The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided proposals, plan drafts, or directions to be processed by multiple independent agent perspectives. This represents an ingestion point for untrusted data. The risk is mitigated by a multi-round procedural structure (independent findings, cross-attack, and distillation) which ensures the input is scrutinized before being distilled into a structured bundle. The output is strictly formatted into Hard Constraints, Decisions, Risks, and Open Questions, intended as input for a separate planning step rather than for direct execution.
[COMMAND_EXECUTION]: The skill utilizes the platform's omh CLI tool (e.g., omh hermes plan --record) to store planning artifacts in a local .omh/plans/ directory within the repository. This is standard operational behavior for the Oh-My-Hermes ecosystem to maintain plan history and does not involve arbitrary or unsanitized shell execution from external sources.
[SAFE]: The skill demonstrates safe practices by enforcing separation of duties between the 'attacker' roles and the 'planner' roles. No obfuscation, unauthorized network operations, credential harvesting, or persistence mechanisms were detected. The metadata and workflow instructions are consistent with the stated purpose of multi-perspective planning and red-teaming.

omh-adversarial-consensus