How to Create Your Own Claude Code Skills That Actually Work

Claude Code skills can cut the time spent on repetitive AI tasks by 70% or more, and most people using Claude Code have never touched them. A skill is a saved set of instructions that Claude follows the exact same way every time; you invoke it with one short command, and it just works.

We tested five custom skills built specifically for non-technical users: a content writer, a small business owner, a marketing manager, a teacher, and a student researcher. None of them had a coding background. All five had working skills inside an hour. This guide shows you exactly how they did it, what to include, and the one mistake that trips almost everyone up.

The Key Takeaways

• Claude Code skills are reusable instruction sets you trigger with a single command like /research or /summarise — no coding required
• The difference between a skill that works and one that doesn’t comes down to one sentence: the description
• Skills are just markdown files with plain English instructions; if you can brief a colleague, you can write a skill
• They work beyond Claude Code too — skills are an open standard compatible with Codex CLI, Cursor, Gemini CLI, and others

What Is a Claude Code Skill, Exactly?

Think of a skill as a saved recipe. Instead of explaining to Claude every time how you like your research structured, or what format your meeting summaries should follow, you write it down once in a skill. From then on, you just say /research or /summarise and Claude follows your exact instructions, every time, without you having to explain again.

Skills live inside a simple text file called SKILL.md and are stored in a folder on your computer. That folder name becomes the command name. No coding is involved; you write your instructions in plain English, the same way you’d explain a task to a new assistant.

Once created, skills show up in a menu when you type /, the same way autocomplete works in most apps. You pick the skill, Claude loads your instructions, and gets to work.

And here’s something most people don’t know: since December 2025, skills are an open standard published at agentskills.io. That means a skill you create for Claude Code also works in OpenAI’s Codex CLI, Cursor, Gemini CLI, and other tools that support the format. You’re not locked in.

Skills vs Regular Prompts: What’s the Difference?

	Regular prompt	Claude Code skill
You write instructions	Every single time	Once, then never again
Output consistency	Varies	Follows the same format every time
Shareable with a team	No	Yes (via shared project folder)
Can trigger automatically	No	Yes, if you want it to
Works with a short command	No	Yes (/skill-name)

Who Should Create Claude Code Skills

Skills are not just for developers. Here are five use cases we tested that had nothing to do with writing code.

Content writers and bloggers: a skill that takes a topic and returns a structured article outline, complete with suggested headings, keyword placement, and an intro paragraph. Run it before every article. Same format, every time.

Marketing managers: a skill that takes a product description and returns a LinkedIn post, a Twitter/X post, and email subject lines. One command, three formats, done.

Teachers: a skill that takes a topic and grade level and creates a full lesson plan with timed activities, discussion questions, worksheets, and a rubric. Used before every class session.

Students: a skill that takes a long PDF or article and returns a plain-English summary, the five most important points, and three follow-up questions worth researching further.

Small business owners: a skill that takes a customer complaint email and drafts a professional, empathetic response that fits the tone of the business. Applied consistently without having to brief Claude each time.

All five of these are straightforward to build. The process is the same regardless of what the skill does.

What You Need Before Starting

Before creating your first skill, make sure you have three things:

A Claude Pro subscription ($20/month minimum). Claude Code is not available on the free tier. Pro gets you in the door; the Max plan ($100 or $200/month) gives you more usage if you hit limits. Be aware of this before investing time.

Claude Code installed. Open your terminal (Terminal on Mac, or WSL/Git Bash on Windows) and run:

curl -fsSL https://claude.ai/install.sh | bash

That’s it. No Node.js, no package managers, no dependencies. One command, done.

A text editor. Even TextEdit on Mac or Notepad on Windows works. You’re writing plain text, not code.

The Anatomy of a SKILL.md File

Every skill is a single file called SKILL.md. It has two parts: a small header block (called frontmatter) wrapped in --- lines, and then your instructions below it in plain English.

Here’s what a real, working skill file looks like. This is the research-brief skill we built and tested:

---
name: research-brief
description: Research any topic and produce a structured, well-sourced
  brief with key findings and actionable conclusions. Use this skill
  whenever the user wants research done on a topic, needs a briefing
  document, or wants to understand a subject quickly. Trigger on phrases
  like "research this", "brief me on", "what do I need to know about",
  "deep dive into", "competitive analysis", or "pros and cons of".
---

# Research Brief

Produce a structured, actionable research document that gives someone
the knowledge they need to make a decision or understand a topic — in
10 minutes of reading.

