RAPTOR
A security research framework on Claude Code: static analysis, LLM validation of findings, exploits and patches
Install
git clone https://github.com/gadievron/raptor.git
cd raptor
pip install -r requirements.txt
pip install semgrep
export PATH="$PATH:$PWD/bin"
raptorThe raptor launcher runs pre-flight checks and loads the coverage plugin.
This is third-party code. Review the repository files before installing.
What it does
RAPTOR turns Claude Code into an agent for finding and validating vulnerabilities. /scan runs Semgrep, plus Coccinelle and CodeQL when available, deduplicates and writes a SARIF report. /agentic adds multi-stage LLM validation of each finding, exploitability assessment, exploit and patch generation with a per-run cost cap. Skills cover crash analysis with rr, gcov and function tracing, Frida, and open source forensics from GitHub archives. Untrusted code runs in a sandbox using Linux namespaces, Landlock and seccomp.
Who it is for. For application security engineers and vulnerability researchers auditing their own or authorized code.
Good fit when
- You need static analysis with LLM triage of false positives
- You need to assess whether a finding is exploitable and draft a patch
- You need to analyze a binary crash or investigate GitHub history
Not a fit when
- You lack authorization to analyze the target code or system
- You need a polished product rather than a research tool
Example request
/agentic /path/to/codeLimitations
The authors describe it as unpolished. Requires Claude Code with a subscription or Anthropic API key, Python 3.10+, Node.js 18+ and Semgrep; CodeQL is optional and forbids commercial use. The container image is about 6 GB. The full pipeline on a mid-size project takes tens of minutes and consumes tokens; other models need their own keys.
How to disable. Run plain claude outside the RAPTOR directory, remove bin from PATH or delete the container.
Security check
- Executes model-generated code and exploits
- Processes untrusted repositories and binaries
README in short
The README presents RAPTOR as an autonomous offensive and defensive research framework built on Claude Code by well-known security researchers. It covers prerequisites, manual install with the raptor launcher, and running in Docker or a devcontainer. It explains what /scan and /agentic do and their time and cost. A separate section covers the security model and sandbox.
FAQ
How much does a full run cost?
With Claude Code as the analysis model there is no extra cost beyond the subscription. External models bill per token, with a default per-run cost cap.
Why run it in a container?
The agent executes generated code and processes untrusted repos; a container limits filesystem access and supply chain blast radius.
Related
NVIDIA's open stack for running OpenClaw, Hermes and LangChain Deep Agents in OpenShell sandboxes with network policy and managed inference
Static code analysis with rules that look like source code, plus a built-in MCP server for AI agents
Trail of Bits skills
Trail of Bits Skills Marketplace
Trail of Bits plugin marketplace for security work: smart contracts, C/C++ and Rust review, Semgrep, CodeQL and fuzzing
A cybersecurity skill library: forensics, incident response, pentesting, cloud and malware analysis mapped to MITRE ATT&CK and NIST