donimas/ai-routing ? reverse-engineered prompt

Build me a small starter kit for AI model routing in Claude Code, something that helps me stop using the biggest model for every task and instead pick the cheapest model that still gets the job done.

I want a simple setup I can copy into another repo, with a main Claude Code config, a few routed subagents for things like searching, editing, testing, and reviewing, plus a clear guide in the repo that explains when each one should be used. Include an optional safety check that blocks any subagent launch unless a model is explicitly set, and make it easy to turn that on or off.

Also include a tiny command line router example that takes a task description, classifies it, then sends it to the right model and effort level, plus an offline test for it that runs without an API key. Keep the instructions practical and real, like something I could follow in about 10 minutes. If you need to check current Claude Code docs or model names, look them up online first.