ai-product-photography
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 installation instructions and documentation hosted on a public GitHub repository (inference-sh/skills) and image assets from the service's domain (inference.sh). These references are used for standard configuration and user guidance.
- [COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute 'belt' CLI commands. This access is properly restricted in the skill metadata ('allowed-tools: Bash(belt *)'), ensuring the agent can only execute commands related to the intended service.
- [INDIRECT_PROMPT_INJECTION]: The skill demonstrates an attack surface for indirect prompt injection where user-supplied text is interpolated into shell commands.
- Ingestion points: User-defined product names and prompt descriptions within shell script examples.
- Boundary markers: Prompts are encapsulated within structured JSON strings in the CLI arguments, which mitigates simple command injection attempts.
- Capability inventory: The skill uses the 'belt' command to communicate with external AI models via an API.
- Sanitization: Uses JSON formatting for arguments, though explicit validation of the inner prompt text is handled by the downstream model provider.
Audit Metadata