Compare UI screenshots against reference images and output JSON visual QA verdicts.
npx skills add https://github.com/xz1220/oh-my-kimi --skill visual-verdict-xz1220
Please help me install this Agent Skill. Skill: visual-verdict Source: https://github.com/xz1220/oh-my-kimi/tree/main/skills/visual-verdict Command: npx skills add https://github.com/xz1220/oh-my-kimi --skill visual-verdict-xz1220
It solves the problem of uncertain, subjective visual review by comparing a generated UI screenshot against one or more reference images and producing deterministic pass/fail guidance.
category_match.Use case example: Validate a generated dashboard screen against reference designs; if the score is below 90, drive the next edit cycle with targeted difference descriptions (e.g., spacing/typography shifts) until the UI passes.
Provide your generated UI screenshot and the reference images, then ask the skill to return a strict JSON verdict with differences and edit suggestions.
High-intent search queries and answers about installing and using this skill.
Visual regression testing for generated UI screenshots is automated by comparing them against reference images to produce a structured JSON verdict. This includes a 0-100 score, pass/revise/fail outcome, category matching, explicit differences, and actionable edit suggestions.
A JSON verdict in UI QA is a strictly formatted output that scores visual fidelity from 0-100 and flags layout, spacing, typography, color, and hierarchy mismatches. It guides iterative design refinement by providing concrete next-step edits to reach a 90+ threshold.
You validate dashboard layout fidelity and component styling by providing your generated UI screenshot and reference images for comparison. The evaluation checks category matching and returns explicit visual differences with actionable remediation steps to correct deviations.
Yes, you can use screenshot comparison for component-level styling verification without external dependencies. The evaluation logic processes the provided UI screenshots and reference images internally to determine structured visual pass/fail outcomes.
When UI typography and spacing differences keep failing visual validation, use the actionable remediation output from the JSON verdict. It converts specific visual mismatches into concrete next-step edits, driving the next iteration loop until the score reaches 90+.
Visual pass/fail validation works effectively for iterative UI reproduction tasks such as feed pages and dashboards. It evaluates layout fidelity, flags category matching status, and outputs targeted difference descriptions to drive subsequent editing cycles.
huyhung9630Compare UI screenshots against reference images and generate a JSON visual QA verdict.
marcmunoz-unoCompare generated UI screenshots against reference images and output structured JSON verdicts.
Kirankumar2604Recreate React UI components from static screenshot images.
CodeAtCodeCompare UI screenshots against baselines with Playwright and report visual diffs.