higgsfield-seedance-vfx
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructions require the agent to ingest and process untrusted external data provided by the user to generate video prompts.
- Ingestion points:
SKILL.mdandreferences/dialogue-timing.mdinstruct the agent to inspect source clips, extract frames for context, and quote dialogue verbatim from user-provided media or transcripts. - Boundary markers: While the skill uses specific tags like
@sourceand@creatureto structure the generated prompt, it lacks explicit instructions for the agent to ignore or sanitize potentially malicious commands embedded within the user-provided video frames, audio, or text data. - Capability inventory: The skill produces complex instructions for the Seedance 2.0 engine and interacts with multiple other system skills (audio, camera, FACS) to perform its task.
- Sanitization: There is no evidence of sanitization, filtering, or validation of the content extracted from the source media before it is interpolated into the final output.
Audit Metadata