The Agent Skills Directory

[DYNAMIC_EXECUTION]: The Mermaid library is initialized with securityLevel: 'loose' in both assets/deck-template.html and assets/document-template.html. This setting enables the rendering of HTML within diagrams. If the agent populates these diagrams with unsanitized external data, it creates a vector for local Cross-Site Scripting (XSS).\n- [INDIRECT_PROMPT_INJECTION]: The 'Interactive' profile creates a feedback loop where user interaction with a generated HTML page produces Markdown to be pasted back to the agent. This is a multi-step injection surface.\n
Ingestion points: External data from research logs and codebases is used to populate the HTML templates.\n
Boundary markers: The templates do not include explicit boundary markers or warnings to the agent to ignore instructions embedded in the generated markdown summary.\n
Capability inventory: The skill produces HTML files with JavaScript for state management, clipboard access, and rendering.\n
Sanitization: No explicit data sanitization or escaping is implemented before content is interpolated into the HTML templates.\n- [COMMAND_EXECUTION]: The skill instructs the agent in SKILL.md to run python3 -m http.server 8000 to serve the generated artifacts. While standard for local development, it exposes the current directory's contents to the local network.\n- [EXTERNAL_DOWNLOADS]: The HTML templates include a reference to the Mermaid library hosted on the jsDelivr CDN to enable diagram rendering.

rich-html