wt - Git Worktree Manager

A fast, simple Git worktree helper written in Go. Inspired by haacked/dotfiles/tree-me.

Features

• Configurable worktree strategies: global, sibling-repo, parent-branches, and more
• Clone into organized locations — wt clone acquires a repo under repo_root (<host>/<owner>/<repo>/<branch>), ready to inspect
• Simple commands for common worktree operations
• Interactive selection menus with fuzzy matching for checkout, cd, remove, pr, and mr commands
• GitHub PR support via wt pr command (uses gh CLI) — checks out the PR's actual branch name
• GitLab MR support via wt mr command (uses glab CLI) — checks out the MR's actual branch name
• Built-in file copy — declare .env and friends in [files] or a committed .worktreeinclude; copied with a reflink on APFS/Btrfs/XFS, so even node_modules costs metadata rather than disk
• Pre/post command hooks — run custom scripts on create/checkout/remove/clone (e.g. launch AI assistants, share build caches, assign dev server ports, init submodules)
• Stale worktree detection — find worktrees with deleted remote branches or inactive commits (wt cleanup --stale)
• Color-coded status output — green (clean), red (dirty), yellow (ahead/behind), bold cyan (current); respects NO_COLOR=1 and auto-strips colors when piped
• CI/CD status integration — wt status --ci shows pipeline status (✓/✗/●) per branch via gh or glab CLI
• Per-repo .wt.toml config — override global settings (strategy, hooks, etc.) on a per-repository basis
• git config support — keep personal settings in .git/config or ~/.gitconfig, with no extra file to gitignore
• Shell integration with auto-cd functionality
• Tab completion for Bash, Zsh, and Fish

Quick Start

brew install timvw/tap/wt   # or: go install github.com/timvw/wt@latest
wt init                      # configure shell integration

See docs/installation.md for all platforms (Scoop, WinGet, Linux packages, from source).

Usage

Clone

# Acquire the main repository, ready to inspect. Placed under repo_root as
# <repo_root>/<host>/<owner>/<repo>/<default-branch>, left on its default branch
# (that trailing segment makes the clone a normal worktree slot, so wt create
# later puts siblings next to it).
wt clone timvw/wt                              # owner/repo resolved via gh/glab
wt clone git@github.com:me/dotfiles.git        # full URL used as-is
wt clone acme/api ~/src/api                    # explicit destination

Two settings control placement: repo_root (default ~/dev/repos) and repo_pattern. Grouping levels like "work" vs "personal" are yours to define — put an env var in the pattern with a :- default so it works even when unset:

repo_pattern = "{.repoRoot}/{.env.WT_CATEGORY:-personal}/{.repo.Owner}/{.repo.Name}/{.branch}"

WT_CATEGORY=work wt clone acme/api   # ~/dev/repos/work/acme/api/main
wt clone timvw/wt                    # ~/dev/repos/personal/timvw/wt/main (default)

To set the category for a whole tree of repos instead of per command, add a [[context]] rule:

[[context]]
when_path = "~/dev/repos/work"
env = { WT_CATEGORY = "work" }

Every wt command operating on a repo under that path then resolves work, including wt create from a worktree in a different tree. The same rule can go in ~/.gitconfig instead, if you would rather not keep a config file:

git config --global wt.context.work.whenpath "~/dev/repos/work"
git config --global --add wt.context.work.env "WT_CATEGORY=work"

See Setting the category per directory.

Checkout & Create

# Checkout existing branch in new worktree
wt co feature-branch
wt co                             # interactive: fuzzy-search from available branches

# Create new branch in worktree (defaults to main/master as base)
wt create my-feature
wt create my-feature develop      # specify base branch

Switch Between Worktrees

# Switch to a worktree that already exists (never creates one)
wt cd feature-branch
wt cd                             # interactive: fuzzy-search from existing worktrees
wt sw                             # alias for wt cd

Unlike wt co, the wt cd list contains only branches that already have a worktree — including the main checkout — so it stays short in repositories with many branches.

PRs & MRs

# Checkout GitHub PR (requires gh CLI)
wt pr 123                                          # looks up branch for PR #123
wt pr https://github.com/org/repo/pull/123         # GitHub PR URL
wt pr                                              # interactive: fuzzy-search from open PRs

# Checkout GitLab MR (requires glab CLI)
wt mr 123                                          # looks up branch for MR !123
wt mr https://gitlab.com/org/repo/-/merge_requests/123  # GitLab MR URL
wt mr                                              # interactive: fuzzy-search from open MRs

Untracked files

A new worktree has everything git tracks and nothing else — no .env, no .envrc, no editor state. Declare what a usable checkout needs and wt puts it there on create, checkout, pr and mr:

# ~/.config/wt/config.toml, or a repo's .wt.toml
[files]
copy = [".env", ".claude/settings.local.json"]
link = ["node_modules"]

wt copy                           # re-run for the current worktree
wt copy feature-branch --dry-run  # show what would happen, change nothing
wt copy feature-branch --force    # overwrite files already there
wt create feature --no-copy       # skip it just this once

Or commit a .worktreeinclude at the repo root — same gitignore syntax, one pattern per line — so every contributor gets working worktrees without configuring anything. Only untracked, git-ignored files are ever candidates; a tracked file is already in the worktree and is never touched. See Files.

