worktrunk: Git Worktree CLI for Parallel AI Agents

TL;DR: max-sixty/worktrunk is a 7,248-star Rust CLI that wraps Git worktree management for the specific case of running multiple AI coding agents in parallel, each on its own branch and working copy.

max-sixty/worktrunk is described as “a CLI for Git worktree management, designed for parallel AI agent workflows.” That single sentence captures both the technical scope and the target audience. The repo carries 7,248 stars and 256 forks as of this writing, with the most recent push on September 12, 2026. The project page lives at worktrunk.dev.

The pain point is familiar to anyone running multiple coding agents simultaneously: each agent needs its own working directory, its own branch, and ideally its own merge queue so they do not stomp on each other’s changes. Without worktrees, you end up stashing, switching branches, and losing context every time a second agent starts. With native git worktree, you can have multiple branches checked out at once, but the command line surface is verbose enough that scripting it correctly takes a half-day the first time.

Worktrunk packages the common patterns into a single CLI. You can spin up a new worktree for an agent task, run the agent in that worktree, capture the diff, merge or discard based on outcome, and tear down the worktree — all without typing out the full git worktree add incantation each time. Topics on the repo include agents, claude-code, codex, developer-tools, git, and worktrees, which confirms the target audience: developers running Claude Code, Codex, or similar AI coding assistants.

The implementation is Rust, so the binary installs quickly, starts instantly, and behaves predictably on macOS, Linux, and Windows. There is no runtime dependency on Python or Node, which matters when you are running it inside an agent loop where shell latency compounds across hundreds of invocations.

For someone setting up an AI-assisted development workflow, the value is removing a class of friction. Instead of hand-rolling worktree scripts, you can configure worktrunk once and have every agent invocation land in a clean, isolated working copy. The agent’s diff becomes reviewable in isolation, and merging back into the main branch is a deliberate step instead of an accident.

If you are not running AI agents in parallel, the tool still has value as a friendlier interface to git worktree itself. Commands like creating a worktree for a quick experiment, naming it after the task, and cleaning it up afterward are all simpler than the underlying Git commands. But the project’s framing — and its roadmap — assumes the AI agent use case is primary.

The license is NOASSERTION, which means the maintainer has not explicitly declared a license in the GitHub API response. That is worth investigating before you vendor the code into a commercial project. The source repository is available and the maintainer is responsive, so reaching out for clarification is reasonable.

For comparison, similar tools include git-branchless for stack-based branch management and Graphite for stacked diffs. Worktrunk’s differentiation is the AI agent angle — it assumes your “developer” running commands might be a model, not a human, and optimizes the ergonomics accordingly.

Source and Accuracy Notes

• Project page: https://github.com/max-sixty/worktrunk
• Source repository: max-sixty/worktrunk (verified via GitHub API html_url)
• License: NOASSERTION (verified via GitHub API license.spdx_id — contact maintainer before vendoring)
• Stars at time of writing: 7,248