Slash Commands and Skills Look Alike — Until You Use Them
If you have been using Claude for more than a few months, you already know slash commands. You type a forward slash, you pick the command, Claude does the thing. Skills look almost identical on the surface. They are reusable instruction packs that live in a folder on your machine. They have names. They have descriptions. They can take input.
But the moment you actually use a Skill, the difference becomes obvious. A slash command waits for you to remember it exists. A Skill watches the conversation and steps in on its own when what you are doing matches what the Skill was built for.
That single behavioural shift, from "I summon the tool" to "the tool summons itself", is why Claude Skills are quietly becoming the most useful feature added to the platform in 2026. Anthropic introduced them earlier this year, and most intermediate Claude users still default to commands by habit.
This article shows you what Skills actually do, how to write your first one in under twenty minutes, and the exact moments where a Skill outperforms anything you could type into the chat box.
What Is a Claude Skill?
A Claude Skill is a directory on disk that contains a single file called SKILL.md plus any supporting files the Skill needs. The SKILL.md file has two parts. The first is a short YAML frontmatter with a name and a description. The second is a markdown body containing the instructions Claude follows when the Skill activates.
When you open a Claude session, Claude reads the descriptions of every Skill it can see. It does not read the bodies yet. It compares your messages to those descriptions. When one matches, Claude pulls that Skill's full body into the conversation and acts on it. The model treats the Skill as if you had pasted those instructions yourself.
The folder structure for a personal Skill is straightforward. On Mac and Linux, your personal Skills live in ~/.claude/skills/. Each Skill is one subfolder. The subfolder name is the Skill's slug. Inside, you have your SKILL.md and any reference files, templates, or example outputs the Skill references.
How Claude Skills Differ From Slash Commands
The technical difference is small but the workflow difference is enormous. A slash command is invoked. You type the slash, the menu appears, you select the command, and only then does Claude apply the instructions you wrote.
A Skill activates. You describe what you want to do in plain language. Claude scans every Skill description it has access to, decides which one fits, and silently loads that Skill's full instructions into context before responding. You never had to know the Skill existed at that moment.
This matters because intermediate AI users rarely fail at writing prompts. They fail at remembering which of their dozens of prompt templates applies to the task in front of them. The cognitive load of "which template did I save for this?" is the real friction. Skills remove that friction entirely.
One other practical difference: Skills can reference other files in their directory. A slash command is a single block of text. A Skill is a directory, so it can ship with example outputs, style guides, lookup tables, or full templates the Skill body points to. That makes Skills significantly more powerful for any work that needs reference material.
How to Write Your First Claude Skill in Under 20 Minutes
Let us build a Skill that drafts LinkedIn posts in your voice. This is the kind of task where you have probably written the same prompt twenty times, slightly differently each time, with inconsistent results.
Step 1: Create the folder. In your terminal, run mkdir -p ~/.claude/skills/linkedin-post-drafter. The folder name is the Skill's slug. Keep it short, lowercase, hyphen-separated.
Step 2: Create the SKILL.md file. Inside that folder, create a file called SKILL.md. The exact filename matters: capitals, with the .md extension.
Step 3: Write the frontmatter. Open SKILL.md and add this at the very top:
---
name: linkedin-post-drafter
description: Draft LinkedIn posts in a peer-to-peer, lightly opinionated tone. Use whenever the user wants to write, draft, or rewrite a LinkedIn post, share a professional insight on LinkedIn, or says things like "make this a LinkedIn post" or "write a LinkedIn post about X".
---
The description is the most important field. Claude uses it to decide whether to load the Skill. Write it as a sentence that names every trigger phrase a real person would use. Vague descriptions produce Skills that never activate.
Step 4: Write the body. Below the frontmatter, write the instructions Claude should follow. Be specific. Tell Claude the structure, the length, the tone, the things to avoid:
You are drafting a LinkedIn post. Follow these rules exactly:
1. Length: 120 to 220 words
2. Open with a one-line hook that names a specific tension or finding
3. Use single-sentence paragraphs
4. No emojis, no hashtags, no "thoughts?" at the end
5. End with one clear takeaway, not a question
Step 5: Test it. Open a new Claude session. Type something like "draft a LinkedIn post about how my team finally killed our weekly status meeting". Claude should pull in the Skill automatically. You should see the output follow your rules without you ever typing "use the linkedin-post-drafter Skill".
