Security Audit: Cloudflare's agent-run code security review
cloudflare/security-audit-skill
Your AI coding assistant becomes a careful security reviewer for your code. Ask a security question and it answers with the relevant checks. Ask for a full audit and it runs a team of helper agents in six steps. A fresh agent tries to disprove each finding, and the report gives proof and the smallest fix.
Pack description by its author: “Your AI coding assistant becomes a careful security reviewer for your code. Ask a security question and it answers with the relevant checks. Ask for a full audit and it runs a team of helper agents in six steps. A fresh agent tries to disprove each finding, and the report gives proof and the smallest fix.”
DevOps, cloud and securityKnowledge and rules for your assistantGives your assistant new abilities with scripts
1. Changes your files
None of its skills edits your files.
github.com/cloudflare/security-audit-skill/blob/c1c8a8c/2. Runs code on your machine
Install: Markdown guides, a JSON schema and two Node.js checker scripts. No hooks, no install scripts. Full audit, after the coverage ledger is created or updated: The assistant runs validate-coverage-ledger.cjs on coverage-ledger.json. It only reads the file and the schema and prints PASS or errors; no network, no writes.
github.com/cloudflare/security-audit-skill/blob/c1c8a8c/3. Sends data
The skill reads your source and, in full audits, has the assistant run two read-only Node checker scripts. It forbids running your code without an OS-enforced sandbox, installing packages, and probing live systems. Its domain guides name attack classes such as prompt injection as things to look for; they are reference text, not instructions to your assistant. Install only from github.com/cloudflare/security-audit-skill.
github.com/cloudflare/security-audit-skill/blob/c1c8a8c/Checked against the pack's files at commit c1c8a8c, on Oct 30, 2026. Tap a verdict to see its source.How we checkedReport an error
Good for you if
- You want your own code checked for real security holes before you launch or share it
- You have a security question or a suspected problem and want a focused, source-based answer
- You use a coding assistant that can run several helper agents at once
Probably not for you if
- You want it to fix the code: the audit describes fixes and does not change your code
- You want your live site or servers tested: it reads source and runs only small local checks
- You want a quick, cheap check: a full audit launches many helper agents and costs accordingly
What you'll notice
- Answers by default; the full audit starts only when you ask to audit or pen-test the code or ask for a report
- A full audit writes its files to a folder in your home folder, not into your project
- Three kinds of result: confirmed (with severity), needs validation (no severity, the missing fact named), rejected
- Reports: REPORT.md, FINDINGS-DETAIL.md and NEEDS-VALIDATION.md
What you need
- Coding agents with tool use and parallel sub-agents (agent-neutral; installs through the Skills tool)
- Free, MIT license
- One command with the Skills tool, or copy the skill folder
- Node.js for the two checker scripts; a locked-down sandbox if the assistant is to run your code during checks
Words on this page
- Skill
- A set of written instructions your AI assistant follows. Installing one is like giving it a playbook.
- Subagent
- A helper the assistant starts for one part of a task.
| Repository | cloudflare/security-audit-skill |
|---|---|
| GitHub stars | 72,809 |
| Skills | 1 · 1 with scripts |
| Category | DevOps, cloud and security |
| License | MIT |
| Last updated | Sep 14, 2026 · card checked Oct 30, 2026 at commit c1c8a8c |
| Install counts | cloudflare/security-audit-skill on skills.sh, Vercel's skills directory, which shows installs and security audits per skill |
cloudflare/security-audit-skill popularity over time
What Security Audit does
- Two modes: guidance (answers, focused reviews, no files) and full audit (all six phases, report files)
- 1 Reconnaissance: maps the architecture, trust boundaries and inputs, and builds a coverage ledger of what to check
- 2 Hunting: separate helper agents each take ledger units; coverage critics look for gaps
- 3 Validation: every candidate goes to a fresh agent that tries to disprove it
- 4 Structured output: confirmed, needs_validation and rejected records in findings.json, checked by a bundled script
- 5 Record verification: fresh agents re-check the final source claims
- 6 Reports written from the verified records only
- Run profiles: quick, standard (default) or deep; an optional budget caps the number of agent calls
- Domain guides for native code, AI and LLM apps, web and auth, browser code, supply chain, cloud, messaging, data isolation, desktop and mobile
Never cut, per the skill: a concrete boundary and result for every finding, independent verification of confirmed records, no live or shared-environment testing, no severity for unproven leads.
| Level | Behaviour |
|---|---|
| quick | One hunter wave, one final critic pass, one verifier per candidate |
| standard | The workflow as written (default) |
| deep | Finer units, critic waves until a clean pass, separate fresh agents for each verification |
From the pack's SKILL.md files, summarised by AgentGid.
