video-restyle
Pass
Audited by Gen Agent Trust Hub on Jul 21, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses local shell commands via
ffmpegandffprobefor video processing tasks including creating analysis proxies, cutting segments with frame accuracy, and conforming restyled clips to match source durations. These commands are integral to the stated purpose of video editing and are constructed using parameters derived from video analysis. - [EXTERNAL_DOWNLOADS]: The workflow requires downloading video files from user-provided URLs or local paths for processing and restyling. This behavior is documented and necessary for the skill's functionality.
- [PROMPT_INJECTION]: The skill includes instructions to the agent to prioritize identity preservation and lip-sync during visual transformation. It uses transcript data from the source video to anchor visual events in the generated prompts. This is a form of indirect injection surface as the agent is processing untrusted video data, but the risk is mitigated by the highly structured nature of the visual-only restyling task and the human-in-the-loop approval gate.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted video content and interpolates transcript data into prompts. Ingestion points: Video transcripts from source files. Boundary markers: Transcribed text is enclosed in double quotes. Capability inventory: Access to AI video generation and local command-line tools. Sanitization: The skill does not explicitly describe sanitization of the transcript text before interpolation.
Audit Metadata