[BUG] Desktop App disappears without warning and cannot reopen. Reinstall Required to work again

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?

What's Wrong?

Claude Desktop becomes unresponsive and closes without warning during a synchronous subagent call (run_in_background: false) that uses WebSearch/WebFetch against a site that stalls or resists automated requests. No crash dialog appears — the window simply disappears.

Task Manager shows no CPU, memory, or disk spike at the moment it happens, which points to the process being blocked on network I/O rather than compute-bound. Windows Event Viewer logs this as an Application Hang (Event ID 1002, Hang type: Quiesce) every time — meaning Windows itself killed the process because it stopped responding to the OS's routine liveness check, not a normal crash.

After this happens, relaunching frequently fails with "This app can't open. There's a problem with Claude. Reinstall the application from its original install location or contact your administrator." Get-AppxPackage confirms the package status as Modified, NeedsRemediation. Recovering requires fully removing the package (Remove-AppxPackage), clearing leftover AppData folders, and reinstalling — a repair-in-place (Add-AppxPackage -Register) is not always sufficient; a full wipe has sometimes been necessary.

This has reproduced 4+ times over two days.

Possibly related to #85314 — same version, same platform, and the exact same downstream symptom (app disappears, package reports corrupted, full reinstall required). The trigger there is different (a Cloudflare check firing when the Browser pane opens an external site) versus a stalled synchronous subagent fetch here, so I'm filing separately per the checklist above rather than assuming the same root cause — but both point at the same failure pattern: some blocking operation hangs the main process long enough that Windows kills it, and the kill leaves the MSIX package in a corrupted/unregistered state. Worth a maintainer checking whether both trace back to a missing timeout somewhere in the blocking-call path.

What Should Happen?

A stalled synchronous subagent call (or any blocking network operation) should time out and fail gracefully rather than hanging the entire application indefinitely. Separately, an unclean process termination shouldn't be able to leave the MSIX package registration corrupted — at minimum the app should be able to self-repair or clearly report why it won't launch, rather than requiring manual Get-AppxPackage/Remove-AppxPackage intervention.

Error Messages/Logs

Log Name: Application
Source: Application Hang
Event ID: 1002
Level: Error
Hang type: Quiesce

The program claude.exe version 1.26832.0.0 stopped interacting with
Windows and was closed.
Faulting package full name: Claude_1.26832.0.0_x64__pzs8sxrjxfjjc
Faulting package-relative application ID: Claude

Steps to Reproduce

Session transcript (~/.claude/projects/<project>/<session-id>.jsonl) confirms the last action before the hang each time was launching a synchronous subagent, e.g.:

{"type":"tool_use","name":"Agent","input":{
  "description":"Trace [business] owner + profile",
  "subagent_type":"general-purpose",
  "run_in_background": false,
  "prompt":"...WebSearch and WebFetch [county assessor / Secretary of State portal]..."
}}

No further entries follow in the transcript until the app is relaunched — the turn never resolves.

Steps to Reproduce

1. In the Code tab, start a session and ask Claude to run a subagent with run_in_background: false that uses WebSearch/WebFetch against a site known to rate-limit or stall automated requests (in my case: a county property assessor portal and a Secretary of State business search).
2. Wait for the subagent call to be issued (visible in Verbose transcript view).
3. If the target site stalls rather than returning an error, the session appears to hang with no visible resource usage.
4. After some time, the app window disappears with no dialog.
5. Relaunching frequently produces the "This app can't open" error described above.

What I've Already Ruled Out

• Disk corruption — chkdsk /f /r, sfc /scannow, and DISM /Online /Cleanup-Image /RestoreHealth all came back clean before the most recent recurrence.
• Antivirus interference — checked Windows Security Protection History; no threats, quarantines, or blocks around any of the crash timestamps.
• Drive failure — Get-PhysicalDisk | Get-StorageReliabilityCounter shows zero uncorrected read errors.
• System-wide memory exhaustion — checked Microsoft-Windows-Resource-Exhaustion-Detector/Operational; no 2004/2005 events, so this is not Windows force-killing processes under system-wide memory pressure.

Claude Model

claude-sonnet-5 (subagent), claude-fable-5 (main session)

Is this a regression?

I don't know

Last Working Version

No response

Claude Code Version

Claude 1.26832.0.0 (bundled Claude Code CLI 2.1.222)

Platform

Anthropic API (Claude Pro/Max subscription via Desktop)

Operating System

Windows 10, build 10.0.19045

Terminal/Shell

Claude Desktop (MSIX install)

Additional Information

Session JSONL transcripts and full Event Viewer exports are available on request if it helps triage.

Claude Model

Not sure / Multiple models

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

Claude 1.26832.0.0 (bundled Claude Code CLI 2.1.222)

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

PowerShell

Additional Information

<img width="715" height="221" alt="Image" src="https://github.com/user-attachments/assets/8aa5cdf3-c82b-4def-a5f6-802a24dc836f" />

<img width="568" height="344" alt="Image" src="https://github.com/user-attachments/assets/e96d16d4-0274-40c7-affd-2cc4525a9611" />