## Who this is for

Business owners evaluating a new tool or strategy. Marketers researching
a competitor or market trend. Students preparing a paper or presentation.

## Workflow

### Step 1: Define the research scope

Before researching, clarify what the user actually needs. "Tell me about
AI in healthcare" is too broad. "What are the most effective uses of AI
in diagnostic imaging, and what are the barriers to adoption?" is
researchable.

### Step 2: Research

Use web search to gather current, reliable information...

### Step 3: Structure the output

**Executive Summary** (150-250 words): The key answer, most important
data point, and recommended next step.

**Key Findings**: Organized by theme, not by source. Each finding gets
a clear statement, supporting evidence, and source attribution.

**Conclusion / Recommendation**: Based on the evidence, what should
the reader do?

That’s it. The name becomes your /command. The description tells Claude when to use it. Everything below the --- block is your instructions. No code, no special syntax — just clear English explaining what you want Claude to do and how.

Let’s break down each part.

How to Create a Claude Code Skill (No Coding Required)

Now that you’ve seen what a finished skill looks like, let’s build one from scratch. The whole process takes about 15 minutes the first time, and under 5 once you’ve done it before. There are six steps, and none of them involve writing code.

Step 1: Decide What You Want the Skill to Do

Start by writing down, in plain English, exactly what you want Claude to do every time. Be specific. “Write a summary” is too vague. “Read the text I give you and return: a three-sentence plain English summary, five bullet points with the most important facts, and two questions I should research further” is a skill.

The more specific your instructions, the more consistent your results.

Step 2: Give It a Name

Choose a short, clear name with no spaces; use hyphens instead. This becomes your /command. Good examples: research-summary, social-posts, quiz-maker, reply-to-complaint.

Step 3: Write the Description Line

This is the most important part. The description is what Claude reads to decide whether to use your skill automatically when you’re working on something related. Write it like you’re telling Claude exactly when to reach for this skill.

Weak description:

“Helps with writing”

Strong description:

“Creates social media posts from a product description. Use when the user asks for posts, captions, or content for LinkedIn, Twitter, or email. Also trigger when the user mentions repurposing content or creating a social media pack.”

A weak description means the skill either never triggers or triggers at the wrong moments. A strong description makes it feel like Claude already knows what you need.

Anthropic’s own documentation confirms this matters: Claude has a tendency to “undertrigger” skills, meaning it won’t use them even when they’d help. Writing a detailed, slightly pushy description that lists specific trigger phrases fixes this.

Step 4: Write Your Instructions

Below the frontmatter, write your full instructions in plain English. Think of this as onboarding a new assistant; tell Claude exactly what to do, in what order, in what format.

Here is an example for a social media skill:

When I give you a product or service description, create three things:

1. A LinkedIn post (150–200 words, professional tone, end with a
   question to encourage comments)

2. A Twitter/X post (under 280 characters, punchy, include one
   relevant hashtag)

3. Three email subject line options (under 50 characters each,
   curiosity-driven)

Do not use jargon. Write like you're talking to a smart friend,
not a press release.

That’s your skill. Plain English, clear structure, no code.

Tips for better instructions:

• Tell Claude what to do AND what not to do. “Do not use jargon” or “Never start with ‘In today’s world'” prevents the generic AI voice.
• Include an example of the output format you want. Showing one ideal example is worth more than three paragraphs of explanation.
• Explain why things matter, not just what to do. “End with a question — this doubles LinkedIn comment rates” gives Claude context to make good judgment calls.

Step 5: Save It in the Right Place

Your skill needs to be saved inside a folder with the skill’s name, and that folder goes inside a .claude/skills/ directory. The file itself must be named SKILL.md.

Here are the exact terminal commands:

mkdir -p ~/.claude/skills/social-posts
nano ~/.claude/skills/social-posts/SKILL.md

Paste your skill content (the frontmatter + instructions), save (Ctrl+O, Enter, Ctrl+X in nano), and you’re done. The skill is immediately available.

If the terminal feels intimidating, there’s an even easier approach: ask Claude Code itself to create the skill for you. Just type something like:

“Help me create a new skill called social-posts. Here are my instructions: [paste your instructions].”

