The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates navigation to external, potentially untrusted URLs (e.g., demo links or local dev servers) to capture visual artifacts. This ingestion of external content constitutes an indirect prompt injection surface where a target website could include instructions designed to subvert the agent's behavior or safety guidelines.
Ingestion points: URLs provided via --demo-url or --url arguments, or detected from project files like package.json or README.md.
Boundary markers: None identified to mitigate instruction following from target page content.
Capability inventory: Browser automation via Playwright, local file system writes, and FFmpeg execution.
Sanitization: Standard URL resolution is performed without deep content sanitization.
[COMMAND_EXECUTION]: The record-readme-tour.mjs script executes the FFmpeg binary via the Node.js spawn method to process recorded WebM videos into GIF format. The path to the executable is managed by the @ffmpeg-installer/ffmpeg package.
[EXTERNAL_DOWNLOADS]: The skill utilizes the playwright and @ffmpeg-installer/ffmpeg libraries. These packages are configured to download external browser binaries (Chromium) and the FFmpeg executable respectively during installation or initialization, which are necessary for the skill's media capture and processing functions.

enhance-readme