The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user content to generate HTML cards, creating a potential injection surface.
Ingestion points: User-provided text content is processed according to the instructions in SKILL.md and example.md.
Boundary markers: The instructions specify HTML structures and CSS styles but do not define explicit delimiters or escaping protocols for user-supplied strings before they are interpolated into the HTML template.
Capability inventory: The skill invokes skills/shared/scripts/render_card.py, which utilizes Playwright (a headless browser engine) to render the HTML. This capability could be exploited if malicious scripts are embedded in the user content and executed by the renderer.
Sanitization: No sanitization or filtering logic is documented for handling external user content.
[DYNAMIC_EXECUTION]: The skill dynamically generates HTML code by combining design templates with user-supplied text. This generated content is then executed/rendered in a browser environment via Playwright.
[COMMAND_EXECUTION]: The skill directs the agent to execute shell commands to run local Python scripts (render_card.py and card_audit.py) and to install the playwright dependency to fulfill its card generation functionality.

card-xiaohongshu