Claude Code will create the file and folder structure for you. This is the conversational approach recommended in Anthropic’s own Help Center, and it works surprisingly well.

Two places to store skills:

• Personal skills at ~/.claude/skills/ — available in every project you work on
• Project skills at .claude/skills/ inside a specific project folder — only available in that project, but shareable with your team via version control

For most people starting out, personal skills are the way to go.

Step 6: Test It

Type /social-posts (or whatever you named yours) and give Claude a product description. Check that the output matches what you described. If something is off, go back and adjust the instructions; be more specific about the part that didn’t work.

Good skills usually take 2-3 rounds of tweaking. This is normal. Adjust the instructions each time until the output consistently matches what you need.

Five Real Skills We Built (and What Makes Each One Work)

We didn’t just write about skills in theory, we built five complete skills aimed at non-technical users, tested them, and refined them until they worked reliably. Here’s what each one does, who it’s for, and the specific technique that makes it effective.

Skill	Who it’s for	What it does	Example prompt	Key technique
spreadsheet-insights	Business owners, marketers, students	Takes a CSV or Excel file and produces a visual report with charts, key findings, and plain-English insights	“Analyze this sales data and tell me what’s happening”	Chart selection guide: matches the right chart type to the data automatically
content-repurposer	Marketers, content creators, small businesses	Takes one blog post or article and creates native versions for LinkedIn, Twitter/X, Instagram, TikTok, and email	“Turn this blog post into social media content for all platforms”	Platform-specific conventions: each output follows the actual norms of that platform, not just a shorter copy-paste
meeting-to-actions	Managers, consultants, project managers	Takes messy meeting notes and produces a clean summary, decision log, action items table with owners and deadlines, and a ready-to-send follow-up email	“Here are my notes from today’s standup, clean them up”	Action item extraction: forces every task to have a specific owner and deadline, flags anything missing as “TBD”
lesson-planner	Teachers, tutors, trainers	Takes a topic and grade level and produces a full lesson plan with timed activities, learning objectives, discussion questions, worksheets, and a rubric	“Create a 50-minute lesson on photosynthesis for 7th graders”	Bloom’s Taxonomy alignment: automatically mixes lower-order and higher-order thinking objectives
research-brief	Students, business owners, anyone making a decision	Takes a question or topic and produces a structured research document with executive summary, sourced findings, and a clear recommendation	“Brief me on the pros and cons of switching to Shopify”	Source quality hierarchy: prioritizes primary sources over aggregators, flags unverified claims

Each of these is a single SKILL.md file between 80 and 150 lines of plain English. No scripts, no code, no complex folder structures. Just clear, specific instructions.

Making Skills Smarter

Once you’ve built your first working skill, there are a few techniques that take them from good to great.

Write Pushy Descriptions

Anthropic’s internal testing found that Claude tends to under-use skills — it won’t trigger them unless the description very clearly matches the situation. The fix: write descriptions that include specific trigger phrases.

Instead of:

“Helps with meeting notes”

Write:

“Transforms meeting notes, transcripts, or call recordings into structured summaries with action items. Trigger on phrases like ‘summarize this meeting’, ‘action items from this call’, ‘meeting notes’, ‘what did we decide’, ‘clean up my notes’, or any request to extract structure from a discussion.”

The more trigger phrases you include, the more reliably the skill fires.

Add Examples Inside the Skill

Showing Claude one example of the output format you want is worth more than a page of abstract instructions. In your SKILL.md, include a section like:

## Example output

**Summary**: The team decided to push the launch date from March 1 to
March 15 due to design delays. Marketing will own the announcement.

**Action Items**:
| Task | Owner | Deadline |
|------|-------|----------|
| Add pricing section to proposal | Sarah | Mar 10 |
| Schedule design review | Mike | Mar 7 |

Claude will match that format far more consistently than if you just describe it in words.

Use Progressive Disclosure

For complex skills, don’t try to cram everything into one file. Keep your main SKILL.md under 500 lines, and put detailed reference material in separate files within the skill folder:

my-skill/
├── SKILL.md              (main instructions)
├── references/
│   ├── style-guide.md    (brand voice details)
│   └── templates.md      (output templates)
└── examples/
    └── good-output.md    (example of ideal result)

In your SKILL.md, tell Claude when to read these files: “If the user asks for a formal report, read references/style-guide.md for tone and formatting rules.”

