[BUG] GPU process crashes repeatedly when Browser pane preview screenshot fails on a backgrounded session

Status Closed — duplicate
Reported on v2.1.221
Maintainer reply None cached
Activity 3 comments · opened Aug 4, 2026 · closed Aug 25, 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 version: 1.25927.0.0 (Windows, MSIX) · OS: Windows 11 Pro 10.0.26200 · GPU: RTX 3060 Laptop (driver 32.0.16.1088) + Intel Iris Xe hybrid

Summary: The whole app crashes and Windows offers the MSIX "repair" dialog. Traced via main.log: the GPU/Chromium process dies with the same exit code every time, always seconds after a failed preview-screenshot capture of the Browser pane.

Ruled out: GPU driver (current, and crash persisted after disabling NVIDIA Overlay entirely, same exit code).

Suggested fix: guard capturePreviewScreenshot/WarmLifecycle:preview so a timeout on a non-composited pane fails gracefully instead of taking down the GPU process.

What Should Happen?

The preview-screenshot capture (capturePreviewScreenshot / WarmLifecycle:preview) should fail gracefully when the Browser pane isn't composited (as the warning message already indicates it detects), without crashing the GPU process. A failed/skipped thumbnail for a backgrounded session should never take down the whole app.

Error Messages/Logs

[warn] [Preview] capturePreviewScreenshot failed: Screenshot timed out after 5s: the Browser pane is not displayed, so the page is not compositing frames.
[info] GPU process gone: { reason: 'crashed', exitCode: 101457950, serviceName: 'GPU' }

Timestamps: 23:09:14, 23:13:39, 23:28:48 — all exitCode 101457950, all preceded by the same WarmLifecycle:preview timeout warning.

Steps to Reproduce

Start a session using the Browser pane, background/minimize the window while it's still active, wait for the periodic preview warm-up to fire → capture times out → GPU process crashes → app dies.

Claude Model

Sonnet (default)

Is this a regression?

Yes, this worked in a previous version

Last Working Version

1.25927.0.0

Claude Code Version

2.1.221

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Non-interactive/CI environment

Additional Information

Update: Independently confirmed on entirely different hardware by @rbenhaga — same exitCode: 101457950 on Intel Arc integrated graphics only (no NVIDIA), ruling out NVIDIA-hybrid/driver behavior as the root cause.

Two distinct failure paths lead to the same exit code:

  1. CONTEXT_LOST_WEBGL logged twice, no Crashpad dump
  2. Crashpad dump generated (34 MB), no WebGL loss logged anywhere

This means the fix suggested above (guarding capturePreviewScreenshot against the timeout) likely only covers path 1. The guard probably needs to sit lower in the stack, handling GPU process termination generically via the GPU process gone event rather than assuming a screenshot timeout as the sole origin.

Also related: this crash can escalate the MSIX package state from Modified to Modified, NeedsRemediation (see #83318), forcing a reinstall that wipes local sessions — so this is a potential data-loss path, not just a crash.

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