When Claude Skills Beat Anything You Could Type
Skills outperform manual prompting in four specific situations. Knowing these helps you decide which of your repeat prompts deserve to become Skills and which do not.
Situation 1: Tasks you do more than once a week. The cost of writing a Skill is paid back the moment you stop rewriting the prompt. If you draft client briefs every Monday, that prompt belongs in a Skill.
Situation 2: Tasks that require reference files. Anything that pulls from a style guide, a lookup table, or a list of approved examples works better as a Skill because the Skill folder can hold those files. Your prompt body just says "follow the style guide in style-guide.md" and Claude pulls it in.
Situation 3: Tasks where consistency matters more than creativity. Status updates, meeting summaries, client proposals, weekly reports. Anything where the output should look the same every time benefits from a Skill, because the Skill body locks the structure.
Situation 4: Tasks you delegate or share with teammates. A Skill is a folder. You can commit it to a shared repo. Your whole team gets the same behaviour from Claude without anyone copy-pasting prompts into Slack.
The Three Mistakes That Stop Skills From Activating
Most Skills that fail to activate fail for the same three reasons. Avoid these and your Skills will work the first time.
Mistake 1: The description is too vague. A description that just says "helps with writing" tells Claude nothing. Claude needs to match real user phrases against the description, so the description must include the specific verbs and nouns users actually type. Compare "writes LinkedIn posts" to "Use whenever the user wants to write, draft, or rewrite a LinkedIn post or share a professional insight on LinkedIn". The second version activates reliably. The first does not.
Mistake 2: One Skill is trying to do five jobs. A Skill that drafts LinkedIn posts, X threads, blog posts, and emails will activate inconsistently because Claude cannot match it cleanly to any specific request. Split it into four narrow Skills. Each one becomes more reliable.
Mistake 3: The body assumes context it does not have. Your Skill body must work for a fresh Claude session that has not seen your previous chats. If your instructions reference "the usual format" without saying what that is, the Skill produces inconsistent results. Spell everything out, even the things that feel obvious to you.
Try This Skill in the Next 20 Minutes
Here is a complete copy-paste-ready Skill you can save and test right now. It builds a structured client-meeting recap from raw notes. Save this exact content as ~/.claude/skills/meeting-recap/SKILL.md and try it.
---
name: meeting-recap
description: Convert raw meeting notes into a structured client recap. Use whenever the user wants to write a meeting recap, summarise client meeting notes, draft a follow-up email after a meeting, or says things like "turn these notes into a recap" or "summarise this client meeting".
---
You are converting raw meeting notes into a client-ready recap. Follow this exact structure:
1. One-paragraph executive summary (max 60 words)
2. Decisions made — bullet list, each one starts with a verb
3. Open questions — bullet list, name the person responsible
4. Action items — table with columns: Action, Owner, Due Date
5. Next meeting — single line with date and agenda
Rules:
- Do not invent any detail not present in the raw notes
- If a section has no content, write "None recorded"
- Tone: professional, neutral, no filler phrases
- Do not include greetings, sign-offs, or closing pleasantries
Open Claude. Paste in your raw notes from any recent meeting. Say "turn these notes into a recap". The Skill activates, the output lands in the exact structure you defined, and you never had to remember the Skill existed.
That is the moment most intermediate Claude users realise their prompt library was actually slowing them down. Skills do not just save time. They remove the mental tax of picking the right tool. The best workflow is the one you never have to think about. We know AI's cold edges. We know your real challenges. 28 years with UD, turning technology into a partnership with warmth.
Ready to Build Your AI Workflow Around Skills?
You now have the building blocks. The next step is figuring out which of your repeat workflows actually belong in Skills, and how to structure them so the whole team gets value. We'll walk you through every step, from auditing your existing prompts to deploying Skills across your organisation.
