video-download
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill relies on running shell commands with the
yt-dlptool to download files, list formats, and extract metadata.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted metadata and text from external video URLs, creating an indirect prompt injection attack surface.\n - Ingestion points: External data enters the agent context through video metadata, descriptions, and subtitles retrieved via
yt-dlp --dump-jsonor subtitle download flags inSKILL.md.\n - Boundary markers: No specific boundary delimiters or isolation techniques are described to instruct the model to treat the video metadata purely as data rather than instructions.\n
- Capability inventory: The skill utilizes command execution to run
yt-dlpcommands on the host environment as documented inSKILL.md.\n - Sanitization: No mechanism is present to filter, sanitize, or escape the retrieved video descriptions or titles before they are processed by the agent.
Audit Metadata