The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes local documentation files (MDX) to extract and render SVG figures, creating a potential surface for indirect injection if these files contain malicious content.
Ingestion points: Local MDX files passed as command-line arguments to the scripts/render.ts utility.
Boundary markers: The rendering script does not utilize specific delimiters or instructions to ignore malicious directives embedded within the figure markup.
Capability inventory: The render.ts script can read local files, write HTML files to the /tmp directory, and execute the agent-browser tool to render and screenshot content.
Sanitization: The script performs basic JSX-to-HTML string replacement but lacks robust sanitization to prevent potential script execution (XSS) within the browser environment when rendering extracted figure tags.
[COMMAND_EXECUTION]: The skill includes a TypeScript utility designed to invoke external platform tools.
Evidence: scripts/render.ts uses the node:child_process.spawn function to execute the agent-browser CLI tool for file preview and screenshotting operations.
[DYNAMIC_EXECUTION]: The skill generates and writes executable-like content (HTML) at runtime to facilitate its preview functionality.
Evidence: The documentFor function in scripts/render.ts dynamically constructs a complete HTML document string containing figure markup and styles, which is saved to a temporary file and subsequently loaded into a browser.

figure