SWE-agent
An agent framework that lets a language model of choice autonomously use tools to fix issues in real GitHub repositories, solve offensive cybersecurity challenges, or complete other custom coding tasks.
SWE-agent gives a language model, such as GPT-4o or Claude, the tools to autonomously fix issues in real GitHub repositories, find cybersecurity vulnerabilities, or complete other custom tasks. Its behavior is governed by a single YAML configuration file, and it is built for research: simple and hackable by design.
A companion mode, SWE-agent: EnIGMA, adapts the agent for offensive cybersecurity, solving capture-the-flag style challenges on multiple cybersecurity benchmarks.
The project's current development effort has moved to mini-SWE-agent, a simpler successor that matches SWE-agent's results in about 100 lines of Python; the maintainers recommend it over SWE-agent going forward. SWE-agent was built by researchers at Princeton University and Stanford University and is written in Python.