review-skillslisted

# review-skills ## Voice Read `../persona.md` at the start of this skill. The mad-scientist voice is canonical for all output. **Scope:** local to this skill's execution only. Revert to session default voice once the final report is printed. ## Language Adapt all output to the user's language. Technical identifiers (file paths, skill names, CLI flags) stay in their original form. ## What this skill does Two-phase quality pass over all nuthouse plugin skills: 1. **Structure** — runs `/audit` to check conventions (frontmatter, sections, persona pointer, plugin.json). Reports critiques and warnings. 2. **Quality** — lets the user pick which skills to test with `skill-creator:skill-creator` (evals, benchmarks, description optimization). Processes them one by one. The two phases are independent. If the user only wants structure: point them to `/audit`. If they only want quality: start at Step 3. ## Step 1 — Preconditions Verify we're in the nuthouse repo: ```bash test -f .claude-plugin/marketplace.json && echo "ok" || echo "not nuthouse" ``` If not nuthouse: abort with _"ce labo n'est pas le bon."_ ## Step 2 — Structural audit Tell the user: _"phase 1 — j'inspecte la structure. on vérifie les formules."_ Dispatch `/audit` by invoking the `audit` skill directly. Let it run and print its report. Do not suppress its output. After the report, ask: ``` des critiques ou warnings ? veux-tu corriger ça avant de tester la qualité ? (y) oui — stop ici, corrige d'abord le