# Work folder checkpoints no longer pause running commands

> Checkpoints now flush only files still being written, so a dev server keeps running, and a program run from the work folder no longer makes saves fail.

Oct 20, 2026. Topics: Agents, Platform. https://www.generalinput.com/changelog/2026-10-20-work-folder-checkpoints-without-pauses

A specialist's [work folder](/changelog/2026-10-08-lazy-workspace-filesystem) is checkpointed while it works, and each checkpoint briefly pauses everything running in its sandbox so the saved copy is consistent. That pause used to grow with the size of the folder: a JavaScript project of about 21,000 files paused for 10 to 12 seconds on every checkpoint, and a folder past 100,000 files came close to the one minute limit. Running a program stored in the work folder, such as a dev server from a package store, could also make every checkpoint fail, so edits made after the last good one were lost when the sandbox was replaced. Both are fixed.

### How it works

- **Only files still being written are flushed.** A file is saved for good the moment the program writing it closes it, so a checkpoint now flushes only files a program still holds open for writing. In our tests, the pause for a 20,000 file folder fell from 3.3 seconds to 7 milliseconds.
- **Running programs are skipped.** A program running from the work folder can't be written to while it runs, so it has nothing left to flush.
- **Never less safe than before.** When a checkpoint can't see every program's open files, it flushes every file in the folder, as it did until now.

### Also shipped

- Storage from old checkpoints is reclaimed again for folders of any size, including ones holding a Next.js build.
- A work folder can hold about twice as many files before a checkpoint asks the specialist to remove rebuildable folders like `node_modules`.