Trail of Bits Skills
trailofbits/skillsTrail of Bits' marketplace of 97 security skills: smart-contract scanners, C/C++ and Rust review, Semgrep and CodeQL workflows, variant analysis and audit helpers.
Recommended
Broad access Depth 4/5
Cloudflare security-audit Claude skill
A single multi-agent skill from Cloudflare that audits a codebase in six phases and writes validated, independently verified findings.
Our verdict Recommended
One of the most carefully specified audit skills available, with verifiers, schemas and sandbox rules; budget for a large token spend on full runs.
What Cloudflare security-audit does
security-audit is one skill, but a large one: a 22 KB SKILL.md, fourteen companion markdown files and two zero-dependency Node validators with tests. It has two modes. In guidance mode it answers security questions or does focused reviews without creating files. In full audit mode, triggered only by an explicit request for a codebase audit, pen test or report, it runs six phases: reconnaissance (architecture notes plus a coverage ledger), coverage-led hunting by isolated agents, validation of each unique candidate by a fresh verifier that tries to disprove it, structured output into findings.json checked against a JSON schema, independent re-verification of source claims, and derived reports (REPORT.md, FINDINGS-DETAIL.md, NEEDS-VALIDATION.md). Findings are confirmed, needs_validation or rejected. Companion files hold attack classes for memory safety, web and auth protocols, client-side code, supply chain, cloud and deployment, RPC and messaging, resource exhaustion, data isolation, desktop and mobile IPC, and LLM-backed targets. Repeat runs add to earlier ledgers. Target code may only be executed inside an OS-enforced sandbox with no external network.
Good for
- Source-level security audits of a repo where you want evidence-backed, machine-readable findings
- Teams building their own agent-driven vulnerability review process
- Repeated audits that should build on earlier coverage ledgers
Not for
- Quick lint-style checks or single-file reviews (guidance mode is lighter)
- Live penetration testing of deployed systems; the skill forbids probing production endpoints
What's inside depth 5/5
22 KB SKILL.md, 14 companion method files, a JSON schema, two validators with unit tests and a phase-by-phase artifact contract, though no skill-level evals.
- Skills
- 1
- Reference docs
- 14
- Scripts
- 4
- Tests / evals
- 2
- Templates
- 0
- Avg SKILL.md
- 21.5 KB
Counted from the repository file tree. How we score depth.
Dependencies and costs
- Node.js (the validators are .cjs scripts)
- An OS sandbox with no external network if the agent is to run target code
- An agent that supports sub-agents
Security review
Review before installingCloudflare's source-first security audit skill: Markdown methodology plus two local Node validators for the findings and coverage-ledger JSON. The skill defaults to guidance mode and only writes audit files in an explicit full-audit run; it tells the agent to run target-controlled code only in a network-isolated sandbox with dummy secrets. No network calls, credentials handling or telemetry were found in the scripts.
What we checked
The repository (22 files) on Oct 29, 2026. Not a full review: we read 2 of 4 code files; it relies on code outside the repository: skills/security-audit/validate-findings.test.cjs and validate-coverage-ledger.test.cjs (grepped: they spawn `node` on the validators with temp files and `mkfifo`; not read in full).
NVIDIA SkillSpector static scan: 13 pattern hits (Privilege Escalation 6, MCP Rug Pull 3, Excessive Agency 2, Memory Poisoning 1, Tool Misuse 1). SkillSpector hits are phrases in the audit methodology text (credentials.json, keychain, world-writable, 'delete memory', '/etc/passwd' in a validator test) and an 'npx skills' install mention; none are instructions to harm the user's system. SKILL.md requires target-controlled code to run only inside an OS sandbox without external network.
A review, not a certification. Skills change after we check them — read the files before installing.
License
MIT — commercial use allowed.
Can it run unattended?
Yes. Full mode is a resumable, phase-gated pipeline with a coverage ledger, independent verifiers and schema validation, and repeated runs build on earlier results; expect a long, expensive run.
Alternatives to Cloudflare security-audit
Snyk Agent Scan
snyk/agent-scanSnyk's CLI scanner that discovers installed agents, MCP servers and skills on your machine and checks them for prompt injection and other risks.
With caveats
Check first Depth 3/5
Cisco Skill Scanner
cisco-ai-defense/skill-scannerCisco AI Defense's open-source scanner for agent skills: YARA and YAML rules, dataflow analysis and an optional LLM judge, with SARIF output and a pre-commit hook.
Recommended
Check first Depth 4/5
SkillSpector
NVIDIA/SkillSpectorNVIDIA's open-source security scanner for agent skills, with 71 detection patterns, optional LLM review and a skill-inspector skill.
Recommended
Check first Depth 4/5
FAQ
What is Cloudflare's security-audit skill?
An agent skill that orchestrates a multi-phase security audit of a codebase: reconnaissance, hunting, validation by fresh verifiers, structured findings, independent verification and reporting. Cloudflare says it seeded their vulnerability discovery harness.
How do I install security-audit?
Run npx skills add https://github.com/cloudflare/security-audit-skill --skill security-audit, adding --global for a user-level install. Then ask your agent to security audit the codebase.
Is security-audit safe to run on my code?
The skill is guidance by default and only runs the full workflow when you explicitly ask. It requires an OS-enforced, no-network sandbox before executing target-controlled code and forbids touching deployed systems.
security-audit vs Trail of Bits skills?
Cloudflare's skill is a single end-to-end audit pipeline with verification and schemas. Trail of Bits publishes a collection of separate skills for vulnerability detection and audit workflows that you pick from.
Sources: GitHub repository (stars, license, file tree, last push Sep 14, 2026); install counts from skills.sh via Best Skills (CC BY 4.0). Review based on the README and SKILL.md files as published.