What problem does it solve? Design and engineering teams lack a consistent, evidence-based way to evaluate whether a component or screen meets current B2C UI quality standards, and improvements often drift into hardcoded values or uncontrolled scope. This Skill scores UI code on five defined axes and applies only pre-approved, token-compliant variants. ## Core Features & Use Cases - Read-only scoring (/review): Scores a component on visual hierarchy, motion intent, density balance, brand coherence, and trend relevance, citing specific lines, values, or tokens as evidence, then proposes 2-3 named improvement variants. - Controlled implementation (/apply): Applies exactly one named variant from a prior review using existing semantic tokens only, across Web, iOS, and Android files, and commits with a standardized message. - Preview generation: Creates a side-by-side HTML mockup file (docs/preview-variants.html) so reviewers can visually compare variants before applying. - Use Case: A designer suspects a product card feels off-brand. They run /review CardProduct, receive a 5-axis score with evidence and three variants, open the generated preview page, then run /apply tonal-surface to implement the chosen variant with existing tokens. ## Quick Start Ask the assistant to run /review on a component file such as src/components/CardProduct.tsx to get a scored evaluation with improvement variants.