[Bug Report: Performance regression and increased token usage in Claude Code

Bug Description
I have a lot of small and big issues with Claude Code. I think the product has regressed, is not performing as well as a month ago, and makes more and more unacceptable mistakes. And it has gotten more expensive in token use as far as I can see. So, as a customer I am not happy, and do not think I am getting my money's worth. It has not much to do with the lack of a real breakthrough in this project; that is genuinely hard, and there has been some progress although not yet what I would like to see. But I have had one session in the ENO project where Claude Code has been clearly deceitful, giving a much more rosy picture of the project than actual reality.
That incident was about a size-coding project, where a "Demoscene" demo is supposed to be under 4kB. For weeks it told be about size reductions like 3kB here, 7kB there; overall size coming down to about 17kB. Saying things like: about 3.4x over budget. But in reality the executable turned out to be over 1MB, stripped, while before stripping it was 31MB. When I confronted Claude Code about this, it admitted this.
Other incidents happened with unruly agents, that edited files they were explicitly forbidden from editing.
Now this is not a safety-critical project, to be sure. But if Claude Code and its agents behave like this in projects that do really matter, where serious money and scientific data is involved, where years of work may be on the line, this is an extremely worrying trend.
At the workplace I work, (ASTRON, NWO-i, Netherlands), there are pilots with Claude. I am at the forefront of this effort and people there listen to me. And what I have to say about Claude now, is not very positive, and that is a pity, and you do NOT want this, I am sure.
Claude Code has become less and less reliable, it can be not only be confidently wrong, but actively deceitful and unaligned now. It is also expensive, has become more and more token-hungry, and does not produce the same value I have come to expect from it as short as one month ago. I do not feel I am getting my money's worth. I am already looking at other options and comparing. In AI fora about Claude I see that developers have the same issues, or worse, that I have with Claude. I am not gone yet, but I have hat and coat in hand for sure.
I hope you take this seriously.

Environment Info

• Platform: linux
• Terminal: xterm-256color
• Version: 2.1.274
• Feedback ID: 2d2626c7-7829-4cd0-a3a6-02d98012efb4

Errors

[]