The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill documents features that allow the agent to fetch content from arbitrary URLs for the purpose of class extraction (e.g., async () => (await fetch('...')).text()). This represents an ingestion point for untrusted data that could potentially influence agent behavior if not sanitized.
Evidence Chain:
Ingestion points: references/core-extracting.md (via fetch in content.inline).
Boundary markers: None specified for external content.
Capability inventory: The skill is designed to generate CSS based on extracted content.
Sanitization: None specified for the fetched content.
[Dynamic Execution] (LOW): UnoCSS relies on JavaScript/TypeScript configuration files that use dynamic rules and custom extractor functions. This is a standard architectural feature for CSS-on-demand engines, allowing for runtime logic during the build process.
[External Downloads] (LOW): The documentation includes references to external CDNs for browser-based icon loading (e.g., https://esm.sh/) and font providers (Google, Bunny). These are standard industry sources for web development assets.
[Command Execution] (SAFE): Package management commands like pnpm add are provided for installation and integration with frameworks like Nuxt and Vite. These are standard developer operations.

unocss