design-eval - Author a frozen UI design-fidelity evaluation
Creates and installs a frozen `score_design_fidelity` component that captures app and reference screenshots and evaluates fidelity with a vision judge.
Tags
Updated: 2026-10-19Capabilities
Typical Inputs
Typical Outputs
What this skill does
- Scaffold evaluation runtime
- Capture application screenshots
- Configure reference screens
- Define scoring policies
- Judge visual fidelity
- Freeze evaluation configuration
- Install oracle component
Inputs
- Servo-scaffolded target
- Design mockups
- Screen definitions
- Reference PNG paths
- Screen setup modules
- Application URL
- Viewport configuration
- Sampling parameters
- Scoring policy
- Vision model ID
- Judge transport
- Capture configuration
Outputs
- Design evaluation runtime files
- Evaluation configuration file
- Captured screenshots
- Fidelity score
- Evaluation ledger
- Installed oracle component
Requirements
- Existing oracle.sh
- Existing .servo/install.json
- Playwright project dependency
- Chromium browser installation
- Browser-renderable design mockups
- API key or Claude CLI for gated scoring
- Vision-capable judge model
- Node.js on the scoring machine
- Python 3