home / skills / willsigmon / sigstack / claude-vision-qa

claude-vision-qa skill

/plugins/testing/skills/claude-vision-qa

This skill helps you QA apps using Claude Vision to detect visual bugs, layout issues, accessibility concerns, and design system violations from screenshots.

npx playbooks add skill willsigmon/sigstack --skill claude-vision-qa

Review the files below or copy the command above to add this skill to your agents.

Files (1)
SKILL.md
4.2 KB
---
name: Claude Vision QA
description: Use Claude's vision to QA apps like a human analyst - screenshots, design review, bug detection
allowed-tools: Read, Edit, Bash, mcp__plugin_playwright_playwright__browser_take_screenshot, mcp__xcode__build
model: sonnet
---

# Claude Vision QA Expert

Use Claude's vision capabilities to QA your app like a human analyst would.

## Perfect for Vibe Coders
No coding needed to do QA - just take screenshots and ask Claude to analyze them. Claude can see:
- Layout issues
- Alignment problems
- Missing elements
- Accessibility concerns
- Design system violations
- Visual bugs

## Quick QA Workflow

### 1. Capture Screenshot
```bash
# iOS Simulator
xcrun simctl io booted screenshot ~/Desktop/ios-screen.png

# Android Emulator
adb exec-out screencap -p > ~/Desktop/android-screen.png

# Web (Playwright MCP)
# Use browser_take_screenshot tool

# macOS Screenshot
screencapture -x ~/Desktop/mac-screen.png
```

### 2. Ask Claude to Review
Just drag the screenshot into Claude and ask:

**For Bug Detection:**
> "Look at this screenshot of my app. Find any visual bugs, misalignments, or things that look broken."

**For Design Review:**
> "Review this UI design. Is anything off? Are the colors, spacing, and alignment consistent?"

**For Accessibility:**
> "Check this screen for accessibility issues. Are there contrast problems? Missing labels? Touch targets too small?"

**For Comparison:**
> "Here are before/after screenshots. What changed? Did any regressions appear?"

## Automated QA Script

### Capture & Analyze Loop
```bash
#!/bin/bash
# qa-screenshot.sh

# Capture iOS simulator
xcrun simctl io booted screenshot /tmp/qa-screen.png

# Open in Claude Code (drag and drop works)
echo "Screenshot saved to /tmp/qa-screen.png"
echo "Drag into Claude and ask for visual review"
```

### Playwright + Vision
```typescript
import { chromium } from 'playwright';
import Anthropic from '@anthropic-ai/sdk';

const browser = await chromium.launch();
const page = await browser.newPage();
await page.goto('http://localhost:3000');

// Capture screenshot
const screenshot = await page.screenshot({ fullPage: true });
const base64 = screenshot.toString('base64');

// Send to Claude Vision
const claude = new Anthropic();
const response = await claude.messages.create({
  model: "claude-sonnet-4-5-20250929",
  max_tokens: 1024,
  messages: [{
    role: "user",
    content: [
      { type: "image", source: { type: "base64", media_type: "image/png", data: base64 } },
      { type: "text", text: "You are a QA analyst. Review this webpage screenshot and list any visual issues you find. Be specific about locations." }
    ]
  }]
});

console.log(response.content[0].text);
```

## QA Prompts by Category

### Layout Check
> "Check this screenshot for layout issues: overlapping elements, cut-off text, improper spacing, elements going off screen."

### Responsive Design
> "I took this on iPhone 15 Pro. Does everything look properly sized? Any horizontal scrolling? Text readable?"

### Dark Mode
> "This is dark mode. Check for: hard to read text, missing dark backgrounds, icons not visible, inconsistent colors."

### Form Validation
> "Look at this form. Are error messages visible? Are required fields marked? Is the submit button accessible?"

### Loading States
> "This is during loading. Is there a loading indicator? Is the layout stable? Any flash of unstyled content?"

## Learning Over Time

### Document Patterns
After Claude finds issues, save them:
```markdown
## Known Issue Patterns
- [ ] Button text gets cut off at < 375px width
- [ ] Dark mode: secondary buttons hard to see
- [ ] Long usernames overflow profile card
```

### Create Test Checklist
```markdown
## Visual QA Checklist for Each Screen
1. Light mode screenshot - review
2. Dark mode screenshot - review
3. Small device (iPhone SE) - review
4. Large device (iPad) - review
5. With long text content - review
6. Empty state - review
7. Error state - review
```

## Integration with MCP

### Playwright MCP
Use `mcp__plugin_playwright_playwright__browser_take_screenshot` to capture web pages directly in Claude.

### Xcode MCP
Use `mcp__xcode__build` to build, then capture simulator screenshots.

Use when: Visual bug hunting, design review, accessibility checks, regression testing without code

Overview

This skill uses Claude's vision capabilities to perform human-like visual QA on apps and webpages. It inspects screenshots for layout problems, design inconsistencies, accessibility issues, and regressions so you can find visual bugs without heavy manual testing. The workflow supports single screenshots, before/after comparisons, and automated capture loops.

How this skill works

Provide screenshots (simulator, emulator, Playwright, or device captures) to Claude and instruct it to act as a QA analyst. Claude examines the image for alignment, spacing, missing elements, contrast, touch targets, and other visual defects, then returns a prioritized, specific list of issues with locations. It can be integrated into scripts or Playwright flows to automate captures and send images programmatically.

When to use it

  • After UI changes to catch visual regressions before release
  • During design reviews to validate spacing, color, and component consistency
  • For accessibility checks (contrast, labels, touch target size)
  • When comparing before/after screenshots for regressions
  • As part of CI or local dev loops to catch layout problems early

Best practices

  • Capture screenshots in consistent device sizes and modes (light/dark, small/large)
  • Provide context in the prompt: device type, state (loading, error), and which areas to prioritize
  • Use before/after pairs to highlight regressions and ask for diffs
  • Save recurring issues as patterns to build a visual QA checklist
  • Automate capture with Playwright or emulator scripts to reduce manual work

Example use cases

  • Run visual QA on a checkout flow after a frontend refactor to detect misaligned buttons or cut-off text
  • Compare nightly build screenshots to a baseline to flag regressions automatically
  • Review mobile screenshots in dark mode to find low-contrast text and invisible icons
  • Scan form screens to confirm error messages and required-field indicators are visible
  • Use Playwright + Claude Vision to embed visual checks in end-to-end test runs

FAQ

What screenshot types work best?

Full-page or full-screen PNGs from simulators, emulators, Playwright, or device screenshots work best for a comprehensive analysis.

Can this run automatically in CI?

Yes. Capture screenshots via Playwright or emulator scripts, convert to base64, and send to Claude Vision as part of your CI pipeline.

How specific will the feedback be?

Claude provides targeted observations and locations (e.g., top-right button misaligned, label truncated near footer), plus suggestions for what to check or fix.