Worktree

---
name: worktree
description: Create a work branch in a ready-to-use git worktree. The normal way to start a branch, so as not to cross another session in the main tree.
argument-hint: "<type/branch-name>"
---

Branch asked for: `$ARGUMENTS` (`feat/…`, `fix/…`, `chore/…`, like the repo's history). Without
an argument, propose one from the task.

1. Find the root of the repo (`git rev-parse --show-toplevel` of the main tree) and its
   integration branch: the one the project's `CLAUDE.md` names, otherwise `develop` if it exists
   on `origin`, otherwise the default branch (`git symbolic-ref refs/remotes/origin/HEAD`).
2. From the root, never switching the branch of the main tree: `git fetch -q origin
   <integration>`, then `git worktree add --no-track .claude/worktrees/<name> -b <branch>
   origin/<integration>` (`<name>` = the branch with `/` replaced by `-`). Without `--no-track`,
   the branch would track `origin/<integration>`: a bare `git push` refused, a `git pull` that
   pulls the integration branch. If the branch already exists, attach it without `-b`. Check
   that `.claude/worktrees/` is ignored by git; otherwise add it to `.git/info/exclude`.
3. Prepare the worktree: `bash "${CLAUDE_PLUGIN_ROOT}/scripts/prepare-worktree.sh" <path of the
   worktree>` (copies the ignored `.env*` files of the main tree, installs the dependencies of
   each `pnpm-lock.yaml` / `package-lock.json`). If the repo also provides
   `.claude/worktree-setup.sh`, run it next with the same argument.
4. Then work only in that worktree (absolute paths) and give its path.