design-critique

name	design-critique
description	Evaluate design from a UX perspective with Nielsen's heuristics scoring, anti-pattern detection, persona-based testing, and actionable feedback. Report-only. Triggers on critique, UX review, design feedback, how does this look, evaluate the design, component review.
argument-hint	[area (feature, page, component...)]

Evaluate the design quality of a feature or component. Report-only — this skill documents issues but does not fix them.

Preparation

Read ~/.claude/skills/minoan-frontend-design/SKILL.md for aesthetic principles and anti-pattern guidance. If no .design-context.md exists in the project root, ask the user about audience and intended feel before proceeding — context determines whether a design choice is right or wrong.

Resolve Target and Load Ignore List

1. Resolve the primary artifact. Map the user's phrasing to a concrete file path: "the homepage" → src/pages/index.tsx, "the settings modal" → src/components/Settings.tsx. File paths are stable across runs; dev-server ports drift.

2. Read the ignore list at .design-critique/ignore.md if it exists. Each non-empty, non-comment line is something the user has marked as "do not re-raise" (deferred tradeoffs, designer-intended deviations, accepted false positives). When a finding matches a line (case-insensitive substring), drop it silently — do not mention it in the report.

Step 1: LLM Design Review

Read the relevant source files (HTML, CSS, JS/TS) and visually inspect if browser automation is available. If using browser automation, create a new tab — never reuse an existing tab, to prevent state interference between assessments.

Evaluate as a design director:

AI Slop Detection: Does this look like every other AI-generated interface? Check against the rejection guidelines in the minoan skill. Look for: AI color palette, gradient text, glassmorphism, hero metric layouts, identical card grids, generic fonts (including second-tier defaults like Fraunces, Instrument Sans, Outfit), side-tab accent borders. The test: if someone said "AI made this," would they believe immediately?

Holistic Design Review: Visual hierarchy (eye flow, primary action clarity), composition (balance, whitespace, rhythm), typography (hierarchy, readability, font choices), color (purposeful use, cohesion, accessibility), states & edge cases (empty, loading, error, success).

Cognitive Load (consult ~/.claude/skills/minoan-frontend-design/references/cognitive-load.md): Run the 8-item checklist. Report failure count: 0-1 = low (good), 2-3 = moderate, 4+ = critical. Count visible options at each decision point — if >4, flag it. Check for progressive disclosure: is complexity revealed only when needed?

Emotional Journey:

• What emotion does this interface evoke? Is that intentional?
• Peak-end rule: Is the most intense moment positive? Does the experience end well?
• Emotional valleys: Check for anxiety spikes at high-stakes moments (payment, delete, commit). Are there design interventions (progress indicators, reassurance copy, undo options)?

Nielsen's Heuristics (consult ~/.claude/skills/minoan-frontend-design/references/heuristics-scoring.md): Score each of the 10 heuristics 0-4.

Step 2: Systematic Anti-Pattern Check

Walk through ~/.claude/skills/minoan-frontend-design/references/anti-patterns.md checklist item by item against the code. For full technical dimension scoring (a11y, performance, responsive, theming, anti-patterns), recommend /design-audit — this step focuses on design-quality anti-patterns only, not technical implementation.

Step 3: Combined Report

Synthesize both assessments. Note where they agree, where systematic checks caught issues the holistic review missed, and vice versa.

Design Health Score

#	Heuristic	Score	Key Issue
1	Visibility of System Status	?
2	Match System / Real World	?
3	User Control and Freedom	?
4	Consistency and Standards	?
5	Error Prevention	?
6	Recognition Rather Than Recall	?
7	Flexibility and Efficiency	?
8	Aesthetic and Minimalist Design	?
9	Error Recovery	?
10	Help and Documentation	?
Total		??/40

Be honest with scores. A 4 means genuinely excellent. Most real interfaces score 20-32.

Anti-Patterns Verdict

Does this look AI-generated? Specific tells found. For full technical dimension scores (a11y, performance, responsive, theming), recommend /design-audit.

What's Working

2-3 things done well. Be specific about why they work.

Priority Issues

3-5 most impactful problems, ordered by importance. For each:

• [P0-P3] What: Name the problem
• Why it matters: How this hurts users
• Fix: Concrete suggestion
• Suggested command: /design-audit, /design-polish, /minoan-frontend-design, etc.

Severity definitions: P0 = blocks task completion, P1 = significant difficulty or WCAG violation, P2 = annoyance with workaround, P3 = nice-to-fix.

Persona Red Flags

Consult ~/.claude/skills/minoan-frontend-design/references/personas.md. Select 2-3 personas most relevant to this interface. For each, walk through the primary user action and list specific red flags:

Alex (Power User): No keyboard shortcuts. Form requires 8 clicks. High abandonment risk. Jordan (First-Timer): Icon-only nav. Technical jargon in errors. Will abandon at step 2.

Be specific — name exact elements and interactions that fail each persona.

Recommended Actions

List recommended commands in priority order. End with /design-polish as the final step if fixes were recommended.

For technical quality issues (contrast ratios, performance, responsive breakpoints), run /design-audit.

Critique Persistence

After generating the report, save it so /design-polish can read prior findings and trends can be tracked across runs.

1. Write the report to .design-critique/<timestamp>__<slug>.md where <timestamp> is YYYYMMDD-HHmmss and <slug> is a kebab-case identifier derived from the resolved target path (e.g., src-pages-index from src/pages/index.tsx).

2. Include frontmatter with structured metadata:

   ---
   target: "<user's original phrasing>"
   resolved: "<file path>"
   total_score: <heuristic total /40>
   p0_count: <n>
   p1_count: <n>
   date: "<ISO 8601>"
   ---
   

3. Append a trend line after the report: read all snapshots for this slug, extract total_score from each, and present the trajectory:

   Trend for <slug> (last 5 runs): 24 → 28 → 32 → 29 → 32 If this is the first run, say: "First run for this target, no trend yet."

Persistence is fire-and-forget — failures should not block the report. The .design-critique/ignore.md file is user-maintained (never auto-generated).

Re-run /design-critique after fixes to see your score improve.