intelligent-router

What This Skill Does

The intelligent-router is a critical infrastructure component for OpenClaw that automates model selection for sub-agent tasks. By evaluating task complexity—ranging from SIMPLE maintenance tasks to CRITICAL security operations—the router assigns the most cost-effective and capable model available. This implementation replaces inefficient default behaviors, such as using expensive top-tier models (Sonnet) for trivial tasks, effectively reducing infrastructure expenditure by 80-95% without compromising output quality.

Installation

The intelligent-router is installed via the OpenClaw skill ecosystem. Run the following command in your terminal to activate the routing policy enforcer and scripts: clawhub install openclaw/skills/skills/bowen31337/intelligent-router After installation, run bash skills/intelligent-router/install.sh to configure the router_policy.py enforcer, which prevents non-compliant agent payloads from executing.

Use Cases

• Automated Monitoring: Route heartbeat checks and status reporting to low-cost, high-speed local models.
• Production Engineering: Ensure high-stakes tasks or production patching utilizes the CRITICAL tier (Claude Opus) to mitigate risk.
• Research & Development: Direct complex architecture planning and multi-file debugging to intelligent models like Claude Sonnet.
• Cost Governance: Automatically block unauthorized or high-cost model usage for background cron jobs to prevent budget overrun.

Example Prompts

1. "Classify the following task: 'Run a 24/7 background heartbeat check on the primary API endpoint.'"
2. "Generate the payload for a new cron job that monitors system disk usage, optimized for the cheapest available model."
3. "Validate this agentTurn JSON payload to ensure it adheres to the router's model recommendations: {"kind":"agentTurn","message":"Refactor the auth middleware"}"

Tips & Limitations

• Avoid Hardcoding: Never hardcode models into your agent payloads. Always use the spawn_helper.py script to generate the appropriate model ID dynamically.
• Understand the Hierarchy: The router balances effective_params, context_window, and cost_input. If you are performing a complex proof or formal logic task, use the REASONING tier to leverage models with chain-of-thought capabilities.
• Blocking Rules: Note that the ollama-gpu-server is strictly blocked for network reasons; the enforcer will reject these assignments automatically. Always verify your payload using the validation tool before deployment.