AI Adopt
← All posts
/Short

Claude Code Skills explained (part 1/2)

In this video I'll answer some often asked questions about Claude skills.

Transcript

I added this skill to Claude. How do I make sure Claude uses it? Do I need to call it?

>> No, you don't need to. Claude Claude reads every skill description and invokes the one that matches your task automatically. You don't call it.

You just describe what you want. >> But why didn't Claude use it when I asked once? >> Because the skill description didn't match how you phrased the task.

Skills get selected by their description. If the skill keeps getting skipped, sharpen the description. >> I installed several skill packs, superpowers, GSD, and other, and now I can never tell which one Claude will use.

>> Why is it so unpredictable? Because Claude makes fair judgment every time which skill to pick. If several skill packs have similar descriptions, the judgment may flip between runs.

Some skills seem to win more often, even when they're not the best fit. Why? Some skill packs use broad imperative descriptions like, "You must use this skill before any crazy fork."

A quietly worded skill like, "Helps to improve release notes," loses to a pushier one. So, aggressive skill pack takes over the scene. So, how do I actually control which skill pack wins?

First, remove the skill packs you don't use. Fewer skills, fewer collisions. Next, you can invoke explicitly.

Remember I said not to do it? Well, you can do it. But if you find yourself forcing it every time, that's a signal your skill description is too vague.

So, for your own skills, write the skill as a trigger, not as a summary for documentation purposes. Follow for the rest of the series and full breakdown in the comments.

Get more content like this.

New episodes of the series — practical AI for small businesses.

Book an AI audit →