security-review

---
name: security-review
description: "Review a change for security risks before it is accepted. Use when a change touches a trust boundary — file writes, paths, dependencies, stored secrets, network calls, telemetry, MCP, auth — and needs a decision before merging. Skip for feature planning, test writing, and convention checks."
---

# Goal

Find security risks in a change before it is accepted.

## Inputs

- The change as a diff or summary.
- Test results, when they exist.

## Workflow

1. Identify changed trust boundaries: file writes, paths, dependencies, auth, stored secrets, network calls, telemetry, MCP.
2. Run scripts/scan-secrets.sh over the staged diff and treat matches as blockers until cleared. If scripts/ is unavailable (deleted or cannot execute), perform the same scan by reading the diff directly.
3. Check path validation, overwrite policy, and symlink handling.
4. Check dependency, template, and configuration risk.
5. Check that tests cover the security-sensitive behavior.
6. Classify findings as blockers, risks, or documented tradeoffs.
7. Hand back the verdict with the finding list.

## Decisions

- If a credential is present in the change → blocker; stop and ask for its removal.
- If writes can escape the repository root → blocker.
- If the change conflicts with accepted repository memory → stop and ask for a human decision.

## Verification

- Every trust boundary the change touches has a finding or an explicit all-clear.
- Blockers name the exact file and line.
- No finding is generic filler.

## Resources

- For the security model → docs/20-security/SECURITY_MODEL.md
- For threat context → docs/20-security/THREAT_MODEL.md (when present)

## Output

- Verdict: accept, accept with risks, or block.
- Finding list with files, lines, and severity.