The Agent Skills Directory

[COMMAND_EXECUTION]: The lesson-screen.mjs script utilizes node:child_process.spawn to manage its background web server and to invoke a local web browser (Chrome, Chromium, or Edge) for generating screenshots of learning materials via the shot command.
[EXTERNAL_DOWNLOADS]: The skill fetches visualization configuration and supplementary components from the author's GitHub repository and the project's official website (aieng-zh.cn). It also loads legitimate libraries such as Mermaid.js and Three.js from cdn.jsdelivr.net for browser-based rendering.
[DYNAMIC_EXECUTION]: The screenshot functionality involves rendering HTML and executing JavaScript within a headless browser instance. This behavior is confined to the specific purpose of capturing visual representations of the course content.
[INDIRECT_PROMPT_INJECTION]: The skill implements a feedback loop where interactions within the browser (such as clicking quiz options or exploring diagrams) are sent back to the agent as text events. This ingestion of external data is documented and intended for interactive teaching, though it represents a surface for indirect instructions.
Ingestion points: Reads docs/zh.md and quiz.json from the current workspace or remote repository; receives events from the browser via the /event endpoint.
Boundary markers: None identified for individual event strings.
Capability inventory: File system writes (screens), subprocess spawning (browser automation), and network requests (fetching course data).
Sanitization: Uses standard HTML escaping for titles and JSON strings in templates.

lesson-screen