The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local binaries ffmpeg and ffprobe to perform technical analysis of video files. As implemented in scripts/review.py, all subprocess calls use list-based arguments and the default shell=False setting. This is a secure pattern that prevents command injection by ensuring that user-provided file paths or metadata are not interpreted as shell commands.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted media files, which represents a potential attack surface. However, the risk is effectively mitigated through multiple layers of isolation. 1. Ingestion points: The skill reads video streams and metadata from target files and directories provided as arguments. 2. Boundary markers: The analysis results are returned to the agent as a structured JSON object, preventing raw file content from being misinterpreted as instructions. 3. Capability inventory: The skill is limited to media analysis tools and local file writes to a specific review/ subdirectory. 4. Sanitization: Inputs are never executed as code or interpreted as prompts; they are only analyzed for technical properties (e.g., luma values, decibel levels).

video-review