Cloudflare open-sources security-audit-skill: parallel agents run six-phase vulnerability audits
tom_doerr · x · 2026-08-27
Cloudflare has open-sourced security-audit-skill (3.1k stars on GitHub), a coding-agent skill that turns your agent into a security auditor.
- Six-phase pipeline: Recon (parallel agents map architecture, trust boundaries and input surfaces) → Hunt (parallel agents attack from angles like injection, access control, business logic, crypto, feature abuse and chained attacks, spawning sub-agents) → Validation → Reporting → structured output → independent verification.
- Machine-readable: findings are structured and independently verified, focused on exploitable vulnerabilities with real impact.
- Origin: it's the single-repo starting point that seeded Cloudflare's fleet-wide vulnerability discovery harness, described in their "Build your own vulnerability harness" post.
More from coding & agent
- Your Agent Doesn't Need to Walk the Graph: Architecture Misconceptions — growth_man · 2026-08-27
- Speculative Programmatic Tool Calling (sPTC) explained — CShorten30 · 2026-08-27
- Yacine Proposes 'Burnout Bench': Let LLM Agents Run Free Until They Break — yacinelearning · 2026-08-27
- AGENTS.md Guidelines for Optimizing Agent Output — dbreunig · 2026-08-27
- DIY Mobile ComfyUI Dashboard with Ollama Prompt Generation — Z0mboyy · 2026-08-27
- Claude's sandbox test runs rm -rf, wipes developer's entire machine — PMinervini · 2026-08-27