ai-image-generation
Pass
Audited by Gen Agent Trust Hub on Aug 28, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the 'belt' CLI tool to interface with remote AI models. It utilizes the 'allowed-tools' configuration to restrict the agent's Bash capabilities exclusively to the 'belt' command, adhering to the principle of least privilege.
- [EXTERNAL_DOWNLOADS]: Fetches installation instructions from a GitHub repository (raw.githubusercontent.com/inference-sh/skills). This is a standard practice for distributing documentation and setup scripts from a well-known service.
- [INDIRECT_PROMPT_INJECTION]: The skill acts as a surface for processing user-defined prompts for image generation. While it lacks explicit boundary markers or sanitization within the instructions, this is a functional requirement for image generation models and the risk is mitigated by the scope of the 'belt' tool.
- Ingestion points: User prompts are interpolated into the 'belt app run' commands in SKILL.md.
- Boundary markers: None identified in the instruction templates.
- Capability inventory: Limited to executing the 'belt' CLI tool via Bash.
- Sanitization: Not present in the skill instructions; sanitization is expected to be handled by the 'belt' CLI or the inference service provider.
Audit Metadata