[BUG] Cowork Windows: EXDEV cross-device rename failure causes machine crashes on bundle update — affects all MSIX installs

Resolved 💬 5 comments Opened Apr 7, 2026 by KRMOtech Closed Jun 3, 2026

Preflight Checklist

  • [x] I have searched existing issues and this hasn't been reported yet
  • [x] This is a single bug report (please file separate reports for different bugs)
  • [x] I am using the latest version of Claude Code

What's Wrong?

After Claude Desktop auto-updated on March 31 2026 (bundle version changed from fb30784d to 5680b11b), Cowork became permanently broken and caused 4 complete machine crashes in a single day due to an infinite retry loop.

Root cause: cowork-svc.exe downloads rootfs.vhdx to AppData\Local\Temp\wvm-XXXXX\ then attempts fs.rename() to move it to AppData\Roaming\Claude\vm_bundles\claudevm.bundle\. This fails with EXDEV: cross-device link not permitted because MSIX filesystem virtualization treats these as separate devices.

When the rename fails, CoworkVMService retries immediately with no backoff or retry limit, spinning up new VM instances every 2-7 minutes. Each attempt consumes increasing amounts of RAM until the system runs out of memory and crashes. This happens even when the user has NOT opened the Cowork tab.

Additional bug on top: Windows username containing non-ASCII characters (Danish ø) is corrupted to KristianMøller in the bundlePath log, causing an additional path resolution failure on top of the EXDEV error.

What Should Happen?

  1. rootfs.vhdx should be downloaded and placed in the bundle directory successfully
  2. If the rename fails with EXDEV, the app should fall back to fs.copyFile() + fs.unlink() which works across device boundaries
  3. If download fails, CoworkVMService should implement exponential backoff and a maximum of 3 retries — not an infinite loop
  4. A failing CoworkVMService should never be able to crash the host machine
  5. Non-ASCII characters in Windows usernames should be handled correctly in all path construction

Error Messages/Logs

From cowork_vm_node.log:

[VM:start] Beginning startup, bundlePath=C:\Users\KristianMøller\AppData\Roaming\Claude\vm_bundles\claudevm.bundle
[VM:steps] download_and_sdk_prepare started
[VM:steps] download_and_sdk_prepare failed (3014ms): Download failed. Check your internet connection and try again.

Loop repeating every 2-7 minutes:
[13:46:11] rootfs.vhdx not found, downloading...
[13:52:59] rootfs.vhdx not found, downloading...  <- new VM instance
[13:54:35] rootfs.vhdx not found, downloading...  <- another VM instance

Windows Event Log:
- 4x "previous system shutdown was unexpected" on 31/03/2026
- "The Claude service terminated unexpectedly. It has done this 1 time(s)."

Memory: 27.7GB committed out of 23.6GB visible RAM, forcing heavy page file use and system crash.

Steps to Reproduce

  1. Install Claude Desktop on Windows 11 Pro (MSIX install)
  2. Have Cowork working with an existing rootfs.vhdx bundle
  3. Allow Claude Desktop to auto-update to a version that ships a new bundle version hash
  4. Observe: Claude invalidates existing rootfs.vhdx and attempts to re-download
  5. Observe: Download starts but fails with EXDEV on the rename step
  6. Observe: CoworkVMService enters infinite retry loop consuming all system RAM
  7. Machine crashes — even without opening the Cowork tab

Note: The EXDEV error can also be reproduced on a fresh install on any Windows MSIX install.

Claude Model

Not sure / Multiple models

Is this a regression?

Yes, this worked in a previous version

Last Working Version

1.1.9493 (before March 31 2026 update)

Claude Code Version

1.1.9669 (Claude Desktop)

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

PowerShell

Additional Information

Related issues: #36642, #30584, #34605, #38316, #38457 (all report EXDEV, none fixed)
Non-ASCII username issue: #38208, #35366 (closed as invalid — incorrectly dismissed)

Fix is straightforward: replace fs.rename() with fs.copyFile() + fs.unlink() as fallback when EXDEV is caught. This is standard Node.js practice for cross-device moves.

Machine was rendered unusable 4 times in one day. CoworkVMService had to be permanently disabled via registry to prevent further crashes:
reg add "HKLM\SYSTEM\CurrentControlSet\Services\CoworkVMService" /v Start /t REG_DWORD /d 4 /f

This bug will affect ALL Windows MSIX users the next time Anthropic ships a bundle version update.

View original on GitHub ↗

This issue has 5 comments on GitHub. Read the full discussion on GitHub ↗