Definition
MIT-licensed agent-skills package from cloudflare that orchestrates coding agents through a six-phase security audit: reconnaissance, coverage-led hunting, validation, structured JSON output, independent verification, and reporting.
Key Points
- Evolved from Cloudflare’s internal vulnerability discovery harness (2026-09-17-cloudflare-security-audit-skill-clauday)
- Install:
npx skills add https://github.com/cloudflare/security-audit-skill --skill security-audit - Findings schema: confirmed / needs_validation / rejected with independent agent re-verification
- Attack modules: memory safety, LLM, web, client-side, supply chain, cloud, RPC, resource exhaustion, data isolation, desktop/mobile IPC
- Requires OS-enforced sandbox for target-controlled builds when executing exploit validation
Related
- agent-skills
- cloudflare
- llm-security
- autonomous-penetration-testing
- ai-red-teaming
- supply-chain-security
- coding-agents