security-reviewlisted

# Security Review Gate skill backing the security-engineer. This body is the Rule block; the expanded checklist lives in `references/security-checklist.md`, read on demand. Review the diff, not the intentions. User instructions always override this skill. All factual output follows the anti-hallucination skill: cite file:line or mark UNKNOWN, never guess. agentic-signals: goal_intake, refusal_gate, verification, loop_discipline, close_out ## Rule block (every reviewed diff is checked against all six) 1. Secrets: no credential, token, key, or connection string in code, fixtures, logs, or CI config -- placeholders only. Never open a secret-bearing file to verify it; name the missing variable instead. 2. Injection: every input reaching a shell, SQL/query builder, template engine, parser, or deserializer is validated or parameterized AT the trust boundary, not upstream of it. 3. Unsafe constructs: eval/exec on external input, unpinned curl-to-shell installs, YAML/pickle load on untrusted data, path traversal on user-supplied paths, disabled TLS verification. 4. Dependencies: a new dependency is a finding by default; it must name what it replaces and why an existing rung of the minimization ladder cannot do it. 5. Authorization: changed endpoints and handlers keep (or add) their permission checks; privilege boundaries are asserted in a test. 6. Data handling: PII is not newly logged, persisted, or sent to third parties; vault writes pass the memory-p