TraceMend

TraceMend is an evidence-backed debugging and verification system for AI-generated code. It helps developers inspect failing code, request a bounded AI-generated repair, compare the before-and-after code, and independently verify the result inside a constrained Docker sandbox. The current MVP focuses on a real Python debugging workflow. A developer can open an investigation, inspect the original failing implementation, review the AI-generated correction, and then examine the verification evidence, including passing tests, regressions, and execution details. Instead of asking users to blindly trust an LLM patch, TraceMend helps them decide whether the result is actually trustworthy. The problem it solves is simple but important: AI coding agents can generate fixes quickly, but developers still need reliable evidence that those fixes work.