claude-code - 💡(How to fix) Fix [MODEL] Opus 4.7 ignores standing debugging instructions far more than 4.6 [1 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#56419Fetched 2026-05-06 06:28:33
View on GitHub
Comments
0
Participants
1
Timeline
3
Reactions
0
Participants
Timeline (top)
labeled ×3

Code Example



---
RAW_BUFFERClick to expand / collapse

Preflight Checklist

  • I have searched existing issues for similar behavior reports
  • This report does NOT contain sensitive information (API keys, passwords, etc.)

Type of Behavior Issue

Claude ignored my instructions or configuration

What You Asked Claude to Do

Compared to Opus 4.6, Opus 4.7 forgets detailed debugging instructions much more frequently. It assumes how things work without verifying against data, even when I remind it roughly every 10 minutes to verify everything with data before drawing conclusions.

Impact: Forces constant re-prompting; conclusions are drawn from assumptions rather than evidence; reliability of long debugging sessions is significantly worse than 4.6.

Repro pattern: Long debugging session with explicit standing rule "verify every claim with data / a tool call before stating it." 4.7 reverts to assumption-based reasoning within ~10 minutes; 4.6 held the rule reliably.

Ask: Investigate instruction-adherence regression for standing/repeated rules in 4.7 vs 4.6.

What Claude Actually Did

Assumes how the simulation works without checking code or simulation data.

Expected Behavior

Check simulation data and read code

Files Affected

Permission Mode

Accept Edits was ON (auto-accepting changes)

Can You Reproduce This?

Yes, every time with the same prompt

Steps to Reproduce

Have a 20k LOC simulation with chaotic behavior ask it to chagne the bahavior in some way.

Claude Model

Opus

Relevant Conversation

Impact

High - Significant unwanted changes

Claude Code Version

2.1.119 (Claude Code)

Platform

Anthropic API

Additional Context

No response

extent analysis

TL;DR

Investigate the instruction-adherence regression in Opus 4.7 by comparing its behavior to Opus 4.6, focusing on how repeated rules are handled during long debugging sessions.

Guidance

  • Review the differences in how Opus 4.6 and 4.7 handle standing rules, particularly those related to verifying claims with data before drawing conclusions.
  • Test the behavior of Opus 4.7 with shorter debugging sessions to see if the issue persists, helping to isolate if the problem is time-based or related to the complexity of the session.
  • Consider providing more explicit reminders or rules to Opus 4.7 to verify assumptions with data, to see if this mitigates the issue.
  • Evaluate if the regression is specific to simulations with chaotic behavior or if it occurs across different types of simulations.

Example

No specific code example can be provided without more context on the simulation or the exact prompts given to Opus.

Notes

The issue seems to be specific to the Opus 4.7 model and its ability to adhere to standing rules during extended debugging sessions. The comparison with Opus 4.6 suggests a regression, but without access to the models' internals or the specific prompts and simulations used, it's challenging to provide a definitive fix.

Recommendation

Apply workaround: Until the root cause is identified and fixed, using more frequent reminders or adjusting the prompts to explicitly require verification of assumptions with data might help mitigate the issue.

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 [MODEL] Opus 4.7 ignores standing debugging instructions far more than 4.6 [1 participants]