The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches video metadata and transcript data from YouTube's official domains (youtube.com). These operations target a well-known service and are necessary for the skill's primary function.
[PROMPT_INJECTION]: As the skill ingests external text from YouTube captions, it presents an indirect prompt injection surface. Maliciously crafted transcripts could contain instructions intended to influence the agent's behavior.
Ingestion points: youtube-transcript.js fetches external content via HTTP GET requests.
Boundary markers: The output is formatted (JSON, SRT, etc.), but no explicit delimiters or instructions are provided to the agent to disregard embedded commands in the transcript text.
Capability inventory: The skill has the ability to write files to the local filesystem using the --output flag.
Sanitization: The script performs XML unescaping and whitespace normalization but does not filter for potentially malicious natural language instructions.
[DATA_EXPOSURE]: The skill implements a local cache in /tmp/youtube-transcript-cache/ to store retrieved transcripts for 24 hours. While standard for performance, content stored in shared temporary directories may be accessible to other local users depending on environment configuration.
[COMMAND_EXECUTION]: The script provides a file-writing capability via the --output parameter. While intended for saving transcripts, this allows the agent to write data to arbitrary paths provided in the arguments.

youtube-transcript