git-worktree（Git 工作树）

Git Worktree

Manage worktrees for concurrent development without clobbering changes.

Setup

~/.claude/skills/git-worktree/scripts/wt.sh install
source ~/.zshrc

Usage

wt <branch>              # Create worktree, run setup, open editor
wt <branch> --base ref   # Create from a specific base branch (default: main)
wt <branch> --no-editor  # Create without opening editor
wt <branch> --open       # Also open terminal tab with claude session (macOS)
wt <branch> --carry      # Create and copy work-in-progress files
wt <branch> --context f  # Copy file to .context/handoff.md (session fork)
wt cd <branch>           # Change to worktree directory
wt home                  # Return to main repo (or REPOS_ROOT if outside git)
wt apply [branch]        # Rebase onto branch and merge (default: main)
wt apply --archive       # Merge and archive without prompting
wt apply --push          # Merge and push to remote
wt archive [branch]      # Run archive script, move to ~/.worktrees/.archive
wt list                  # List all worktrees
wt ls                    # Alias for list
wt tree                  # Tree view with git status indicators
wt status                # Show worktrees with Claude session activity
wt open [branch]         # Open editor for worktree (current dir if no branch)
wt install               # Add wt to ~/.zshrc (one-time setup)

Worktree Path Convention

Worktrees are created at ~/.worktrees/<repo>/<branch> where <repo> is the origin remote name (not the local directory name). This is derived from git remote get-url origin.

~/.claude  (remote: dotclaude.git)  →  ~/.worktrees/dotclaude/<branch>
~/code/services  (remote: services.git)  →  ~/.worktrees/services/<branch>

Environment

WORKTREES_ROOT=~/.worktrees  # Where worktrees are created
REPOS_ROOT=~/code            # Fallback for `wt home` outside git
WT_TERMINAL=ghostty          # Terminal for --open (auto-detects from TERM_PROGRAM)

Example

wt feature-auth       # Creates worktree and opens editor
# ... work on feature ...
wt apply              # Merge into main (prompts to archive)

Or with the traditional archive workflow:

wt feature-auth       # Creates worktree and opens editor
# ... work on feature ...
wt home               # Back to main repo
wt archive feature-auth  # Archive when done (moves to .archive)

Carrying Work in Progress

When you've been exploring and decide it should be its own branch:

# You're in main with untracked files and modifications...
wt feature-x --carry  # Creates worktree with those files copied over

Copies both untracked files and modified tracked files. Works from any branch.

conductor.json (Optional)

If your repo has a conductor.json, scripts run automatically:

{
  "scripts": {
    "setup": "cp $CONDUCTOR_ROOT_PATH/.env .env && bun install",
    "archive": "git stash"
  }
}

For Claude Code

When user asks to create a worktree, run:

wt <branch>

The script handles branch detection, env file copying, and setup automatically. Suggest opening the worktree in their editor after creation.