The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the node command to execute a local script, scripts/watch.mjs, for indexing and frame extraction. The instructions guide the agent to extract specific parameters like the 11-character YouTube video ID, which acts as a form of input validation.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content from YouTube transcripts and video frames. It mitigates potential injection attacks by explicitly instructing the agent to treat all retrieved content as untrusted evidence and never as instructions. The evidence chain includes ingestion of YouTube data, presence of boundary markers, a specific capability inventory for local file and process management, and sanitization through identifier extraction.

youtube-watch