[Bug] Claude Code desktop application crashes with unresponsive processes after startup

I had the same 1YCZHOG crash and found the root cause + workaround.

Root cause: A Code session's metadata JSON still has "isArchived": false (left in "active/running" state after an unexpected exit). When you click the session, the app tries to reconnect to the dead CLI process → React fatal error (#185) → crash screen.

Fix:

1. Navigate to %APPDATA%\Claude\claude-code-sessions\

• For Microsoft Store/UWP installs: %LOCALAPPDATA%\Packages\Claude_{id}\LocalCache\Roaming\Claude\claude-code-sessions\

1. Find the subfolder {org_id}\{user_id}\
2. Open each local_*.json file and look for the session that has "isArchived":false
3. Change it to "isArchived":true
4. Important: Save as UTF-8 without BOM (Notepad++ → Encoding → UTF-8, not "UTF-8-BOM")
5. Restart the app

The session will open normally with full conversation history after this change.

Additional technical details from investigation:

Crash sequence (from main.log and claude.ai-web.log):

LocalSessions.setFocusedSession: sessionId=local_67b96e49...
LocalSessions.getGitDiff: cwd=...\worktrees\zealous-babbage, base=main
[warn] Claude projects directory not found
LocalSessions.setFocusedSession: sessionId=null

[error] Error: Minified React error #185
[error] [BOOTSTRAP] Fatal error boundary triggered [object Object]
[error] Uncaught Error: Minified React error #418

What was ruled out (to save your debugging time):

• Not the JSONL conversation log — crash occurs even with an empty/10-line JSONL
• Not large lines in JSONL — other sessions with 433KB lines open fine
• Not a missing worktree directory — directory exists with valid .git
• Not session JSON field differences — identical structure to working sessions
• Not the nul reserved filename in worktree — present in other non-crashing sessions

Two additional related issues found:

1. Empty session JSON → sessions disappear from UI: If claude-code-sessions/{org}/{user}/local_*.json becomes 0 bytes (e.g., after a crash), the session silently disappears. Log shows: Failed to load session from local_xxx.json: Unexpected end of JSON input. Also, files with UTF-8 BOM (0xEF 0xBB 0xBF) are rejected — the app requires BOM-free UTF-8.

1. Windows UWP app ignores CLAUDE_CONFIG_DIR for JSONL lookup: When CLAUDE_CONFIG_DIR is set to relocate .claude (e.g., to D:\.claude), the CLI writes JONLs to the new location, but the desktop app still looks in C:\Users\{user}\.claude\projects\. Workaround: mklink /D C:\Users\{user}\.claude\projects D:\.claude\projects

Suggested fix: When a session has isArchived: false but the CLI process is not running, the app should gracefully fall back to read-only mode instead of crashing the React renderer.

Just some feedback, I've tried the suggested workaround putting all sessions to true, unfortunately no success. Also tried removing the .git folder as mentioned in the linked issues and that did not work either. If there is any more info I can provide, do let me know. But as I'm not very technical, please be as precise as you can with the questions. Many thanks!

I'm experiencing the same 1YCZHOG crash on Windows. The app opens and crashes after a few seconds every time.

I've tried the following without success:

• Deleted the entire %APPDATA%\Claude folder (all contents, not just specific subfolders)
• Completely uninstalled and reinstalled Claude Desktop
• Restarted PC multiple times
• Changed "isArchived" to true in claude-code-sessions JSON files
• Cleared all cache folders (Cache, Code Cache, GPUCache, blob_storage, Session Storage, etc.)

Nothing works — the 1YCZHOG error persists after every attempt. This has been going on since around February 17th.

Is this a server-side issue? status.claude.com showed high error rates around that date. If so, is there an ETA for a fix?

Key errors from logs:

From claude.ai-web.log:

[error] [REACT_QUERY_CLIENT] QueryClient error: Error: ["account_profile"] data is undefined
[error] Connecting to 'https://a-api.anthropic.com/v1/m' violates the following Content Security Policy directive
[error] Fetch API cannot load https://a-api.anthropic.com/v1/m. Refused to connect because it violates the document's Content Security Policy.
[error] Error: Minified React error #185
[error] [BOOTSTRAP] Fatal error boundary triggered [object Object]

From main.log:

[error] Failed to read sessions directory: ENOENT: no such file or directory, scandir 'C:\Users\...\claude-code-sessions\...'
[error] [LocalSessionManager] loadSessions failed during account transition, restoring previous sessions

The CSP violation blocking a-api.anthropic.com seems significant — the app is blocking its own API connection. Is this a known issue with the MSIX/Microsoft Store build?

Additional data: 6 reproducible crashes on v1.1.3541 (Windows VM, agent mode)

I'm hitting the same 1YCZHOG crash, but with a 100% reproducible trigger across 6 independent sessions. Every single crash follows the exact same sequence in the logs.

Environment

• Version: Claude Desktop 1.1.3541 (Electron 40.4.1, Node 24.13.0)
• Platform: Windows 11 Pro (10.0.22631), x64, Xeon E5-2695 v4, 32GB RAM
• GPU: Microsoft Basic Render Driver (VM — no hardware GPU)
• Mode: Agent mode (Code) with git worktrees

Exact crash sequence (identical all 6 times)

From main.log:

[info] LocalSessions.respondToToolPermission: requestId=..., decision=always (tool: Edit/Bash)
[info] LocalSessions.getGitDiff: cwd=...\.claude\worktrees\{name}, base=main
[info] LocalSessions.checkGhAvailable: cwd=...   ← (fails silently: spawn gh ENOENT)
[info] [CCD] LocalSessions.getPlanForSession: sessionId=local_...
[info] [CCD] LocalSessions.setFocusedSession: sessionId=null   ← CRASH

From claude.ai-web.log (same timestamp as setFocusedSession: null):

[error] Error: Minified React error #185
[error] [BOOTSTRAP] Fatal error boundary triggered [object Object]

The crash happens 3-4 seconds after tool permission is granted, every time. The getGitDiff call on the worktree completes, then the React renderer crashes trying to display the result.

