nano-banana
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided descriptions to create image generation prompts, which constitutes an attack surface where untrusted input could attempt to manipulate the agent or the image model.
- Ingestion points: User input fields for image generation and editing requests described in SKILL.md and references/social-recipes.md.
- Boundary markers: Lacks formal delimiters between user-supplied content and the prompt logic, relying on natural language reasoning.
- Capability inventory: Accesses local configuration files (brand-profile.md) and references tools for media upload and post scheduling.
- Sanitization: Incorporates specific pushback instructions to decline unsafe or non-compliant content and requires human verification of all rendered text and data before publication.
- [SAFE]: The skill adheres to security best practices for AI agents by emphasizing transparent AI disclosure, SynthID watermarking, and clear separation of concerns between prompt generation and API/integration logic.
Audit Metadata