gpt-image
Pass
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references documentation and installation scripts hosted on GitHub (inference-sh organization) and the inference.sh domain. These references are used for setting up the required CLI environment.
- [COMMAND_EXECUTION]: The skill utilizes the
beltcommand-line utility via a restricted Bash environment (allowed-tools: Bash(belt *)) to run image generation tasks. This prevents the execution of unauthorized shell commands. - [INDIRECT_PROMPT_INJECTION]: The skill exposes an attack surface by accepting untrusted user data (prompts and image URLs) that are processed by the underlying AI model.
- Ingestion points: User-supplied
prompt,images, andmaskfields in SKILL.md. - Boundary markers: Data is passed as structured JSON via the
--inputflag to the CLI tool. - Capability inventory: Command execution is limited to the
belttool; no file writing or broad network access is granted to the skill itself. - Sanitization: Input is wrapped in single quotes for the shell command, mitigating basic command injection.
Audit Metadata