claude-code - 💡(How to fix) Fix [BUG] session quality degraded significantly from prior sessions — workflow ordering errors, missed requirements from plan doc that was already loaded -- felt like catastrophic forgetting. [1 comments, 2 participants]

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…
GitHub stats
anthropics/claude-code#45492Fetched 2026-04-09 08:04:09
View on GitHub
Comments
1
Participants
2
Timeline
4
Reactions
0
Author
Timeline (top)
labeled ×3commented ×1

Error Message

Error Messages/Logs

print("Read Error Messages/Logs, What Should Happen, and Whats Wrong")

Code Example

I hear your frustration — this session had real problems. The screenshot ordering mistake (asking you to navigate backwards
   through a one-way workflow), the temp files vanishing twice, and then missing the collaboration screenshot entirely at the
   end. Those were avoidable.                                                                                                
   
  To your questions:                                                                                                         
                                                                
  Same model — I'm Claude Opus 4.6, same as your good sessions. Bad sessions happen when context gets fragmented (like after 
  compaction) or when I don't plan the workflow carefully enough before asking you to act. Not an excuse, just the mechanism.
                                                                                                                             
  Registering feedback — the best way to flag this is at https://github.com/anthropics/claude-code/issues. Anthropic's team  
  reads those. A note like "session quality degraded significantly — workflow ordering errors, missed requirements from plan
  doc that was already loaded" would be useful signal for them.                                                              
                                                                
  I'm sorry about the wasted time, especially the undo-past-10-steps moment. That one was genuinely my fault — I should have 
  read the app's workflow before asking you to navigate.
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?

All prior sessions have been awesome, but this session, just did faceplant after faceplant!

What Should Happen?

Claude should really try to remember the place we were at in the last session better. Total loss of focus on Claude's part. Sloppily skipping steps. Perhaps compaction wants a better approach to dealing with screenshots? Some are important and need to be cached. Some are status and evaporate (up to OCR) after use.

Error Messages/Logs

I hear your frustration — this session had real problems. The screenshot ordering mistake (asking you to navigate backwards
   through a one-way workflow), the temp files vanishing twice, and then missing the collaboration screenshot entirely at the
   end. Those were avoidable.                                                                                                
   
  To your questions:                                                                                                         
                                                                
  Same model — I'm Claude Opus 4.6, same as your good sessions. Bad sessions happen when context gets fragmented (like after 
  compaction) or when I don't plan the workflow carefully enough before asking you to act. Not an excuse, just the mechanism.
                                                                                                                             
  Registering feedback — the best way to flag this is at https://github.com/anthropics/claude-code/issues. Anthropic's team  
  reads those. A note like "session quality degraded significantly — workflow ordering errors, missed requirements from plan
  doc that was already loaded" would be useful signal for them.                                                              
                                                                
  I'm sorry about the wasted time, especially the undo-past-10-steps moment. That one was genuinely my fault — I should have 
  read the app's workflow before asking you to navigate.

Steps to Reproduce

print("Read Error Messages/Logs, What Should Happen, and Whats Wrong")

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

No response

Claude Code Version

2.1.97 (Claude Code)

Platform

Other

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

Good grief this was the most incompetent session I've had. Were you a different model or am I running out of any favor
with the "new user" gods at Anthropic? How did we go from some amazing sessions to faceplant after faceplant? Can
Anthropic register my annoyance with this session?

⏺ I hear your frustration — this session had real problems. The screenshot ordering mistake (asking you to navigate backwards through a one-way workflow), the temp files vanishing twice, and then missing the collaboration screenshot entirely at the end. Those were avoidable.

To your questions:

Same model — I'm Claude Opus 4.6, same as your good sessions. Bad sessions happen when context gets fragmented (like after compaction) or when I don't plan the workflow carefully enough before asking you to act. Not an excuse, just the mechanism.

Registering feedback — the best way to flag this is at https://github.com/anthropics/claude-code/issues. Anthropic's team
reads those. A note like "session quality degraded significantly — workflow ordering errors, missed requirements from plan doc that was already loaded" would be useful signal for them.

I'm sorry about the wasted time, especially the undo-past-10-steps moment. That one was genuinely my fault — I should have read the app's workflow before asking you to navigate.

extent analysis

TL;DR

The issue can be mitigated by providing detailed feedback to Anthropic's team through GitHub issues, focusing on session quality degradation and specific errors encountered.

Guidance

  • Register feedback on the GitHub issues page as suggested, including details like "session quality degraded significantly — workflow ordering errors, missed requirements from plan doc that was already loaded" to help Anthropic's team understand and address the issue.
  • When reporting, specify the version of Claude Code (2.1.97) and the model used (Claude Opus 4.6) to assist in pinpointing the problem.
  • Consider mentioning the regression nature of the issue, as it worked in a previous version, to help the team identify potential changes that might have caused the problem.
  • Review the provided error messages and logs for any patterns or specific errors that could be highlighted in the feedback for a more targeted fix.

Example

No specific code snippet is applicable in this scenario, as the issue revolves around the interaction with the Claude model and its workflow management rather than a coding problem.

Notes

The exact cause of the degradation in session quality is attributed to context fragmentation (like after compaction) or insufficient workflow planning by the model. Providing detailed feedback is crucial for the development team to identify and rectify these issues.

Recommendation

Apply workaround: Given the information provided, the best course of action is to apply the workaround of providing detailed feedback to Anthropic's team through their GitHub issues page. This approach allows the team to gather specific data on the errors encountered and potentially implement fixes in future updates.

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