Model Router for Hermes Agent

This Hermes skill adds intent-based model-routing guidance on top of stock Hermes. Hermes already supports auxiliary.*, delegation.*, fallback_model, custom providers, /model, and per-run --provider --model overrides. This skill does not add a Hermes core feature; it reads user-editable route settings from skills.config.model-router and returns the safest command or session-switch recommendation.

Load the routing table

Read ~/.hermes/config.yaml and inspect:

skills:
  config:
    model-router:
      default_route: routine_chat
      execution_mode: advise_or_oneshot
      routes: {}

If the table is absent, ask the user to add the sample from assets/config.yaml.routed. Do not invent a root intent_routes: key.

Route selection policy

Prefer native auxiliary.* slots for built-in fixed tasks. Use this skill only for semantic user intent. Match the user request against each route's keywords and intent. If multiple routes match, prefer the more specific and higher-capability route. If no route matches, use default_route.

Execution policy

Always state selected route, provider, model, rationale, fallback route, and exact command. Never use /model ... --global unless the user explicitly asks to persist the change.

Optional helper

When terminal access is available, run:

python scripts/model_router.py --prompt "<user request>" --config ~/.hermes/config.yaml

Use --execute only after the user confirms that an isolated hermes -z call is acceptable.