model-showdown
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides a comprehensive
ffmpegcommand for video processing, including scaling, cropping, and overlaying graphics to create a final comparison reel. - [INDIRECT_PROMPT_INJECTION]: The skill processes external data (briefs, headlines, and logos) which are interpolated into prompts for other models and rendered into video overlays.
- Ingestion points: User-supplied briefs, headlines, and model names; local
brand.mdandMOTION.mdfiles. - Boundary markers: Explicit instructions are provided to sub-models to ensure deterministic HTML output and to prevent external network requests.
- Capability inventory: Shell command execution via
ffmpeg, file reading/writing, and HTML rendering. - Sanitization: Textual inputs are rendered into an image overlay (
overlay.png) before being passed to shell tools, reducing the risk of shell injection.
Audit Metadata