SWE-agent

Princeton's autonomous GitHub issue solver, SWE-bench SOTA

SWE-agent from Princeton NLP achieves state-of-the-art results on SWE-bench, the benchmark for AI software engineering. It uses a custom Agent-Computer Interface (ACI) to interact with codebases through a purpose-built shell environment. SWE-agent autonomously reads files, edits code, runs tests, and submits pull requests to fix real GitHub issues.