delegatelisted

# Delegate - Route Bounded Sub-Tasks to a Local Ollama Model Help a controller agent decide whether to hand a **narrow, well-scoped, reviewable** sub-task to a LOCAL Ollama model, or keep it with the controller - then route it when appropriate. Local delegation is for cheap, bounded transforms on material the controller **has already fetched**. The controller always stays in charge and always verifies the result. This skill is **MCP-aware but MCP-optional**. The active-routing path depends on the `ollama-telemetry` MCP server (server name `ollama-delegate`) being registered in this host. When it is not registered, or when the fleet is degraded, the skill falls back to **static guidance** and tells the controller to do the task itself. **MCP tools used (when registered):** `ollama_usage_guidance`, `ollama_readiness`, `ollama_delegate`, `ollama_batch_delegate`, and optionally `ollama_fleet_status`. --- ## The Delegation Philosophy Use Ollama as a **bounded local helper**, not as the agent in charge. The normal loop: 1. Controller decides what to inspect or fetch. 2. Controller gathers the files, logs, API output, or web results. 3. Ollama performs one narrow transformation on that material. 4. Controller **verifies** the result and makes the real decision. `fetch`, `plan`, and `decide` always stay with the controller. Local delegation is never for architecture, security, deployment, rollback, arithmetic, code tracing, or multi-step reasoning. --- ## Commands | Comman