omh-model-setup
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to interact with the host system using specific CLI utilities, including
omh model-chains set,hermes auth add, andomh runtime record. These are used for updating model routing preferences and recording workflow events. - [INDIRECT_PROMPT_INJECTION]: The skill processes external data from local filesystem sources which could theoretically be influenced by untrusted content.
- Ingestion points: Configuration data is ingested from
~/.omh/routing/model-chains.json,~/.omh/routing/providers.json, and general session metadata stores. - Boundary markers: The instructions define a strict scope, requiring the agent to only read 'allowlisted Hermes config metadata' and explicitly forbidding the emission or reading of transcript prose or tool results from these sources.
- Capability inventory: The skill allows for the modification of local configuration files and the execution of authentication-related commands through the
omhandhermesbinaries. - Sanitization: Security is managed through the use of metadata allowlists and a requirement for 'digest-checked' diff approval, where the exact configuration change must be previewed and approved by the user before being applied.
Audit Metadata