| name | orchestrator-overrides |
| description | Index the EPYC orchestrator x_* request overrides exposed through Hermes command skills. |
| version | 1.0.0 |
| metadata | {"hermes":{"tags":["routing","orchestrator","overview"]}} |
Orchestrator Overrides
Use this as the entry point for Hermes commands that shape EPYC orchestrator
routing and stack lifecycle control.
Override Inventory
| API field | Type | Effect | Consumed by |
|---|
x_orchestrator_role | string or null | Force a role from GET /v1/models and bypass frontdoor routing. | /use |
x_max_escalation | string or null | Cap escalation at A, B1, B2, or C. | /escalation |
x_force_model | string or null | Force one registry model and bypass role routing. Wins over x_orchestrator_role. | No dedicated command; advanced API override only |
x_disable_repl | boolean | Skip REPL execution and request direct text. | /nocode |
x_show_routing | boolean | Include routing metadata in the response for debugging. | No dedicated command; advanced API override only |
Command Skills
| Command skill | Primary field | Use when |
|---|
/use | x_orchestrator_role | A conversation should stay on a known current role. |
/escalation | x_max_escalation | A conversation needs a latency or quality ceiling. |
/nocode | x_disable_repl | A conversation should avoid REPL/tool execution. |
/stack | local launcher | A conversation needs direct stack status/start/stop/reload control. |
Drift Check
Keep this inventory aligned with OpenAIChatRequest:
python3 scripts/hermes/skills/check_drift.py
The checker compares the request-side x_* fields declared in
/mnt/raid0/llm/epyc-orchestrator/src/api/models/openai.py with the documented
fields under scripts/hermes/skills/.