Security Audit

Before you start: read this whole install prompt, tell me in a few lines what this skill does and what installing it involves, and wait for my go-ahead before you do anything else. Prerequisites — Security subject, stated plainly first: this is a defensive audit and review skill, judged on its content; it does not attack live systems. No account, no key and no payment are needed to install it, and nothing in the folder is run during installation — its two Node (.cjs) validators and their tests are copied as-is only. At use time: source inspection is read-only; target-controlled code may be run only inside an OS-enforced sandbox with no external network, an allowlisted empty environment and strict resource limits, otherwise the skill must not execute it; it never probes deployed endpoints, production identities or other people's data, and it creates audit output files only in full audit mode, in a new directory outside the target. Install "Security Audit" for me. It is Cloudflare's defensive security review skill: a source-first methodology that separates guidance mode (security questions, focused reviews, triage) from a full audit mode that runs only when explicitly requested and then follows six phases — reconnaissance and a deterministic coverage ledger, coverage-led hunting waves by delegated research/general agents, validation of every candidate by a fresh verifier, structured output (findings.json checked against report-schema.json by validate-findings.cjs, the coverage claim checked by validate-coverage-ledger.cjs), independent record verification, and a final report set (REPORT.md, FINDINGS-DETAIL.md, NEEDS-VALIDATION.md). A candidate becomes a confirmed finding only with a concrete crossed trust boundary, an affected principal or resource and bounded local evidence; severity is calibrated and needs-validation items get none. Its folder holds the main SKILL.md, 14 companion guides (attack classes, hunting, reconnaissance, validation/reporting and domain guides for web/auth, AI/LLM, cloud, client-side, supply chain and more), the report schema and the two validators with their test files. Repository: https://github.com/cloudflare/security-audit-skill (the skill is the whole skills/security-audit/ folder at the pinned commit: 20 files in total) Pinned version: install the code at commit c1c8a8c1471069fb0e188eeaff69b8e8db6564a8 exactly. Do not substitute a newer version. If this commit is unreachable, stop and tell me. 1. Clone https://github.com/cloudflare/security-audit-skill into a temporary folder, check out commit c1c8a8c1471069fb0e188eeaff69b8e8db6564a8, and copy the skills/security-audit/ folder from that checkout. Then compare every file in the folder with the fingerprints recorded for this listing on 2026-10-22: [FINGERPRINTS] - AI-AND-LLM.md: SHA-256 7335a95c43554eadefcf03ece639291bb0e038c398a179299e3887ffe06bd2ef - ATTACK-CLASSES.md: SHA-256 3ee4f00c6e8c9dcf6dd02dd5ac2a4ceb322e2085f3e9fbe79cf0c6330bc7ef83 - CLIENT-SIDE.md: SHA-256 613f6b98f86083e045525c5b60154ffdab122bac35f1e05eb17488e6e453370c - CLOUD-AND-DEPLOYMENT.md: SHA-256 3c80f3088be22345e07b057c056d6ed99fe2b0304f794398964f7ccd5a496ca3 - DATA-ISOLATION-AND-LIFECYCLE.md: SHA-256 307dcb3e24d292715c4a041190e7bb135646a37eb6187619ae01d000a634a4ad - DESKTOP-MOBILE-AND-LOCAL-IPC.md: SHA-256 58eae8ee0ca6a46611c501ec5a3ab47fdff88dd7db0d8d74789bbf8bb19d8cb2 - HUNTING.md: SHA-256 71c121decea1322c47c718f6e110f1a892980bb4266bd7ac55ef043a213ade09 - MEMORY-SAFETY-AND-BINARY.md: SHA-256 7c8b0dccfb35315cca2455509674d49632b2fe37e1f3aac16c1a496a7e9924a7 - PROTOCOLS-RPC-AND-MESSAGING.md: SHA-256 8b73816fa5b3697c9c40977f0c836b40f01d6e81929ff2720fe83ce089e9a619 - RECONNAISSANCE.md: SHA-256 02ca887aca84de8ae137f1891bd2f0a54f28d3fa6da92a27f54734382b3381a2 - RESOURCE-EXHAUSTION-AND-AVAILABILITY.md: SHA-256 