| name | security-review |
| description | Use for auth, authorization, tenant isolation, public/private payloads, secrets, supply-chain, source safety, and security review. Do not run production-impacting scans, read secrets unnecessarily, weaken controls, auto-configure tools, or mark metadata-only scanners as approved. |
Security Review
Use this as the canonical final security review skill for application and toolkit work.
This skill is a review and routing layer. It does not authorize production scans, secret access, policy weakening, external installs, CI edits, MCP setup, global config changes, or tool activation.
Review Focus
- Trust boundaries, auth, authorization, public/private payloads, and tenant data are explicit.
- Secrets, tokens, credentials, cookies, and private environment values are not exposed.
- External sources and tools remain untrusted until reviewed.
- Tenant isolation, RLS, API payloads, source safety, and supply-chain metadata are checked when in scope.
- Metadata-only tool records, skipped scans, dry-runs, unavailable scanners, and partial reviews are labeled honestly.
- Route application security readiness through
.ai-toolkit/methods/security/application-security-readiness.md. Canonical toolkit source: methods/security/application-security-readiness.md.
Tool Boundaries
Use scanner metadata only as routing intelligence unless the target project already owns the scanner command or the owner separately approves a scoped scan. Do not use Socket, ZAP, Harden-Runner, deep secret scanners, CI permission changes, or production-impacting scans without explicit approval.
Route WebView, native bridge, token/cookie, deep-link, local-file, and embedded-content risks through .ai-toolkit/methods/security/webview-boundary-review.md. Canonical toolkit source: methods/security/webview-boundary-review.md.
Completion Evidence
Report findings by severity, coverage, commands run, skipped checks, WARN output, and residual security risk.
