UXD

/design-review

5-perspective review of a generated screen set — UX researcher, accessibility auditor, engineer, brand guardian, end-user voice. Blocks handoff on any P0.

Hat Stage Upstream gate Output Blocks
UXD Design review design/outputs/screens/PRD-NNN/index.md exists design/outputs/handoffs/PRD-NNN-review.md design handoff on any P0

When to Use

Prerequisites

Reviewers (Dispatched in Parallel)

# Reviewer Lens Output shape
1 UX Researcher Flow match persona, cognitive load, first-time friction ✅ Works · ⚠️ Concerns (H/M/L) · 💡 · ❓
2 A11y Auditor (WCAG 2.1 AA) Contrast, keyboard nav, SR/ARIA, touch targets ≥44px, focus indicators, prefers-reduced-motion, form labels/errors ✅ · ❌ Failures (WCAG criterion + severity) · 💡 Fixes (code)
3 Engineer Reusability, CSS complexity, responsive effort, DS alignment, perf ✅ Easy · ⚠️ (S/M/L/XL) · 💡 · 🔴 Blockers
4 Brand Guardian Colour / typography / spacing / style / copy voice vs brand docs ✅ On-brand · ⚠️ Deviations · 💡
5 End-User Voice (persona, first-person) “My first reaction is…”, delight, confusion, elevator pitch First impression · 😍 · 😕 · 🤔 · 💬

Protocol

  1. Load targets in parallel — PRD, brief, index.md + every screen file referenced, design/context-library/brand/*, design/context-library/personas/* (when present).
  2. Dispatch 5 reviewer subagents in parallel (single message, 5 Task calls). Each gets PRD + brief + screen file + brand + personas + own rubric. Parallel dispatch keeps context lean.
  3. Synthesise to design/outputs/handoffs/PRD-{NNN}-review.md with sections 1–5 (verbatim agent output), Synthesis (Agreed / Conflicts), Priority fix list (🔴 P0 / 🟡 P1 / 🟢 P2).
  4. Gate — count P0s. > 0 → block. == 0 → clean, print publish prompt.

Flags

Do Not

Source

skills/design-review/SKILL.md