nano-banana-build
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill provides links to the official Google GenAI Python SDK source code for deep inspection of model parameters.
- Evidence: Links to the
googleapis/python-genairepository on GitHub are included inreferences/source_code.md. - [INDIRECT_PROMPT_INJECTION]: The skill includes functionality to process external data sources, such as reference images and Google Search results, which could serve as vectors for indirect instructions.
- Ingestion points: Reference images are ingested via local file reads and Google Cloud Storage URIs in
references/image_editing.mdandreferences/recipes.md. Real-time data is ingested through the Google Search tool inreferences/recipes.md. - Boundary markers: The provided code examples do not explicitly define delimiters for untrusted content.
- Capability inventory: The skill has the capability to read local files, interact with the Gemini API for image generation, and perform web searches through the SDK.
- Sanitization: The skill relies on the underlying Gemini model's safety filters and the SDK's internal handling of multi-modal inputs.
Audit Metadata