Ultra: The Vibe Coder's Dream I Don't Recommend

A complete course for developers who are serious about making AI part of how they actually work.
One last funky thing before we leave the commands behind: ultra. The idea is simple. You're telling Claude to do everything it can to make sure the thing you asked for gets done to the nth level, and you're telling it to go ahead and spend the tokens doing it. People reach for it when they're about to ship something, or when they've inherited a project and want to know what they're actually dealing with.
To try it I switched the model over to Fable 5.1:
/model
Anthropic gives you a lot out of the box: packaged skills (plugins, which we'll get to), design (which I wouldn't use, but I have, and it's decent), and code review. There are two ways to get an ultra code review. The first is the built-in review, where you can set the level of thinking all the way up to ultra:
/code-review
Or you can go straight to it, and at the time of recording I had three free runs left:
/ultrareview
Before you ultra anything, understand what you're buying. When you ask Claude to think things through, it gyrates on the problem: am I going down the right path? What about this? What about that? That can be useful. It's also expensive, and it ups the randomness. You can't be sure what you're going to get back. For a code review I'd much rather have my procedures locked down, with exact instructions and templates for what good code looks like, and I already review every PR and issue that goes into this code base, so I didn't run it here.
Over on X, folks like Dan McAteer are setting reasoning effort to ultra (ultrathink, ultracode, who knows, they change the names all the time) and letting it blast out a whole project from a spec while they sleep. A vibe coder's dream! You might wake up to something amazing, or to a big bill and a maxed-out Fable allowance. Do I recommend it? No. I think we still need to be involved in the process. Up to you, but now you know it's there.
A complete course for developers who are serious about making AI part of how they actually work.