Skip to content
Back to skills

Design Audit

ASecurity

Full project design audit that orchestrates all agents, scores across 12 dimensions, generates prioritized action plan

  • 7 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added May 26, 2026
ai-agentsrustawsperformance

Works with

  • mcp

Security analysis

A100/100

Scanned May 27, 2026

npx -y skills add Aboudjem/ui-ux-suite --skill design-audit --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Design Audit?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Design Audit
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/aboudjem-design-audit/badge)](https://www.skillsdirectory.com/skills/aboudjem-design-audit)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: design-audit
description: Full project design audit that orchestrates all agents, scores across 12 dimensions, generates prioritized action plan
trigger: "audit design|design review|UX audit|how does my design look|review my UI|design audit"
---

# /design-audit: Full Project Design Audit

Run a comprehensive design audit on the current project or a specified path.

## Usage
```
/design-audit              # Audit current project
/design-audit /path/to/project  # Audit specific project
```

## Flow

1. **Load knowledge base** - Read the knowledge index and all 19 knowledge files:
   - Use `uiux_knowledge_query` with `listFiles: true` to get the full inventory
   - Every agent reads its assigned knowledge files (see `knowledge/INDEX.md` for the mapping)
   - The evidence base, principles, anti-patterns, and trends files are required reading for ALL agents
2. **Scan project** - Use `uiux_scan_project` to detect framework, styling, component library, theme system
3. **Display profile** - Show the user what was detected, ask for confirmation
4. **Dispatch agents** - Launch all relevant specialist agents in parallel (each reads its knowledge files first):
   - `color-analyst` - Color system diagnosis
   - `typography-analyst` - Typography system diagnosis
   - `layout-analyst` - Layout, spacing, grid analysis
   - `component-reviewer` - Component quality and consistency
   - `accessibility-auditor` - WCAG compliance check
   - `interaction-analyst` - Motion, transitions, feedback
   - `psychology-analyst` - Cognitive load, hierarchy, trust
   - `visual-style-advisor` - Style direction evaluation
   - `platform-advisor` - Platform convention adherence
   - `ux-flow-analyst` - Navigation and user flows
   - `performance-ux-analyst` - Loading and perceived speed
4. **Aggregate scores** - Use `uiux_score_overall` to calculate weighted total
5. **Generate report** - Full markdown report with:
   - Project design profile
   - Score card (12 dimensions)
   - Top 20 findings ranked by impact
   - Detailed per-dimension reports
   - Prioritized action plan (quick wins / medium / major)

## Agent dispatch strategy

Launch agents in parallel waves:

**Wave 1** (always): color-analyst, typography-analyst, layout-analyst, accessibility-auditor, component-reviewer
**Wave 2** (always): interaction-analyst, psychology-analyst, visual-style-advisor
**Wave 3** (if applicable): platform-advisor (if mobile/cross-platform), ux-flow-analyst (if has routing), performance-ux-analyst (if has data fetching/images)

## Output

Full audit report saved to `design-audit-report.md` in the project root.
Score card displayed inline for quick overview.

Each finding may cite one or more UX laws inline (for example, `[!] psychology: 12-item nav menu - violates Hick's Law, Choice Overload`). When any finding carries laws, the report ends with a `## Laws of UX Coverage` markdown table summarizing violation counts and worst offenders per law. Agents look up law display names and primary-source citations from `knowledge/laws-of-ux.md` and the `uiux_laws_query` MCP tool.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.