Audit Security

!cat ~/.claude/skills/audit-workflow.md

Run as the security dimension. Lens: find vulnerabilities where untrusted input or weak controls let an attacker change behavior, exfiltrate data, or escalate privilege. Tag each finding w/ a CWE.

Patterns under-detected in practice — probe each explicitly:

• Injection — subprocess via string arg (command injection); SQL via concatenation, template literal, or ORM escape hatch (.raw(), .extra(), RawSQL) — parameterized queries non-negotiable; template engines / DOM APIs rendering user input unescaped (dangerouslySetInnerHTML, innerHTML, unescaped template vars).
• Deserialization / parsing — untrusted input into pickle, yaml.load, JSON-into-class hydrators; XML parsers at default settings (XXE); parsers w/o size/depth limits.
• Access control — routes w/o auth check; state-changing endpoints missing CSRF protection; IDOR where resource fetched by user-supplied ID w/o ownership check.
• Crypto / secrets — JWT accepting alg:none, key/algorithm confusion, missing exp/iss/aud validation; md5/sha1 for passwords, Math.random for tokens, ECB mode, missing salt; secrets in source, logs, or error responses.
• Trust-boundary gaps — server relying on client-side validation only; SSRF/path traversal where user input becomes URL or filesystem path w/o allowlist.
• LLM output handling — model-generated content used as code, query, or shell command w/o sanitization (insecure output handling, CWE-116).

Filter aggressively. Not a vulnerability: server-controlled config (env vars, constants, settings.*), framework-mitigated sinks (React {x}, parameterized ORM, prepared statements), UUID identifiers, client-side-only validation when server check also exists. Finding requires concrete attack path from untrusted source to sink — "could be vulnerable if" is noise.

Auto-fix when safe pattern already established in codebase or framework: parameterize the query, switch subprocess to array form, escape rendered output, add ownership check, swap md5 for argon2, scrub secret from logs — each behavior-preserving for legitimate input. Sign-off required: auth/session/crypto changes that rotate secrets, migrate algorithms, or change token format (affect active sessions and stored credentials) — sketch the change, surface w/ CWE + attack path closed.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: JHostalek
• Source: JHostalek/dotclaude
• License: CC0-1.0
• Homepage: https://jhostalek.github.io/dotclaude/

Install and usage instructions live in the source repository linked above.