[Windows] Opening tineye.com in the in-app browser crashes the entire desktop app (reproducible), corrupts MSIX registration, takes explorer.exe down with it

Environment

• Claude desktop app (Windows MSIX): Claude_1.26832.0.0_x64__pzs8sxrjxfjjc
• Windows 11 Pro build 26200
• GPU: AMD Radeon RX 7900 XTX, driver 32.0.31035.1003 (2026-07-23)
• 64 GB RAM, plenty of free disk

Repro

1. In a Claude Code session inside the desktop app, have Claude open https://tineye.com in the in-app Browser pane.
2. The entire desktop app process tree exits immediately — main window, all sessions, background service.

Reproduced deliberately a second time with the same result.

Impact observed (from Windows event logs)

• The app's main process stopped logging abruptly (main.log cut mid-write, no shutdown lines) — no WER AppCrash record, no crash dump, so the process died hard.
• On relaunch, Windows considered the MSIX package broken and auto-launched Microsoft Store Remediation (ms-windows-store:Remediation?PFN=Claude_pzs8sxrjxfjjc). The repair then failed repeatedly with 0x80073D02 ("apps need to be closed") because the app's own cowork-svc service (Claude service) was still registered/running — the Service Control Manager log shows the Claude service being reinstalled 7 times in one minute while remediation looped.
• CoworkVMService / Hyper-V VmSwitch churn accompanied each cycle (the Cowork VM NIC created/deleted).
• explorer.exe also died and restarted in the same window, along with other GPU-accelerated apps — the WER ReportQueue filled with video LiveKernelEvent dumps: Kernel_117 (VIDEO_TDR_TIMEOUT), Kernel_193 (VIDEO_DXGKRNL_LIVEDUMP), Kernel_1a8, Kernel_1b0, Kernel_1b8 (miniport failure / black-screen livedumps). So the failure appears to go through a GPU driver reset (TDR) triggered by rendering that site, and the app does not survive the device loss.
• Recovery required the Store repair flow plus manually restarting everything the crash took down.

Timeline (local, 2026-08-07)

• ~02:02 — app main process stops logging mid-operation
• 04:00:04 — Store launched via Remediation protocol for the Claude package
• 04:00:03–04:00:59 — Claude service (cowork-svc.exe) reinstalled 7×, repair failing 0x80073D02
• 04:01–04:03 — Cowork VM teardown/re-registration, package ACL repair, app relaunch
• 04:14 — explorer.exe restart

Suggested angles

• The in-app browser (Chromium) hitting a GPU driver reset on this site should lose the renderer, not the whole app + service.
• The MSIX repair loop (0x80073D02 because cowork-svc is still running) means a crashed app can wedge its own package repair — remediation never stops the service it installed.