I route Claude Code steps by decision type, not difficulty A developer built a Claude Code plugin called maddog that routes each step of a task to a subagent on a model chosen by the kind of decision the step requires rather than by task difficulty. In 49 advisor-mode sessions compared against 30 sessions without it, the developer found that routing by decision type reduced the share of output tokens coming from top-tier models such as Opus and Fable, though small changes cost more to dispatch than to perform directly and Haiku failed on tasks that only appeared mechanical. The plugin is MIT-licensed and available via the /plugin command in Claude Code. In my Claude Code sessions, 69% of the output came from the top models Opus, Fable , including reading a changelog, applying an edit I'd already decided on, and running tests. "Hard task → strong model" didn't fix it. Difficulty is the wrong signal: once a refactor is decided, applying it needs no judgment, however complex the code. What worked was routing by the kind of decision a step needs: I built this as advisor-mode in a plugin called maddog . Your main session splits the goal and sends each piece to a subagent a separate Claude instance with its own context on the right model. Only a short result comes back, and the main session checks it before accepting. Merges and pushes need your go-ahead. That's a rule the agents follow, backed by a hook in subagents, not a hard lock. My numbers, from my own 49 advisor-mode sessions against 30 sessions without it over the same weeks output tokens, not cost or quality : In my use, small changes cost more to dispatch than to just do, and Haiku fails on tasks that only look mechanical. The plugin also has section-by-section , which reviews a skill or agent file with you one section at a time. Video ~100 s : https://youtu.be/b mFbVY5jzk https://youtu.be/b mFbVY5jzk Try it: in Claude Code, run /plugin and search "maddog" on the Discover tab, then start with /maddog:advisor-mode