The skill, explained
security-audit
Reviews your code for security holes and reports them with proof and the smallest fix
You give
- The code to check (start the assistant in that folder or point it there)
- Optional: a folder for the results, a profile (quick, standard, deep), the paths in scope, a budget of agent calls
You get
- architecture.md and coverage-ledger.json: the map and what was checked
- findings.json: confirmed, needs_validation and rejected records
- REPORT.md: scope, posture summary, table of confirmed findings by severity, needs-validation table, hardening notes, coverage summary
- FINDINGS-DETAIL.md: full trace and local reproduction for medium, high and critical findings
- NEEDS-VALIDATION.md: unresolved leads with the exact missing fact and a safe plan
See it in action
The skill defines what its answer looks like. Its own example:
1. Run profile, scope, budget, source ref, prior runs, deferred and out-of-scope coverage 2. One short security posture summary 3. Confirmed findings table: severity, title, affected boundary, observed result 4. Each confirmed finding: source location, attacker, local reproduction, result, impact, smallest fix 5. NEEDS VALIDATION table: title, trace, exact blocker, next local step (no severity) 6. Hardening notes and positive source patterns 7. Coverage summary from the ledger
When to reach for it
| You want | Use |
|---|---|
| A security question answered or one suspected issue checked | Ask it directly (guidance mode, no files written) |
| A full review of a codebase with a report | "security audit this codebase" (full audit, standard profile) |
| A fast first look or a re-run | Ask for the quick profile |
| A high-stakes or large target | Ask for the deep profile |
| To check only one folder or a change between two versions | Ask for a scoped run (named paths or a diff) |
| To cap cost | Give a budget: a maximum number of agent calls |
Skills in cloudflare/security-audit-skill
| Skill | In plain words | Changes files? | What it does (author) |
|---|---|---|---|
| security-audit | Reviews your code for security holes and reports them with proof and the smallest fix | No | Security guidance and vulnerability review for codebases, APIs, services, CLI tools, libraries, and daemons. Use for security questions, focused reviews, vulnerability research, security audits, or pen tests. Run the complete workflow only for explicit codebase audit or pen-test requests,… has scriptsnpx skills add cloudflare/security-audit-skill --skill security-audit |
The author's descriptions come from each skill's SKILL.md; the plain-words column is AgentGid's.
Good to know before you install
- It switches on by itself for security requests (security audit, find vulnerabilities, pen-test the code), but only answers until you clearly ask for a full audit or a report.
- A full audit writes outside your project by default, to ~/security-audit-skill/<repo-name>/run-<N>. It writes inside the project only to a folder you pick that version control ignores.
- Running your code needs a locked-down sandbox (no network, limited writes and resources). Without one the assistant does not run your code and marks the lead as needs validation.
- It never installs packages or tests live systems: no deployed endpoints, no shared services, no paid API quota.
- A full audit launches many helper agents; set a budget (maximum agent calls) to cap cost. The skill says when a budget cannot cover the plan.
- Runs add up: a new run reads earlier ledgers and findings to target gaps and re-check changed code.
Before you start
Works in
- Any coding agent with tool use and parallel sub-agents: skills CLI
You will need
- A coding agent with parallel sub-agents: The full audit delegates work to many helper agents
- Node.js: Runs the two findings and coverage-ledger checker scripts
- OS-enforced sandbox: Needed only to run your code during checks
Steps only you can do
- Set up an OS-enforced sandbox with no external network, an allowlisted environment, resource limits and writes only to scratch folders (only if the assistant should run your code during checks) (README.md (Requirements))
Then: The assistant runs minimal local checks inside it to confirm or reject leads
Mistakes to avoid
Ask a security question and expect report files
Ask explicitly for a full audit, a pen-test of the codebase or a report
The skill answers in guidance mode by default and writes no files until a full audit is requestedPoint the output folder inside the project that git tracks
Keep the default folder in your home directory, or pick a project folder that version control ignores
The run stops and asks for an outside path otherwiseExpect it to build and run your tests on a normal machine to prove a finding
Provide an OS-enforced sandbox (no external network, allowlisted environment, resource limits, scratch-only writes)
Without one it does not execute your code and leaves the lead as needs validationSet a very small agent budget for a full audit
Give a budget that covers reconnaissance, critics and verifiers, or narrow the scope or choose the quick profile
If the budget cannot fund the minimum, no agent is launchedTreat one run as complete coverage
Run it again on the same repo; later runs target gaps and re-check changed code
Runs are additive and reuse earlier ledgers and findings
Which one to pick
| If you need | Pick |
|---|---|
| You want a full multi-agent audit with independently verified, machine-readable findings and a report | This pack |
| You want a broad set of security research, vulnerability detection and audit skills | trailofbits/skills |
| Your app was built with an AI coding assistant and you want a check for the common holes such code has | raroque/vibe-security-skill |
| You want separate skills that look for vulnerabilities in web or mobile apps | utkusen/sast-skills |
Works well with
- semgrep: A rule-based static scanner for a fast first pass; this skill does the deeper, agent-run review
May I use it like this?
