AI Adopt
← All posts
/Short

Claude Skills Demystified

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

Questions & Answers

QI added a skill to Claude. How do I make sure Claude uses it — do I need to call it?
A

No. Claude reads every skill's description and automatically invokes the one that matches your task. You don't call it — you just describe what you want.

QThen why didn't Claude use it when I asked once?
A

Because the skill's description didn't match how you phrased the task. Skills are selected by their description, so if a skill keeps getting skipped, sharpen its description.

QI installed several skill packs (superpowers, GSD, and others) and now I can't tell which one Claude will use. Why is it so unpredictable?
A

Claude makes a fresh judgment every time about which skill to pick. If several packs have similar descriptions, the choice can flip between runs.

QSome skills seem to win more often even when they're not the best fit. Why?
A

Some packs use broad, imperative descriptions like “You must use this skill before any task.” A quietly worded skill like “Helps improve release notes” loses to a pushier one, so the aggressive pack takes over.

QSo how do I actually control which skill wins?
A

First, remove the skill packs you don't use — fewer skills, fewer collisions. Second, you can invoke a skill explicitly (yes, even though I said you don't need to). But if you find yourself forcing it every time, that's a signal your skill description is too vague.

QAny advice for writing my own skills?
A

Write the skill's description as a trigger, not as a documentation summary. Follow the rest of the series for the full breakdown.

Go deeper on Claude Code:

Get more content like this.

Frameworks, hard-won lessons, and tools for engineering managers — in your inbox.

Score your company's AI adoption →