4326fab5c92a0b4c420a70196254040105cd794026562130b59bf1a6db3e4baa - SKILL.md: SHA-256 5e3e96a1e438d8f35fef0a1e38f02d4f00e2bdf910401b7dfe00e6779c6dac85 - SUPPLY-CHAIN-AND-RELEASE.md: SHA-256 f372e919940e335468ba1ae09f99026c3986a345f7fe26f8068f132e199f52a5 - VALIDATION-AND-REPORTING.md: SHA-256 b892470cb89bb31f33403e050d4bdfbe2cdaef940ade1496fa65e32b2e129449 - WEB-PROTOCOL-AND-AUTH.md: SHA-256 4f3066dffa23c5635859187be9b96c32f2aff29a688186b91ff0c63e47d62b5c - report-schema.json: SHA-256 6575c9de4a62255699ea052ebf9a49ebfb373a32891bcd89af6f55cc84b35519 - validate-coverage-ledger.cjs: SHA-256 2eb280bd2e33c002916f86db9b29f9edfedb6454e5db973c766290465b868ac8 - validate-coverage-ledger.test.cjs: SHA-256 cfb9df26b03469a4279a70e7664a2a5434dd6f8c2c204e91f9b0aeb9dd71e648 - validate-findings.cjs: SHA-256 e85f232e36bfac0f866f244da50376692d6b45c690beadcb00608b0e6c8b5283 - validate-findings.test.cjs: SHA-256 b92001117a92c4161d43777a0ba2fa182a8a4d97f2304cc8c0cee5331497b73b If a file does not match, it changed since the listing was prepared: do not install it yet, tell me what you found and let me decide. 2. Read SKILL.md in full, and tell me in plain words: what the skill does when used, what it connects to, whether it reads or sends any personal information or credentials, and anything that seems unusual. Stop on any red flag and tell me before going further. Confirm these points from the files: - The skill folder holds 20 files: SKILL.md, 14 companion Markdown guides, report-schema.json, and 4 Node files (validate-findings.cjs, validate-coverage-ledger.cjs and their two .test.cjs test files). Nothing in the folder is run during installation; the validators and tests are copied as-is. - Nothing is sent anywhere during installation itself; the only download at install time is the skill folder from GitHub at the pinned commit, and no code of mine is read or executed. - At use time this is a defensive review workflow: source inspection is read-only, target-controlled code runs only inside an OS-enforced sandbox with no external network and dummy principals/fixtures, and the skill forbids probing deployed endpoints, external services, production identities or other users' data. It describes fixes; it does not modify target source. - File creation at use time happens only in full audit mode (explicit audit/pen-test/full-review request): run files such as run-metadata.json, findings.json and REPORT.md go to a new output directory outside the target, with per-agent write isolation. Guidance mode creates no output directory and no audit artifacts. - Do not start any audit or review, do not run any validator or test file, and do not use the skill in any way during this install: using it is a separate step that only starts when I ask for it later. 3. When I say "yes, install", install it as a skill: copy the whole security-audit folder into the agent's skills directory, in a folder named exactly "security-audit", keeping its 20 files in their original relative layout (SKILL.md at its root). 4. Verify with no network calls: frontmatter valid (name: security-audit), the 20 files in place with matching fingerprints. Do not run anything from the folder and do not use the skill to verify. 5. Report what was installed, where, and what I still need to do myself (for example: that using it in guidance mode needs nothing extra, that a full audit runs only on my explicit request and writes its artifacts to a new directory outside the target, that target code is executed only inside a no-network OS sandbox (otherwise not at all), and that no live system is ever probed). GitHub is optional: if I have a GitHub account or the gh CLI, you may use it; otherwise public access is fine. Never require it unless it's in the prerequisites above. Rules: don't touch anything outside the temp folder and the install target. If anything looks off, stop and ask me.