Frontend Design
Production-grade frontend generation and audit. Composite of anthropics/skills/frontend-design (818k installs) + vercel-labs/web-interface-guidelines (576k).
When to use
- Building or modifying any user-facing UI (React, Vue, Svelte, Next.js, Astro)
- Auditing an existing UI for Vercel Interface Guidelines violations
- Generating interfaces that must avoid generic "AI slop" aesthetics
- Reviewing design tokens, layout, or component architecture
Workflow
1. Understand intent (ask one question at a time if underspecified)
- What is the UI for? Who is the user? What is the success criterion (e.g., "feels premium, not template")?
2. Generate / audit
Generation path:
- Pick a distinctive visual direction (reference
brand-guidelinesif present) — do not default to centered card + purple gradient. - Structure with semantic layout: header, hero, feature grid, proof, CTA. Use design tokens (spacing scale, type scale, color tokens).
- Build with React + Tailwind + shadcn/ui. Keep components small and focused.
- Ensure WCAG 2.1 AA (contrast ≥4.5:1, keyboard operable, alt text).
Audit path (Vercel guidelines):
- Fetch
https://raw.githubusercontent.com/vercel-labs/agent-skills/main/skills/web-design-guidelines/SKILL.mdif needed and compare. - Output violations table:
| Guideline | Severity | Location | Fix |
3. Verify
- Visual: does the UI avoid the 3 generic patterns (centered hero + 3 cards, purple gradient, blob illustration)?
- Accessibility:
axeor manual keyboard tab + contrast check. - Tokens: spacing uses 4/8 scale, type uses capped scale, colors are semantic (not hex litter).
Verification
- UI renders without generic AI aesthetics (distinctive direction chosen)
- WCAG AA contrast and keyboard checks pass
- Spacing/type/color use design tokens, not hardcoded values
- Audit (if run) lists violations with severity and fix
References
- Source:
anthropics/skillsfrontend-design,vercel-labs/agent-skillsweb-design-guidelines - See also:
design/system-architecturefor non-UI architecture