AI RESEARCH
Understanding Automated Program Repair Agents Through the Lens of Traceability: An Empirical Study
arXiv CS.AI
•
ArXi:2506.08311v2 Announce Type: replace-cross Automated Program Repair (APR) agents leverage Large Language Models (LLMs) to autonomously diagnose and fix software bugs through reasoning, planning, and tool use. Despite impressive leaderboard gains on benchmarks such as SWE-bench, little is understood about how these agents take actions, where they fail, and how their behavior compares to that of human developers.