Carl Yeh

Make it every single time.

No, no, no.

Oh, well, you can put it back up and I can just...

So, yeah, I would be, like, I'm not that, like, in this case, will I check every single time the skill and the learning?

Hell no.

I just want to make sure it executes.

Really, the whole point of this whole program is to execute what I'm looking to do.

Now, if there is an issue that we saw last time and it's not executing, because I get our final report after the execution on what,

the reviews are what um what the confidence score based on the the agent execution is and as like for me it's not for me it's for my the client and we haven't heard any issues with it because they use it every week at a hundred pdfs a week and they're just ripping this through so they do their spot checks haven't heard from them so it's like okay you can yeah do that um

Correct.

Each agent will be pulling from the skill.

Well, so here's my thing, too.

Like, I have to update my Git on this.

The question I have, like, I'm thinking is, do I switch all my agents now?

Because it's all reviewing PDFs.

Do I switch it all to Gemini 3 Flash versus...

Opus 4.5 and it was okay.

I think that what makes this a little easier is a sub-agent is only looking for one thing versus one agent looking for six different things.

One agent is just the whole thing, right?

But then I can have in the skill, as you can see here,