codebase-security-auditlisted
Install: claude install-skill Quant-Off/skills
# Codebase Security Audit
Finding the dangerous sink is the mechanical step. The real work is proving an
attacker reaches it with data they control, and that no validator breaks the
chain. **A pattern match is a worklist entry, not a vulnerability**, and
reporting one as a finding is the primary failure mode of this skill.
Operate under zero trust: assume every external input is hostile and every trust
boundary is crossed. Then prove it, per finding, with a traced flow.
## When to Use
- A security audit of a repository, service, module, or single file.
- A pre-release or pre-merge security pass on a diff.
- Threat-modeling an application's attack surface.
- Triaging whether a suspected issue is real before filing it.
- Hardening review for a high-assurance, regulated, or air-gapped deployment.
## When NOT to Use
- **Cryptographic implementation review.** Constant-time behavior, zeroization,
and primitive misuse need `crypto-source-audit`.
- **Compiled-artifact verification.** Use `binary-crypto-verify`.
- **Runtime testing.** This skill reads code. It does not fuzz, scan a live
host, or execute exploits.
- **Pure code quality review.** Style, naming, and architecture are out of scope
unless they create a security consequence.
## Rationalizations to Reject
| Rationalization | Why it is wrong | Required action |
| --- | --- | --- |
| "This pattern is dangerous, so it is a vulnerability." | Pattern recognition is not analysis. Most `exec` and `query` hits take trus