creativeclaw-gemini-omni
Pass
Audited by Gen Agent Trust Hub on Sep 11, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill defines a dependency on a Model Context Protocol (MCP) tool located at
https://app.creativeclaw.co/mcp/chatgpt. This endpoint is a vendor-owned resource used to provide media generation and feedback capabilities. - [INDIRECT_PROMPT_INJECTION]: The skill functions by ingesting external media, creating a surface for indirect prompt injection through data interpolation.
- Ingestion points: The workflow processes external data via
image_url,image_urls, andvideo_urlsfields within theSKILL.mdinstructions. - Boundary markers: The skill employs a specialized reference syntax (e.g.,
<IMAGE_REF_N>) to identify and delimit specific assets within the generation prompt. - Capability inventory: The skill utilizes tools for fetching model parameters (
get_model_params), generating video content (generate_video), monitoring job status (check_job), and submitting feedback (submit_feedback). - Sanitization: The documentation instructs users to set
agentic_prompting: falsewhen specific tokens or authored constraints are present, which serves as a mitigation to prevent the model from re-interpreting or overriding the provided instructions.
Audit Metadata