The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data (video titles, channel names, and descriptions) fetched from YouTube. This content is presented to the agent and could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: Video metadata is captured via yt-dlp and processed by jq in SKILL.md.
Boundary markers: No explicit delimiters or instructions are provided to the agent to treat the video metadata as untrusted data.
Capability inventory: The skill utilizes shell execution via yt-dlp and jq to process and format results.
Sanitization: Metadata is truncated and formatted using jq, but no specific sanitization is performed to prevent prompt injection attacks embedded in the video descriptions.
[COMMAND_EXECUTION]: The skill template interpolates user-controlled variables (QUERY and COUNT) directly into shell command strings (e.g., yt-dlp "ytsearch${COUNT}:${QUERY}"). While the examples use double quotes, this pattern represents a shell injection surface if the agent platform does not strictly validate or escape the input before execution.

youtube-search