gemini-imagegen

Pass

Audited by Gen Agent Trust Hub on Feb 20, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [Indirect Prompt Injection] (LOW): The skill ingests untrusted user content which may contain instructions targeting the model or agent. \n
  • Ingestion points: User prompts passed to the contents parameter and local image files (e.g., input.jpg) read via Path.read_bytes(). \n
  • Boundary markers: None present; user input is interpolated directly into model requests. \n
  • Capability inventory: The skill writes generated image data to the local file system using open() in write-binary mode. \n
  • Sanitization: No explicit sanitization or validation of input data is performed, relying instead on API-side safety filters. \n- [External Downloads] (LOW): The skill suggests installing the google-genai package. Per [TRUST-SCOPE-RULE], this finding is downgraded to LOW/INFO as 'google' is a Trusted External Source. \n- [Data Exposure & Exfiltration] (SAFE): Sensitive credentials like GEMINI_API_KEY are managed via environment variables rather than being hardcoded or requested in cleartext.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 20, 2026, 04:59 PM