The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted transcript data from YouTube, presenting a surface for indirect prompt injection attacks.
Ingestion points: Untrusted transcript text is ingested through scripts/fetch_transcript.py.
Boundary markers: The skill does not use explicit delimiters or specific instructions to isolate external transcript content from the agent's core logic.
Capability inventory: The skill possesses the capability to execute shell commands (via yt-dlp) and perform file system writes (via scripts/analyze_video.py).
Sanitization: While the video ID is validated via regex, the actual transcript content is not sanitized before being processed by the language model.
[COMMAND_EXECUTION]: The skill executes the yt-dlp tool via the Python subprocess module to fetch video data.
Evidence: scripts/fetch_transcript.py uses subprocess.run to call yt-dlp for metadata and transcript extraction.
Mitigation: The skill validates the video ID using a strict regex ([A-Za-z0-9_-]{11}) in scripts/utils.py before inclusion in the command, which effectively prevents shell command injection.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of external Python libraries.
Evidence: SKILL.md identifies youtube-transcript-api and yt-dlp as required dependencies.

youtube-analysis