Security-audit-skill
lastmod 2026-11-13created 2026-11-131 edits
Cloudflare's security audit skill that turns a coding agent into a security auditor
About
- A public release of the skill that served as the starting point for Cloudflare's vulnerability detection harness. Install it in a coding agent such as Claude Code and ask "do a security audit of this code", and it runs automatically
- Runs a six-stage audit with isolated agents
- Reconnaissance: Records the architecture, trust boundaries, and entry points in
architecture.mdandcoverage-ledger.json - Coverage-driven search: Assigns hunter agents per ledger unit, and a coverage critic looks for gaps
- Candidate validation: A fresh validation agent tries to refute each candidate
- Structured output: Writes confirmed / needs_validation / rejected records to
findings.jsonand validates them against a schema - Independent record verification: A fresh agent rechecks the final source evidence
- Reporting: Generates
REPORT.md,FINDINGS-DETAIL.md, andNEEDS-VALIDATION.md
- Reconnaissance: Records the architecture, trust boundaries, and entry points in
npx skills addhttps://github.com/cloudflare/security-audit-skill--skill security-audit (add --global for a per-user install)