The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill processes untrusted external data (YouTube transcripts) and instructs the agent to perform file writes and text manipulation based on that data.
Ingestion points: scripts/get_transcript.py fetches content via YouTubeTranscriptApi().fetch(video_id).
Boundary markers: None. The transcript text is interpolated directly into the agent's context without delimiters or warnings against embedded instructions.
Capability inventory: The SKILL.md instructs the agent to "save the transcript to a specific file" and "clean it up so that it is arranged by complete paragraphs". These capabilities (filesystem access and reasoning/modification) are high-risk when applied to untrusted data.
Sanitization: None. The script returns raw text from the YouTube API.
[External Downloads] (LOW): The skill relies on an external Python package youtube-transcript-api>=1.0.0.
Evidence: Defined in the /// script metadata block of scripts/get_transcript.py.
Status: This is a common third-party package. While not on the pre-approved trusted list, it is a standard dependency for this functionality.

youtube-transcript