Right. So I'm going to tell you what I tell everyone.

1. Turn off memory
2. Don't argue with bots or correct them, instead start over when the results are not up to expectations. Repeat the prompt with fixes.
3. Clear the context window often, re-provide the important context

it was not bad. but what really frustrated me is that when it refused, I asked it to make a command line tool that would be helpful in constructing generic results, and it refused to build that as well because it claimed in the context of our conversation I would likely use it to do something bad.