The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from YouTube video descriptions without adequate security measures.
Ingestion points: External data enters the agent context via the media:description field in YouTube RSS feeds fetched by scripts/get_updates.py.
Boundary markers: The skill documentation does not provide delimiters or instructions to treat the fetched video metadata as untrusted content.
Capability inventory: The agent is instructed to execute shell commands to run the update script and interact with a downstream podcast-workflow skill.
Sanitization: While the generate_summary function performs basic text cleaning (removing timestamps and URLs), it does not include filters or escaping mechanisms to neutralize potential prompt injection payloads.
[COMMAND_EXECUTION]: The skill's primary operation involves executing a local Python script (scripts/get_updates.py) with various flags to manage the feed.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch RSS feeds from YouTube and uses Jina AI's Reader service (r.jina.ai) as a fallback to scrape video metadata. It also references an external data source at zara.faces.site.

youtube-feed