Model Choice Router

You checked your credit usage last night. 74% gone and it's only Wednesday. You look at the logs: your agent used GPT-5.4 to rename a file. It used the premium model to write a one-line commit message. It ran a $0.12 query to check if a directory exists. Meanwhile, the one task that actually needed premium reasoning — a complex debugging session — failed because you hit the rate limit from all the wasted calls. This is how most OpenClaw users bleed credits: not from one expensive task, but from hundreds of cheap tasks routed to expensive models.

The problem isn't your agent being wasteful on purpose. It's that there are no routing rules. Every task gets the default model, regardless of complexity, risk, or budget. Model Choice Router gives your agent a production-tested routing discipline for choosing the right model tier by task type, risk level, and remaining budget. It covers premium vs mini vs nano decisions with clear rules for each, fallback chains for rate limits or outages, hard boundaries for when weak models must never be used (security audits, financial calculations, complex debugging), stop conditions when credits drop below safe thresholds, and concrete routing examples for build work, research, briefings, maintenance, and emergency fallback mode. Install it once, and your credits last the full week instead of running dry by Wednesday.

What's Included

• Complete SKILL.md with routing discipline and decision framework
• Routing matrix organized by task type, risk level, and complexity
• Quality-risk rules — which tasks require premium and which never should
• Fallback chain policy for rate limits, outages, and model unavailability
• Budget threshold system with stop conditions at configurable credit levels
• Emergency fallback mode rules for operating below 10% credits
• Example routing decisions showing the framework applied to real tasks
• Model tier definitions with cost-per-query benchmarks

Why This is Better

• OpenClaw-specific — not generic API cost optimization theory
• Built from operating a real production stack where credit waste was measured daily
• Balances quality, cost, and failure recovery — not just "use the cheap model"
• Includes hard rules for when premium is required, preventing quality failures on critical tasks
• Stop conditions prevent the worst outcome: running out of credits mid-week with no fallback
• Pairs naturally with Skill Stack Router — right skill AND right model for every task
• Works on first install with zero configuration