[BUG] Windows desktop app main process does ~95,000 file ops/sec → kernel non-paged pool leak (~14 GB/hr) → full system freeze. Regression in 1.11847.5

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?

Environment

• Claude Desktop (MSIX/Store) 1.12603.1, bundled Claude Code CLI 2.1.165
• Windows 11 Pro 25H2 (build 26200), 16 GB RAM, HP ProDesk 400 G5, i5-8500

Summary
Since the 1.11847.5 auto-update (installed 2026-06-10 07:09, per AppX deployment log), the app's main Electron process (the Claude.exe instance with no --type flag) issues ~95,000 file operations/sec continuously. Breakdown: ~94,500/sec are "IO Other" operations (metadata/control: stat, directory-change notifications, pipe peeks) vs only ~750 reads + ~700 writes/sec — i.e. a runaway polling loop, not real file work. It is 99.3% of all file I/O on the machine.

System-level impact (the severe part)
Every op traverses the filesystem minifilter stack, and NTFS leaks an Ntfc-tagged non-paged pool allocation (~44 bytes/op) whenever any AV minifilter is attached — reproduced identically with AVG and with stock Windows Defender (WdFilter), and with scan exclusions on the app's data dir. Result: pool grows ~14 GB/hour, exhausts 16 GB RAM in ~1–1.5 h, and the machine hard-freezes with no bugcheck (Kernel-Power 41, Bugcheck=0). 8 freezes in the first 2 days post-update.

Regression window

• 1.11187.4 (installed 2026-06-06): no issue, 4 days clean
• 1.11847.5 (installed 2026-06-10): onset day
• 1.12603.1 (2026-06-12): still reproduces, measured live

Ruled out (tested individually): June Windows update KB5094126 (rolled back, no change), AVG (uninstalled, no change), Defender exclusions (no change), RAM (memtest clean), NIC/thermal.

Survives full app restart; persists while the app is simply open with an active session.

What Should Happen?

idle/working desktop app should be doing orders of magnitude fewer file ops (file-watching via change notifications, not a hot poll loop).

Error Messages/Logs

system halts after memory is exhausted

Steps to Reproduce

Repro / measure

Get-Counter '\Process(claude*)\IO Other Operations/sec' -SampleInterval 2 -MaxSamples 3
Get-Counter '\Memory\Pool Nonpaged Bytes'   # watch growth over 60s

Claude Model

Not sure / Multiple models

Is this a regression?

Yes, this worked in a previous version

Last Working Version

1.11187.4

Claude Code Version

1.11847.5 and 1.12603.1

Platform

Other

Operating System

Windows

Terminal/Shell

PowerShell

Additional Information

_No response_