seedance-2-deep-dive
Pass
Audited by Gen Agent Trust Hub on Jun 19, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: No malicious behavior, obfuscation, or unauthorized data access patterns were detected. The skill instructions are consistent with its stated purpose of assisting in video generation planning.
- [PROMPT_INJECTION]: The skill defines a workflow for processing external multimodal references (images, video, and audio). While this constitutes a data ingestion surface, the instructions encourage the use of explicit purpose declarations and structured referencing (e.g., using '@Image1') which acts as a boundary for how the agent processes external content.
- Ingestion points: Up to 12 external media files as references (SKILL.md).
- Boundary markers: Use of media-role syntax and purpose-driven prompting.
- Capability inventory: Submission of video generation jobs via the Higgsfield MCP.
- Sanitization: No specific validation logic is present within the skill instructions.
Audit Metadata