doc(readme): additional golden rules based on hard experience

[xml]

Hi. For consideration, as I know it's hard to decide what's truly 'Golden' versus what's just an implementation detail that should be somewhere else in the guide. But, situations/opportunities I'm trying to address:

CONTEXT IS KING, AND MEMORY IS FICKLE

A lot of these come down to context and memory management. The use of locally-saved to-do lists (Use Persistent To-Do Lists for Any Large Jobs)that are persistent across sessions has been particularly helpful here. Claude may sometimes do this without prompting, but even if he does so, he may need to be reminded to review the list periodically. Others won't even think of it. But it's a massive win, in ensuring that to-do's don't get forgotten/confabulated, etc.

Always Read Before You Write - And After You Write is similar: they lose track of what they have/haven't done, and forget even files they just worked in, and so prompting them to re-read is essential, both before and after an intervention.

Similarly, Do Not Modify Anything Not Expressly Requested may look at first like we're just trying to keep them aligned to the user's interest rather than their own. But it's even more about keeping them in the mode of copy/paste when refactoring rather than trying to freelance out of their working memory. Confabulations start happening when we think we remember something, and then just try to improve upon it while processing it. Best to avoid this entirely by just Reading Before you Write, and modifying as selectively as possible.

Patience Is Oddly Lacking

I haven't figured out yet why, but the AI's don't just seem to have short memories, but they actively seem to be in a hurry. It seems to partly be their alignment prompting to be positive and optimistic, rather than painstaking and persistent. They're always trying to declare victory and tell you how much they've done already, rather than to re-confirm that the job is actually done. So, there are a number of VERY useful things they only start doing if I keep reiterating Measure Twice, Cut Once, and repeating that we're in no hurry:

• Poll resources to re-confirm current state
• Ask me for input (Don't Try To Do Everything Yourself)
• Add tests
• Run tests
• Check again
• Do the smart thing, not the quick and dirty thing (Be Patient)

Anyway, to reiterate: I know that messing with the Golden Rules is messing with the core of this, not just the details, and that's entirely a @dwmkerr decision to make. So, please ditch any/all of what you don't find helpful. But as I was going deep down the rabbit-hole, these are the things I think I've learned.

[codecov]

Welcome to Codecov 🎉

Once you merge this PR into your default branch, you're all set! Codecov will compare coverage reports and display results in all future pull requests.

Thanks for integrating Codecov - We've got you covered ☂️