The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill includes capabilities to browse external websites for design inspiration and process their structure and tokens.
Ingestion points: External URLs visited via WebSearch and analyzed using Playwright as described in reference/inspiration.md.
Boundary markers: The documentation does not specify explicit delimiters or "ignore instructions" warnings when processing third-party CSS or DOM data.
Capability inventory: The skill utilizes sub-agents to read web content, extract design patterns, and take screenshots using Playwright (reference/inspiration.md).
Sanitization: There is no mention of sanitizing or escaping the content retrieved from external sites before it is synthesized by the model.
[DYNAMIC_EXECUTION]: The skill instructs the agent to generate and run temporary JavaScript driver scripts to handle complex browser automation scenarios.
Evidence in reference/audit.md guides the creation of a local .mjs file that imports internal skill modules to drive the browser manually when standard tools cannot reach an authenticated or complex target.
These scripts dynamically import modules from the local filesystem using absolute paths.
[EXTERNAL_DOWNLOADS]: The skill references external design assets and tools.
Instructions in reference/brand.md use curl to fetch font configurations and assets from Google Fonts, which is a well-known and trusted service.
The documentation suggests using Python packages like pymupdf (fitz) for PDF rasterization during the validation of printed artifacts.

michelangelo