Mini-SWE-Agent: A 100-Line AI Coding Agent Trending on GitHub
Mini-swe-agent is a newly released, radically simple AI coding agent written in Python in just around 100 lines of code. It requires no massive configurations or giant monorepos while capable of solving real software engineering issues. This project represents a major shift toward lightweight, transparent AI agent architectures that strip away unnecessary complexity. By achieving high performance with minimal code, it makes autonomous software engineering tools much more accessible to developers. Despite its extreme conciseness, mini-swe-agent achieves over 74% accuracy on SWE-bench verified. It is designed to solve GitHub issues or assist users directly from the command line.
## BACKGROUND
SWE-bench is a widely recognized evaluation benchmark designed to test the ability of generative AI models and agents to resolve real-world software engineering issues in code repositories. Agent harnesses like mini-swe-agent act as wrappers or execution environments that connect language models to software development tools, allowing them to interact with codebases, run tests, and fix bugs autonomously.