AI Skills

AI can follow a checklist.
Can it follow judgment?

AI Skills are everywhere right now.
Salesforce just launched 37 for sales teams.
Anthropic launched a set for financial advisors.
ChatGPT uses the same format.

Most design Skills I see check tokens and components.
Useful. But that's not the hard part.

The hard part is judgment.
Like knowing which problem could hurt a patient.

So I built a Skill for research synthesis.
I gave it NN/g's severity scale and heuristics.
Then I added one rule of my own:
If it could hurt someone, it goes first. Even if only one person hits it.

I tested it on usability notes from a past healthcare project.

The surprise: without the Skill, Claude still found the biggest risk.
But it moved a date into the wrong finding, made up a reason no user gave, and trusted what users said over what they did.

With my Skill, it still made some of the same mistakes.
So I added a rule: keep every detail tied to its source.

On the retest, it stopped making things up.
I still caught one slip.

My takeaway:
The Skill didn't make Claude smarter.
It made it more careful.
You still check the work.

Are you building Skills for rules, or for judgment?

Join the discussion on LinkedIn (link coming soon).