nano-banana-2
Pass
Audited by Gen Agent Trust Hub on Aug 7, 2026
Risk Level: SAFE
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches installation guidelines and skill components from the vendor's official GitHub repository (inference-sh).
- [COMMAND_EXECUTION]: Utilizes the
beltCLI tool for model interaction. The skill's execution environment is restricted to this specific tool via theallowed-toolsconfiguration. - [PROMPT_INJECTION]: Processes user-provided text prompts and images for generative tasks. This represents a standard surface for indirect prompt injection common to generative AI tools, but uses structured JSON for input delivery.
- Ingestion points: User-provided
promptandimagesparameters inSKILL.mdexamples. - Boundary markers: Uses JSON structure (
--input '{"prompt": ...}') to delimit user input from command parameters. - Capability inventory: Uses the
beltcommand for remote model execution. - Sanitization: None explicitly defined in the instructions; relies on the underlying CLI and model provider's safety filters.
Audit Metadata