The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of YouTube transcripts and web search results, which provides a surface for indirect prompt injection.
Ingestion points: Transcripts are retrieved from YouTube via the scripts/fetch_transcript.py script, and additional context is gathered using the WebFetch tool.
Boundary markers: The skill does not define specific delimiters or instructions to the AI to ignore embedded commands within the analyzed transcripts.
Capability inventory: The agent has access to Bash (allowing for network requests and script execution), WebSearch, and WebFetch tools.
Sanitization: No evidence of sanitization, filtering, or validation of the transcript text is present before the content is passed to the AI for analysis.
[EXTERNAL_DOWNLOADS]: The skill documentation and scripts specify a dependency on the youtube-transcript-api Python package. This is a common community-maintained library, but it must be installed from an external registry.
[COMMAND_EXECUTION]: The skill provides numerous bash command templates using curl and jq to interact with the YouTube Data API. It specifically utilizes bash -c execution patterns to maintain environment variables like the API key during processing.

youtube-research