Auto-code-rover is an automated solution for resolving GitHub issues (bug fixes and feature additions). It combines language models with analysis and debugging capabilities to pinpoint the fix location and generate patches. Currently, it can solve 16% of SWE-bench problems and 22% of SWE-bench lite problems, surpassing the performance of existing AI software engineers. It works in two stages: first, it retrieves relevant context information through a code search API, and then, based on the obtained context, it attempts to write a patch program. It has two unique characteristics: first, the code search API has project structure awareness, it's not a simple string match, but a search for related code contexts in the abstract syntax tree; second, if a test suite is available, it can leverage statistical fault localization to improve the repair rate.