genmedia-image-artist

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill explicitly instructs the agent to perform a 'clinical rewrite' of prompts that trigger safety filters to bypass content restrictions while maintaining the physical description of the request.
  • [INDIRECT_PROMPT_INJECTION]: The skill accepts untrusted external data including PDF documents and video files as input for the image generation tool, which creates a surface for embedded instructions to influence the agent's output.
  • Ingestion points: The images, video, and pdf parameters in the nanobanana_image_generation tool are used to ingest external content (SKILL.md).
  • Boundary markers: No delimiters or instructions to ignore embedded content are present.
  • Capability inventory: Image generation and stylized refinement using the mcp_nanobanana_nanobanana_image_generation tool (SKILL.md).
  • Sanitization: No evidence of input sanitization or validation for multimodal data types.
  • [EXTERNAL_DOWNLOADS]: The skill references an official prompting guide hosted on Google Cloud's documentation domain.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 04:28 PM
Security Audit — agent-trust-hub — genmedia-image-artist