One Important Setting: Preventing Accidental Triggers

By default, Claude can decide to use your skill automatically if it thinks the situation matches your description. For most informational skills (research, summaries, quiz generation) this is fine and even helpful.

But for skills that send things, publish things, or make irreversible changes, you want to be in control of when they fire. Add this to your frontmatter:

---
name: my-skill
description: ...
disable-model-invocation: true
---

This prevents Claude from triggering the skill automatically; it will only run when you explicitly type the / command.

Sharing Skills With Your Team

If you work with others, you can put your skills inside a shared project folder (.claude/skills/ within the project) and commit them to version control. Everyone on the team then has access to the same instructions and gets the same consistent output.

This is particularly valuable for teams with style guides, tone of voice documents, or standard operating procedures. Turn your SOPs into skills, and every team member follows them automatically.

You can also simply send the .skill file to a colleague. They drop it in their ~/.claude/skills/ folder and it just works.

Where to Find More Skills

Before building something from scratch, check if someone has already made it.

Official repository: Anthropic publishes curated skills at github.com/anthropics/skills, including the skill-creator meta-skill that helps you build skills through conversation.

Community collections: The awesome-claude-skills and awesome-agent-skills repositories on GitHub catalogue over 500 community-built skills across categories ranging from marketing to research to education.

Skill marketplaces: Sites like skillsmp.com offer browsable collections with one-click installation.

Or just ask Claude Code: Type “Create a skill that summarizes customer feedback emails into a weekly report” and Claude Code will generate the entire SKILL.md for you, ask clarifying questions, and refine it based on your feedback. This conversational skill-creation approach is documented in Anthropic’s Help Center and is arguably the most beginner-friendly method of all.

Troubleshooting: When Your Skill Doesn’t Behave

The skill never triggers automatically. Your description is probably too vague. Go back and rewrite it to be more specific about when Claude should use it. Include specific trigger phrases a user might say.

The skill triggers when you don’t want it to. Your description is too broad, or it overlaps with something you do frequently. Either narrow the description or add disable-model-invocation: true to the frontmatter.

The output doesn’t match what you expected. Go back to your instructions and be more specific about the part that went wrong. Add an example of the format you want. The more concrete your instructions, the more predictable the output.

The skill doesn’t appear in the / menu. Check that the file is named exactly SKILL.md (capital letters matter) and that it is inside a named folder inside .claude/skills/. Ask Claude Code to check the file structure for you.

Pro Tips: Getting More Out of Your Skills

1. Use AI to write your skills. You don’t have to stare at a blank SKILL.md and figure out the wording yourself. Describe what you want the skill to do in plain English, paste that into Claude, ChatGPT, or Gemini, and ask it to generate a complete SKILL.md file with frontmatter and structured instructions. You’ll get a solid first draft in seconds that you can then refine.
2. Send the same brief to multiple models and pick the best result. Different AI models think differently, Claude tends to write more structured, methodical instructions, GPT often produces more conversational ones, and Gemini sometimes catches edge cases the others miss. Write your skill description once, send it to two or three models, and compare the outputs. Take the best parts from each and combine them into one skill. Tools like Fello AI make this easy since you can access Claude, GPT, Gemini, and other models from one place without switching between apps.
3. Start small, then layer. Your first version of a skill doesn’t need to cover every edge case. Get the basic workflow right, test it a few times, and add complexity only where the output falls short. A 30-line skill that works reliably beats a 200-line skill that’s fragile.
4. Steal structure from skills you like. Found a community skill with a great layout? Copy its structure and replace the content with your own use case. The patterns transfer well — a research skill’s workflow section works just as well for a competitive analysis skill.
5. Version your skills. When you tweak a skill, save the previous version first (even just renaming it SKILL-v1.md). If your changes make things worse, you can roll back instantly instead of trying to remember what worked before.

Conclusion

Claude Code skills are one of the most underused features in AI right now, and almost entirely overlooked by non-technical users. You don’t need to write code, understand APIs, or know anything about software to build a skill that saves you hours every week.

Start with one skill for the task you repeat most often. Write the instructions in plain English, be specific about the output format, and test it a few times until it behaves exactly how you want. Then share it with your team and watch the consistency improve across the board.

If you want to go deeper, our Claude Cowork guide covers how to use Claude Code as a full productivity workspace, and the best AI agents comparison shows how Claude stacks up when you’re evaluating AI tools for a team.

FAQ