nano-banana-2
Pass
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references the installation of the
belt-sh/clitool and theinferenceshPython package, which are necessary for the skill's core functionality of interacting with the inference.sh API. - [COMMAND_EXECUTION]: The instructions utilize the
beltCLI to run image generation apps. The skill's configuration limits the agent's shell access specifically to this command. - [INDIRECT_PROMPT_INJECTION]: The skill has a data ingestion surface where it processes user-provided prompts and image URLs.
- Ingestion points: The
promptandimagesinput parameters in thebelt app runcommand and the Python SDK methods. - Boundary markers: Not present.
- Capability inventory: Execution of the
beltcommand and network access to the inference service. - Sanitization: No explicit sanitization of input text or image links is described in the provided documentation.
Audit Metadata