gadievron/raptor

Raptor turns Claude Code into a general-purpose AI offensive/defensive security agent. By using Claude.md and creating rules, sub-agents, and skills, and orchestrating security tool usage, we configure the agent for adversarial thinking, and perform research or attack/defense operations.

What it solves

RAPTOR is an autonomous security research framework designed to automate the discovery, validation, and patching of vulnerabilities in software. It replaces manual, fragmented security audits with a unified pipeline that combines static analysis, binary reverse engineering, and LLM-driven reasoning to find real bugs rather than just pattern-matching noise.

How it works

RAPTOR orchestrates a multi-stage pipeline. It begins by mapping the attack surface and trust boundaries of a target. It then uses static analysis tools like Semgrep and CodeQL to identify potential vulnerabilities. These findings are passed through a multi-stage validation pipeline (Stages A-F) where an LLM analyzes reachability, feasibility, and exploitability. For compiled binaries, it integrates with tools like Ghidra, Frida, and radare2 to build evidence-backed maps and perform root-cause analysis on crashes. The framework also includes a Z3 SMT solver to mathematically prove if certain code paths are reachable before involving an LLM.

Who it’s for

It is built for security researchers, penetration testers, and developers who need to perform deep, autonomous vulnerability research on source code or compiled binaries.

Highlights

  • Autonomous Pipeline: Chains scanning, validation, exploit generation, and patch writing into one workflow.
  • Binary Analysis: Includes black-box investigation, Ghidra integration, and autonomous crash analysis for C/C++.
  • Multi-Stage Validation: Uses a rigorous A-F stage process to filter out false positives and verify exploitability.
  • SMT Integration: Employs Z3 to pre-screen CodeQL paths and analyze binary gadget constraints.
  • Software Composition Analysis (SCA): Normalizes dependencies across manifests and lockfiles to identify supply-chain risks.
  • Sandboxed Execution: Runs LLM-generated code in isolated environments using Linux namespaces and seccomp to prevent system compromise.

Related

  • Project
    anthropics/defending-code-reference-harness

    A reference implementation for autonomous vulnerability discovery and remediation that uses Claude to find, verify, and patch security bugs in source code.

  • Project
    visa/visa-vulnerability-agentic-harness

    An agentic application security pipeline that automates the discovery, adversarial verification, and remediation of source code vulnerabilities to reduce the time from disclosure to validated fix.

  • Project
    cloudflare/security-audit-skill

    A coding-agent skill that implements a structured, six-phase security audit workflow to systematically discover and verify vulnerabilities in a codebase.

  • Project
    knostic/OpenAnt

    OpenAnt (knostic/OpenAnt) is a genuine open-source software project that applies LLMs to security: it scans code repositories to find and verify vulnerabilities, and can generate and vet trusted patches. It's a practical, system-level tool with a CLI, configurable LLM adapters, and a web UI — not a list or tutorial.

  • Project
    adshao/flounder

    An autonomous white-hat security auditor that uses LLMs to map attack surfaces, construct exploits, and verify vulnerabilities through sandboxed execution.