Skip to main content
Set design=true on a snapshot or screenshot to analyze the page’s authored design system.
Design extraction loads stylesheets while continuing to block font binaries, images, and media. It costs two credits because it performs additional stylesheet capture and browser evaluation.

​
Result areas

  • Tokens: CSS custom properties, resolved values, kinds, and referenced status.
  • Color: observed palette entries ranked by frequency.
  • Typography: font families, faces, weights, styles, and sanitized delivery signals.
  • Breakpoints: width media queries ranked by rule count.
  • Motion: keyframe summaries, transition durations, easing curves, reduced-motion support, and scroll behavior.
  • Libraries: confidence-ranked signals for common animation and interaction libraries.
  • Coverage: stylesheet counts, bytes and rules inspected, truncation, and omitted counts.
Raw data: and blob: font URLs are dropped. HTTP(S) sources are length- and count-capped for the extraction response, while the projection sent to a model for blueprint interpretation removes font URLs completely and retains only delivery kinds such as self-hosted, google-fonts, or cdn.
Design evidence reflects the page’s initial authored state. It is collected before click or scroll interactions, so scroll-triggered variants may not be sampled.

​
Bounded extraction

Scrinly inspects at most 60 stylesheets, 40,000 rules, and 2 MiB of captured CSS. The result reports truncation and coverage rather than silently implying complete inspection.