Claude Skills Tutorial for Beginners: The Complete 2026 Guide

Verdict: Claude Skills are the highest-leverage feature most people aren't using. They're folders containing a SKILL.md file — plain-text instructions that tell Claude exactly how to handle a specific task, every time, without re-prompting. You don't need to code. You start with Anthropic's built-in skills, modify them for your brand, connect them to tools like Slack and Canva, and eventually build your own. The entire system runs on an open standard (the Agent Skills specification), so what you build works across Claude.ai, Claude Code, and the API.

Last verified: 2026-08-01 · Skills available on Pro, Max, Team, Enterprise plans · 17 official skills in anthropics/skills · Agent Skills spec adopted by Claude Code, Codex CLI, Cursor, Gemini CLI

What Are Claude Skills?

Claude Skills are directories containing a SKILL.md file with YAML frontmatter and markdown instructions that teach Claude how to handle specific tasks in a consistent, repeatable way. Think of them as procedural memory: you write the instructions once, and Claude loads them automatically whenever your request matches what the skill does (claude.com/skills).

A skill is fundamentally just a folder:

my-skill/
├── SKILL.md          # Required: metadata + instructions
├── scripts/          # Optional: executable code
├── references/       # Optional: documentation
└── assets/           # Optional: templates, resources

The SKILL.md file has two parts: YAML frontmatter (at minimum, name and description) and a markdown body with instructions. That's it. No special syntax, no programming language — just structured plain text (agentskills.io/specification).

Here's what the format looks like:

---
name: internal-comms
description: Write internal communications using company formats. Use when asked to write status reports, newsletters, FAQs, incident reports, etc.
---

## When to use this skill
- Status reports
- Company newsletters
- FAQ responses

## How to use this skill
1. Identify the communication type
2. Load the appropriate template
3. Follow formatting instructions

How skills load: progressive disclosure

Skills use a three-tier loading system called progressive disclosure that keeps Claude's context window clean (claude.com/docs/skills/overview, agentskills.io/specification):

1. Discovery (~100 tokens per skill): At startup, Claude loads only the name and description of every installed skill. This is how it knows what's available.
2. Activation (full SKILL.md): When your task matches a skill's description, Claude reads the entire SKILL.md file into context. The body should be under 500 lines / 5000 tokens.
3. Resource loading (as needed): Any bundled scripts, reference files, or templates load only when the skill's instructions call for them.

This means you can have 20+ skills installed with minimal context overhead, because Claude only loads the full instructions for the one skill it actually needs.

Do you need a paid plan?

Skills are available to users on Claude Pro ($20/month), Max ($100-$200/month), Team ($25/seat/month), and Enterprise plans. They are not available on the free plan as of mid-2026. Skills also require code execution to be enabled in your Claude settings — go to Settings > Capabilities and make sure the toggle is on (claude.com/docs/skills/overview, claude.com/pricing).

How to use a skill: two methods

There are two ways Claude picks up a skill, and you should understand both.

Method 1 — Automatic activation. You write a normal prompt with no mention of skills. Claude reads the descriptions of all installed skills at startup and decides whether one is relevant. If you have a PowerPoint skill installed and you say "turn this research paper into a deck," Claude recognizes the request calls for slides and loads the skill automatically. You don't invoke anything.

Method 2 — Direct invocation. You type a forward slash (/) in the chat box. Claude shows a list of all your installed skills. You click one, type your prompt, and Claude uses that skill's instructions. This is the method you'll use most often, because it gives you control over which skill is active.

The difference matters: if your skill has a vague description, automatic activation won't trigger reliably. If the description is specific — listing exact trigger words like "slides," "presentation," "pptx" — Claude activates it without being asked. This is why writing a good description is the single most important part of skill design.

Level 1: Start with Claude's built-in skills

The easiest way to understand skills is to use one before you build one. Claude ships with built-in skills for common document tasks: creating Excel spreadsheets, PowerPoint presentations, Word documents, and PDFs. These are available in the Skills browser under Settings > Capabilities > Skills.

