security-audit is a coding-agent skill that turns an agent into a security auditor with a method. It runs six phases: reconnaissance that maps the architecture, trust boundaries and input surfaces into a coverage ledger; coverage-led hunting by isolated sub-agents with critics looking for gaps; validation where a fresh verifier tries to disprove every candidate; structured output checked against a JSON schema; independent verification of the final source claims; and a target-neutral report. Findings land as confirmed, needs_validation or rejected, and a confirmed one has a complete source trace and an observed result.
It is the skill that seeded Cloudflare's own vulnerability discovery harness, the one described in its Build your own vulnerability harness post, and the repository is the single-repo starting point that system grew from. That pedigree is why 14 commits have drawn about 55,000 stars. The playbooks cover web and auth protocol bugs, client-side injection, supply chain and release, cloud and deployment, memory safety, RPC and messaging, data isolation, desktop and mobile IPC, and prompt injection in LLM-backed targets.
The design principles are the useful part: only confirm established boundary failures, never let the agent that found a bug be the one that checks it, severity needs impact, and a missing second layer of defence is a hardening note rather than a vulnerability. Cloudflare says a single run found roughly half of what repeated runs found, so it is built to be run more than once.
- Repository: github.com/cloudflare/security-audit-skill
- Licence: MIT (MIT License)
- Language: JavaScript. Stars: 55.3K. Forks: 3,607. Last push: Sep 14, 2026.
- Scan: safe, Nov 1, 2026, commit c1c8a8c
Who it is for
Developers who want a structured second look at a codebase before a release, security teams building an AI-assisted review, and open-source maintainers who cannot afford a pentest.
Getting started
1. Install the skill with the Skills CLI (add --global for a user-level install)
npx skills add https://github.com/cloudflare/security-audit-skill --skill security-audit2. Start your agent in the codebase and ask for the audit
security audit this codebase3. Or point it at a folder and an output directory
do a security review of ./src, output to ~/audits/my-projectIt needs an agent that can run parallel sub-agents and Node for the validators. Without an OS-enforced sandbox for building and running the target, leads that need execution stay as needs_validation instead of being run, which is the safe default.
Safety scan
We cloned cloudflare/security-audit-skill at commit c1c8a8c on Nov 1, 2026 and ran the checks described on the GitHub Tools page: credential patterns, decode-and-execute code, install-time scripts, committed binaries, risky CI workflows, every host the code talks to, known vulnerabilities in pinned dependencies, and project hygiene. A person read every hit. This is what we found.
- No secret hits and no pattern hits in 22 files and 5,403 lines: 3,041 lines of JavaScript (the two validators and their tests), 1,900 of Markdown and 462 of JSON. No installers, hooks or binaries.
- No workflows at all, so nothing runs in CI, and no lockfiles, so there were no dependencies to query. The validators are written without packages on purpose.
- The code reaches no network hosts. The audit it runs stays inside your agent, and the skill tells the agent to build or execute target code only inside a sandbox with networking disabled.
- A licence is present; there is no security policy, contributing guide or Dependabot. Last push 2026-09-14, 14 commits and four contributors: this is a published artefact that changes rarely, not a busy project, which for a skill is fine.
What the scanner counted
| Check | Result |
|---|---|
| Secrets | None found. |
| Suspicious code | None found. |
| Install-time code | None: nothing runs at install beyond the package manager itself. |
| Committed binaries | None. |
| CI workflows | No GitHub Actions workflows. |
| Network hosts | 0 distinct hosts referenced from source. No URLs to bare IP addresses. |
| Known vulnerabilities | No lockfile to check: dependencies are declared as ranges, so what gets installed is whatever is current on the day. |
| Project hygiene | Has licence file. Missing security policy, automated dependency updates, CodeQL, contributing guide. |
| OpenSSF Scorecard | Not scored: the project is not in Scorecard's weekly index. |
By the numbers
| Stars | 55.3K |
|---|---|
| Forks | 3,607 |
| Contributors | 4 |
| Commits | 14 |
| Open issues | 38 |
| Open pull requests | 70 |
| Releases | 0 |
| Latest release | none tagged |
| Licence | MIT |
| Main language | JavaScript |
| Project age | 4 months |
| Last push | Sep 14, 2026 |
| Tracked files | 22 |
| Lines of code | 5,403 |
| Checkout size | 323 KB |
Lines by language: JavaScript 3,041, Markdown 1,900, JSON 462.
Questions
Is the security audit skill free?
Yes, MIT licensed. The cost is your agent's tokens: a full audit spawns many sub-agents across six phases, so expect a long, expensive run on a big codebase. Guidance mode, for focused questions, is cheaper.
Which coding agents can run it?
Any agent with tool use and parallel sub-agents that reads Agent Skills, such as Claude Code, Codex or Cursor. The Skills CLI handles installation and agent selection.
Does it replace a penetration test?
No. It finds code-level issues with evidence and marks what it could not prove. Cloudflare's own numbers say one run finds about half of what repeated runs find. Use it before a release and alongside human review, not instead of either.
This post is part of GitHub Tools, where every repository is cloned and scanned before it is written up. The scan is a snapshot of one commit on one day; the repository has moved on since, so check it before you install.