Skill

visual-rosetta

Map a page's visual regions to its DOM, and compare a reference against an implementation.

When a Claude Design handoff doesn’t match your implementation and CSS tweaks keep failing to close the gap, the issue is usually structural: a grid where a flex should be, an extra wrapper, a missing region. visual-rosetta enumerates both pages as region maps — DOM subtrees tagged with their visible bounding boxes — then runs a mechanical comparison that surfaces primitive-level differences (display, grid template, nesting) separate from cosmetic differences (spacing, color).

It has four modes:

  • inspect — enumerate one page, see its regions
  • capture — gather a reference + implementation pair
  • prepare — bundle a package for handoff
  • review — consume a package and run a remediation discussion

Built during the Michi work to bridge Claude Design → Claude Code. Used here on Save The Tokens too — the observation report under this epic’s verification folder was authored by this skill.

© 2026 · Christian Sepulveda