The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a shell script render.sh that executes Google Chrome or Chromium from the local system to perform headless rendering. It uses the CHROME environment variable if present, which allows for local path overrides.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided inputs such as eyebrow and title to populate the template.html file. This creates a surface for indirect prompt injection if the inputs contain malicious HTML or SVG tags that could influence the rendering process or browser behavior.
Ingestion points: User-provided strings for the 'eyebrow' and 'title' fields (SKILL.md).
Boundary markers: None explicitly defined for the interpolation of text into the SVG <text> nodes.
Capability inventory: The skill executes a headless browser (render.sh) which can render HTML and access local files via the file:// protocol.
Sanitization: No specific sanitization or escaping of the user-supplied strings is mentioned in the workflow instructions.
[DYNAMIC_EXECUTION]: The rendering workflow involves dynamic generation of HTML content that is subsequently processed by a browser engine. While intended for SVG rendering, this mechanism executes the browser's rendering logic on potentially non-static content.

vector-cover