Worktrees in Codex, Cursor and Claude Code compared

Compiled from each tool’s official documentation as of November 2026: Claude Code, Codex, Cursor. These features change often.

At a glance

	Claude Code	Codex (ChatGPT desktop app)	Cursor
How to start	claude --worktree <name>, or ask Claude mid-session	Pick “Worktree” under the composer	Agents Window; /worktree in the IDE
Location	.claude/worktrees/<name>/ in the repo	$CODEX_HOME/worktrees (configurable)	Managed by Cursor
Branch	New branch worktree-<name>	Detached HEAD; “Create branch here” when ready	Separate checkout per agent
Starting point	Remote default branch (or local HEAD via setting)	Branch you choose, including uncommitted changes	Your project
Copy .env	.worktreeinclude	.worktreeinclude	Setup commands, e.g. cp $ROOT_WORKTREE_PATH/.env .env
Install dependencies	Ask Claude or run it yourself	Local environment setup script	setup-worktree in .cursor/worktrees.json
Cleanup	Prompt on exit; clean unnamed sessions removed automatically	Keeps the 15 most recent by default, snapshot before deleting	/delete-worktree

Claude Code

Run claude --worktree feature-auth and Claude Code creates .claude/worktrees/feature-auth/ on a branch named worktree-feature-auth, branched from your remote’s default branch. Subagents can get their own temporary worktrees with isolation: worktree. On exit, Claude removes clean worktrees from unnamed sessions and asks about everything else. While a session runs in a worktree, Claude Code blocks edits to the main checkout. The full guide: Claude Code worktrees.

Codex

In the ChatGPT desktop app, choose Worktree and a starting branch, and Codex creates a worktree under $CODEX_HOME/worktrees. It works in a detached HEAD state, so it doesn’t create branches you didn’t ask for. When you want to keep the work, use Create branch here, then commit, push and open a PR, or Hand off the chat to your local checkout. Codex keeps your 15 most recent managed worktrees by default and saves a snapshot before deleting one.

The detached HEAD has a cost. Commits made there belong to no branch. If a worktree is deleted before you create a branch, those commits are reachable only through Codex’s snapshot or git’s reflog.

Cursor

Cursor’s worktrees are built around its Agents Window: start or move an agent into a worktree and it gets its own checkout. In the IDE, /worktree runs a task in an isolated checkout, /apply-worktree brings the result into your main checkout, and /delete-worktree cleans up. Setup is the most scriptable of the three. .cursor/worktrees.json holds commands such as npm ci and cp $ROOT_WORKTREE_PATH/.env .env, with separate keys for Unix and Windows. Cursor advises against symlinking dependencies into worktrees.

Which one should you use?

• You live in the terminal: Claude Code’s --worktree is the most transparent. The folders and branches are ordinary git ones you can see and script.
• You start lots of throwaway tasks: Codex’s detached, auto-pruned worktrees create the least clutter. Just remember to create a branch for work you want to keep.
• Your setup is heavy: Cursor’s worktrees.json runs your install and migration steps automatically.
• You use more than one: that’s common. Many developers run Claude Code and Codex side by side, which means worktrees in two different places with two different naming schemes.

One view across all of them

Each tool shows only its own worktrees. Orchard shows all of them: every worktree in every project on your Mac, whether Claude Code, Codex, Cursor Agent, Gemini CLI, Aider, opencode, Amp or Goose made it. It tracks commits on a detached HEAD as unlanded work, so Codex’s branchless commits aren’t lost. It also shows which agent is working where and which worktrees are already in main and can be cleaned up.

FAQ