visual-verdict

UI visual regression is detected by comparing generated screenshots against reference images to identify pixel-level differences. The skill evaluates layout, typography, and styling concerns across screen sizes and themes to produce a deterministic pass or fail outcome.

You compare UI screenshots against reference images by applying the visual-verdict skill to evaluate pixel-level differences. It returns a strict JSON payload including score, verdict, category_match, differences, suggestions, and reasoning to guide the next edit iteration.

Screenshot diff results can automatically guide UI remediation through the structured JSON output. The returned payload contains specific suggestions and reasoning fields that provide actionable feedback for correcting layout, typography, and styling regressions.

Visual regression testing works across different screen sizes and themes by applying the evaluation to UI components with layout, typography, and styling concerns. Multiple reference images and optional category hints can be supplied to improve contextual evaluation accuracy.

The best way to get a deterministic pass/fail verdict for UI visual QA is to compare current renders against reference images using a structured evaluation method. This approach returns a strict JSON payload with a score and verdict to ensure consistent, reliable results.

JSON output for visual regression testing includes the fields score, verdict, category_match, differences, suggestions, and reasoning. This strict JSON payload ensures deterministic evaluation and provides structured data to guide the next edit iteration.