Cloudflare open-sources security-audit skill: multi-agent repo security audits, 71.5k stars
bibryam · x · 2026-11-13
Cloudflare released security-audit-skill, an open-source (MIT) coding-agent skill that turns your agent into a security auditor — already at 71.5k GitHub stars.
The audit runs in six structured phases:
- Reconnaissance: maps architecture, trust boundaries, and input surfaces
- Parallel reviewer subagents hunt for broken access controls, unsafe input handling, and cross-tenant data leaks
- Coverage critics flag unreviewed gaps and trigger revisits
- Fresh agents independently verify findings to cut false positives
- Output is a machine-readable coverage ledger plus a findings report
It's the single-repo starting point that grew into Cloudflare's fleet-wide vulnerability discovery harness — pitched as "a whole security review team in a single skill" for auditing AI-generated code.
More from coding & agent
- How do you give agents a real inbox? Sending is easy, receiving replies is not — seckin-gueler · 2026-11-13
- Malini: open-source macOS app shows before/after screen recordings of Claude Code UI edits — Szymeo · 2026-11-13
- Gemini 4 Hydrogen builds 400K-voxel Chinese mountain temple in Blender, zero manual input — iamfakhrealam · 2026-11-13
- Lorebase: git-style branch/merge for agent knowledge bases, no more silent overwrites — aantenore · 2026-11-13
- Open-Source Bot Fills Arrival Cards for 14 Countries, China Card Done in 20 Minutes — juntao · 2026-11-13
- Dev builds deskcrew with Claude Code: coding agents share your desktop, you keep the mouse — acus9345 · 2026-11-13