The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documentation includes examples of fetching external content and references standard package installations from well-known registries.
Evidence: Reference to fetching content from 'https://example.com' as an extraction source example in references/core-extracting.md.
Evidence: Installation instructions for unocss and @unocss/nuxt from official registries in references/integrations-nuxt.md.
Evidence: Usage of well-known CDNs like https://esm.sh/ for icon loading in references/preset-icons.md.
[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and analyze potentially untrusted project data, such as configuration files and source code, to generate CSS.
Ingestion points: The agent is instructed to check for project-specific uno.config.* files and scan source files (e.g., .vue, .tsx, .html) in SKILL.md and references/core-extracting.md.
Boundary markers: The documentation does not provide specific delimiters for separating configuration logic from agent instructions.
Capability inventory: The agent reads project files to understand the design system and utility rules defined by the user.
Sanitization: No sanitization methods are described for handling potentially malicious content within project files.

unocss