ai-automation-workflows
Pass
Audited by Gen Agent Trust Hub on May 5, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The documentation references installation instructions hosted on a GitHub repository associated with the inference.sh service.
- [COMMAND_EXECUTION]: The skill provides Bash and Python script templates that perform system operations, including CLI execution, file system management, and cron job scheduling for automation.
- [PROMPT_INJECTION]: Example templates for data processing and conditional workflows ingest untrusted external data into LLM prompts, creating a surface for indirect prompt injection.
- Ingestion points: Command-line arguments ($1) in
conditional_workflow.shand local file contents (cat $file) indata_processing.sh. - Boundary markers: The templates do not implement delimiters or instructions to ignore embedded commands within the processed data.
- Capability inventory: The scripts utilize the
beltCLI, file system commands, and network utilities likecurl. - Sanitization: There is no evidence of input validation or sanitization in the provided examples.
- [DATA_EXFILTRATION]: A monitoring script example demonstrates how to send error logs to an external webhook URL using
curl.
Audit Metadata