# Visual QA Evaluator

> Compare matched before and after renders for task clarity and visual regressions after a product change.

- Skill: `shuaimxu/visual-qa-evaluator` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx get-skill add shuaimxu/visual-qa-evaluator`
- Raw SKILL.md: https://api.skillmd.com/api/skills/shuaimxu/visual-qa-evaluator/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- Author: ShuaiMXu (https://skillmd.com/u/shuaimxu)
- Updated: 2026-11-16
- Page: https://skillmd.com/skills/shuaimxu/visual-qa-evaluator

---


# Visual QA Evaluator

Prepare an independent visual decision from matched evidence.

## Required evidence

- same-state desktop and mobile captures
- critical task
- declared design intent

## Decisions

- Compare hierarchy, grouping, contrast and reading order
- Report regressions separately
- State uncertainty rather than inventing a score

## Boundary

The deterministic handler only prepares matched artifacts; screenshot-aware review is required.

Return structured evidence, findings, proposed updates and limitations. Use `needs-evidence` when required inputs or professional review are missing. Never invent an Experience Score or promote a project observation into a universal rule.

## Design language continuity

Read [the shared design-language rubric](../visual-hierarchy-reviewer/rubric.json). Apply it to the confirmed change and introduced regressions, preserving supplied brand choices. Put the visible element, task/brand conflict and minimal remedy in each finding's observation. Existing out-of-scope opportunities belong in limitations, not automatic failure of a narrow fix. The Workbench's independent AI review receives this same versioned rubric and foundation context.

## Bundled OJO methodology

Use the [Harness OJO adapter](../../vendor/ojo/ADAPTER.md), then its linked original Skill and relevant references. The runtime records a verified guidance snapshot in this result; design execution and independent QA receive the same pinned source with mode-specific reading routes. Preserve the existing evidence and approval boundaries.

## Execution contract

The bounded handler prepares source-bound matched originals and a review brief; it does not view the images or execute an independent visual judgment. Keep missing states and mismatched configurations visible. Use the existing independent ai-review/attest record for actual task, hierarchy, interaction, design-system and regression conclusions. A successful evidence pairing cannot stand in for an independent review or final acceptance.

