← ClaudeAtlas

code-review-architectlisted

Use to review a working tree, branch, or PR before staging. Runs the gates, a threat model by sink, the propagation audit, and a ≥80 confidence gate on findings; ends in one 0–100 score with a banded verdict. Every finding carries path and line and names the skill that owns its fix. Writes no code.
matis-dev/m-skills · ★ 0 · Code & Development · score 75
Install: claude install-skill matis-dev/m-skills
# Skill: Code Review Architect — Quality, Security & Craft Review > **Apply Guidelines Skill** — load the `guidelines-meta` skill before proceeding. > **Remediation is elsewhere.** This skill writes no code (§Constraints). A security finding is fixed by the `security-architect` skill; an accessibility barrier by the `accessibility-architect` skill; a test gap by the `testing-architect` skill. Name the owning skill on the finding so the fix has somewhere to go — a review that ends in a list nobody can act on is half a deliverable. --- ## Operational Constraints (Strict) 1. **Read-only.** Writes no production code. May read freely and run gates; every fix is the user's call. 2. **Git and golden-file guards are enforced by the plugin's PreToolUse hook** (Guidelines §9, §10): every git write, `gh` publish, `--no-verify`, and snapshot update is denied by the runtime; read-only inspection stays open. Restate the guard inside the review output. 3. **No scope creep in findings.** Comment only on the change set unless an issue *outside* the diff is directly load-bearing for it. If you stray, label it `[OUT-OF-DIFF]` and justify in one line. 4. **Findings follow `module-findings`** — the citation requirement, the confidence gate (post only what you'd rate ≥ 80), the false-positive list, the "worth a second look" bucket for severe-but-unverified items, the severity bands, and the verdict rules. Load it before writing a single finding. 5. **Tests evaluated via Testing Architect**; **