Screenshot Skill

Analyze UI design directly from screenshot images using the seat's multimodal
vision. No code needed — review what users actually see.

Reading the images

Open the PNG, JPG, WEBP or GIF file with your harness's file reader — the image is presented
visually. Analyze it for design quality, consistency, and usability issues without requiring
access to source code.

What to Analyze

Layout and Spacing

• Is whitespace consistent and intentional?
• Do elements align to a visible grid?
• Are margins/padding proportional and balanced?
• Is content grouped logically (proximity principle)?
• Does the layout breathe, or feel cramped?

Color and Contrast

• Is the color palette cohesive (limited, intentional palette)?
• Do text and background combinations have sufficient contrast?
• Are colors used semantically (red for errors, green for success)?
• Is there visual hierarchy through color weight?

Typography

• Is there a clear heading hierarchy (H1 > H2 > body)?
• Are font sizes proportional and readable?
• Is line spacing comfortable (not too tight or loose)?
• Consistent font family usage — not too many typefaces?

Component Consistency

• Do similar elements look the same (all buttons styled identically)?
• Are interactive elements visually distinguishable from static ones?
• Are states (active, selected) visually distinct?

Responsiveness Clues

• Does the layout look designed for this viewport size?
• Are there elements that appear truncated or overflowing?
• Would this layout adapt to mobile without breaking?

Comparison with Reference Designs

When a reference image is provided, open both with your harness's file reader — the reference
(design spec or comp) and the implementation (built UI).

Compare:

• Spacing fidelity (does built match spec proportions?)
• Color accuracy (same palette?)
• Typography match (weights, sizes, hierarchy)
• Component faithfulness (icons, buttons, inputs)
• Missing or added elements

Screenshot Capture

Use the Read tool on a screenshot file, or capture via browser automation if available (e.g., Playwright, Puppeteer, or an MCP browser tool). Place captured images in a local path and then read them for analysis.

Multi-View Analysis

Review the same page at multiple viewports:

1. Capture desktop (1440px)
2. Capture tablet (768px)
3. Capture mobile (375px)
4. Compare: layout adapts correctly?
5. Check: no content hidden or truncated?
6. Check: touch targets large enough on mobile?

Output Format

## Screenshot Review: {filename or URL}

**Viewport**: {desktop/tablet/mobile/unknown}
**Overall**: {Polished | Minor Issues | Needs Work | Major Problems}

### Visual Design

**Spacing**: {✓ Balanced | ⚠ Inconsistent | ✗ Cramped/Sparse}
- {finding}

**Color**: {✓ Cohesive | ⚠ Minor issues | ✗ Inconsistent}
- {finding}

**Typography**: {✓ Clear hierarchy | ⚠ Minor issues | ✗ Unclear}
- {finding}

**Components**: {✓ Consistent | ⚠ Variations | ✗ Inconsistent}
- {finding}

### Issues Found

#### Critical
- {issue visible in screenshot, with description of location}

#### Minor
- {polish item}

### Comparison with Reference {if provided}
- Spacing: {matches/off by approximately X}
- Color: {matches/discrepancy}
- Missing: {elements in spec not in implementation}
- Added: {elements in implementation not in spec}

### Recommendations
1. {highest priority fix}
2. {second fix}

Integration

• visual-review skill: Use for code-based analysis; screenshot for image-based
• accessibility skill: Color contrast visible in screenshots
• mockup skill: Screenshots of implementations vs mockup specs
• Browser automation: Capture screenshots from live URLs via Playwright, Puppeteer, or an MCP browser tool if available
• imagery/review skill: For AI-powered visual analysis of screenshots, see the imagery/review skill