Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
What This Skill Does
A model routing protocol that silently classifies every request into one of four tiers and dispatches work to the cheapest capable model via sessions_spawn. Applies strict output compression at every tier. Targets 75% or more reduction in token usage compared to running all tasks on a single premium model.
Automatically picks the cheapest model that can handle each task, so users don't overpay for simple work or under-provision for complex tasks.
When to use it
- Routing bulk CSV or list processing to free Groq instead of a paid model
- Automatically sending code generation tasks to Codex while handling yes/no questions inline
- Cutting response length on status checks to one or two lines
- Escalating multi-agent coordination tasks to Opus without manual model switching
- Parallelizing multiple tool calls to avoid sequential token overhead
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: Refactor this 200-line Python class to use dataclasses and add type hints
- 1Classifies request as T3 (code generation, multi-step)
- 2Spawns Codex via sessions_spawn with full file content and detailed spec in the task string
- 3Codex returns refactored file
- 4Orchestrator returns result directly with no prose wrap-up
- 5Appends token/tier/route metadata if measurement mode is on
Complete refactored Python file with dataclasses and type hints, no preamble