GitHub Security Lab released an open-source Taskflow fuzzing agent that can point itself at a public C/C++ repository, select fuzzing entry points, write AFL++ harnesses, and run fuzzing campaigns without human supervision. The LLM-driven agent iteratively improves its harnesses using structure-aware mutators, triages and deduplicates crashes, and generates vulnerability reports with suggested fixes as unified diffs. The tool is built on GitHub’s Security Lab Taskflow Agent framework, defaults to Claude Sonnet 5, and includes a live dashboard for monitoring fuzzing campaigns.
