SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsmobile-design-review

mobile-design-review

Design & creation

View repository

Updated 10-15 21:12

About this skill

mobile-design-review: Audit, Score, Baseline

Report-only by default. --fix (or the user asking you to fix what you find) enables the fix loop. Never both silently.

The method is ../_shared/references/review-rubric.md (passes, narration, Trunk Test, finding format, ten categories, scoring math, baseline file) and ../_shared/references/tells.md (the grep checklist + tell catalog). Fix recommendations cite ../_shared/references/redesign-protocol.md (lever order) and sections of ../mobile-taste/SKILL.md. Read all four before starting.

Flow

  1. Stack sniff. package.json / app.json / theme files / MOBILE-DESIGN.md / design-baseline.json. If a baseline exists, this run is a REGRESSION run - deltas lead the report.
  2. Scope. Whole app by default; a named screen/flow if the user scoped it. List the screens you will cover (from the route tree) so coverage is explicit.
  3. Static pass (always): the tells.md grep block, then code inspection per the ten rubric categories. Triage every hit.
  4. Live pass (when reachable): simulator/emulator → Expo web phone-viewport → user-provided screenshots, in that order. Per screen: screenshot, first-impression narration, Trunk Test, touch targets, dark mode + large fonts when practical. If nothing is reachable, say so and continue static-only.
  5. Findings. review-rubric.md format: FINDING-NNN [impact | category], evidence, tell number when it maps, concrete fix.
  6. Scores. Design Score + AI Slop Score per the mechanical letter math. Per-category letters in a table. No rounding up.
  7. Baseline. Write/refresh design-baseline.json. On a regression run, the delta table (category changes, new, resolved) comes FIRST.
  8. Report. Narrations, Trunk Test results, findings by impact, scores, the explicit unverified list ("no device reachable: keyboard, splash, scroll perf, font scale, edge-to-edge NOT verified"), and - report-only mode - a recommended fix order (the modernization-lever order from ../_shared/references/redesign-protocol.md).

Fix mode (opt-in only)

Runs the fix loop from redesign-protocol.md: one finding per atomic commit (style(design): FINDING-NNN ...), before/after screenshots when a live pass exists, re-check to close, risk governor (stop > 20%, hard cap 20 fixes), findings marked fixed in the baseline. Constraints:

  • Never touches navigation structure, route names, deep links, tab order, token names, or form field semantics - those are redesign-skill territory and approval-gated there.
  • High-impact findings first, then medium; polish only if the user asked for polish.
  • Stopping at the governor is a success state: report fixed / remaining / risk math.

Rules that bind this skill

  • Report-only means ZERO app-file edits. The one sanctioned write is design-baseline.json (the review's own artifact); skip even that if the user asked for zero writes and inline the JSON in the report instead.
  • Every score follows the math; every dismissed grep hit gets its one-line reason.
  • If the project declares design commitments - MOBILE-DESIGN.md, or any equivalent committed design doc - findings also flag DRIFT from them (a declared accent used inconsistently, a Nav Read the code no longer matches). Name the doc you used.