template-auditlisted

Use when auditing rendered templates via local advisory review. Triggers on keywords «template», «audit», «render», «generated docs», «AGENTS.md», «paraphrase», «cue placement», «local advisory», «template-render», «audit-template». Session-bound under Claude Code subscription — no API key, no CI gate, no blocking. Step 1 runs deterministic probes; Step 2 asks the current session to semantic-check for paraphrase / cue-placement bugs (P2/P3/P5 from Wave 6 D-2 taxonomy, not in CI per Decision 3).
artyhoo/getff · ★ 0 · Code & Development · score 56

Install: claude install-skill artyhoo/getff

# template-audit — local advisory skill > **Authoritative for:** local advisory template audit skill — trigger keywords, two-step procedure, P2/P3/P5 advisory checks, promotion trigger. > **NOT authoritative for:** project goal — see [README.md#why-this-exists](../../../README.md#why-this-exists). Deterministic CI gate — see [template-render.audit.ts](../../../packages/core/audit-self/template-render.audit.ts). Session-bound advisory audit. **FREE under Claude Code subscription.** No API key. Not blocking. ## Procedure **Step 1 — Deterministic probes (CI-equivalent)** ```bash npm --prefix packages/core run test:template-render ``` If this fails — stop and fix before Step 2. **Step 2 — LLM advisory checks (session-bound, P2/P3/P5)** Ask the current Claude session (no API call): - **P2 — Paraphrase fidelity**: Does rendered `AGENTS.md` convey "every rule is an executable test that fails CI"? Or has framing drifted to "guidelines/suggestions"? - **P3 — Cue placement**: Is the session-bootstrap cue in the first 10 lines of rendered `AGENTS.md`? - **P5 — Synonym coverage**: Does the P1 synonym list in `template-render.audit.ts` still match current template phrasing (no synonym drift)? Report P2/P3/P5 as **advisory** (not blocking). Open follow-up issue if drift detected. ## Promotion trigger Promote P2/P3/P5 → CI gate when: - Deterministic CI PASS rate <80% over 30 days, OR - Consumer reports goal-phrase miss not caught by P1/P4/P6. Promotion path: re-evaluate