CodepitChoose the model for every subagent
Before a subagent starts, the tuner sets its model and reasoning effort: by asking you, by a recommendation for the kind of task, or by a default you saved. A review can run on Opus while a file search doesn't.
Scan in CodepitScan checks all of Kruhlikau/subagent-tuner on your computer, before anything installs.
Codepit runs on your computer. Open this page there to scan it before installing.
Aleh KruhlikauMade by Kruhlikau
2Stars on GitHubStars
MITLicense
15 OctLast update
See it running
Real recordings from the author's repo.
What you get
A band above the prompt holds the mode and your saved default. Built-in agents get a recommendation by task: reviews and audits on Opus at high effort, builds and fixes on Sonnet at medium, searches on Haiku at low.
- Manual: asks first
- Auto: never asks
- Saved default
- Pick pane
- Recommendations by task
Before you start
- Claude Code 2.1.295 or later.
- A terminal 144 columns wide or more for the side pane. Narrower ones get a short chain of dialogs instead.
Try saying
In Manual, a pane asks for the model and effort before the subagent starts. In Auto, your saved default is applied and a short message says what was changed.
Worth knowing
It reads each Agent call's type, description, model, effort and prompt length, and changes only the model and effort fields. It never sends anything anywhere.
Manual is the mode after install, so every subagent launch asks first. An unanswered pane goes ahead with what it shows after ten minutes; switch to Auto before you leave a session to run on its own.
In Auto your saved default wins, even over a model a rule in CLAUDE.md told the agent to use. Custom agents with no saved default keep their own model and effort.
To wait for your answer in the pane it runs sh (macOS, Linux) or powershell.exe (Windows), and writes empty marker files in its own plugin folder.
