The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external, untrusted sources which could potentially contain malicious instructions intended to influence the agent's planning output.
Ingestion points: The skill fetches YouTube video titles, metadata via the search_youtube_videos tool, and video captions using yt-dlp as described in SKILL.md.
Boundary markers: The instructions do not specify the use of delimiters or clear separation between ingested content and agent instructions to prevent accidental instruction following.
Capability inventory: The skill has the capability to execute shell commands (yt-dlp, curl) and run local Python scripts (hookscore.py, title.py), as well as access YouTube account information via platform tools.
Sanitization: No explicit validation or filtering mechanisms are mentioned for the text retrieved from YouTube titles or captions before it is processed for planning.
[COMMAND_EXECUTION]: The skill invokes external command-line utilities to retrieve media and metadata.
Evidence: The skill uses curl to download video thumbnails and yt-dlp to extract captions and playlist information from YouTube URLs.
[DYNAMIC_EXECUTION]: The skill executes external scripts to perform specialized analysis on video content.
Evidence: The skill calls ../yt-script/scripts/hookscore.py and ../yt-package/scripts/title.py to analyze and score video hooks and titles during the planning phase.

yt-benchmark