worktrunk: A Git Worktree CLI for Running 5 to 10 AI Agents in Parallel

A CLI tool that creates isolated work environments to prevent conflicts when running multiple AI agents, such as Claude Code or Codex, simultaneously. It is built on Git's native worktree functionality but designed to address the drawbacks of existing commands being long and cumbersome. It minimizes the commands required from worktree creation to cleanup, lowering the barrier to entry for parallel work.

1 / 2

It manages the entire worktree lifecycle with four core commands: wt switch, wt list, wt merge, and wt remove. wt switch creates a worktree and moves to the corresponding directory by simply entering a branch name; using the -x flag allows external agent processes, such as Claude, to be executed immediately upon switching. wt list outputs a table that provides an at-a-glance view of the current changes, commit status, and CI status for each worktree.

It also provides additional features to improve the efficiency of parallel work environments. It shares copy-on-write-based build caches on APFS, btrfs, and XFS file systems, so there is no need to separately copy or build target/ or node_modules/ even with 10 worktrees. It also allows setting hooks that run at worktree creation to automate dependency installation or starting development servers, and supports automatically generating diff-based commit messages using an LLM.

Written in Rust, it is available on macOS, Linux, and Windows. It is suitable for developers who manually manage Git worktrees or teams operating multiple instances of AI coding agents and experiencing bottlenecks. It allows quickly setting up and cleaning up workspaces for each agent without memorizing complex Git commands, making it useful for building large-scale parallel development workflows.