[BUG] Claude CLI 1.0.96 randomly/predictably freezes on Windows - affects all commands, persists across sessions

Resolved 💬 16 comments Opened Aug 29, 2025 by nsxdavid Closed Jan 6, 2026
💡 Likely answer: A maintainer (ant-kurt, collaborator) responded on this thread — see the highlighted reply below.

Environment

Other: Claude CLI (standalone terminal usage)
• Claude CLI version: 1.0.96 (Claude Code)
• Operating System: Windows 11 Pro Build 26100
• Terminal: PowerShell 7.5.2 (tested in both Warp and standard PowerShell)

Bug Description
Claude CLI version 1.0.96 randomly (sometimes or predictably (with certain commands like doctor) freezes and becomes completely unresponsive on Windows, affecting all commands including claude doctor, claude -c, and normal interactive usage. The freeze is so severe that standard interrupt signals (Ctrl+C, Ctrl+Z) cannot terminate the process, requiring force-kill via Task Manager. Once a freeze occurs with claude -c, the issue persists in all subsequent sessions, suggesting session state corruption. This is a critical regression - version 1.0.85 works reliably.

Steps to Reproduce
Method 1 (Most Reliable):

  1. Install Claude CLI 1.0.96: npm install -g @anthropic-ai/claude-code@1.0.96
  2. Run: claude doctor
  3. Command freezes immediately after output "Checking installation status..." and hangs indefinitely

Method 2 (Random but Common):

  1. Use claude for interactive sessions
  2. After some usage, Claude randomly freezes during startup or mid-conversation
  3. claude -c, will load previous context but be immediately frozen
  4. Once frozen with -c, all subsequent claude -c attempts also freeze
  5. Pattern suggests persistent session state corruption

Method 3 (General Usage):

  1. Run any Claude command during normal usage
  2. Freezing can occur randomly at various points
  3. No specific trigger pattern identified

Expected Behavior
• All Claude commands should execute normally and complete
• Interactive sessions (claude -c) should start reliably
• Commands should be interruptible with standard terminal signals
• Previous session issues should not affect new sessions

Actual Behavior
• Random freezing during any Claude command execution
• Complete terminal freeze - no response to Ctrl+C, Ctrl+Z, or any input
• Persistent session corruption - once claude -c freezes, all future -c attempts freeze
• High resource consumption - process shows high CPU (281+ seconds) and memory (390MB+)
• No error output - process hangs silently with no diagnostic information
• Only recovery - force-kill process via Task Manager

Additional Context

Critical Pattern - Session State Corruption:
• When claude -c freezes once, ALL subsequent claude -c attempts freeze immediately
• This suggests corrupted session state that persists across process restarts
• Pattern indicates the issue may be related to configuration files or cached state

Process Information When Hanging:
System Details:
• Windows 11 Pro x64 (Build 26100)
• Node.js: v22.16.0
• NPM: v11.5.2
• RAM: 130GB total
• PowerShell: 7.5.2

Freeze Triggers Observed:
• claude doctor - 100% reproducible freeze
• claude -c - random freezes, becomes persistent once it happens
• General usage - random occurrence during various commands
• No specific content trigger identified - seems related to internal state

Regression Information:
• Last Working Version: 1.0.85 (completely stable)
• First Broken Version: Unknown (likely 1.0.95 or 1.0.96)
• Workaround: Downgrade to 1.0.85, but auto-updates re-break it
• Scope: Affects all Claude CLI functionality on Windows

Impact:
• Severity: Critical - makes Claude CLI completely unreliable on Windows
• Randomness: Unpredictable failures during normal usage
• Persistence: Session corruption makes problem permanent until config reset
• Platform: Windows-specific (similar patterns reported on macOS in #6775)

Investigation Priorities:

  1. Session state management - corruption persists across restarts
  2. Configuration file handling - may be corrupting saved state
  3. Shell snapshot/permission system - likely culprit based on #6775
  4. Windows-specific process/file handling changes in recent versions

Related Issues:
This appears related to #6775 but is more severe and affects direct terminal usage with persistent session corruption, not just programmatic spawning.

This regression makes Claude CLI fundamentally unreliable and unusable on Windows. The random nature combined with persistent corruption makes it impossible to use for any serious work.

View original on GitHub ↗

16 Comments

github-actions[bot] · 10 months ago

Found 3 possible duplicate issues:

  1. https://github.com/anthropics/claude-code/issues/6705
  2. https://github.com/anthropics/claude-code/issues/6757
  3. https://github.com/anthropics/claude-code/issues/6796

This issue will be automatically closed as a duplicate in 3 days.

  • If your issue is a duplicate, please close it and 👍 the existing issue instead
  • To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

ftToddSSG · 10 months ago

My agent is outing me as a summary to my investigation. Don't give me the eye rolls for my conversation sizes. They have worked for me so far ;)

Follow-up: Confirmed Reproduction with Additional Details

I can confirm this exact issue on Windows with Claude Code 1.0.96. After investigating what I initially thought was a conversation size problem, I've isolated this to be a 100% reproducible bug with specific commands.

