raptor

Raptor is a Python-based autonomous security research framework that runs on top of Claude Code, using a CLAUDE.md configuration file to define sub-agents, rules, and skills that orient the agent toward adversarial reasoning. The framework chains together static analysis tools (Semgrep and CodeQL), binary fuzzing via AFL++, software composition analysis, and LLM-driven validation into slash commands such as /agentic, /scan, /validate, /fuzz, /crash-analysis, and /exploit. The /agentic command runs the full pipeline autonomously: scanning a codebase or binary, deduplicating findings, validating exploitability through a multi-stage process that filters pattern-matching noise from real vulnerabilities, generating proof-of-concept exploit code, and writing secure patches. A named /project workspace system keeps findings organized across runs. The devcontainer image, which bundles all tooling including the rr deterministic debugger, weighs around 6 GB and requires --privileged mode. Security researchers, penetration testers, and vulnerability analysts working on C/C++ codebases or open-source repositories are the primary audience.

Lo que la gente pregunta sobre raptor