What are Claude Skills?
A skill is a folder of instructions you hand to Claude so it already knows your task. Here is what is actually inside one, and what people mix it up with.
updated Sep 9, 2026
A Claude Skill is a folder of instructions you give to Claude so it acts like it already knows your task. Instead of pasting the same prompt every time, you save the know-how once as files, and Claude reads them when the work calls for it.
What a skill actually is
At minimum, a skill is a file named SKILL.md with YAML frontmatter at the top
and instructions below. The frontmatter's description field is the part that
matters most: it is what Claude compares against your request to decide whether
to load the skill, so it needs to say both what the skill does and when to use
it. You can add extra files next to SKILL.md, like a detailed reference doc or
a script, and Claude only reads those when the task actually needs them.
What people confuse it with
A skill and a plain custom slash command are close enough that Claude Code
treats them as the same thing under the hood: a file at .claude/commands/deploy.md
and a skill at .claude/skills/deploy/SKILL.md both give you /deploy, and a
skill just adds the option of a supporting-files folder and a description Claude
can match automatically. Skills also get confused with MCP servers, but they
solve different problems: a skill is know-how you package, an MCP server is a
live connection to an outside system. See
Claude Skills vs MCP servers for that
distinction in full.
How Claude decides to use one
Claude compares your request against every installed skill's description. When
one matches, it reads the full SKILL.md into the conversation and follows it.
You do not have to call it by name, though you can always invoke one directly by
typing / followed by its folder name.
When a skill helps
Skills pay off when you do the same kind of work often: reviewing code, writing in a house style, running a multi-step procedure you would otherwise re-explain every time. They also let a team share one approach, since a project skill committed to a repo gives everyone the same result. For a one-off question, a plain prompt is still faster; packaging it as a skill only pays back after repeat use.
Lorenzo Meola
I use agentic AI daily in my day job to build tools, and I've spent months learning how these systems actually work under the hood. I'm not an AI expert by title, just someone building this directory so other developers can find AI assets that are genuinely useful and verified to work, not just indexed.
Project on GitHub