ai-automation-workflows
Pass
Audited by Gen Agent Trust Hub on Nov 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface where untrusted data could be used to override agent instructions.
- Ingestion points: The
data_processing.shscript reads arbitrary content from all.txtfiles in a directory, and theconditional_workflow.shscript processes input from command-line arguments. - Boundary markers: Prompt templates in the provided scripts (e.g., in
data_processing.shandconditional_workflow.sh) do not use delimiters or instructions to ignore embedded commands within the interpolated data. - Capability inventory: Scripts within the skill demonstrate the ability to execute CLI commands (
belt), perform network operations (curl), and write to the local file system. - Sanitization: There is no evidence of sanitization, escaping, or validation of the ingested content before it is passed into the AI model prompts.
Audit Metadata