A defensive security audit skill released by Cloudflare, applicable to codebases, APIs, services, command-line tools, libraries, and daemons. By default, it only provides guidance for security Q&A and targeted reviews; it runs the full six-stage process and generates audit files only when the user explicitly requests a complete audit, penetration test, or report. It requires an OS-level sandbox before executing target code: network isolation, empty environment variables, read-only target access, and limits on CPU, memory, and execution time. If these conditions cannot be met, it does not run the code and only provides a verification plan; it does not probe live services, shared infrastructure, or other people's data.
Why It Matters Only reports vulnerabilities with specific affected objects, accompanied by source code evidence and safe reproduction steps, to reduce false positives.
skills.sh installs 92590 (+1881 / 24h)
Security Audit: Risk warnings present
- Initial screening only: Scanned only SKILL.md, no full repository audit performed, installation command not provided for now
- Audit scope: GitHub cloudflare/security-audit-skill/skills/security-audit/SKILL.md (only scanned SKILL.md text)
Rationale
- skills.sh 24 hour installs 1881 (indicates attention level only)
- SKILL.md separates "guidance mode" from "full audit mode," avoiding automatic directory creation and file writing whenever a security question is asked
Usage
- Tell the agent "perform a complete security audit on this repository" to trigger the full workflow; when only asking specific security questions, it provides guidance only and does not create files
Risks
- The automated audit only scanned SKILL.md, no full repository audit was performed, installation command not displayed for now; the skill dispatches sub-agents and runs target code in a sandbox, which aligns with its security audit purpose
- No account required; full audits consume significant time and resources