Environment Details

  • Claude CLI version: 1.0.96 (npm installation via @anthropic-ai/claude-code)
  • Operating System: Windows 11
  • Terminal: PowerShell (both built-in and external)
  • Installation method: npm global install
  • nvvm used and powershell runs npm via npm.ps1

Reproduction Steps - 100% Reliable

Method 1: claude doctor (Always hangs)

``` powershell
claude doctor

Output: "Checking installation status..."

--> Immediate hang, no further output

--> Completely unresponsive to Ctrl+C, Ctrl+Break (vs code terminal, pwsh Ctrl+Break goes to powershell debug and I can 'exit')

--> Requires taskkill /f /im node.exe

### Method 2: claude --resume (Always hangs after conversation cleanup)
```powershell
claude --resume
# --> Hangs during conversation loading/processing
# --> No output, completely frozen
# --> Same behavior: unresponsive to interrupt signals

Key Observations

  1. Not conversation-size related: Initially suspected large conversation files (I had some 11MB+ transcripts), but after archiving ALL conversations to leave only fresh/small ones, the hang persists identically.
  2. Process behavior: When hung, the node.exe process shows:
  • High CPU usage (sustained)
  • Memory consumption continues to grow
  • No response to standard interrupt signals
  • Only recoverable via force kill
  1. Session state corruption: After a hang occurs, subsequent claude --resume attempts hang immediately, suggesting persistent state corruption as reported in the original issue.
  2. Clean claude works: Fresh interactive sessions (claude with no flags) work normally, confirming the issue is specific to diagnostic/resume code paths.

Additional Context

This appears to be an infinite loop in the diagnostic/health check systems rather than a network or configuration issue. The fact that both claude doctor and claude --resume trigger the same hang suggests they share common code paths for system validation or conversation processing.
The Windows-specific nature aligns with the original report - this seems to be a regression in recent versions affecting Windows process/file handling during these specific operations.

Workaround (not really)

So far, using claude (fresh sessions only) and avoiding --resume and doctor commands entirely. Suspect this would be short lived.
This confirms the critical nature of this regression - these are core diagnostic and workflow features that are completely broken on Windows 11 in 1.0.96.

nsxdavid · 10 months ago

I would add that it is both claude -r and -c (I guess -c is a subset of -r, but just being padantic)

ftToddSSG · 10 months ago
I would add that it is both claude -r and -c (I guess -c is a subset of -r, but just being padantic)

I'm not entirely sure how it all works (I should probably read the source code at some point). It definitely has to do with a scan of the current conversations, and something that --resume does automatically (maybe to build a summary to present to the user in the resume function?). It's why I cleaned out the conversation history as a first step. I thought something in them was causing the failure until I tried to resume a previous conversation after doing some fixes and testing.

nsxdavid · 10 months ago
(I should probably read the source code at some point)

I didn't think the source code for claude code was available. Have I been sleeping on this?

ant-kurt collaborator · 10 months ago

claude doctor has been resolved as of 1.0.98

For continue / resume - can you confirm your %UserProfile%\.claude.json is well-formed? Do you see any relevant-looking errors when running with flag --debug or env var ANTHROPIC_LOG=debug?

nsxdavid · 10 months ago

Testing .98, seems better so far.... claude doctor working now. Will report back if it seems resolved in the less predictable situations.

nsxdavid · 10 months ago

@ant-kurt Unfortunately I can get it to hang still using claude -c. Nothing in --debug stands out. A failing MCP server is only issue but unrelated, far as I can tell. %UserProfile%\.claude.json is verified well-formed JSON.

Version: 1.0.98

Anything else I can do to help debug?

nsxdavid · 10 months ago

@ant-kurt Following up, seems pressing ENTER does cause it to become interactive again as per a mention by UsernameZero from earlier (which I strangely don't see anymore, but is in my email).

ghost · 10 months ago

I deleted my earlier comment because I initially thought it wasn’t relevant, but it turns out it does provide a usable workaround. Reposting for clarity:

When you run claude -r, you need to press ENTER afterwards for the session picker to become responsive.

I’m not sure if the same applies to other commands, but this behavior seems specific to Windows PowerShell/Terminal.

blimmer · 10 months ago

I encountered this on MacOS, too. It looks like I got stuck on version 1.0.96. I upgraded to latest (1.0.109) via:

curl -fsSL https://claude.ai/install.sh | bash

and the issue was resolved

MikeWise2718 · 10 months ago

Getting this on my Windows 11 desktop, but not on a very similarly configured Windows 11 laptop. claude -c will often hang without any way of getting out of it. claude alone always works fine. claude -r seems to work when I enter return, but I think it wasn't yesterday, not sure anymore.

I installed it both with node and with bun, and I think it was worse and harder to get out of with bun. So staying with node for now. I looked with procmon, and it was doing things, trying to create files, etc. while it was hung.

Also when it is hung the mouse cursor disappers when you hit a key, and then comes back after awhile, so some key processing is going on, but the keys pressed never appear and you can't actually do anything.

Using 1.0.109 today which has it. It started when I upgraded 1.0.72 to 1.0.9x - can't remember exactly.

Happens on Windows Terminal, but I also tried it using the old Windows Console Host, and there was no difference.

Also tried deleting all my settings, conversations, etc. Didn't help.

PCJones · 10 months ago
I deleted my earlier comment because I initially thought it wasn’t relevant, but it turns out it does provide a usable workaround. Reposting for clarity: When you run claude -r, you need to press ENTER afterwards for the session picker to become responsive. I’m not sure if the same applies to other commands, but this behavior seems specific to Windows PowerShell/Terminal.

This also fixed it for me with claude -c

github-actions[bot] · 7 months ago

This issue has been inactive for 30 days. If the issue is still occurring, please comment to let us know. Otherwise, this issue will be automatically closed in 30 days for housekeeping purposes.

github-actions[bot] · 6 months ago

This issue has been automatically closed due to 60 days of inactivity. If you're still experiencing this issue, please open a new issue with updated information.

github-actions[bot] · 6 months ago

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.