Worktrees - three versions of an idea, zero collisionsHands-on

A git worktree is a second working copy of the same repo, on its own branch, in its own folder — so you can build three versions of an idea side by side and never let a half-finished experiment touch main.

Normally a repo has one working directory, and switching branches swaps the files under you — which means you can't have branch A and branch B open at once. A worktree breaks that limit: `git worktree add` creates a SECOND folder checked out to a different branch, sharing the same underlying history. Two folders, two branches, one repo. Build variant-A of a landing page in one, variant-B in the other, compare them running at the same time, keep the winner, delete the rest — and main never saw either experiment.

For parallel work this is the difference between safe and reckless. Tuesday's lesson was about two Claude sessions fighting over one working tree; the real fix isn't 'be careful,' it's 'stop sharing the tree.' Point session one at the main checkout and session two at a worktree and there is no shared file to clobber — each has its own copy, and git reconciles them at merge time, which is exactly what git is for.

It's wired into your tooling. The superpowers:using-git-worktrees skill carries the discipline for spinning them up and tearing them down; and some agents can be launched with isolation:'worktree' so they automatically get a scratch copy of the repo that's cleaned up if they didn't change anything. The mental model to keep: a worktree is disposable scratch space with full git power — cheap to create, cheap to throw away, and it keeps your experiments off the branch you actually ship.

Worked example

Create one, prove it doesn't touch the original, tear it down — all safe.

cd /Users/tom/Claude/PDB/output   # any project repo
git worktree add ../scratch-experiment -b scratch-experiment
echo 'throwaway' > ../scratch-experiment/EXPERIMENT.txt
git -C . status --short              # ← the ORIGINAL folder: still clean
git worktree list                    # both trees, each on its own branch
git worktree remove ../scratch-experiment --force

• `worktree add … -b <name>` makes a new folder AND a new branch in one step — the experiment lives there, not here.
• The status check on the original folder shows nothing changed: the new file only exists in the worktree, proving the isolation.
• `worktree remove` deletes the scratch copy cleanly; the branch stays in history if you want it, or delete it too.

Gotchas

• You can't check out the same branch in two worktrees — each needs its own; that's the constraint, and `-b` creating a fresh branch is how you satisfy it.
• A worktree is a real folder taking real disk — remove it when done (`git worktree remove`), don't just `rm -rf` it or git keeps a stale reference (fixable with `git worktree prune`).
• Uncommitted work in a worktree is as clobberable as anywhere else — the isolation is between trees, not a substitute for committing inside each one.