[BUG] No longer productive to code with Claude Code > signif' output degradation over last week

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?

For the last week at least, Claude Code (latest Opus 4.5) is massively degraded in it's responses.

CC has stopped following instructions accurately, made guesses when specifically told not to and to be careful to check original code to be worked on, failed to implement very basic frontend layout requests (including inability to simply create an accurate copy of existing code, even in small sections) and repeatedly lies about solutions or claims something does not exist when I can then prove to the model that it does.
Tasks which would have taken several hours are now bleeding into days through error finding/fixing and regular guessing - despite being told specifically how to get/check variables, CC repeatedly choose to make things up instead of do the very specific work requested.
These are problems which did not occur, or more rarely until around a week ago.

The problem is exacerbated by the loss of trust - not only does CC have less ability, it also lies about what it has done or can see in the code or files - constant double checking suspicions adds to the extra wasted time.

The new timeframes to complete work (typically more than double previous durations) means it is no longer possible to work productively with Claude Code.

What Should Happen?

Find out why the model is being dumbed down, or has new training to ignore commands and make more guesses (eg where effort/work to do it properly is detected).
Is this a resource issue? Is it a UK resources issue?
This is urgent, as working with CC is no longer viable, with current error rates + un-trustable output requiring so much checking and fixing.

Error Messages/Logs

Steps to Reproduce

Try to copy any small section of frontend code.
Take any small section of eg https://www.sno.co.uk/ and ask Claude to rebuild that section with an exact copy of the code, then remove the bootstrap dependency.
We have been successfully refactoring frontend layout modules by asking CC NOT to guess and simply copy html and css verbatim for several weeks, and then remove Bootstrap dependencies while maintaining naming conventions, and changing only font family and colour/weight/opacity to our new styles.
Now CC cannot even accurately do the initial stage of copying now, and is completely unable to stop itself making an interpretation with guesses for variables etc, even though I simplify and ask it to copy a very small section at a time into our new page (ie even without any refactoring and being very specific that it must copy the 'exact' codes first so I can see the section in working as original, in the new page, before we update it.

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

claude-opus-4-5-20251101

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

VS Code integrated terminal

Additional Information

_No response_