omh-adversarial-consensus
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill implements a workflow for multi-perspective technical review and planning, focusing on structured analysis and independent critique.
- [COMMAND_EXECUTION]: Employs platform-specific CLI commands (
omh hermes plan,omh runtime record) for workflow state management and recording plan inputs. These tools are documented as native parts of the Hermes environment and do not perform arbitrary shell execution or exfiltration. - [INDIRECT_PROMPT_INJECTION]: The skill ingests user-supplied technical proposals for review, presenting an attack surface for indirect prompt injection.
- Ingestion points: User-provided technical proposals, plan drafts, or architectural directions.
- Boundary markers: The protocol enforces independent rounds and utilizes subagent delegation (
delegate_task) to isolate perspectives from each other. - Capability inventory: No general subprocess execution; limited to platform recording tools and workflow metadata management.
- Sanitization: Results are processed through a strict distillation phase that maps findings to a closed set of predefined buckets (Hard Constraints, Decisions, Risks, Open Questions), preventing unvalidated content from escalating beyond planned constraints.
Audit Metadata