The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted transcript data from YouTube videos and interpolates it directly into a prompt for AI summarization.
Ingestion points: The get_transcript method in youtube_transcript_summarizer.py fetches text from external YouTube APIs based on a provided URL.
Boundary markers: The transcript text is inserted into the prompt template without specific delimiters or instructions to ignore embedded commands, which could allow a malicious transcript to influence the agent's behavior.
Capability inventory: The script performs file writing operations via open().write() and executes a specific subprocess via subprocess.run().
Sanitization: There is no content-based sanitization or escaping performed on the transcript text before it is sent to the LLM.
[COMMAND_EXECUTION]: The script uses subprocess.run to call the yt-dlp command-line utility for title extraction.
Evidence: The fetch_video_title method in youtube_transcript_summarizer.py executes yt-dlp to retrieve video metadata. Although the video ID input is strictly validated using a regex ([a-zA-Z0-9_-]+) to prevent shell injection, this pattern involves external binary execution.

youtube-transcript-summarizer