Worktrunk：为并行 AI Agent 简化 Git Worktree

AI coding agents like Claude Code and Codex can now work on longer tasks without supervision. That means you can realistically run several of them at once, each on a different feature or bug. The catch: if they all work in the same folder, they overwrite each other's changes. Git has a built-in answer for this called worktrees. A worktree is a separate working directory attached to the same repository, so each branch can live in its own folder at the same time. The problem is that git's worktree commands are clunky. Just starting a new one looks like this: git worktree add -b feat ../repo.feat cd ../repo.feat Enter fullscreen mode Exit fullscreen mode You type the branch name three times. Cleaning up afterward takes another three commands. Do this ten times a day and it gets old fast. What Worktrunk is Worktrunk is a command-line tool, written in Rust, that makes worktrees as easy to use as regular branches. You refer to everything by branch name, and Worktrunk works out the folder paths for you. It's open source (MIT or Apache-2.0), was released at the start of 2026, and already has over 5k stars on GitHub. The core commands Here's how Worktrunk compares with plain git: Task Worktrunk Plain git Switch to a worktree wt switch feat cd ../repo.feat Create and start an agent wt switch -c -x claude feat git worktree add -b feat ../repo.feat && cd ../repo.feat && claude Clean up wt remove cd ../repo && git worktree remove ../repo.feat && git branch -d feat List with status wt list git worktree list (paths only) Four commands cover most of what you'll do day to day. Getting started Install with Homebrew (macOS and Linux): brew install worktrunk && wt config shell install Enter fullscreen mode Exit fullscreen mode Or with Cargo: cargo install worktrunk && wt config shell install Enter fullscreen mode Exit fullscreen mode The wt config shell install step matters. It adds shell integration so wt can actually change your current directory. Create a worktree for a new feature: wt switch --create feature-auth Enter fullscreen mode Exit fullscreen mode This creates the branch and the worktree, then moves you into it. Check all your worktrees and their status with: wt list Enter fullscreen mode Exit fullscreen mode When you're done, you can open a PR as usual and run wt remove after it merges. Or merge locally with one command: wt merge main Enter fullscreen mode Exit fullscreen mode That commits your changes, rebases onto main, merges, and removes the worktree and branch. Running agents in parallel This is where Worktrunk shines. Start three agents on three tasks, each in its own isolated worktree: wt switch -x claude -c feature-a -- 'Add user authentication' wt switch -x claude -c feature-b -- 'Fix the pagination bug' wt switch -x claude -c feature-c -- 'Write tests for the API' Enter fullscreen mode Exit fullscreen mode The -x flag runs a command after switching, and everything after -- is passed to that command. No agent touches another agent's files. Extras worth knowing Beyond the basics, Worktrunk includes: Hooks that run commands when a worktree is created or merged, such as installing dependencies or starting a dev server. LLM-generated commit messages based on your diff. Build cache copying so a new worktree can reuse node_modules/ or target/ instead of starting cold. PR checkout with wt switch pr:123. A unique port per worktree so multiple dev servers can run side by side. Is it worth a try? Yes, if you: Run more than one AI coding agent at a time. Juggle several branches and hate stashing or re-cloning. Already use git worktrees and are tired of the long commands. A few honest caveats. It's a young project, so expect frequent releases and occasional changes. Every worktree is a full copy of your working files, so large repos use more disk space (the cache-copying feature helps, but doesn't remove this). On Windows, wt clashes with the Windows Terminal command, so it installs as git-wt instead unless you disable that alias. And if you only ever work on one branch at a time, plain git switch is all you need. If parallel agents are part of your workflow, Worktrunk removes most of the friction of worktrees. Install it, spin up two worktrees, and you'll know within ten minutes whether it sticks. Links GitHub: https://github.com/max-sixty/worktrunk