video-review-loop
Pass
Audited by Gen Agent Trust Hub on Oct 2, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill's Python scripts (review_video.py and strict_video_signals.py) execute system binaries including ffmpeg and ffprobe via subprocess.run. These calls use list-based argument arrays rather than shell strings, which is a secure practice that prevents command injection vulnerabilities.
- [DYNAMIC_EXECUTION]: The scripts utilize importlib.util.spec_from_file_location to load gpu_policy.py and reference_fidelity.py from a sibling directory (motion-director). While this is dynamic code loading, the paths are hardcoded relative to the skill's own installation directory to facilitate code sharing between related tools, and they do not incorporate untrusted external input into the loading path.
- [INDIRECT_PROMPT_INJECTION]: The skill processes external video files (MP4, MOV) and parses their metadata via ffprobe to generate JSON reports for the agent. This represents an indirect prompt injection surface where malformed media metadata could theoretically attempt to influence the agent's behavior. The risk is minimized by the use of structured JSON output and objective technical reporting.
Audit Metadata