Here's how to get started:

1. Open Claude.ai and go to Settings > Capabilities.
2. Confirm Code Execution and File Creation are toggled on.
3. Click Skills and then Browse.
4. You'll see a set of pre-built skills designed by Anthropic for common tasks. Click any skill to read its description.
5. Click the + icon to add a skill to your library.
6. In a new chat, type / to see it in your skill list.

See the difference a skill makes

The best way to feel the impact is a before-and-after test. Try this:

• Without a skill: Ask Claude "I want to learn about convolutional neural networks." Claude gives you a generic explanation — a definition, some core concepts, a few resources. Fine, but not personalized.

• With a skill: Install a learning-focused skill (one that's designed to teach rather than just answer). Activate it with /learn and ask the same question. Instead of dumping information, Claude now asks you questions: "What's your starting point? Do you already know neural networks? Is there something specific pulling you toward CNNs?" It's trying to understand where you are in the learning process so it can teach accordingly.

The same prompt, completely different response — just by adding a skill. And this is only Level 1.

Level 2: Modify a built-in skill for your brand

Once you're comfortable using built-in skills, the next step is modifying one for your own needs. Instead of building from scratch, you take an existing skill and adapt it.

A practical example: Anthropic ships a brand-guidelines skill that applies Claude's brand colors (orange and gray) to documents and presentations. But your company has its own brand colors, fonts, and tone. Rather than writing a brand skill from zero, you can duplicate the built-in one and replace Anthropic's brand specs with yours.

Here's the process:

1. Make sure the built-in skill is installed and active.
2. In a chat, invoke the skill and tell Claude: "Create a new skill by duplicating this one. Replace the brand colors and typography with mine (uploaded as a reference document). Call the new skill 'brand-book.' Give me an option to save it."
3. Upload your brand guidelines document.
4. Claude generates the modified skill. Click Save to add it to your skill library.
5. In your next chat, type / and you'll see "brand-book" in the list.
6. Now when you ask Claude to make a deck or document, invoke /brand-book and the output will follow your brand identity instead of Anthropic's.

The principle: modify before you create. Starting from an existing skill saves time and teaches you the structure by example.

Level 3: Use partner skills and connectors

Level 2 deals with modifying your skills. Level 3 is about connecting Claude to the external tools you already use — Slack, Canva, Notion, HubSpot, Figma, and more.

Under Settings > Capabilities > Skills, click Plugins to browse skills organized by domain (marketing, sales, productivity, data, finance). Each domain contains a collection of relevant skills. For example, the Marketing plugin includes skills for brand reviews, campaign planning, competitive briefs, and content creation.

Connectors: the superpower

The most powerful part of Level 3 is connectors. Think of Claude as the brain — it thinks and decides, but it can't take action without hands. Connectors are those hands. They connect Claude to your external tools so it can actually do things, not just suggest them.

Within each domain's plugin page, click the Connectors tab. For marketing, you might see Slack, Canva, Figma, HubSpot, Notion, Amplitude, Klaviyo, and Supermetrics. For sales, you might see Slack, HubSpot, Close, Monday.com, and Fireflies.

To connect a tool:

1. Click the + icon next to the connector.
2. You'll be redirected to that tool's authentication page.
3. Log in and authorize Claude to access your account.
4. Once connected, Claude can perform real actions in that tool (e.g., resize images in Canva, pull CRM data from HubSpot, post messages in Slack).

Partner skills: written by the companies themselves

Under the Partners tab, you'll find skills written by the companies that make these tools — Slack, Canva, Notion, Atlassian, Box, and others. These aren't community-built approximations; they're the official skills the companies maintain so Claude works with their products optimally (claude.com/skills).

For example, Canva offers six partner skills: brand-check, implement-feedback, design-feedback, edit-design, resize-for-social-media, and bulk-create. Once you install the Canva partner skill and connect your account, you can upload an image and say: "Convert this 16:9 image into 1:1, 9:16, 3:2, and 4:3." Claude executes the resize inside Canva without you leaving the chat.

