Cowork VM crashes after switching windows / system sleep, causing app unusable error and apparent loss of session history

Environment

• OS: Windows 11 Home, build 10.0.26200
• Claude Code: recently updated 2.1.220 -> 2.1.223 (auto-update log, 2026-08-06)
• Feature: Cowork (VM-backed sandbox), Hyper-V-based VM (HCS)

Summary

The Cowork VM repeatedly crashes/disconnects when the app is backgrounded (switching to another window, or the machine going to sleep) and later resumed. When this happens, the app shows an in-app error stating it can no longer be used, and the only recovery path is to reinstall the app. After reinstalling, the app appears to have no memory of prior conversations, which reads to the user as total data loss. This has happened at least 4 times.

Evidence

From C:\ProgramData\Claude\Logs\cowork-service.log, recurring since at least 2026-07-17 through 2026-08-06 (representative sample):

2026/07/26 13:35:54.776121 [VM] Graceful shutdown failed, forcing termination: HcsShutdownComputeSystem failed with HRESULT 0x0/0x80070032: {"Error":-2147024846,"ErrorMessage":""}
2026/07/26 13:35:54.837633 [RPC] Read error: failed to read length: An established connection was aborted by the software in your host machine.
2026/07/26 13:35:54.887258 [Console] Daemon console read error: No process is on the other end of the pipe.
2026/08/06 13:36:10.330794 [VM] Graceful shutdown failed, forcing termination: HcsShutdownComputeSystem failed with HRESULT 0x0/0x80070032: {"Error":-2147024846,"ErrorMessage":""}
2026/08/06 13:36:10.341241 [RPC] Read error: failed to read length: An established connection was aborted by the software in your host machine.
2026/08/06 13:36:11.306043 [Console] Daemon console read error: No process is on the other end of the pipe.

Also recurring throughout the same window:

[Server] Persistent RPC: connection ended: failed to read length: EOF
Warning: failed to open service for recovery config: Access is denied.

Steps to reproduce

1. Start a Cowork session in the Claude Code desktop app on Windows.
2. Switch away to another window/app for a while, or let the machine go to sleep.
3. Return to Claude Code.
4. App reports it can no longer be used; only recovery is reinstalling.

Expected behavior

• The Cowork VM connection should recover gracefully after the host resumes from background/sleep, without forcing a full app reinstall.
• Even if a reinstall is required, prior local session/chat history should not appear lost to the user.

Additional (secondary) observation

Separately, local session transcripts under ~/.claude/projects/ are in fact preserved on disk, but appear to be scoped by the exact working-directory path. Since the working directory name changed slightly across sessions (e.g. desktop folder renamed), the app UI seemed to show no memory of prior work even though the underlying data was not deleted. This compounds the impression of "everything lost" when the VM-crash issue above forces a fresh start.

Impact

User has experienced this 4 times, each time believing all prior work/conversation context was permanently lost.