The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/vtd.js uses node:child_process.spawn to execute yt-dlp and ffmpeg. It accepts arbitrary arguments through a -- flag, which are passed directly to the yt-dlp binary. This pattern allows for the injection of dangerous flags, such as --exec, which can be used to execute arbitrary shell commands on the host system if the input source is untrusted.
[EXTERNAL_DOWNLOADS]: The skill's primary function is to fetch content from external URLs. It uses the youtube-transcript-plus library and yt-dlp to download media files, transcripts, and subtitles from various online platforms at runtime.
[PROMPT_INJECTION]: The skill exhibits vulnerability to indirect prompt injection (Category 8) because it ingests and processes untrusted external transcript data.
Ingestion points: Untrusted data enters the agent context through transcripts fetched by youtube-transcript-plus and yt-dlp in scripts/vtd.js.
Boundary markers: The skill lacks clear delimiters or protective instructions (e.g., "ignore embedded instructions") when returning fetched transcript text to the agent.
Capability inventory: The script scripts/vtd.js possesses the ability to execute shell commands via spawn and perform file system operations like writing and copying files.
Sanitization: The script performs basic formatting cleanup (removing HTML tags and brackets), but it does not implement sanitization to prevent natural language instructions embedded within transcripts from influencing the agent's behavior.

video-transcript-downloader