[BUG] Claude Desktop crashes on switching to Code mode (Windows 11 Home, MSIX)

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?

Claude Desktop (v1.3109.0.0, MSIX, Windows 11 Home) crashes every time I switch from Chat mode to Code (Agent) mode. Chat mode works fine. The crash is
100% reproducible, even on a completely fresh install with all data wiped. The last log entry before crash shows getSupportedCommands called with
sessionId=undefined, cwd=undefined, suggesting the session fails to initialize. No error is logged — the Electron renderer silently dies, leaving only a
~35MB Crashpad dump. The VM itself starts and runs successfully; it's the client-side app that crashes. This started after auto-update from ~v1.1062 to
v1.2773+.

What Should Happen?

Title: Claude Desktop crashes on switching to Code mode (Windows 11 Home, MSIX)

Body:

Bug Description

Claude Desktop crashes immediately when switching from Chat mode to Code (Agent) mode. Chat mode works fine. The crash is 100% reproducible.

Environment

• OS: Windows 11 Home (Build 26200)
• App version: 1.3109.0.0 (MSIX)
• GPU: Intel Iris Xe Graphics (Driver 30.0.101.2079)
• RAM: 15.7 GB

Steps to Reproduce

1. Open Claude Desktop
2. Use Chat mode normally (works fine)
3. Switch to Code mode
4. App crashes instantly

What I've Tried

• Clean uninstall + reinstall (all data wiped)
• --disable-gpu flag
• hardware_acceleration_mode: false in Preferences
• Deleting VM bundles / stopping CoworkVMService
• Setting ELECTRON_EXTRA_LAUNCH_ARGS=--disable-gpu
• None of the above fixed the crash

Log Analysis

The last entries in main.log before every crash:

[CCD] LocalSessions.setFocusedSession: sessionId=null
LocalAgentModeSessions.getSupportedCommands: sessionId=undefined, cwd=undefined

The VM service (cowork-service.log) shows the VM actually starts and runs successfully:

[HCS] HcsCreateComputeSystem returned: hr=0x0
[VM] VM started successfully
[RPC] sdk-daemon is ready
[VM] API reachability: REACHABLE

Then the Electron client dies (connection ended: failed to read length: EOF) while the VM remains in Running state. A ~35MB Crashpad dump is generated
each time.

Notes

• This started after updating from ~v1.1062 to v1.2773+. Older versions (used ~1 week ago) did not have this issue.
• The crash happens even on a completely fresh install with no prior data.
• sessionId=undefined, cwd=undefined suggests the Code mode session initialization fails to resolve these values before calling getSupportedCommands,

leading to an unhandled exception in the renderer process.

Error Messages/Logs

● # main.log (last entries before crash)
  2026-04-17 11:58:01 [info] [CCD] LocalSessions.setFocusedSession: sessionId=null
  2026-04-17 11:58:01 [info] LocalAgentModeSessions.getSupportedCommands: sessionId=undefined, cwd=undefined
  2026-04-17 11:58:02 [info] Config file written
  2026-04-17 11:58:02 [info] [CCD] LocalSessions.setFocusedSession: sessionId=null
  # App dies here — no error logged, only a ~35MB Crashpad dump generated

  # main.log (Sentry capture on startup)
  2026-04-17 11:04:56 [error] Sentry caught: {
    eventId: '13a17c242cb74bd1877ddf72e5545839',
    type: 'Unknown',
    value: 'No message',
    stack: undefined
  }

  # cowork-service.log (VM is fine, client dies)
  2026/04/17 11:37:32 [Server] Persistent RPC: connection ended: failed to read length: EOF
  2026/04/17 11:37:32 [Server] Session USER:...\Claude has no active connections, idle timeout starts
  # VM remains Running after client crash:
  2026/04/17 11:37:36 [HCS] Properties result: {...,"State":"Running"}

Steps to Reproduce

1. Install Claude Desktop v1.3109.0.0 (MSIX) on Windows 11 Home (Build 26200)
2. Launch the app and log in
3. Use Chat mode — works normally
4. Click on Code mode (Agent mode) in the sidebar
5. App crashes immediately — every time, 100% reproducible

Additional context:

• Fresh install with no prior data still crashes (tested after full uninstall + wiping %APPDATA%\Claude, %LOCALAPPDATA%\Packages\Claude_pzs8sxrjxfjjc, and

registry entries)

• The VM successfully starts (HcsCreateComputeSystem hr=0x0, VM started successfully, API reachability: REACHABLE) — the crash is in the Electron

renderer, not the VM

• Last log entries before crash (main.log):

[CCD] LocalSessions.setFocusedSession: sessionId=null
LocalAgentModeSessions.getSupportedCommands: sessionId=undefined, cwd=undefined

• A ~35MB Crashpad dump is generated at Crashpad/reports/ on each crash
• Attempted workarounds (none worked): --disable-gpu, ELECTRON_EXTRA_LAUNCH_ARGS, deleting VM bundles, stopping CoworkVMService,

hardware_acceleration_mode: false in Preferences

• Previously working on ~v1.1062 (around April 10). Crash started after auto-update to v1.2773+

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

1.3109.0.0

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Other

Additional Information

_No response_