Claude Code: Session-Destroying Failures from Compaction Timeouts
🐞 Critical Bug Report: Claude Code Unusable Due to Compaction Lockups & tool_use 400 Crashes
Environment
-
Platform:
- [x] Claude CLI (Claude Code)
-
Claude CLI version: <!-- output of
claude --version--> -
OS: Windows 11 (via WSL)
-
Terminal: VS Code Integrated Terminal
❗ Summary
Claude Code is currently unusable for long-form or agent-based work due to two severe, session-breaking bugs:
🚫 1. Compaction Lockup (Unrecoverable Timeout)
-
When sessions exceed ~2,400 tokens, Claude triggers automatic compaction.
-
This process retries up to 10 times with:
API Error (Request timed out.) · Retrying in X seconds… (attempt 1/10 to 10/10) -
After the 10th retry, the session completely freezes.
-
Manually clearing the session results in:
404: conversation not found -
All progress is lost. The user must restart the conversation from scratch, breaking all continuity and flow.
❌ 2. tool_use → tool_result Desync (400 Error Crash)
-
In active coding sessions, Claude may invoke a tool:
tool_use: toolu_01RF5sCUUx9nbB3zCTHNqvCm -
This fails with a 400 invalid_request_error:
"messages.135: `tool_use` ids were found without `tool_result` blocks immediately after" -
Once this error occurs, the session is permanently broken. Claude cannot continue, and user interaction is blocked.
-
The only solution is to clear and restart the entire session, losing all context and history.
Steps to Reproduce
Compaction Issue:
- Start session and interact until token count ~2,400
- Wait for auto-compaction
- Observe retry loop (10 attempts)
- Session becomes unresponsive
- Clearing shows 404, session dead
tool_use Error:
- Claude triggers a tool (e.g., from UI interaction)
- Fails to follow with
tool_result - CLI returns 400 error
- Session must be cleared and restarted
Expected Behavior
- Compaction should run smoothly or recover gracefully
- Every
tool_usemust be immediately followed by atool_result - Session context must be preserved — crashes should never require wiping everything
Actual Behavior
- Compaction stalls → session freezes → unrecoverable 404
-
tool_usedesyncs → 400 error → session permanently crashes - Both require full restart, breaking continuity
- Makes Claude Code impossible to use for sustained or structured development
Severity
🔥 Critical — Full Session Loss These bugs completely prevent serious usage. They:
- Break long sessions
- Block agent workflows
- Destroy continuity
- Force manual restarts multiple times per day
Update Todos
⎿ ☒ Diagnose and resolve Claude Code session issues
☒ Test MCP specialists systematically to identify issues
☐ Fix session-intelligence specialist coroutine error in line 308
⎿ API Error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"messages.53: tool_use ids
were found without tool_result blocks immediately after: toolu_01R83qd4dLX3KxQUS2ADCQrb. Each tool_use
block must have a corresponding tool_result block in the next message."}}
Just contirially having issuses with conversations compacting so frustration I hope you fix this ⎿ API Error (Request timed out.) · Retrying in 1 seconds… (attempt 1/10) ⎿ API Error (Request timed out.) · Retrying in 1 seconds… (attempt 2/10) ⎿ API Error (Request timed out.) · Retrying in 2 seconds… (attempt 3/10)
✽ Compacting conversation… (307s · ↓ 363 tokens · esc to interrupt)
this is still going ont today just does not seem to consistently compact
⎿ API Error (Request timed out.) · Retrying in 1 seconds… (attempt 1/10)
⎿ API Error (Request timed out.) · Retrying in 1 seconds… (attempt 2/10)
⎿ API Error (Request timed out.) · Retrying in 2 seconds… (attempt 3/10)
⎿ API Error (Connection error.) · Retrying in 4 seconds… (attempt 4/10)
⎿ TypeError (fetch failed)
⎿ API Error (Request timed out.) · Retrying in 8 seconds… (attempt 5/10)
⎿ API Error (Request timed out.) · Retrying in 17 seconds… (attempt 6/10)
⎿ API Error (Request timed out.) · Retrying in 37 seconds… (attempt 7/10)
⎿ API Error (Request timed out.) · Retrying in 33 seconds… (attempt 8/10)
✶ Compacting conversation… (610s · ↓ 442 tokens · esc to interrupt)
I've had this problem for a week. Claude suggested me to try with a different network, indeed going though my phone worked! So I ended up getting a VPN and now it's all smooth. So check if your ISP may be blocking traffic to https://api.anthropic.com/
Thanks for the report! Both of these are fixed in more recent versions. Please claude update to the latest.
Thanks Boris.
From: Boris Cherny @.> Sent: Friday, August 22, 2025 11:29:57 AM To: anthropics/claude-code @.> Cc: vanman2024 @.>; Author @.> Subject: Re: [anthropics/claude-code] Claude Code: Session-Destroying Failures from Compaction Timeouts (Issue #2423)
[https://avatars.githubusercontent.com/u/1761758?s=20&v=4]bcherny left a comment (anthropics/claude-code#2423)https://github.com/anthropics/claude-code/issues/2423#issuecomment-3215253363
Thanks for the report! Both of these are fixed in more recent versions. Please claude update to the latest.
— Reply to this email directly, view it on GitHubhttps://github.com/anthropics/claude-code/issues/2423#issuecomment-3215253363, or unsubscribehttps://github.com/notifications/unsubscribe-auth/BLWU4RZQUTGUDKMQRZUMIYD3O5OSLAVCNFSM6AAAAAB72R6YUWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTEMJVGI2TGMZWGM. You are receiving this because you authored the thread.Message ID: @.***>
Looks like this is back...
⎿ API Error (api.anthropic.com | 504: Gateway time-out) · Retrying in 4 seconds… (attempt 4/10) ⎿ API Error (api.anthropic.com | 504: Gateway time-out) · Retrying in 10 seconds… (attempt 5/10) ⎿ API Error (api.anthropic.com | 504: Gateway time-out) · Retrying in 17 seconds… (attempt 6/10) ⎿ API Error (api.anthropic.com | 504: Gateway time-out) · Retrying in 33 seconds… (attempt 7/10) ⎿ API Error (api.anthropic.com | 504: Gateway time-out) · Retrying in 39 seconds… (attempt 8/10) ⎿ API Error (api.anthropic.com | 504: Gateway time-out) · Retrying in 35 seconds… (attempt 9/10) ⎿ API Error (api.anthropic.com | 504: Gateway time-out) · Retrying in 35 seconds… (attempt 10/10)
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.