claude-code - 💡(How to fix) Fix Recurring mid-session freeze on Windows — Opus 4.7, no consistent trigger

Symptom

Claude Code's desktop UI freezes mid-session and does not recover; the session has to be force-quit and restarted. The in-flight tool call and any unsaved scratch state are lost.

Recurrence

Six occurrences recently. Most recent: 2026-05-23. The most recent crash coincided with closing a Microsoft Word document, but earlier occurrences did not correlate with Word activity — that's an n=1 observation, not a confirmed trigger. No consistent trigger pattern has been identified across the six events.

Most-recent crash specifics (from local logs / screenshot the user took before force-quitting):

• Session was ~15 minutes in, ~7.2k tokens, mid Bash tool call (reading IBKR XML attribute mapping in a Python file). Nothing unusual about the tool call itself.
• Earlier in the session, no error / warning markers in main.log.

Environment

• OS: Windows 11 Pro 26200
• Claude Code app version: 1.6608.2 (x64, Electron isPackaged)
• CCD version: 2.1.128
• Node: 24.15.0
• Model: Opus 4.7 (1M context), Extra-high effort
• Shell: PowerShell

Local diagnostic artefacts available

The user has the following locally and can share on request:

• %APPDATA%\Claude\logs\main.log (~3.4 MB) — main process log spanning recent sessions
• %APPDATA%\Claude\logs\cowork_vm_node.log (~1.7 MB)
• %APPDATA%\Claude\logs\claude.ai-web.log (~319 KB)
• %LOCALAPPDATA%\Claude\Logs\chrome-native-host.log (~108 KB)
• Windows Application Event Log entries for the last 14 days (filtered to provider "Application Error" / ".NET Runtime" / Claude — most are benign profiler-attach failures, no crash dumps from Claude Code itself)
• No fresh entries in %LOCALAPPDATA%\CrashDumps\ — suggesting the process is hanging, not crashing-out cleanly

A quick grep of main.log for the obvious keywords (error / abort / fatal / crash / hang / oom / sigsegv / exception / unhandled / out of memory) surfaces only benign noise: "coder CLI not found" loops, EBUSY on chrome-native-host copy. Nothing in the log marks the freeze itself, consistent with a UI / event-loop hang rather than a process exit.

Impact

Each crash costs ~30–60 min to rebuild context from PICKUP.md + transcript. Six occurrences in recent weeks is past the threshold where it needs upstream attention. The user has adopted a workaround where Claude writes a defensive PICKUP file before any risky activity — contains the blast radius but doesn't fix the underlying bug.

What would help diagnose

1. Any preferred way to attach the log files (gist, direct upload, send to support email)? Happy to provide whichever is most useful.
2. Is there a verbose / trace logging mode that would capture the seconds before a freeze on the next occurrence?
3. Any known issues around Electron renderer-process hangs on Windows 11 26200 + app version 1.6608.x?

Filed via gh CLI on behalf of @drpaulvarcoe.