gpt-image
Pass
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references external installation instructions and documentation hosted on the service provider's official GitHub repository (inference-sh).
- [INDIRECT_PROMPT_INJECTION]: As an image generation tool, the skill naturally processes untrusted user prompts and external image URLs, which is an inherent surface for indirect prompt injection.
- Ingestion points: Processes user-supplied
prompt,images, andmaskparameters in thebelt app runcommand. - Boundary markers: No explicit delimiters are used in the command-line JSON payload to separate user content from instruction keys.
- Capability inventory: Executes the
beltCLI tool via shell to communicate with a remote inference API. - Sanitization: Input sanitization and safety filtering are handled by the destination API service (inference.sh).
Audit Metadata