The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references guidelines and design principles from Vercel Labs' official GitHub repository (vercel-labs/web-interface-guidelines), which is a well-known service.
[COMMAND_EXECUTION]: The skill uses browser_evaluate to run a local script (layout-audit.js) and standard tools like python3 -m http.server, npx serve, and Playwright for rendering and screenshotting UI during the design audit process. These are standard development tools and are used as intended for the skill's primary purpose of UI verification.
[DATA_EXPOSURE]: The skill includes instructions to measure scrollWidth and clientWidth and perform pixel-density analysis on rendered artifacts. These operations are performed locally on the generated UI to prove design balance and do not involve exfiltration of sensitive user data.
[INDIRECT_PROMPT_INJECTION]: The skill ingests rendered UI and interaction traces for evaluation. While this is an ingestion surface, the skill utilizes a "fresh-eyes" subagent judge pattern and provides explicit heuristics and deterministic metrics (like WCAG contrast ratios and Apple HIG tap target sizes) to ground the evaluation in factual data rather than untrusted content.

deterministic-design