What problem does it solve? Interfaces built by AI agents often look templated and generic, and code that reads correctly frequently renders incorrectly. This Skill applies concrete design laws (color strategy, type scale, spacing, motion, component rules) and enforces screenshot-based visual QA so design work is verified against rendered output rather than assumed from code. ## Core Features & Use Cases - Design system work: define and audit design tokens, OKLCH color ramps, fluid type scales, dark mode rules, spacing, and component laws, with a 100-point scoring rubric for design-system audits. - Scoped bans and exceptions: an authoritative list of banned patterns (gradient text, decorative glass panels, icon-card grids, fake metrics) with allowed exceptions and required replacements. - Source-to-implementation visual QA: compare a mock, Figma frame, or reference screenshot against the rendered implementation using Playwright screenshots at desktop (1280px) and mobile (390px) widths, producing a visual-qa-report.md with P0-P3 findings. - Use Case: Ask the agent to restyle a dashboard page. It reads PRODUCT.md and DESIGN.md, states a preflight, applies the design laws, renders desktop and mobile screenshots, and ends with a ship gate verdict of ship, ship-with-caveats, or hold. ## Quick Start Use the suede-design skill to audit and restyle the settings page in this repo, then show me desktop and mobile screenshots of the result.