The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes subprocess.run to invoke external binaries including yt-dlp for video downloads, and ffmpeg and ffprobe for scene detection, frame extraction, and duration analysis.
[EXTERNAL_DOWNLOADS]: Fetches video media and transcript data from third-party platforms such as YouTube and Instagram, and interfaces with the DataForSEO API to retrieve subtitle information.
[DYNAMIC_EXECUTION]: The skill includes logic in scripts/download.py to dynamically install the yt-dlp package via pip if it is not found in the environment at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted video content and transcripts as multimodal input for visual analysis, creating a potential surface for indirect prompt injection. Ingestion points: Video media files and transcripts retrieved from external URLs. Boundary markers: The prompt in scripts/analyze.py provides a structured framework and explicit formatting instructions for the model's analysis pass. Capability inventory: The skill maintains capabilities for file system writes (Obsidian vault management) and command execution (ffmpeg, yt-dlp). Sanitization: No explicit filtering or sanitization is performed on the raw transcript text or visual frames before they are analyzed by the multimodal AI model.

video-watcher