nano-banana
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process user-supplied descriptions and editing requests to generate and refine image prompts for Gemini models.
- Ingestion points: User input describing scenes, subjects, and modifications in
SKILL.mdandreferences/prompt-examples.md. - Boundary markers: The skill lacks explicit delimiters or instructions to isolate untrusted user content from the prompt structure.
- Capability inventory: The skill instructs the agent to generate API calls for Gemini models and suggests using the
google_searchtool for search-grounded imagery. - Sanitization: No evidence of input validation, escaping, or filtering of external content was found.
- [EXTERNAL_DOWNLOADS]: The skill references official documentation and development tools from a trusted source.
- Evidence: Direct links to
aistudio.google.comandai.google.devfor model specifications, API documentation, pricing, and deprecation schedules. Mentions of thegemini-apiCLI tool. - Context: These resources belong to Google and are used to provide technical verification and implementation details for the user.
- [NO_CODE]: The skill is composed of markdown instructional files, reference documents, and evaluation data. No executable scripts, binaries, or automated installation commands were found in the provided files.
Audit Metadata