
Security Audit Skill
A coding-agent skill for multi-phase security audits with independently verified, machine-readable findings
- Category
- AI
- Website
- github.com
- Language
- English
- Listed
- 1d ago
- Stars
- 10,801
- Forks
- 574
cloudflare/security-audit-skill ↗, JavaScript, MIT, Last commit 4d ago
The skill drives a coding agent through six phases: mapping architecture, trust boundaries and input surfaces into a coverage ledger; assigning isolated hunters against units of that ledger; handing every unique candidate to a fresh verifier whose job is to disprove it; writing records into findings.json against a schema; having separate agents re-check the final source claims; and only then deriving the report files.
The three verdicts are kept apart on purpose. A confirmed finding carries a complete source trace and a bounded observed result, needs_validation names the exact unresolved fact and gets no severity, and rejected records a candidate that was disproved. Repeat runs on the same repository build on the earlier ledgers instead of starting over. Cloudflare describes this as the single-repo starting point its fleet-wide vulnerability harness grew out of, and running it needs Node.js plus an OS-enforced sandbox for anything the target controls.