Instead of rewriting yourself have you tried telling the agent what it did wrong and do the rewrite with it? Then at the end of the session ask it to extract a set of rules that would have helped to get it right the first time. Save that in AGENTS.md. If you and your team do this a few times it can lead to only having to rewrite 5% of the code instead of 70%.
It says "of course you're right" and may or may not refactor/fix/rewrite the issue correctly. More often than not it doesn't or misses some detail.
So you tell it again, "of course you are right", and the cycle repeats.
And then the context window gets exhausted. Compaction loses most of the details and degrades quality. You start a new session, but the new session has to re-learn the entire world from scratch and may or may not fix the issue.
