Authority AI Editorial
How to Use Skills in Claude Code: Add One, Name It, Check It Ran
You'd like to try a skill in Claude Code, and the guides you find assume you're a developer. You don't need to be one.
A skill is a written method for one kind of job: what good looks like, the steps in order, the checks between them. Using one in Claude Code comes down to four moves. Put the skill where Claude Code looks. Give it a description it can match. Name it when you want it. Then check the work against the method. This guide walks through each one.
What a skill looks like in Claude Code
Each skill is its own folder, named after the skill. Inside is one document of plain instructions. It opens with two short lines: the skill's name and a description of what it does and when to use it. Everything after those lines is the method itself, written the way you would write it for a colleague.
Anthropic describes skills as "folders that include instructions, scripts, and resources that Claude can load when needed." Most skills you'll write are only the instructions part, and that's enough.
You don't have to type the folder or the document yourself, either. Anthropic's authoring guide notes that Claude already understands how skills are structured, so you can simply ask it to create one. Describe the job, paste in how you do it, and ask Claude Code to save it as a skill. Then read what it wrote, the way you'd check any draft before you rely on it.
Step 1: Put the skill where Claude Code looks
Claude Code looks for skills in two everyday places.
The first is your personal skills folder, which sits inside Claude Code's own settings folder in your home folder. A skill kept there is available in every project you open on that computer. This is the right home for methods you use everywhere, such as how you like meeting notes written up.
The second is a skills folder inside a single project. A skill kept there only loads when you work in that project, and anyone who shares the project gets it too. This suits methods that belong to one client or one piece of work.
If two skills share a name, the personal one wins, so give each skill a distinct name.
You don't need to restart anything after adding a skill. The Claude Code documentation says it watches skill folders and picks up added, edited or removed skills within the current session. The one exception is when the skills folder itself didn't exist when you started. In that case, close Claude Code and open a fresh session.
Step 2: Write a description Claude can match
The description is the most important line in the skill. Claude Code keeps a short list of every skill's name and description in view while it works. It only opens the full method when your request matches a description.
That means a vague description makes a skill invisible. "Helps with notes" will rarely be matched. Anthropic's authoring guide asks for two things in every description: what the skill does, and when to use it. It also asks for the description to be written in the third person, because it's read as part of Claude's own instructions.
Here is an illustration, not a logged run. Say you want a skill that turns rough meeting notes into a clean summary. A weak description says it handles meetings. A strong one says it turns raw meeting notes into a summary listing each decision, the person who owns it and the date it's due, and that it's used when someone pastes meeting notes or asks for a write-up of a meeting.
The Claude Code documentation adds two practical points. Put the key use case first, because long descriptions get cut short in the list. And include the words people naturally say. If you always call them "call notes", put "call notes" in the description.
Step 3: Name it in your request
There are two ways to get a skill used.
The first is to let Claude choose. You paste your notes and ask for a write-up. Claude Code sees that the request matches your meeting-notes description, opens the method and follows it. The documentation puts it simply: Claude uses skills when they're relevant.
The second is to call the skill yourself. Start your message with a forward slash and the skill's name, so a skill called meeting-notes becomes /meeting-notes, then add your request or paste your notes after it. This is the sure route. It doesn't depend on Claude matching your wording to the description.
One detail catches people out. Mentioning the skill's name in the middle of a sentence doesn't run it directly. Claude may still pick it up from the description, but if you want to be certain, put the slash and the name at the start.
Skills you can call by name also appear in Claude Code's slash menu, which is a quick way to check a skill's exact name.
Step 4: Check it ran
A skill is only useful if you can tell it was followed. There are three checks, and the last one matters most.
First, ask Claude Code which skills it has. The documentation's own troubleshooting starts there: ask "What skills are available?" and make sure yours is on the list. If it isn't, Claude Code can't see it, and the problem is where you put it.
Second, watch the work as it happens. Anthropic notes that you'll see skills appear in Claude's reasoning as it works. When a skill is called, its method is added to the conversation, so you can see that it was brought in.
Third, check the result against the method. This is why a good skill contains at least one thing you can verify at a glance. In the meeting-notes illustration, every decision has an owner and a due date. If a decision comes back with no owner, the method wasn't followed, however polished the summary looks. Write one checkable rule like that into every skill, and checking takes seconds.
When it doesn't run
If Claude doesn't use your skill when you expected it to, the documentation suggests working through four steps in order:
- Check the description includes the words people would naturally say.
- Confirm the skill appears when you ask which skills are available.
- Rephrase your request so it matches the description more closely.
- Call the skill directly by name with the forward slash.
The opposite problem happens too. A skill with a broad description can turn up on jobs it wasn't meant for. The fix is a more specific description. You can also set a skill so that it only runs when you call it by name, which suits methods you want to trigger yourself.
Keep it sharp after the first run
Once a skill has been called, its method stays in the conversation for later turns. The documentation is clear that Claude Code doesn't re-read the skill on each turn. Two habits follow from that.
Write the rules that must hold for the whole job as standing instructions, such as "every decision gets an owner", rather than one-off steps. And when you edit a skill partway through a long session, call it again by name so the new version is the one in play. The same applies after a very long session is summarised to free up space: the documentation suggests calling the skill again to restore its full method.
Anthropic's authoring guide recommends improving skills by watching them work on real jobs. Notice where the result fell short and add a line to the method.
Where to start
Pick one small job you do every week and already do the same way each time. Meeting write-ups, a reply to a common customer question, or a checklist before you send a document out are all good first choices.
Ask Claude Code to create a skill from your description of how you do it, and keep it in your personal skills folder. Read the description line and make sure it says what the skill does and when to use it, in the words you actually use. Add one rule you can check at a glance.
Then run your next real job with the slash and the skill's name, check the result against that rule, and fix the method where it fell short.
If you'd rather start from methods someone has already written, the AI Ultimate Skill Collection is 3,029 ready-made skills sorted into 12 packs, one for each kind of work, such as coding, research or finance, and it works with Claude Code and other agents that use skills.