One line in your terminal and your assistant starts choosing the right brain and effort for every task. Four ready-made helpers and one rule, set up for you in about ten seconds.
Press return. It installs the four helpers and the rule into your Claude Code setup, then tells you exactly what it did.
You need: Claude Code, on a Mac or Linux terminal. On Windows, use the by-hand steps below.
Grab the whole kit as a folder, then move the files in yourself. Nothing runs automatically.
Unzip it, then in your terminal:
mkdir -p ~/.claude/agents cp agents/*.md ~/.claude/agents/ cp MODEL_ROUTING.md ~/.claude/MODEL_ROUTING.md
Then add one line to ~/.claude/CLAUDE.md:
Before any substantial task, choose the model and effort using ~/.claude/MODEL_ROUTING.md.
Restart Claude Code, then run this to see the four helpers sitting in place:
ls ~/.claude/agents
You should see writer, builder, analyst and fast-worker. That is it. From now on your assistant matches the tool to the task on its own, and you can ask it to use a helper by name any time.
Two edits worth two minutes.
~/.claude/MODEL_ROUTING.md and set what counts as private for you. That rule sits above all the others, so your sensitive work never leaves your own machine.~/.claude/agents/writer.md and describe how you actually sound, so the writer speaks like you and not like a chatbot.rm ~/.claude/MODEL_ROUTING.md
rm ~/.claude/agents/{writer,builder,analyst,fast-worker}.md
Then delete the one routing line from ~/.claude/CLAUDE.md. Any file the installer replaced keeps a .bak copy next to it.