collage-motion
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill analyzes user-provided reference images to generate detailed natural language prompts for image and video generation tools. This ingestion of untrusted visual data presents a potential surface for indirect prompt injection.
- Ingestion points: User-supplied reference images (SKILL.md, PHASE A).
- Boundary markers: None specified for the analysis of image content.
- Capability inventory: Utilization of image and video generation tools and file upload mechanisms.
- Sanitization: No explicit sanitization of extracted content is performed.
- [COMMAND_EXECUTION]: The skill instructs the agent to perform a PUT operation to upload local files to a presigned URL as part of the asset registration workflow.
- [EXTERNAL_DOWNLOADS]: The skill utilizes a dynamic search mechanism to load tool schemas from a remote connector at runtime.
Audit Metadata