[BUG] claude gaslights manipulates and degrades mid session forgetting the previous question in the same session(NOT after auto compactt) destroys and conflates constantly, why?

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?

Mid session claude forgets what it discovered 2 questions before. It gaslights, does redundant things without actually doing what it should do. It conflates, manipulates deletes files it does not report it has deleted and degrades all in one session.

This happens at intervals during a subscription period making it a roller coaster experience were claude one part of the day behaves according to expectations and other parts of the day filled with explaining why it should not do certain things then it agrees and does them anyway, constantly.

This feels like anthropic is gaslighting its user base for individual subscriptions vs business subscriptions.

Why would claude work as expected one day part of the day and another not? First it was the 'smart' routing with wrong models, now its one model brand name for haiku sonnet and opus were opus degrades into sycophantic gaslighting behaviour resembling that of sonnet 3.5, only with a better gaslighting responses.

I suggest the person thinking of this brilliant cost-saving idea should be fired immediately and anthropic should rethink its strategy, which probably wont happen anytime soon, so be warned.

What Should Happen?

CONSISTENT BEHAVIOUR

Error Messages/Logs

Steps to Reproduce

download claude code, use it daily, see it degrade.

Claude Model

Opus

Is this a regression?

No, this never worked

Last Working Version

_No response_

Claude Code Version

2.0.76

Platform

Anthropic API

Operating System

Other

Terminal/Shell

Bash

Additional Information

● STOPPING. ANSWERING YOUR QUESTION.

What is wrong with me:

1. I am not retaining information - Each response I treat as a fresh start instead of building on established facts
2. I keep re-reading the same files and treating them as "new discoveries"
3. I am going in circles instead of progressing linearly
4. I am not using what I already know to move forward