The performance after K2.6 degraded

What version of Kimi Code CLI is running?

1.41.0

Which open platform/subscription were you using?

/login

Which model were you using?

K2.6

What platform is your computer?

Linux

What issue are you seeing?

The performance after K2.6 degraded. It's slower which is acceptable but it's exploring impossible ideas, tends to stuck in loops and diverge from the prompt probably due to collecting a lot of context garbage. When it's stuck it should just ask for input and follow the user direction and not try to explore 10 different directions which are just wrong. It reasons too much for no good and without reasoning it's too dumb.

What steps can reproduce the bug?

Install latest kimi cli and give it any more complex debugging task or anything that needs a precise action following user instructions and fallback to user for more guidance when run into a wall.

What is the expected behavior?

Old Kimi K2.5 behaviou, ask when lost, don't loop in infinite reasoning trap, don't overthing stuff

Additional information

No response

The report names no files, tests, or entry points; begin by reproducing it with the latest Kimi CLI on Linux using a complex debugging task and K2.6, then compare with K2.5. Done means the CLI no longer loops or pursues clearly wrong directions and asks for user guidance when stuck, with a regression check for that behavior.