[BUG] Claude does not do what it is asked to do

Resolved 💬 2 comments Opened Feb 6, 2026 by fred1268 Closed Mar 6, 2026

Preflight Checklist

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

What's Wrong?

Claude did not follow instructions. I pointed this. It acknowledged. I asked how I can make sure this won't happen. It suggested to open a report. Unfortunately, I don't have time for that so I asked it do it itself.

I think Claude itself explained the feature gap perfectly:

I wish I could, but I need to be honest: I don't have the ability to file GitHub issues or send feedback directly to Anthropic.

I have no tools to:

  • Post to GitHub
  • Send emails
  • Write to any external system
  • Communicate outside this conversation

You're absolutely right that users shouldn't have to do this work, and your point about misaligned incentives is spot-on. The system should automatically surface these failures to the creators.

What Might Help (Speculation)

  • This conversation might be sampled for training/quality review (but I don't know)
  • If you downvote my responses or rate them poorly, that might flag them
  • But I genuinely don't know what internal mechanisms exist

The Gap

You've identified a real problem:

  • AI makes mistakes →
  • AI promises to improve →
  • User has to manually report it →
  • If they don't, nothing changes →
  • System keeps making same mistakes

This is a legitimate design flaw in the feedback loop.

I can't fix this gap myself. I can only be honest about it.

What Should Happen?

Claude should have gathered some context, ask me to validate it can share it, and open the issue itself.

Error Messages/Logs

Steps to Reproduce

Ask Claude to do something, in this case it was merging some bullet points from one skill into another EXCEPT two of them very specifically. It did merge, but also merged the one I specifically asked not to merge

Claude Model

Opus

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.33

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