If you’ve ever run git clone and then started editing files in the resulting folder, you’ve been using a Git worktree all along. Git just doesn’t call it that in day-to-day conversation.

git worktree is the feature that lets you add more checkouts (more folders with files on disk) that all share the same underlying repository history. Done well, it replaces the “stash/switch/stash-back” dance with something simpler: keep your in-progress work open in one directory, and do the urgent/clean/experimental thing in another.

This is a mental-model article: what a worktree is, why Git enforces “one branch per worktree,” and the few lifecycle commands that keep worktrees from turning into confusing “already checked out” / stale-metadata problems.

The Mental Model: Shared History, Separate Working State

Git’s own git worktree docs describe the feature as managing “multiple working trees attached to the same repository.” The twist is that you always start with one: the directory you land in after a normal (non-bare) clone is the main worktree.

Linked worktrees are additional working directories attached to the same repo. They share the object database (commits/trees/blobs), but each worktree has its own working state: checked-out files, its own HEAD, and its own index (staging area).

That’s the whole trick:

• Shared: history and objects (you don’t redownload Git data like a second clone).
• Per worktree: files-on-disk + HEAD + index (you don’t stomp on your other checkout).

One nuance that matters: a worktree isn’t its own independent “repo.” Most refs and remote configuration are shared. When you git fetch, you’re updating the shared repository data that all worktrees see. That’s what makes worktrees lighter than multiple clones—but it’s also why Git has to be careful about how a branch name maps to a checkout.

If you want a deeper explanation of why “objects live once” makes this efficient, the Git book’s “Git Objects” chapter is a good reference: Git Internals: Git Objects.

What Actually Changes on Disk when You Add a Worktree

Git keeps the shared repository database in one place, then tracks per-worktree state separately. That’s why it can add a second checkout without cloning.

If you’ve ever noticed that .git sometimes looks like a file (not a folder), that’s often Git pointing your working tree at the real “gitdir.” Worktrees lean on this idea heavily: each worktree has its own admin area, but the objects are shared.

The Rule that Surprises People: One Branch per Worktree

If you try to attach the same branch to two worktrees, Git will usually refuse with “branch is already checked out.” That refusal is protection.

The core reason is that a branch name is just a movable pointer. If two working directories could both “be” feature, the branch pointer could move in one directory while the other directory still has older files on disk. Now you have a branch name that no longer uniquely identifies a single working state.

Said differently: Git is fine with multiple worktrees having different HEADs (each checkout is independent), but a branch ref like refs/heads/feature is shared at the repository level. Letting two worktrees both “own” the same branch would create a footgun where “what commit is feature?” depends on which folder last advanced it.

When you want “two directories, same starting point,” Git’s safer options are:

• Create a new branch name for the second worktree:

git worktree add ../wt-fix -b fix-branch

• Use a detached HEAD when you don’t intend to keep the work long-term:

git worktree add -d ../wt-scratch

Detached HEAD is the “I want another directory at this commit, but I’m not claiming a branch name” mode. That’s why it’s safe: you can still make commits, but you’re not advancing a shared branch ref unless you later create one.

--force exists for edge cases, but it’s opting out of a guardrail. If you don’t already know why you need it, you probably don’t.

Where Worktrees Shine (in Practice)

Worktrees are a productivity primitive whenever parallelism beats context switching. Three common patterns:

1) Hotfix without Touching Your In-Progress Directory

When a production issue lands mid-feature, create a new worktree on a hotfix branch:

git worktree add ../wt-hotfix -b hotfix

You get a clean checkout in a new folder, and your original directory stays exactly as it was. When you’re done, remove it cleanly:

git worktree remove ../wt-hotfix

Tower’s Git worktree FAQ describes this as one of the “default” worktree use cases, and it’s the one most people feel immediately.

2) PR Reviews and Repros in a Clean Sandbox

You can attach a worktree directly to a remote-tracking branch and keep it around for as long as the review takes:

git worktree add ../wt-review origin/some-pr-branch

3) Scratch Experiments You Don’t Want to Keep

Detached worktrees are great for “let me try something” work without committing to a branch name yet:

git worktree add -d ../wt-scratch

If you end up making commits you want to keep, create a branch before you delete the folder so you don’t lose track of them.

A Quick Note on “Why Not Just Clone Twice?”

Separate clones can absolutely solve the same “two directories” problem. The difference is mainly cost and coupling:

• Clones duplicate setup (another fetch/clone, another .git object store).
• Worktrees share Git objects, but still behave like separate checkouts.

If you need complete isolation (different remotes, radically different Git config, or you want to move one copy around freely), a second clone can still be the simplest answer.

Keeping Your Directories Sane

The simplest convention (and the one that avoids the “oops, I edited the wrong checkout” mistake) is: keep linked worktrees outside the main project folder.

repo/            # main worktree
repo-hotfix/     # linked worktree
repo-review/     # linked worktree

If you do this often, encode intent in the directory name so it’s searchable later (wt-hotfix-1234, wt-pr-998, wt-bisect-crash) instead of generic tmp/.

If you want all worktrees contained under one parent folder, a popular layout is “bare repo + many worktrees” (see Nick Nisi and Ahmed El Gabri). For background on bare repositories themselves, the Git book chapter is a good reference: Git Basics: Getting a Git Repository.

If you try the “bare repo + worktrees” approach, keep the tradeoff in mind: you’re saving Git object duplication, not environment duplication. Two worktrees can still mean two node_modules/ folders, two build outputs, and two IDE indexes. If setup dominates your workflow, keep only the worktrees you’re actively using.

The Lifecycle Gotchas (and the 5 Commands that Prevent Them)

Removing vs Deleting (Stale Metadata)

Deleting a worktree folder in Finder isn’t the same as telling Git it’s gone. Prefer:

git worktree remove ../wt-review

If you already deleted the folder, clean up the leftover admin state:

git worktree prune

Git can also garbage-collect stale worktree metadata automatically over time (controlled by gc.worktreePruneExpire). That’s helpful, but it’s not a substitute for removing worktrees intentionally when you’re done.

“This Branch Is Already Checked Out”

This is Git enforcing “one branch per worktree.” Your usual fixes are:

• make a new branch name (-b)
• use detached HEAD (-d)

If you’re debugging “why won’t Git let me…”, list the current attachments:

git worktree list

Moving Worktrees (and Repairing when Paths Change)

If you need to reorganize, prefer Git’s commands over dragging folders around in your file manager.

• git worktree move <worktree> <new-path> relocates a linked worktree (you can’t move the main worktree).
• If you moved things manually and links broke, git worktree repair can reestablish the connection in supported cases.

Intermittent Drives: Lock Before Git Prunes

If a worktree lives on an external drive or network mount, Git may treat it as “missing” when the volume isn’t mounted. Use git worktree lock to prevent surprises:

git worktree lock --reason "On external drive" /Volumes/SSD/wt-release

Cheat Sheet

git worktree add ../wt-name -b my-branch   # new worktree + new branch
git worktree add ../wt-name origin/branch  # worktree from remote-tracking branch
git worktree add -d ../wt-scratch          # detached-HEAD scratch worktree
git worktree list                          # see what's attached
git worktree remove ../wt-name             # remove cleanly
git worktree prune                         # cleanup after manual deletion

The Takeaway

Worktrees aren’t a niche add-on. They’re the name for the checkout you already use, plus a way to add more checkouts without cloning again. Keep the mental model in your head—shared history, separate working state—and Git’s guardrails (like one branch per worktree) stop feeling arbitrary and start feeling helpful.

If you learned something from this post, give me a follow and checkout the AI powered markdown editor that i am building on the side.