gemini-image-gen
Pass
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection by processing untrusted user content (text prompts and image files) and passing it directly to the Gemini API.
- Ingestion points: User-provided text strings and image files are ingested via the
contentsparameter inSKILL.mdand thepromptargument inscripts/generate.py. - Boundary markers: The skill does not employ specific delimiters or "ignore embedded instructions" warnings when interpolating user data into the model context.
- Capability inventory: The skill is granted
Writepermissions to the file system (saving images to./docs/assets/) and access to theBashtool, which could be leveraged if the model is successfully manipulated by malicious instructions embedded in input data. - Sanitization: No sanitization, escaping, or structural validation is performed on the user-supplied content before it is sent to the model.
- [EXTERNAL_DOWNLOADS]: The skill instructs the user to download and install the
google-genaiPython package from the official Python Package Index (PyPI). This is the official SDK from Google and is considered a trusted source.
Audit Metadata