nano-banana-2-image-gen
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute local scripts (
scripts/generate_image.jsorscripts/generate_image.py) to perform the image generation. This is the primary functional mechanism and relies on standard shell execution. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input as the primary prompt for an external image generation model, which is a standard but inherent attack surface for this type of tool.
- Ingestion points: User-provided descriptions and instructions are ingested via command-line arguments in
SKILL.mdand processed by the scripts. - Boundary markers: The skill instructions recommend using the user's exact input as the prompt to maintain fidelity, without explicit delimiters for the model.
- Capability inventory: The scripts have capabilities to perform network POST requests to
api.apiyi.comand write image data to the local filesystem. - Sanitization: The scripts perform basic sanitization on the output filename to remove special characters, but the prompt content is sent as-is to the API provider.
Audit Metadata