claude-code - 💡(How to fix) Fix [BUG] Windows v1.2773.0: silent Electron main-process death ~20min after launch during background event-log flush [1 comments, 2 participants]

What's wrong

After updating to Desktop shell 1.2773.0 on Windows 11, the entire Claude Desktop process dies silently roughly 20 minutes after launch. No crash dialog, no Crashpad dump, no Sentry report — main.log simply stops mid-sentence. The Electron main process exits, not just a renderer or CCD child.

This appears distinct from the related crash reports #48522 / #48926 (UI-switch-triggered) and #48813 (gradual 1-2 hour kernel pool leak), though I'm seeing elevated non-paged pool too (see below).

Environment

• Windows 11 Home 10.0.26200, 32GB RAM
• Claude Desktop shell 1.2773.0 (Microsoft Store install, auto-updated from 1.2581.0 on 2026-04-15 23:17)
• CCD binary 2.1.92 (the deliberate rollback Anthropic shipped with this shell, per #46767-adjacent rollback)
• Default profile, MCP servers: Gmail, Calendar, gdrive, Vercel, qmd, scheduled-tasks, dispatch, Claude in Chrome, computer-use, PDF, mcp-registry
• A long-lived dispatch session (cse_…) plus CCD scheduled tasks (heartbeat-check, qmd-reindex, auto-flush-recovery, etc.)

Repro

1. Launch Claude Desktop on Windows 11 (1.2773.0)
2. Leave it idle / let CCD scheduled tasks run in background
3. ~20 min later, the process dies silently. Task Manager shows no claude.exe; main.log ends mid-flush.

Crash signature (2026-04-16)

08:51:40 [info] Starting app { appVersion: '1.2773.0', nodeVersion: '24.14.0' }
08:51:40 [info] [CCD] Initialized with version 2.1.92
…
09:07:25 Spawning new session for scheduled task heartbeat-check
09:08:10 [CCD CycleHealth] healthy cycle (44s, hadFirstResponse=true)
09:09:58 Spawning new session for scheduled task qmd-reindex
09:10:15 [CCD CycleHealth] healthy cycle (17s, hadFirstResponse=true)
09:10:28 [info] [EventLogging] Flushing 16 events
<<< log ends; file mtime = 09:10:28; no further entries >>>

Both scheduled tasks completed cleanly. The process died 13 seconds after the last Stop hook, mid-event-flush. No exception, no graceful shutdown, no crash dump in %LOCALAPPDATA%\Claude\Crashpad\reports\.

Notes / related observations

• Kernel non-paged pool at the time of writing (after ~2h of partial use): 1.04 GB (normal 0.2-0.5 GB). So #48813's leak is happening here too, but the sudden death at 20 min is too early to be the leak going terminal — looks like a separate failure mode.
• Pre-update Desktop shell 1.2581.0 with the same CCD 2.1.92 was stable in the hours before the auto-update.
• No Code/Chat tab switching took place — purely background activity. So this isn't #48522 / #48926.
• [CCD] Ignoring defaultMode "bypassPermissions" from ~/.claude/settings.json warning visible — unrelated, but flagging in case it interacts with event-flush ordering.

Logs

Happy to share main.log, claude.ai-web.log, cowork_vm_node.log from %APPDATA%\Claude\logs\ on request — they're large and contain personal session content so I'd prefer to share via a private channel rather than attach publicly.

Related issues

• #48522 — Claude Desktop Crashing (Windows, 1.2773.0, area:desktop)
• #48926 — Application hang on Chat ↔ Code tab switch (Windows 11, 1.2773.0)
• #48813 — Cowork/CC-in-Desktop ~19GB kernel non-paged pool leak (Windows)
• #49023 — 1.2773.0 V8 OOM on launch (Windows)
• #46767 — Tool results silently dropped (2.1.101 regression — clean on 2.1.92)