Overview
Research-grade autonomous program repair agent that combines LLMs with abstract syntax tree analysis and spectrum-based fault localization.
Details
AutoCodeRover is an open-source autonomous program improvement agent from a Singapore research group that targets bug fixing and feature additions in real GitHub repositories. It distinguishes itself by combining LLM reasoning with traditional software engineering techniques (abstract syntax tree navigation, code search, and spectrum-based fault localization from failing tests) to ground edits more reliably than pure prompt-driven agents. AutoCodeRover has reported strong results on SWE-bench and is used as a research baseline.