[BUG] 1M context is bullshit. It does not read user prompt carefully, ignores memory and skills

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?

I have been suffering from 1M context update for several days, more than 40hours
Unlike the previous one, it now ignores some of my prompts, for example if I ask 4~5 short tasks in one prompt, it just solves 1~2 and acts as if it never heard of others. And also, memory and skills, it never reads until I ask it to read for 3~4 times in a row. It keeps ignoring my design, but try to solve the problems in much easier way, which results in bad output, and it also does not understand the context anymore. It does not know what was the exact previous conversation, what we were talking about and behave in other way.
It seems it is because of the bigger context given right now, but the algorithm is not adjusted to it. It seems that it is loosing more details than previous compaction, and does not know which point of input vector it must more focus on

What Should Happen?

Please give smaller context option, or revert to previous one.
Why I should suffer from a new, but worse option even though I am a paid user?
I paid for a consistent performance, but the performance is fluctuating depending on your decision, and I have no option to use the previous one. Does it make sense?

Error Messages/Logs

Steps to Reproduce

Just keep chat with the model opus 4.6-1M
You can easily feel it

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.81 (Claude Code)

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

iTerm2

Additional Information

_No response_