Researchers have developed SWE-Doctor, a novel agent designed to improve the accuracy of LLM-based software engineering agents in generating code patches. Unlike previous methods that used bug reproduction tests (BRTs) directly, SWE-Doctor analyzes multi-faceted BRTs to derive runtime diagnoses. These diagnoses, combined with localization information, guide the patch generation process, significantly reducing the occurrence of partial patches. Evaluations on Python bug-fixing issues from SWE-bench Verified and SWE-bench Pro demonstrated SWE-Doctor's superior performance across various LLM backends, achieving resolution rates of 75.7% and 59.4% respectively. AI
IMPACT Enhances LLM capabilities in automated code repair, potentially leading to more efficient software development cycles.
RANK_REASON The cluster contains a research paper detailing a new method for LLM-based software engineering agents. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
