What problem does it solve? Published articles can ship with visual defects—broken CTAs, generic intros, missing hero headlines—that text-only QC never catches. This Skill captures live desktop and mobile screenshots of a published page, runs vision-model analysis on them, and routes only the copy-fixable findings into the rewrite pipeline. ## Core Features & Use Cases - Dual-viewport capture: Takes desktop (1440x900) and mobile (390x844) screenshots via Cloudflare Browser Rendering and archives them to R2 under design-audits/{slug}/. - Vision analysis with classification: Analyzes screenshots with Llava through AI Gateway, dedupes issues, and classifies them as content-addressable (cta, content, hero, nav) versus CSS-only (layout, typography, color, mobile). - Pipeline integration: Runs as Step 11.5 after KV deploy and indexing, feeding content-addressable findings into the QC Review and Polish agents so the article gets rewritten. - Use Case: After publishing an article on catsluvus.com, run the audit to discover the hero headline renders as "Untitled" and the CTA button has no text, then have the Polish agent regenerate that copy automatically. ## Quick Start Run a design audit on https://catsluvus.com/cats/my-article and report the content-addressable visual issues.
