ai-image-generation
Pass
Audited by Gen Agent Trust Hub on Jul 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the 'belt' CLI tool to perform image generation tasks. The instructions demonstrate subcommands such as 'belt login' for authentication, 'belt app store' for discovery, and 'belt app run' for model invocation. These operations are explicitly allowed within the skill's 'allowed-tools' configuration.
- [EXTERNAL_DOWNLOADS]: The skill provides installation instructions for the required CLI tools via a public GitHub repository (raw.githubusercontent.com). It also includes examples that reference external image URLs for processing and provides links to the service's official documentation.
- [DATA_EXFILTRATION]: User input, including text prompts and URLs for source images, is transmitted to the inference.sh hosted platform. This transmission is necessary for the AI models to process the requests and generate the desired visual output.
- [INDIRECT_PROMPT_INJECTION]: The skill ingest untrusted user input that is passed directly into command-line arguments for the 'belt' tool.
- Ingestion points: Prompts and image URLs supplied in the '--input' argument of 'belt app run' commands (SKILL.md).
- Boundary markers: JSON formatting is used to structure the input data.
- Capability inventory: Shell command execution via the 'belt' tool, which has broad access to the inference service.
- Sanitization: No explicit sanitization or filtering of user-supplied prompts or URLs is defined within the skill's instructions.
Audit Metadata