nano-banana-pro
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill utilizes the official
google-genailibrary to interact with external AI services, ensuring that network communication is handled by a reputable and maintained SDK.\n- [SAFE]: The implementation of secret management is consistent with best practices; it supports retrieving the API key from command-line arguments or environment variables and correctly looks for a local.envfile within the skill's own directory to prevent unauthorized access to other workspace secrets.\n- [SAFE]: File system access is scoped to the user's current working directory for saving outputs and only reads specific files when explicitly passed via the--input-imageparameter.\n- [INDIRECT_PROMPT_INJECTION]: The skill possesses an ingestion surface for indirect prompt injection, as is typical for AI-integrated tools.\n - Ingestion points: User-defined text descriptions (
--prompt) and referenced image files (--input-image) are processed byscripts/generate_image.py.\n - Boundary markers: Absent; the skill passes the raw data to the model generation API without additional delimiters.\n
- Capability inventory: The tool has the capability to read local files and perform network requests to the Gemini API.\n
- Sanitization: Absent; the skill relies on the remote API service's internal safety guardrails to mitigate malicious payloads in the input data.
Audit Metadata