Desktop (Windows): GPU process crash (0x060C201E) during Workflow with browser-pane research agents; app unlaunchable until reinstall, twice

Summary

Running a background Workflow with parallel web-research subagents crashed the Claude Desktop app (Windows) twice in a row, ~35 minutes apart, with an identical signature: the Electron GPU process crashes (exitCode: 101457950 / 0x060C201E) about 3 seconds after a [PreviewContext] warning from the in-app browser pane. After each crash the app failed to relaunch entirely and the user had to reinstall it (twice in one morning). The Crashpad reports directory is empty, so no minidump was captured.

Environment

• App: Claude Desktop 1.40609.0 (packaged), CCD 2.1.247, Node 24.18.1, x64
• OS: Windows 11 Pro 10.0.26200
• Hardware: 64 GB RAM, NVIDIA GPU (NVIDIA App present); memory logs show the system far from pressure (~40 GB free)

What was running

A Workflow tool run with 3 parallel research subagents (schema-constrained agent() calls). Two did WebSearch/WebFetch only; the third was asked to check how large EU retail sites (amazon.de, zalando, decathlon, …) present delivery estimates in their carts, and appears to have driven the in-app browser pane to those external sites. Both crashes landed while that third agent was the one still working (its transcript file was last written at the crash minute both times).

The same session had previously run many Workflow code-review fan-outs over local files without any issue — the browser-pane-to-external-sites combination is what separates the crashing runs from the healthy ones.

Log signature (both crashes, %LOCALAPPDATA%\Claude\Logs\main.log)

2026-08-30 09:06:27 [warn] [PreviewContext] Blocked subresource to private-resolving host { resourceType: 'xhr' }
2026-08-30 09:06:30 [info] GPU process gone: {
  type: 'GPU',
  reason: 'crashed',
  exitCode: 101457950,
  serviceName: 'GPU'
}

2026-08-30 09:41:28 [warn] [PreviewContext] Blocked subresource to private-resolving host { resourceType: 'xhr' }
2026-08-30 09:41:31 [info] GPU process gone: {
  type: 'GPU',
  reason: 'crashed',
  exitCode: 101457950,
  serviceName: 'GPU'
}

These are the only two GPU process gone records in the log. Nothing is logged after each one until the fresh Starting app entry that follows the reinstall (09:28:23 and 10:09:54; ClaudeSetup.log timestamps match).

Expected

• A GPU process crash should not take the whole app down un-recoverably; Chromium normally restarts the GPU process (or falls back to software rendering).
• Even if the app dies, the next launch should succeed — instead it would not start at all until reinstalled, which suggests corrupted GPU/graphics cache state (GPUCache?) is read back at startup and kills it again. Clearing that state on a detected GPU-crash exit would likely have avoided both reinstalls.
• Crashpad should have captured a dump; Crashpad/reports is empty.

Repro (as observed)

1. Windows desktop app, a Claude Code session with the browser pane available.
2. Launch a background Workflow whose subagents browse external retail sites via the in-app browser while other agents run web search.
3. Within ~10 minutes: GPU process crash → app gone → app will not relaunch until reinstall.

Happened 2/2 times; the user is understandably unwilling to try a third, so I can't narrow it to the specific site. Happy to pull more from the machine's logs on request — the full main.log around both timestamps is preserved.