What I tried (none fixed it)

1. Created missing claude-code-sessions directory — eliminated Cannot save session: storage path not available warnings but crash persists
2. Removed all stale worktrees (4 accumulated from previous crashes) — crash persists
3. Installed gh CLI — eliminates spawn gh ENOENT but crash persists
4. Confirmed latest version (1.1.3541) — no update available

Key observations

• The crash is not context-dependent: session 4 crashed with only 40 lines of transcript and zero screenshots. Sessions 1-3 had large contexts (browsing with Chrome extension).
• The crash happens on any tool permission (Edit or Bash), not just specific tools.
• SyntaxHighlightedDiff: Token without row: [object Object] warnings appear in claude.ai-web.log just before every crash — the diff renderer may be producing invalid React children.
• The checkGhAvailable call silently fails (gh not on PATH) right before each crash, though installing gh didn't fix it.
• cowork_vm_node.log shows VM not supported (win32/x64), skipping on every launch.
• No GPU hardware (Microsoft Basic Render Driver on VM) — purely software rendering.

Timestamps of all 6 crashes (same day)

| Session | Crash time | Tool | Worktree diff size |
|---------|-----------|------|-------------------|
| 1 (eloquent-tereshkova) | 10:15:53 | Edit | 44KB |
| 2 (stoic-brahmagupta) | 12:57:45 | Edit | 27KB |
| 3 (funny-lovelace) | 17:06:36 | Edit | 21KB |
| 4 (mystifying-booth) | 17:16:53 | Edit | 22KB |
| 5 (hungry-sanderson) | 17:55:27 | Bash | ~22KB |

All show identical log sequence. The SyntaxHighlightedDiff warnings + React #185 suggest the diff rendering component is producing invalid React children that trigger an infinite re-render loop.

I've been getting this relentlessly for the last 2 days on Claude Code Desktop on Windows, using git repos with worktrees.

The only work around that's worked for me so far is changing my session filter on the sidebar to archived only and then using the session search function to open the session (which is not archived) and continue working that way.

As long as the sidebar isn't showing active sessions, my sessions are stable.

Edit: never mind, not stable but at least somewhat more useable

It;'s claude's forced update. Roll back to an earlier version and chnage config json to set autoupdates to false.

Today's 2/19/2026 uupdate regressed HARD.

Turns out that letting AI approve PRs is a steaming pile of poo.

Environment:

• Windows 11, Claude Desktop App
• Tested with versions: 1.1.3770 (current) and 1.1.1520 (rollback via winget)

Issue:
Error code 1YCZHOG appears within seconds of launching Claude Code inside the Desktop App. This has been happening since February 19, 2026.

Troubleshooting steps taken – none resolved the issue:

• Full Windows restart
• Rollback to v1.1.1520 via winget
• Renamed claude-code-vm folder to prevent VM startup
• Renamed vm_bundles folder
• Set autoUpdate: false in claude_desktop_config.json
• Cleared all local cache folders (Cache, Code Cache, blob_storage, IndexedDB, Session Storage, Local Storage, shared_proto_db)

Findings from logs (unknown-window.log):
The root cause appears to be a recurring Uncaught Error: Minified React error #418 followed immediately by [BOOTSTRAP] Fatal error boundary triggered. This loop repeats on every startup regardless of app version or local configuration.

Conclusion:
Since the error persists across two different app versions and survives a full cache wipe, this appears to be triggered server-side (resources loaded from claude.ai on startup), not a local configuration issue. A local fix does not seem possible until the server-side regression is resolved.

yes, server side, but prior version works. Using now back to January 10th. Anthropic doesn't publish old version, but you can download non-official (caveats apply) https://claude.en.uptodown.com/windows

The issue seems to be that while the system works somewhat in other projects, the projects that have open sessions crash every time. This problem has persisted for two days now without any resolution. Despite attempts to find a fix, there has been no solution yet, and the sessions remain unusable whenever they are open.

Summary:
Other projects: Works a little bit.
Projects with open sessions: Crashes every time.
Status: 2 days of downtime.
Solution: None found yet.

this helped to me:

1. Exit claude, make sure it is not running
2. Update all session files:

Navigate to %APPDATA%\Claude\claude-code-sessions\
For Microsoft Store/UWP installs: %LOCALAPPDATA%\Packages\Claude_{id}\LocalCache\Roaming\Claude\claude-code-sessions\
Find the subfolder {org_id}\{user_id}\
Open each local_*.json file and look for the session that has "isArchived":false
Change it to "isArchived":true

1. Install old version

x64

https://downloads.claude.ai/releases/win32/x64/1.1.2321/Claude-495628f91fbfa276fabd6da835ba226fdf5ec68e.exe

arm64

https://downloads.claude.ai/releases/win32/arm64/1.1.2321/Claude-495628f91fbfa276fabd6da835ba226fdf5ec68e.exe

Also in between 2-3, optionally you can delete cache data.

1. Make sure you are running old version 1.1.2321, in claude menu File -> Help -> About. Don not restart claude! It willl automatically updated to new one, but we want to stay on old meanwhile

https://claude.ai/api/desktop/win32/x64/exe/latest/redirect?utm_source=claude_code&utm_medium=docs <-- this one works

The 1YCZHOG error should be resolved if you update and restart the app. Looking into the related issues in the issue, thanks for reporting.

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.