### Preflight Checklist - [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) 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? App UI stops responding a few minutes to ~16 minutes after launch. Web version unaffected. main.log shows every child process — Network Service, GPU, main

[BUG] Windows desktop app: all child processes terminated simultaneously (exitCode 1, reason=killed), UI hangs — 15 occurrences since 9/28, survives reinstall

Status Open

Maintainer reply None cached

Activity 0 comments · opened Oct 2, 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?

App UI stops responding a few minutes to ~16 minutes after launch. Web version unaffected. main.log shows every child process — Network Service, GPU, main webview renderer, git/transcript/file-index workers, and local MCP servers — terminated in the same second with reason: 'killed', exitCode: 1, while the main process keeps running. Example (2026-10-02 09:42:42):

[error] Network Service process gone (reason=killed, exitCode=1) outside app quit — main-process networking is broken until relaunch
[info] GPU process gone: { type: 'GPU', reason: 'killed', exitCode: 1, serviceName: 'GPU' }
[info] Main webview render process gone: { reason: 'killed', exitCode: 1 }
[error] Sentry caught: { eventId: '56a4264ad5b246a086f2f147b41571b8', value: 'Main webview render process gone' }

main.log

What Should Happen?

App should stay responsive indefinitely. Child processes (network service, GPU, renderer, workers, MCP servers) should not be terminated while the main process is still running and the user has not quit.

Error Messages/Logs

2026-10-02 09:42:42 [error] Network Service process gone (reason=killed, exitCode=1) outside app quit — main-process networking is broken until relaunch
2026-10-02 09:42:42 [info] [remote-tools-device] socket closed: 1006 Connection to the network service was lost (net::ERR_FAILED)
2026-10-02 09:42:42 [info] GPU process gone: { type: 'GPU', reason: 'killed', exitCode: 1, serviceName: 'GPU' }
2026-10-02 09:42:42 [info] Main webview render process gone: { reason: 'killed', exitCode: 1 }
2026-10-02 09:42:42 [info] [git] worker exited (1); will refork on next request
2026-10-02 09:42:42 [info] [transcript-read] worker exited (1); will refork on next request
2026-10-02 09:42:43 [error] Sentry caught: { eventId: '56a4264ad5b246a086f2f147b41571b8', type: 'Unknown', value: 'Main webview render process gone' }

Identical signature at every occurrence. Full main.log attached.

Steps to Reproduce

Launch Claude desktop (Windows x64 MSIX) and sign in.
Use the app normally — open a project, run a Cowork session or chat.
Between 3 and 16 minutes after launch, with no specific triggering action, all child processes are terminated simultaneously and the UI stops responding. Main process survives.

Occurrences (local time, UTC-7): 09-28 17:44:12 · 09-29 14:53:15, 15:45:13, 17:15:01, 17:18:24, 17:26:06, 17:39:40, 17:48:09 · 09-30 10:53:40, 12:09:27, 21:55:12, 22:08:05 · 10-01 16:17:23, 22:19:01 · 10-02 09:42:42

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

2.9939.2 (first occurrence was 5 hours after the 2.9939.4 update on 9/28; Windows 11 26H2 installed 9/29 — both changed in the same window)

Claude Code Version

Claude desktop 2.19675.0 (also reproduced on 2.9939.4 and 2.16120.0). Not a Claude Code CLI issue.

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

PowerShell

Additional Information

Environment: Windows 11 Home 26H2 (build 26300.9550, installed 2026-09-29), AMD Ryzen 7 7700, RTX 4070 (Studio driver 616.92), 64 GB RAM. MSIX package Claude_pzs8sxrjxfjjc.

Ruled out: No event-loop stalls, memory pressure, or renderer-unresponsive warnings before any kill (main process is healthy to the instant). Occurs with and without local MCP servers configured, with and without local folders linked (all moved to cloud 10/1), with hardware acceleration disabled, after clean NVIDIA driver reinstall, after clearing app data, and after a full uninstall including the Packages\Claude_pzs8sxrjxfjjc folder. Norton Security History shows no action against claude.exe.

The app logs "[windows-app-job] joined" at every launch. The kill signature — entire child set terminated in the same second, exit code 1, parent survives — is consistent with the job object being terminated.

Sentry event IDs: 56a4264ad5b246a086f2f147b41571b8 (10-02), 2528204c6ca34a5897e765a2ea00b778 (09-28).

Possibly related: #82345 (Windows UI hang surviving reinstall), #86265 (Windows desktop GPU process crash).

View original on GitHub ↗