claw-router

name	claw-router
description	Trait-based intelligent model routing — automatically routes messages to the best AI model

Smart Router Skill

What It Does

Automatically selects the optimal AI model for each message based on complexity tier and task type, matched against user-declared model trait profiles.

How It Works

Message → 8-dimension scoring → Tier (complexity)
        → Keyword classification → TaskType (scenario)
        → Extract traits [Tier, TaskType]
        → Score each model's traits against message traits
        → Select best match (LLM arbitration if tied)

Complexity Tiers

Tier	Scenario
TRIVIAL	Greetings, confirmations, emoji
SIMPLE	One-step Q&A, translations, lookups
MODERATE	Writing, comparisons, summaries
COMPLEX	Coding, debugging, multi-step analysis
EXPERT	System design, formal proofs, architecture

Task Types

Type	Scenario
coding	Programming, debugging, code generation
writing	Writing, creative content, copywriting
chat	Casual conversation, greetings
analysis	Reasoning, comparison, evaluation
translation	Translation
math	Mathematics, formulas, calculations
research	Research, papers, literature review
other	Unclassified (default fallback)

Configuration

Declare a models array in OpenClaw plugin config. Each model specifies an id and its traits:

{
  "models": [
    {
      "id": "anthropic/claude-sonnet",
      "traits": ["coding", "analysis", "COMPLEX", "EXPERT"]
    },
    {
      "id": "openai/gpt-4o-mini",
      "traits": ["chat", "translation", "TRIVIAL", "SIMPLE"]
    },
    {
      "id": "google/gemini-pro",
      "traits": ["writing", "research", "MODERATE", "COMPLEX"]
    }
  ]
}

Trait vocabulary (fixed):

• Tier: TRIVIAL / SIMPLE / MODERATE / COMPLEX / EXPERT
• TaskType: coding / writing / chat / analysis / translation / math / research / other

When models is not configured, all messages use the default model (no routing).

LLM Assist (Optional)

When enabled, LLM is invoked in two scenarios:

1. Score boundary: Rule-based score falls near a tier threshold — LLM refines the complexity judgment
2. Model arbitration: Multiple models tie on match score — LLM picks the best one

{
  "llmScoring": {
    "enabled": true,
    "model": "deepseek-ai/DeepSeek-V3-Chat",
    "apiKey": "sk-xxx",
    "baseUrl": "https://api.siliconflow.cn"
  }
}

Commands

• /route — Show routing status and stats
• /route <message> — Test-route a message
• Agent tool claw_route — Programmatic routing query

Running Tests

npx tsx --test test/engine.test.ts test/scorer.test.ts test/task-classifier.test.ts test/model-matcher.test.ts