# Parking work (/docs/scruff/parking) ```sh scruff park "mid-refactor on the retry logic" scruff unpark ``` `park` commits **the whole dirty tree** (tracked edits, untracked files, deletions) as a single `wip:` commit on the branch only this pane has checked out. `unpark` rewinds that commit and puts the changes back, uncommitted. That is the same thing the pane-close hook does, so a lane you close mid-thought is a lane you can resume mid-thought. Doing it on demand is for the other case: you need a clean tree for a minute, and you want it back. The argument is a **label**, never a lane name. `park` works on the checkout you are standing in, so naming another lane on a clean tree is **exit 2** and the way to reach that one instead; on a dirty tree the park still happens, and says whose tree it took. ## Never `git stash`, once you have two agents [#never-git-stash-once-you-have-two-agents] `git stash` looks like it belongs to the checkout you are standing in. **It does not.** The stash is one ref in the repo's shared `.git` directory, so every worktree of a repo *and* the main checkout push and pop the **same stack**. Two agents stashing at once means either can pop the other's work into a tree that has never seen those files. `stash@{1}` is no safer: it is positional, and the position moved when somebody else pushed. A park has none of that surface. It is a commit, on one branch, that one pane has checked out. Nobody else can pop it, and `scruff` lists the result as a resumable lane rather than as a numbered entry in a shared pile. ## What it will and will not sweep in [#what-it-will-and-will-not-sweep-in] **Ignored files are never included.** A `.gitignore`d secret stays ignored, so parking cannot smuggle one into history. Everything else in the tree goes. If the `wip:` commit is already on the remote (it is visible in an open PR), `unpark` refuses. Rewinding it locally would turn a file restore into a force-push, and by design that is a thing `unpark` can never become. It prints the escape hatch (`git reset --mixed HEAD^`) and leaves the commit alone. Parking on a branch that is not an agent branch works, and says so: a `wip:` commit should not drift onto something you push. ## Where parking happens without you asking [#where-parking-happens-without-you-asking] Every destructive path parks first. That is invariant 1, and it is what makes the rest of scruff safe to automate: | | | | -------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | a pane closes on unlanded work | the tree is parked, the checkout goes, the branch stays | | `scruff reap` meets a dirty lane | it **keeps** it and names what is in the way. An automatic sweep may only ever take *finished* work, so it never parks and sweeps in one motion | | `scruff drop` meets a dirty lane | it refuses with **exit 2**, names the files in the way, and gives you the `cd` into that lane's checkout to park them from. Dropping is deliberate, so it asks rather than deciding for you | The decision "does this dirty tree actually need a commit?" is the [`preserve` seam](/docs/scruff/seams#the-six-shipped-seams). The built-in answer is *yes, unless it is untracked scratch on a branch that already landed*.