The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external video metadata (title, uploader) and transcript content. Maliciously crafted strings within these fields could be interpreted by the agent as instructions.
Ingestion points: Video metadata from video.info.json and transcript segments from VTT files processed in scripts/watch.py.
Boundary markers: Output is structured with markdown headers and code blocks, but no specific safety instructions are provided to the agent regarding the untrusted content.
Capability inventory: The skill executes shell commands (ffmpeg, yt-dlp), manages local files, and performs network requests to transcription APIs.
Sanitization: No filtering or sanitization is applied to the text content extracted from the video sources before being presented to the agent.
[COMMAND_EXECUTION]: The skill uses the subprocess module to call system utilities including ffmpeg, ffprobe, and yt-dlp to process video files and extract frames.
[EXTERNAL_DOWNLOADS]: The skill downloads video and metadata from user-provided URLs using yt-dlp. It also manages the installation of binary dependencies via Homebrew on macOS.

watch