Desktop app hangs after GPU process crash when loading a link preview; requires MSIX Repair to relaunch
Summary
Opening a link from chat (which loads in the app's embedded preview/browser pane) can crash the Electron GPU process. Instead of recovering, the app's main process hangs indefinitely. Windows eventually force-kills it via its hang-detection, and because the process wasn't exited cleanly, the MSIX package registration is left broken - the app then fails to relaunch with "Can't open this app" until a manual Repair is run (Settings > Apps > Claude > Repair).
Environment
- Windows 11 Home (10.0.22000)
- Claude desktop app 1.26832.0 (also reproduced on 1.24012.9)
- MSIX install, package family
Claude_pzs8sxrjxfjjc - GPU: NVIDIA GeForce RTX 3080, driver 32.0.16.1088 (2026-07-21) - current driver
Steps to reproduce
- In a chat, get/click a link to a page that sits behind a Cloudflare "Verifying you are human" interstitial with an animated widget (reproduced with
nexusmods.com/about/vortex/). - The app opens it in its in-app preview/browser pane.
- Shortly after the preview context is created, the GPU process crashes.
- The app UI stops responding. It does not recover or restart the GPU process.
- ~60-90s later, Windows logs an Application Hang (Event ID 1002) and force-closes
claude.exe. - Relaunching the app fails with "Can't open this app" (MSIX). A Repair via Apps & Features is required before it will open again.
Relevant log excerpt (%APPDATA%\Claude\logs\main.log)
09:36:33 [info] [Preview] Created session preview context { previewId: 'preview-local_...', totalContexts: 1 }
09:36:35 [info] GPU process gone: { type: 'GPU', reason: 'crashed', exitCode: 101457950, serviceName: 'GPU' }
-- app goes silent here, no further log lines --
09:37:38 [Application Hang / Event 1002] claude.exe stopped interacting with Windows and was closed.
09:38:53 [info] Starting app { appVersion: '1.26832.0', ... } <- only after manual relaunch
Same pattern also observed at 09:29:53 and on 2026-08-06 21:59:46 (app version 1.24012.9), so it's reproducible across sessions and versions, not a one-off.
Expected behavior
- The app should recover from a GPU process crash (respawn the GPU process / fall back to software rendering) rather than hanging the whole main process.
- Even in a worst case where the process must be killed, relaunching should not require a full MSIX Repair - that suggests state is being left in a broken/inconsistent place on an unclean exit.
Impact
Total loss of the running session/work in progress each time, plus several minutes of downtime to repair the app before it's usable again.
This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