28 October 2026 · Skill-source check · No installation or output test
To match a screenshot, compare the rendered screen
A supplied screenshot is a target, not a request for a new visual style. Start with the project’s components and tokens. Render the build at the image width, compare layout and type by region, then record differences you could not resolve.
UXKIN ui-from-screenshot is a conditional skill for this job. Its current instructions say to measure the image, map it to existing components and tokens, compare the result side by side, and stop after up to three correction rounds. It also asks for responsive and missing-state work. The core skill says it needs no account; its optional UXKIN reference search needs a paid plan. The repository documents an install command for its full skill set, not a pool-tested selective install.
A Claude Code user described repeated manual fixes to spacing, fonts, and borders after asking for a copy of a design. That one account supports the task distinction, not the success of any skill. The author of figma-pixel-check describes a separate Figma-to-browser checker and reports a mutation benchmark, but we did not fetch its exact skill file or reproduce the test. It is not a recommended catalogue listing yet.
The UXKIN file is on a mutable branch. We read its instructions and repository, but did not install it, check agent loading, compare its output with a no-skill brief, or verify a resulting screen. A screenshot may omit exact font data and interaction states; a claimed match needs a rendered check.