ai-automation-workflows
Pass
Audited by Gen Agent Trust Hub on May 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides numerous Bash and Python script templates designed to execute the 'belt' CLI tool for AI operations. These examples include piping model outputs to files, running jobs in parallel, and implementing retry logic.
- [EXTERNAL_DOWNLOADS]: The skill references the installation of the 'belt' CLI and related automation skills from the 'inference-sh' and 'belt-sh' GitHub organizations. These resources are provided by the vendor to facilitate the skill's primary functionality.
- [PROMPT_INJECTION]: The skill demonstrates patterns for interpolating untrusted data from local files (in 'data_processing.sh') or command-line arguments (in 'conditional_workflow.sh') directly into AI prompts. This creates a surface for indirect prompt injection where malicious content in the processed data could influence the AI model's output.
- Ingestion points: 'data_processing.sh' reads file content from './data/raw/'; 'conditional_workflow.sh' reads from the first command-line argument ($1).
- Boundary markers: Absent; external data is embedded directly into the prompt string within the JSON input.
- Capability inventory: The scripts possess the ability to execute AI models via the 'belt' CLI and write output to the local file system.
- Sanitization: No input validation, escaping, or sanitization of the external content is included in the provided templates.
Audit Metadata