What problem does it solve? Reviewing a finished high-fidelity UI design is unreliable when reviewers trust handed-in screenshots or skim the code — stale images, broken renders, off-token values, and contrast failures slip through to the build phase. This Skill acts as an independent acceptance gate that re-renders the code itself and judges the fresh result against a fixed checklist. ## Core Features & Use Cases - Independent re-rendering: Renders the hi-fi code in a headless browser (agent-browser/Playwright), captures fresh screenshots per viewport and state, and vision-reviews them — never trusting supplied images. - Single-sourced 14-condition bar: Judges wireframe coverage, fidelity, polish heuristics, token-backed styling, content realism, rendered states, responsive reflow, numeric WCAG 2.2 AA on the render, and design-system conformance. - Machine-parseable verdict: Emits exactly one VERDICT: approve or VERDICT: revise line plus actionable findings (failed condition, location, concrete fix), including delta-scoped review for amended designs. - Use Case: Before approving a hi-fi Settings screen as a build seed, run this Skill to re-render it at 360/768/1280px, catch a broken mobile reflow, an off-token raw hex color, and a missing accessible name, then return a revise verdict with fixes. ## Quick Start Use reviewing-hi-fi to judge the finished hi-fi code in this directory against the wireframes and design-system, then give me the verdict and findings.
