TraceCoder: A Trace-Driven Multi-Agent Framework for Automated Debugging of LLM-Generated Code

Jiangping Huang, Wenguang Ye, Weisong Sun +3

Large Language Models (LLMs) often generate code with subtle but critical bugs, especially for complex tasks. Existing automated repair methods typically rely on superficial pass/f…