What problem does it solve? Reviewing a personal site's visual quality by hand is inconsistent: reviewers re-measure the same things, findings lack evidence, and recommendations conflict with the site's own design rules. This Skill structures the entire review as a repeatable pipeline with verified evidence and calibrated outcomes. ## Core Features & Use Cases - Evidence-first capture: Builds the site, drives a real browser session per route, and produces a facts sheet of computed styles, geometry, and accessibility trees before any judgment happens. - Parallel critic lanes: Spawns Opus subagents for craft, spec conformance against DESIGN.md, an impeccable-skill audit, and persona walkthroughs (recruiter, engineer, curious visitor), each returning schema-conformant findings. - Synthesis and calibration: Dedupes findings into ranked tiers (clear problems, weak spots, direction questions, doc corrections), publishes one annotated HTML report, and converts accepted direction questions into permanent Rulings in DESIGN.md. - Use Case: After shipping layout changes to the site, run the review to get a verified, prioritized list of presentation defects with concrete fixes, while deliberate design decisions are protected from being flagged as bugs. ## Quick Start Ask the agent to run the design review on the site and produce the HTML report with ranked findings.
