claude-code - 💡(How to fix) Fix [BUG] Repeated API errors, stream idle timeouts, and excessive retry loops causing wasted credits and failed sessions

Official PRs (…)
ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

Helpful · Quick feedback

Loading…

Error Message

  • No silent retry loops burning credits before surfacing an error to the user
  • Graceful failure with clear error message on first timeout, not repeated invisible retries

Error Messages/Logs

  1. Observe: agent reads multiple files, spawns tool calls, then hits API/stream error before responding
RAW_BUFFERClick to expand / collapse

Preflight Checklist

  • I have searched existing issues and this hasn't been reported yet
  • This is a single bug report (please file separate reports for different bugs)
  • I am using the latest version of Claude Code

What's Wrong?

                                                                                                                                                                                                                                                During a single working session, Claude Code repeatedly encountered stream idle timeout / API errors mid-response, forcing session restarts. Each restart consumed credits without completing the task. The agent also exhibited a pattern of excessive multi-step research before answering straightforward questions, further burning credits and time before ultimately    
failing.                                                                                                                                                                                                                                                                                                                                                                      
                                                                                                                                                                                                                                                                                                              
                                                                                                                                                                                                           
Actual behaviour:                                                                                                                                                                                                                                                                                                                                                             
                                                                                                                                                                                                                                                                                                                                                                              
- Stream idle timeout / API errors occurring mid-session, multiple times                                                                                                                                                                                                                                                                                                      
- Agent retried silently, consuming credits, then failed anyway                                                                                                                                                                                                                                                                                                               
- User forced to restart session and re-establish context repeatedly                                                                                                                                                                                                                                                                                                          
- Substantive work lost on each failure                                                                                                                                                                                                                                                                                                                                       
                                                                                                                                                                                                                                                                                                                                                                              
Impact:                                                                                                                                                                                                                                                                                                                                                                       
                                                                                                                                                                                                                                                                                                                                                                              
- Wasted subscription credits (multiple failed sessions)                                                                                                                                                                                                                                                                                                                      
- Lost time re-establishing context across restarts                                                                                                                                                                                                                                                                                                                           
- No visibility into retry behaviour or credit consumption during failures, even with -debug
- excessive development time                                                                                                                                                                                                                                                                                               
                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                             
Platform: Linux (Manjaro), Claude Code CLI                                                                                                                                                                                                                                                                                                                                    
Model: claude-sonnet-4-6                                                                                                                                                                                                                                                                                                                                                      
Date: 2026-05-09

What Should Happen?

Expected behaviour:                                                                                                                                                                                                                                                                                                                                                           
                                                                                                                                                                                                                                                                                                                                                                              
- Stable API connection through a complete response cycle                                                                                                                                                                                                                                                                                                                     
- No silent retry loops burning credits before surfacing an error to the user                                                                                                                                                                                                                                                                                                 
- Graceful failure with clear error message on first timeout, not repeated invisible retries

Error Messages/Logs

Steps to Reproduce

Steps to reproduce:                                                                                                                                                                                                                                                                                                                                                           
                                                                                                                                                                                                                                                                                                                                                                              
1. Open a Claude Code session on a moderately complex codebase (SQLMesh project, PostgreSQL)                                                                                                                                                                                                                                                                                  
2. Ask a question requiring 2–3 file reads and a direct recommendation                                                                                                                                                                                                                                                                                                        
3. Observe: agent reads multiple files, spawns tool calls, then hits API/stream error before responding                                                                                                                                                                                                                                                                       
4. Restart session, repeat — same failure pattern recurs

Claude Model

Sonnet (default)

Is this a regression?

Yes, this worked in a previous version

Last Working Version

No response

Claude Code Version

2.1.136

Platform

Anthropic API

Operating System

Other Linux

Terminal/Shell

Xterm

Additional Information

I note there are many many similar issues already logged, but it's not practical to inspect all of them for a match to my OS, UI, version, etc, especially given that 2.1.136 was released only yesterday (8 May 2026)

Vote matrix · Quick signals

Works
Did the solution work? Tap to confirm.
Easy Fix
Was it a quick fix?
Time Saver
Did it save you time?
Blocking
Was it severely blocking?
Common Issue
Are others likely hitting this too?
Flaky / Intermittent
Is it intermittent?
Verified / Reproducible
Can you reproduce it reliably?
Loading…

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Back to top recommendations

TRENDING

claude-code - 💡(How to fix) Fix [BUG] Repeated API errors, stream idle timeouts, and excessive retry loops causing wasted credits and failed sessions