The Agent Skills Directory

[SAFE]: The skill provides structured documentation and instructions for a standard development workflow involving model onboarding. The requested tools (Bash, Write, Edit) are consistent with the stated purpose of updating configuration files and running local benchmarks.
[COMMAND_EXECUTION]: The skill uses local shell commands such as cat and git grep, as well as project-specific CLI tools like omh and hermes. These commands are used to inspect the codebase, manage routing configuration, and perform token usage measurements within the project scope.
[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read external research data stored in labeled dossiers under .omc/research/ (official docs, vendor harnesses, etc.). While this represents an ingestion of untrusted data, the workflow focuses on extracting specific technical parameters for model calibration rather than executing instructions contained within those dossiers. The risk is minimized by the highly technical and specific nature of the onboarding tasks performed by the agent.

model-onboarding