The Agent Skills Directory

[COMMAND_EXECUTION]: The skill extensively uses the subprocess module to orchestrate media processing workflows. It invokes external CLI tools including ffmpeg, ffprobe, yt-dlp (via its Python module), and node to perform video encoding, metadata extraction, transcription, and Remotion rendering.
[EXTERNAL_DOWNLOADS]: The skill interacts with external network services to acquire media. It fetches metadata from the Invidious API to discover mirror instances and performs multi-threaded, range-based downloads from YouTube and various mirror hosts.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from YouTube video content and Whisper-generated transcripts. Although it does not directly execute text from transcripts, these outputs are used to drive agent decisions for clip selection and image prompt generation, which represents a potential indirect injection surface.
Ingestion points: YouTube URLs and media files, Whisper transcripts (scripts/transcribe.py).
Boundary markers: None implemented for transcript content.
Capability inventory: Subprocess execution (ffmpeg, yt-dlp, node), file system writes, and network operations.
Sanitization: No specific sanitization or filtering is applied to the transcribed text before it is returned to the agent context.
[DYNAMIC_EXECUTION]: The skill utilizes a Node.js-based rendering engine (Remotion) to generate video frames. It also evaluates JavaScript within an isolated browser session via the agent-browser tool to fetch advertised backend routes from media mirrors.

youtube-remotion-clips