The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a local Node.js script (scripts/shot.js) to capture screenshots of the generated animations. It also suggests using standard shell tools like cp and grep for file management.
[DYNAMIC_EXECUTION]: The scripts/shot.js utility dynamically modifies the source code of the HTML animation files (adjusting the virtual clock and injecting status indicators) before executing them in a headless browser to generate screenshots.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided text scripts and visual descriptions, interpolating this untrusted data directly into HTML/JS templates. This creates an attack surface where external content can influence the generated code or subsequent agent actions.
Ingestion points: User-provided script text (口播稿) enters the agent context and is written to HTML files as defined in SKILL.md.
Boundary markers: The templates lack explicit delimiters or warnings to ignore instructions within the user-provided text.
Capability inventory: Subprocess execution (node, cp, grep) and file system write operations across all generated assets.
Sanitization: No specific sanitization or filtering logic is provided for the interpolated text.
[EXTERNAL_DOWNLOADS]: The HTML templates reference external assets from well-known services, specifically Google Fonts (fonts.googleapis.com and fonts.gstatic.com). These are used for rendering visual styles and are documented here as part of the skill's normal operation.

video-shot-demos