Carl Yeh

I have the instructions I gave it is if the confidence score is less than 80%, use the Gemini API.

If it's a picture, use Gemini API.

If it's handwriting, use Gemini API.

If not, just go your regular Opus 4.5.

In that case, now it's 4.6, but that's what it was.

But it's like, do I say, you know what, forget it.

go into gemini flash is just a lot more token consumption every single time right yeah so i i'm not sure that's kind of the question i have and it's so it i don't know it's so i like i think with gemini flash 3 it is the highest rated model to do that kind of stuff

Yeah.

Yeah.

So that would be the only thing left to do.

But hopefully anyone who's watching can kind of get like, oh, in my specific use case, I can engage in...

a retrospective every single time.

So that's what I was wondering in cowork, could we do something like that?

Because it would be nice to go into your plugins and,

and like whatever skill you have set out that you can make it learn.

Because right now he's here, right?

You have to essentially rebuild that skill to add any more instructions or anything like that.

But with the retrospective, you just hit the slash command and it automatically... Now, I guess you could, because what you could do is you could build the slash command

in cloud code with all your skills, package it into a plugin and put it into cowork and kind of do it that way.

I haven't fully done it that way, but that's another option.