A drop-in config that routes each Claude Code subagent launch to the model that should handle it, so you stop overpaying for simple jobs and underpowering the hard ones. Install it once and your agents get faster and sharper, right-sized to each job.
Drops into your global CLAUDE.md and settings. Nothing breaks, nothing moves.
A pre-tool hook blocks the wrong model before it runs. You never have to remember the rule.
Leave your name and email. We send the install link and quick-start steps to your inbox.
Three pieces that work together: a doctrine that classifies tasks, a hook that blocks the wrong model, and a test you watch run in real time before you ship anything.
A model-routing doctrine in your CLAUDE.md classifies every agent task and picks the correct tier before the agent spawns.
A PreToolUse hook intercepts every Agent call. If the model is missing or banned, the call stops. You see why.
After install, a test run triggers the gate on purpose. You see it block the wrong model and pass the right one.
No. You paste one prompt and your own Claude fetches the kit, checks it, and walks you through every step. The routing doctrine and the hook go in exactly the right places for your setup.
Only if they were spawning agents without an explicit model. The hook blocks those calls and tells you exactly what to fix. Agents that already set the model field work without changes.
The kit ships with a two-tier table: reasoning and synthesis tasks go to the large model, retrieval and mechanical tasks go to the mid-tier model. The table is plain text in your CLAUDE.md, so you edit it directly.
Yes, in under a minute. The install doc lists the exact files to delete, and your Claude reads the list before touching anything.
Leave your email above and we send the link. Free, five minutes, removable any time.
Get the free link →