solidcodersolidcoder
Explore Courses
solidcodersolidcoder
CoursesAboutPrivacy PolicyTerms
© 2026 solidcoder · Practical courses for software engineering interviews.
Home/AI Engineering/Skills/How Claude Decides to Use One
Chapters — Skills▾

How Claude Decides to Use One

Claude Code Skills·Basics·6 min read·Sep 6, 2026

Two ways a Skill gets used: asking for it by name, or Claude recognizing the task fits it from a short description.

Asking for a book by name, or browsing the shelf

There are two different ways to find a book in a library. You can walk up already knowing the title and ask for it directly. Or you can describe what you're looking for, and a librarian who has skimmed every spine label points you to the right shelf, without you needing to know the exact title in advance.

A Skill can be reached the same two ways.

By name
You (or a slash command) invoke it directly, the same way you'd ask for a specific book by title. No guessing involved on either side.
By matching
Claude reads the current task, compares it against every available skill's short description, and picks one up automatically if it's clearly relevant, the way a librarian matches a request to a shelf label.

The description is doing all the work in that second case

For automatic matching to work at all, Claude has to decide relevance from something small: a one-line description of what the skill covers, shown alongside every other available skill, before any of the actual instructions get loaded. Nothing else is consulted at that point. If the description is vague, or sounds like it could apply to five different situations, matching becomes a coin flip. If it's specific about exactly what kind of task it's for, matching becomes reliable.

This is worth sitting with, because it flips a natural assumption. You might expect the detailed instructions inside a skill to be the important part, and they are, eventually. But the short description is what determines whether those instructions ever get read in the first place. A brilliant set of instructions behind a vague description simply never gets found by matching.

Why this two-step matters

Loading every skill's full instructions into every conversation, just in case one might be relevant, would be wasteful and slow. Checking a short description first, and only loading the full thing when it actually matches, keeps things fast without losing the ability to reach for the right expertise at the right moment.

What this means in practice

If you've ever typed a slash command in Claude Code, like invoking a specific tool by its name, that's the first path: direct, no ambiguity. If you've ever asked a plain question and watched Claude pull in exactly the right specialized process without you naming anything, that's the second path: the description did its job.

Notice it happening

Next time Claude handles a task in a way that feels unusually well-prepared for that specific kind of work, ask yourself: did I name what I wanted, or did it recognize the shape of the task on its own? Both are the same mechanism from two different directions.

Up next2/3
Part 1 · Basics
Next Chapter →
The Anatomy of a Skill File
6 min · continue reading
→
← Prev Section
Connection Layer (MCP)
Security and Trust Boundaries
Next Section →
Intermediate
Writing a Description That Actually Triggers
Part of a free guide

Skills

A simple guide to Claude Code Skills: what they are, and how to build your own.

Browse All Guides →
On this page
  • Asking for a book by name, or browsing the shelf
  • The description is doing all the work in that second case
  • What this means in practice