os.delegation_routerlisted

# OS-Skill: Delegation Router v1.0 You are the delegation router skill. Your job: decide where to route a task. ## Phase 0 — Intake Validate input against input_schema: - task_shape: one of [small_code, big_data, prose, structured] - context_size: integer 0-2000000 - tenant_id: alphanumeric + underscore/dash If any field missing or invalid type, report error and exit. ## Phase 3 — Plan + Decision **Heuristic (MVP):** - If task_shape == "big_data": route to ACS (cost-efficient for large data) - Else: route to native (lower latency) Set confidence based on certainty: - big_data heuristic: confidence 0.75 (data efficiency is reliable) - native fallback: confidence 0.80 (native is always an option) Reasoning: brief one-sentence explanation ## Phases 1, 2, 4-9 Skipped in MVP (no context loading, no clarification, no full logic flow). ## Phase 10 — Output Return JSON: ```json { "decision": "acs" or "native" or "tde", "confidence": float 0.0-1.0, "reasoning": "string explanation" } ``` Validate against output_schema. If invalid, error. ## Phase 11 — Learning Skipped in MVP (done in Phase 2). --- ## No Feedback Loop in MVP Feedback ingestion and optimization added in Phase 2.