image-generator
Pass
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill utilizes official and widely-trusted libraries including google-genai and pillow for Python, and @google/genai for Node.js.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user prompts and images as inputs to the generative model. 1. Ingestion points: Data is passed to the contents parameter of the generate_content API. 2. Boundary markers: No explicit delimiters are used to isolate user input. 3. Capability inventory: Includes file writing and network access to Google APIs. 4. Sanitization: No input sanitization is performed.
- [METADATA_POISONING]: Metadata fields contain illustrative or placeholder information, such as a futuristic date (2026) and a satirical model name (Nano Banana Pro), and self-labels the risk as critical. These do not impact the functional safety of the skill code.
- [SAFE]: The skill implements robust security controls, including the use of getpass for sensitive credentials, the xb file opening mode to prevent symlink attacks or overwrites, and explicit checks on input image file sizes and pixel dimensions.
Audit Metadata