burnishlisted

# Burnish — Design Review & Improvement (Burnishing polishes finished metal to a better surface — the piece exists; you make it right.) Read `${CLAUDE_PLUGIN_ROOT}/references/creed.md`, `${CLAUDE_PLUGIN_ROOT}/references/memory.md`, and `${CLAUDE_PLUGIN_ROOT}/references/dispatch.md` first. Job slug: active job from STATE.md, or `burnish-<YYYY-MM-DD>` standalone. Log: `bash ${CLAUDE_PLUGIN_ROOT}/scripts/ledger.sh append burnish <slug> audit STARTED -` Requires a runnable UI target (run command + URL from spec/STATE.md, or ask) and Playwright via npx. LOCAL targets only. No target → this skill cannot run honestly; say so and stop. ## Checklist (create a todo per item) 1. Baseline capture (screenshots, key pages × breakpoints) 2. Evaluate against DESIGN.md (or declared heuristics) — proof per finding 3. Present findings + improvement plan; user approves (gate) 4. Fix loop: surgical fix → commit → after-screenshot → verify 5. Re-score, write report (md + json), route leftovers ## 1. Baseline capture Evidence dir: `$SMITHY_MEM/jobs/<slug>/reports/burnish-evidence/`. Screenshot every key page/state the user names (default: main pages plus one form, one empty state, one error state) at 320 / 768 / 1440 widths: `npx playwright screenshot --viewport-size=<w>,720 <url> <dir>/base-<page>-<w>.png` (scratch spec for states needing interaction). These baselines are the before-images for every later fix. ## 2. Evaluate — standard first, then eyes **Standard:** `$SMITHY_MEM/DESIGN.md