List & Remove

wt ls                             # list all worktrees
wt rm old-branch                  # remove a worktree
wt rm                             # interactive: fuzzy-search worktree to remove

Maintenance & Misc

wt migrate                        # migrate worktrees to configured paths
wt migrate --force                # force when target path exists
wt cleanup --stale                # detect stale worktrees (deleted remotes, inactive commits)
wt cleanup --stale --stale-days 7 # custom inactivity threshold (default: 30 days)
wt prune                          # clean up stale worktree admin files
wt version                        # show version
wt examples                       # show practical examples
wt --help                         # show help

Info & Config

wt info                           # show active strategy, pattern, variables
wt config show                    # show effective config with sources
wt config init                    # create a default config file
wt config path                    # print the config file path
# Place a .wt.toml in a repo root to override global config for that repo
# Or keep it out of the working tree entirely, in git config:
git config --local wt.strategy sibling-repo    # this repo only
git config --global wt.root ~/projects/worktrees

Hook trust

wt runs no hook you have not approved — a repo's committed .wt.toml and your own config file alike. It asks once, and remembers the answer until the commands change:

wt trust                          # approve every hook that applies here
wt trust --list                   # show every approval on this machine
wt untrust                        # revoke this repository's approval
wt untrust --path /removed/repo   # revoke stale approvals for a removed repository
wt untrust --global               # revoke your config file's approval

Editing a hook command — or a git pull that adds one — asks again; editing anything else in the file does not. Non-interactive runs (scripts, CI, --format json) decline, unless you opt out with WT_HOOKS_APPROVE_ALL=1. Whitelist a whole tree with [trust] prefix = ["~/src/mine"], or set hooks_policy = "prompt-all" to confirm every hook every time. See Hook trust.

On case-insensitive filesystems such as the default macOS APFS setup, mixed-case branch prefixes can produce confusing worktree paths. For example, Feature/foo and feature/bar both need a first-level directory that macOS treats as the same name. Set separator = "-" to flatten branch paths (Feature/foo -> Feature-foo) and avoid that class of collision. See Configuration.

Status Dashboard

wt status                         # color-coded overview of all worktrees
wt status --ci                    # include CI/CD pipeline status (requires gh or glab)

Shows dirty/clean state, ahead/behind counts, and highlights the current worktree. With --ci, each branch shows ✓ (pass), ✗ (fail), or ● (pending) for its latest CI pipeline. Colors are automatically stripped when piping; set NO_COLOR=1 to disable.

Interactive Selection

When you run wt co, wt cd, wt rm, wt pr, or wt mr without arguments, you'll get an interactive selection menu. Typing filters the results with fuzzy matching, so you can quickly find the branch or worktree you're looking for.

wt co selects from every local and remote branch (creating a worktree when needed), while wt cd selects only from worktrees that already exist.

JSON Output (--format json)

Most commands support machine-readable JSON output:

wt --format json version
wt --format json info
wt --format json config show
wt --format json list
wt --format json examples

In json mode, shell integration does not auto-navigate. For commands that normally prompt interactively, pass explicit arguments when using --format json.

Use with Claude Code

Install the Claude Code plugin to teach Claude how to work with wt-managed worktrees:

claude plugin marketplace add timvw/wt
claude plugin install wt@wt --scope local

Once installed, Claude understands wt commands, worktree strategies, and hooks — so you can ask it to create worktrees, set up hooks for copying .env files or running npm install / uv sync, and follow worktree-based workflows automatically.

See docs/examples.md for hooks that launch Claude Code in tmux per worktree.

Documentation

How It Works

The tool wraps Git's native worktree commands with a convenient interface and organized directory structure:

1. Organized Structure: All worktrees for a repo are kept together
2. Smart Defaults: Automatically detects repo name and default branch
3. Prevents Duplicates: Checks if a worktree already exists before creating
4. Auto-CD: With shell integration, automatically changes to the worktree directory
5. Tab Completion: Makes it easy to work with existing branches

License

MIT

Credits

Based on tree-me by Phil Haack.