I’m a big fan of worktrees when working with agents, and while I know Cursor does support worktrees to an extent, I am unsure if I understand how they are best utilized.
I am a big fan of the way Conductor handles worktrees, I create a worktree that I can use for an entire feature, and then create a PR from that worktree. As far as I know, within Cursor worktrees are meant to be more of a sandbox for individual chat sessions - I create a worktree, let the agent run, and then I approve changes into my main branch.
Is the Conductor-like workflow at all possible right now, or is it not currently supported? Obviously, I can setup a script or something to handle this, but reducing the friction would be ideal
Worktrees are a powerful part of the Cursor workflow, so I get why the Conductor comparison comes up.
Today, local worktrees are primarily session-scoped sandboxes for parallel agents. Each local agent gets its own worktree so it can make changes without conflicting with your working directory or other agents. The usual flow is: agent runs in a worktree → you review → you apply changes back to your branch (or discard). These worktrees are auto-cleaned as limits are hit (default ~20) and aren’t meant to persist across sessions.
If you’re looking for the “one feature = one branch = one PR” model, Cloud Agents are the closest match to Conductor. They clone your repo from GitHub/GitLab, work on a separate branch, and can open merge-ready PRs—without touching your local environment. You can launch them from the IDE (Cmd/Ctrl+E), the web, Slack, or Linear, and monitor progress remotely.
If you want persistent per-feature work locally, a few workable options:
Manual git worktree + separate Cursor workspace: create a worktree (git worktree add ...) and open it as its own workspace. Agents will work there normally; Cursor just won’t manage it via the Parallel Agents system.
Save plans to .cursor/plans/: persist agent plans so you can resume work in a new session even if the original worktree is gone.
Setup scripts via .cursor/worktrees.json: configure per-worktree setup (e.g., npm install) to run automatically when a worktree is created.
There aren’t publicly announced plans for persistent per-feature worktrees in local agent mode right now, but we’re actively iterating on this. If this would be valuable for your workflow, a feature request in Feature Requests is the best way to signal demand.
