What problem does it solve? A clean build and zero console errors only prove a page renders, not that it looks right. This Skill closes the gap between "it compiles" and "it looks good" by requiring real screenshot capture and vision-based review before any web or UI task is declared complete. ## Core Features & Use Cases - Pixel-based verification pipeline: Serve the page over HTTP, screenshot it with Playwright MCP, crop and shrink regions with PIL, host on a public URL, and run vision analysis on the actual pixels. - Interactive DOM verification: Exercise every feature via Playwright's browser_evaluate (clicking calculator keys, reading live-updating values) to prove apps launch, render, and respond. - Deterministic fallback: When no browser is reachable, extract the page's script, run node --check, and unit-test pure functions in Node, then honestly report "logic verified, pixels not captured." - Use Case: You just built a landing page the user asked to look "premium, not AI slop." Instead of reporting done after a DOM probe, you screenshot the hero and each section, run vision with specific quality questions about typography, spacing, and color, and fix issues before presenting. ## Quick Start Use the web-visual-qa skill to screenshot the page I just built and verify it looks right before we call it done.
