The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/ingest-channel.py executes system commands via subprocess.run to interact with yt-dlp and the kb CLI tool.
Evidence: Found in scripts/ingest-channel.py within functions run, resolve_videos, and run_ingest.
Context: The script correctly uses list-based arguments for subprocess.run, preventing shell injection. It also validates the topic-slug input using a strict regular expression to prevent path traversal or command manipulation.
[EXTERNAL_DOWNLOADS]: The skill uses yt-dlp to fetch metadata and video information from YouTube.
Evidence: The resolve_videos function in scripts/ingest-channel.py invokes yt-dlp on user-provided channel URLs.
Context: This behavior is the intended primary purpose of the skill and targets a well-known service (YouTube).
[PROMPT_INJECTION]: The skill processes untrusted external data (YouTube transcripts) which represents an indirect prompt injection surface.
Ingestion points: scripts/ingest-channel.py fetches transcripts by calling kb ingest youtube for each resolved video.
Boundary markers: The script does not implement explicit delimiters for the ingested content, relying on the kb tool's internal handling of transcripts.
Capability inventory: The skill has the capability to write to the local vault and execute CLI tools.
Sanitization: Input validation on slugs ensures data is stored in the correct directory structure, but the content of the transcripts is processed as-is.

kb-yt-channel