The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's HTML templates fetch styling and script resources from well-known public CDNs.
Loads KaTeX CSS and JS for math rendering from cdn.jsdelivr.net.
Fetches typography from Google Fonts (fonts.googleapis.com and fonts.gstatic.com).
[COMMAND_EXECUTION]: The skill executes local Python scripts and manages browser processes for image rendering.
Uses scripts/get_output_dir.py to organize file versions.
Uses scripts/render_images.py which utilizes the html2image library to spawn a headless Chrome or Edge browser instance to capture screenshots of generated HTML files.
[DATA_EXPOSURE]: The skill is designed to read user-provided documents (such as final.md or draft.md) from specified directories to use as input for the rendering process.
[INDIRECT_PROMPT_INJECTION]: There is a potential attack surface where untrusted user content from processed documents is interpolated into HTML templates.
Ingestion points: Reads text from final.md, draft.md, or user-specified custom files via scripts/get_output_dir.py.
Boundary markers: None explicitly defined in the rendering logic; content is parsed by the LLM into a JSON structure (blocks.json).
Capability inventory: The skill can write files to the local filesystem and execute a rendering script that spawns a browser.
Sanitization: While one template includes basic JavaScript escaping, scripts/render_images.py performs raw string replacement of template placeholders. However, the skill instructions explicitly direct the LLM to rephrase and refine all input text, which serves as a natural mitigation layer.

xhs-render