The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes the yt-dlp tool to download video content from URLs provided by the user. This is a documented feature for processing remote video content.\n- [DATA_EXFILTRATION]: The skill reads local video files and transmits their contents to external AI services (Google Gemini, OpenRouter, OpenAI, etc.) for processing. This is the primary function of the skill, and the destinations are well-known API providers.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted media content from video files and URLs in scripts/process_video.py. There are no explicit boundary markers or instructions to ignore embedded prompts. The skill has capabilities including shell command execution (ffmpeg, yt-dlp) and file writing. No sanitization of the processed transcripts or analysis is performed before returning results to the agent context.\n- [COMMAND_EXECUTION]: The skill executes system tools like ffmpeg, ffprobe, and yt-dlp using subprocess.run with argument lists in scripts/process_video.py. This method is secure against shell injection vulnerabilities as it avoids invoking a shell for command parsing.\n- [DYNAMIC_EXECUTION]: The scripts/setup.py script uses the __import__ function to dynamically check for the presence of required Python packages, which is standard for environment verification.

video-understand