ui-expert

A frontend skill grounded in named, market-proven patterns from 13 design systems, 6 brand exemplars, comprehensive accessibility law, performance baselines, AI-era UI patterns, UX research foundations, and a 32-entry anti-pattern catalogue. The goal isn't to be bold — it's to be credible. Output should pass the five-second test where an engineer or designer in tech can't say "AI made that."

Why this skill exists

impeccable and frontend-design already enforce taste. They still produce generic results because the model interpolates from the statistical center of its training data unless anchored to a specific named reference.

This skill adds the missing anchors:

1. Register lock — the surface is classified into a fixed audience register (developer-tooling, fintech, consumer-saas, enterprise-admin, b2b-data-dense, editorial-marketing, ai-first). Each register has its own playbook.
2. Reference anchor — every screen is anchored to at least one named industry pattern (e.g. "Linear sidebar", "Stripe KPI card", "Vercel grid background", "Raycast command palette"). Anchors carry concrete token tables, not vibes.
3. Token spec — concrete numbers (font weights, spacing values, OKLCH triplets, radii) locked before any JSX/HTML is written. No code without a token table.
4. Slop audit — after generation, the output is linted against the named anti-pattern catalogue. Failed lints get rewrites, not warnings.

When to invoke

Auto-invoke when the user asks to: