SWE-agent

Why we included this project

SWE-agent is a good fit for teams that would rather let a language model handle routine bug-fixing than assign it to a developer. Point it at a GitHub issue and your chosen model works in a terminal-like environment, editing files, searching the codebase, and running commands until it produces a patch. It has been a strong performer on the SWE-bench benchmark, and the whole setup is controlled by a single YAML file, so you can adjust the toolset and workflow without modifying the core code. The same loop also works for competitive coding and security exercises, which makes it a general autonomous coding harness rather than a one-purpose tool. One caveat: the maintainers now point new users to the simpler mini-SWE-agent, but this original repository is still the well-documented reference implementation.