gpt-image-1-5
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied prompts and local image files which are then transmitted to external AI models (OpenAI) for generation and editing. This represents an indirect prompt injection surface where instructions embedded in images or crafted prompts could influence model behavior.
- Ingestion points: User-provided strings via the
--promptargument and image files via the--input-imageand--maskarguments inscripts/generate_image.py. - Boundary markers: None explicitly implemented within the script's prompt handling; the skill relies on the target API's (OpenAI) internal safety guardrails.
- Capability inventory: The skill performs network requests to OpenAI's API and writes generated image files to the local file system using the
PILlibrary. - Sanitization: No content validation or sanitization is performed on the prompt text or image metadata before processing.
- [SAFE]: The skill uses standard libraries (OpenAI, Pillow) and communicates exclusively with a well-known service (OpenAI). The use of command-line arguments for API keys is a documented feature for user convenience and does not involve hardcoded credentials.
Audit Metadata