omh-product-discovery-validation
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the vendor's own CLI tool to build metadata from user inputs.
- Evidence: The instruction
omh runtime workflow-artifact product-discovery-validation build --input <json-file-or->is found inSKILL.mdto handle the generation of workflow artifacts. - [INDIRECT_PROMPT_INJECTION]: The skill defines a complex workflow for ingesting and processing untrusted customer data, which presents an attack surface for indirect instructions.
- Ingestion points: Input captured via the
omhtool's--inputflag and hypothesis data defined inSKILL.mdandreferences/procedure.md. - Boundary markers: The
SKILL.mdinstructions explicitly tell the agent to distinguish between 'user's own words' and the 'host's own voice', maintaining OMH's record terms separately. - Capability inventory: The agent's capabilities are restricted to generating planning metadata via the
omhCLI. The skill's safety rules inreferences/full-contract.mdexplicitly forbid participant recruitment, surveys, payments, and code execution. - Sanitization: The workflow uses a series of logical check procedures (e.g.,
discovery_evidence_class_check) to classify and validate data direction and confidence before it is committed to the workflow ledger.
Audit Metadata