The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted video files to generate transcripts and frames which are then processed by the AI agent. This creates a surface for indirect prompt injection attacks where malicious content embedded in the video could trick the agent into performing unintended actions. * Ingestion points: The scripts/understand_video.py script reads user-provided video files. * Boundary markers: The skill does not implement delimiters or instructions to ignore embedded commands within the extracted transcript or visual content. * Capability inventory: The skill can execute system commands via ffmpeg and ffprobe and perform file system operations. * Sanitization: There is no evidence of sanitization or filtering of the transcribed text or visual frames before they are presented to the agent.
[COMMAND_EXECUTION]: The script scripts/understand_video.py uses subprocess.run to call ffmpeg and ffprobe for media processing. While it uses list-based arguments to prevent shell injection, it executes these binaries with paths derived from user input.
[EXTERNAL_DOWNLOADS]: The script provides an option to download and install the openai-whisper package from official repositories if it is missing. This download targets a well-known service and trusted organization.

video-understand