| Use | Answer | Condition |
|---|---|---|
| In a commercial app | Yes | MIT, Cloudflare, Inc. |
| Work for clients | Yes | MIT |
| Passing the files on | With conditions | MIT: keep the copyright and permission notice in copies |
Install Security Audit
Official commands from the repository
npx skills add https://github.com/cloudflare/security-audit-skill --skill security-audit
Add --global for a user-level install. Run npx skills --help for agent selection. Needs Node.js.
Or with the open-source skills CLI: npx skills add cloudflare/security-audit-skill. Skills are instructions your agent follows: read them before installing.
What it installs and runs on your machine
Read in the code at commit c1c8a8c, Oct 30, 2026
| When | What happens |
|---|---|
| Install | Markdown guides, a JSON schema and two Node.js checker scripts. No hooks, no install scripts. |
| Full audit, after the coverage ledger is created or updated | The assistant runs validate-coverage-ledger.cjs on coverage-ledger.json. It only reads the file and the schema and prints PASS or errors; no network, no writes. |
| Full audit, Phase 4 and after every Phase 5 replacement | The assistant runs validate-findings.cjs on findings.json against report-schema.json. Read-only, no network. |
| Full audit | The assistant writes run files (run-metadata.json, architecture.md, coverage-ledger.json, findings.json, the three reports) and per-agent scratch folders to the output folder, by default ~/security-audit-skill/<repo-name>/run-<N>. |
| Checks that run your code | Only inside an OS-enforced sandbox with no external network; otherwise not run. |
The author's numbers, as reported
AgentGid did not run these tests
- In the author's test runs, a single run found roughly half of the vulnerabilities that repeated runs found in total. Source
Quick answers
- Is it free?
- Yes. MIT license, Cloudflare, Inc.
- Will it change my code?
- No. It describes the smallest fix; it does not modify your source.
- Does it attack my live site?
- No. It reads source and, with a sandbox, runs only small local checks. It does not probe deployed endpoints or shared services.
- Where do the results go?
- By default to ~/security-audit-skill/<repo-name>/run-<N>, or another folder you name.
- Why is a lead marked needs validation?
- A fact the source cannot show (deployment settings, proxy, identity policy) decides it, or the sandbox was missing. It gets no severity.
- Is this what Cloudflare uses?
- The README says it seeded Cloudflare's vulnerability discovery harness and is the single-repo starting point that harness grew from.
Checked by AgentGid on Oct 30, 2026 against commit c1c8a8c of the pack's own files. agent.md for AI agents · Report a mistake
More devops, cloud and security skill packs
| # | Skill pack | Skills | GitHub stars | Updated |
|---|---|---|---|---|
| 1 |
ruvnet/RuView A guide for your AI assistant to the RuView project. RuView uses WiFi signals and cheap ESP32 boards to sense people in a room: p…
|
53 | 104K | 2026-11-15 |
| 2 |
mukul975/Anthropic-Cybersecurity-Skills A library of 818 security playbooks for your AI assistant: investigating alerts, analysing malware, checking cloud setups, huntin…
|
818 | 36.2K | 2026-08-31 |
| 3 |
vercel-labs/agent-skills Nine skills from Vercel for your AI coding assistant. Some are rule books for React, Next.js and React Native apps. Two review yo…
|
9 | 34.1K | 2026-11-12 |
| 4 |
NVIDIA/SkillSpector Before you add a new skill to your AI assistant, it checks whether the skill is safe. A scanner looks for known risky patterns, t…
|
30 | 26.1K | 2026-11-14 |
| 5 |
pascalorg/editor Two playbooks that let your AI assistant work in Pascal, a free 3D building editor. It can make and edit a room or a house model,…
|
5 | 25.5K | 2026-11-14 |
| 6 |
google/skills Google's own playbooks for AI assistants: 175 skills on Google Cloud, Google Ads, mobile ads, Analytics and Sign in with Google. …
|
198 | 24.3K | 2026-11-15 |