre> <p>This should also open the corresponding page in your browser. If not, check with the tips above. The default port that is being served is port 3000.</p> <p>Possible errors</p> <p>If you see errors</p> <pre><code>Proxy error: Could not proxy request /socket.io/?EIO=4&transport=polling&t=O-c5kv9 from localhost:3000 to http://localhost:8000.\nSee https://nodejs.org/api/errors.html#errors_common_system_errors for more information (ECONNREFUSED).\n</code></pre> <p>something went wrong with the backend part.</p> <ul> <li> <p> Something broken? Report bug</p> </li> <li> <p> Something unclear? Ask question</p> </li> </ul>"}]}
|
