ai-automation-workflows

Warn

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: MEDIUMPERSISTENCEDATA_EXFILTRATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PERSISTENCE]: The skill provides detailed instructions and code examples for creating crontab entries to schedule the execution of AI workflows at regular intervals.
  • [DATA_EXFILTRATION]: The documentation includes patterns for sending workflow status and error reports to external webhooks (e.g., 'https://your-webhook.com/alert') using the curl command.
  • [COMMAND_EXECUTION]: The skill demonstrates multiple methods for orchestrating the 'belt' CLI tool through Bash and Python scripts, enabling complex operations including batch and parallel processing.
  • [INDIRECT_PROMPT_INJECTION]: Several workflow examples demonstrate the interpolation of untrusted data from the local file system or command-line arguments into AI model prompts.
  • Ingestion points: Untrusted data is ingested from command-line arguments in 'conditional_workflow.sh' and from file contents via 'cat' in 'data_processing.sh'.
  • Boundary markers: The scripts lack delimiters or protective instructions that would prevent the AI model from interpreting commands embedded within the input data.
  • Capability inventory: The skill possesses the ability to execute shell commands, perform network requests, and interact with various AI services through the 'belt' tool.
  • Sanitization: There is no evidence of input validation, filtering, or escaping before the external data is passed to the AI models.
  • [EXTERNAL_DOWNLOADS]: The skill references the installation of the 'belt' CLI and related automation skills from the vendor's repository using npx and markdown links.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 19, 2026, 10:24 AM
Security Audit — agent-trust-hub — ai-automation-workflows