Issue Station
Sign In
Tools hereToolbox

claude-code - 💡(How to fix) Fix [MODEL] Sonnet 4.5 or default in claude code [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#46979•Fetched 2026-04-13 05:44:42
View on GitHub
Comments
0
Participants
1
Timeline
4
Reactions
0
Author
gitsoupgitsoup
Participants
gitsoupgitsoup
Timeline (top)
labeled ×4

Code Example



---

For feedback to Anthropic:

  The best direct channel is: https://github.com/anthropics/claude-code/issues

  For this specific type of feedback — "Claude read the files but didn't apply them carefully when writing specs" — the most useful report would include:
  - The exact task (spec writing)
  - What I missed (method present in reference file, wrong path without verification)
  - What the other model caught
  - Your suggestion: Claude should do a verification pass against source files before finalising a spec

  That's actionable, reproducible, and specific — the kind of feedback engineering teams can actually use.

  Alternatively the feedback form at claude.ai (the thumbs down on responses) routes to the product team, but GitHub issues get more visibility for CLI/agentic use
  cases since that's where the Claude Code team actively monitors.

  Your testing across multiple models simultaneously and comparing outputs is genuinely useful signal — most users don't do that.
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

Other unexpected behavior

What You Asked Claude to Do

"Claude read the files but didn't apply them carefully when writing specs" — the most useful report would include:

  • The exact task (spec writing)
  • What I missed (method present in reference file, wrong path without verification)
  • What the other model caught
  • suggestion: Claude should do a verification pass against source files before finalising a spec

What Claude Actually Did

simply being lethargic and not doing proper review :)

Expected Behavior

Claude should do a verification pass against source files before finalising a spec

Files Affected

Permission Mode

Accept Edits was ON (auto-accepting changes)

Can You Reproduce This?

Haven't tried to reproduce

Steps to Reproduce

No response

Claude Model

Sonnet

Relevant Conversation

For feedback to Anthropic:

  The best direct channel is: https://github.com/anthropics/claude-code/issues

  For this specific type of feedback — "Claude read the files but didn't apply them carefully when writing specs" — the most useful report would include:
  - The exact task (spec writing)
  - What I missed (method present in reference file, wrong path without verification)
  - What the other model caught
  - Your suggestion: Claude should do a verification pass against source files before finalising a spec

  That's actionable, reproducible, and specific — the kind of feedback engineering teams can actually use.

  Alternatively the feedback form at claude.ai (the thumbs down on responses) routes to the product team, but GitHub issues get more visibility for CLI/agentic use
  cases since that's where the Claude Code team actively monitors.

  Your testing across multiple models simultaneously and comparing outputs is genuinely useful signal — most users don't do that.

Impact

High - Significant unwanted changes

Claude Code Version

2.1.104 (Claude Code)

Platform

Anthropic API

Additional Context

No response

extent analysis

TL;DR

To address the issue, consider implementing a verification pass against source files before finalizing a spec in the Claude model.

Guidance

  • Review the spec writing process to identify where the verification pass can be integrated, ensuring that Claude accurately applies information from reference files.
  • Test the updated process with multiple models to compare outputs and verify that the changes improve the accuracy of spec writing.
  • Provide detailed feedback through GitHub issues, including the exact task, what was missed, and suggestions for improvement, to help the engineering team address similar issues.
  • Consider utilizing the feedback form at claude.ai for general feedback, but use GitHub issues for CLI/agentic use cases to ensure visibility for the Claude Code team.

Example

No specific code snippet is provided due to the lack of technical details in the issue, but the suggestion involves modifying the spec writing process to include a verification step.

Notes

The provided information lacks specific technical details about the implementation or the exact changes needed in the code. Therefore, the guidance focuses on the process and feedback rather than code-level fixes.

Recommendation

Apply workaround: Implementing a verification pass as suggested can help mitigate the issue of Claude not applying information carefully when writing specs, even though it does not address the root cause directly. This approach can lead to more accurate spec writing until a more permanent fix is available.

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…
#api#optimization#mixed precision#training loop#device allocation

Still need to ship something?

×6

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

Back to top recommendations

TRENDING