Level 4: Create your own custom skill

This is where skills go from convenient to transformative. A custom skill encodes a workflow you repeat — writing scripts in your voice, generating reports in your format, reviewing code your way — so you never have to explain it again.

Use the Skill Creator

Anthropic ships a meta-skill called skill-creator whose only job is to help you build other skills. It's pre-installed in Claude Desktop and Claude Cowork, and available via the anthropics/skills repository for Claude Code (github.com/anthropics/skills).

To use it:

1. In a chat, type / and select skill-creator.
2. Describe your task with as much detail as possible. Don't be vague — specify the exact format, structure, tone, and any constraints.
3. Upload reference documents if you have them (sample outputs, style guides, templates). The more reference material you provide, the better the output.
4. Answer Claude's follow-up questions. The Skill Creator will ask what inputs you'll typically give, what the output should include, and whether you want test cases.
5. Review the test outputs Claude generates. This is critical: give specific feedback on what's wrong ("the hook is too generic," "push for the real tool name instead of writing around it").
6. Iterate. Keep the feedback loop going until the output matches what you want.
7. Click Save to add the skill to your library.

The three things that make a custom skill great

Based on reading Anthropic's own skill files and the Agent Skills specification, here's what separates a good skill from a great one:

1. A specific, keyword-rich description. The description is the single most important field in your SKILL.md. It tells Claude when to activate the skill. A vague description ("helps with presentations") won't trigger reliably. A specific one ("creates PowerPoint presentations from research papers, meeting notes, or outlines. Use when the user mentions slides, deck, pptx, presentation, or asks to turn content into PowerPoint") will trigger almost every time. List every trigger word someone might use.

2. Document what NOT to do, not just what to do. Reading Anthropic's built-in skill files reveals that most of their content is warnings — what to avoid, common mistakes, edge cases. Great skills don't just list the happy path; they document the failure modes. "Don't use more than 5 bullet points per slide." "Don't invent statistics." "Don't add filler transitions." These guardrails are what make Claude's output consistent rather than variable.

3. Reference the hierarchy of resources. Keep your SKILL.md under 500 lines. If you need more detail, put it in reference files (references/) and point to them from the main file. Claude loads only the main file on activation; reference files load when the instructions tell Claude to read them. This is the progressive disclosure pattern: small core, deep periphery (agentskills.io/specification).

How to read Anthropic's own skill files

Here's a technique that almost nobody talks about — and it's the fastest way to level up your skill writing. The built-in skills that Anthropic ships are actual SKILL.md files sitting in Claude's sandbox. You can read them.

In a Claude.ai chat with code execution enabled, ask Claude to run a directory listing of its skills directory. The exact path may vary, but the prompt looks like:

Run this in bash and show me the output:
ls -la ~/.claude/skills/

You'll see a list of skill folders. Then ask Claude to open one:

Show me the contents of the SKILL.md file inside the pptx skill folder.

What you'll find inside these files is instructive:

• The description lists every trigger keyword — slides, deck, presentation, pptx, PowerPoint, keynote. This is why their skills always trigger and yours might not.
• Most of the file is warnings about what not to do — formatting mistakes, common errors, quality guardrails. Great skills document the mistakes, not just the steps.
• There are actual scripts inside — not just instructions, but executable Python code that Claude runs. That's the ceiling you're aiming for.

Don't guess at how to write a skill. Read the source. Anthropic's files are open-source (Apache 2.0) and available on GitHub. Copy their patterns: the description structure, the warning density, the resource hierarchy.

Skills vs. other Claude features: what's different?

Feature	What it does	When to use
Skills	Task-specific procedures that load dynamically when needed	Repeating a specific workflow (e.g., "write Instagram scripts in my voice")
Plugins	Shareable packages that bundle skills, connectors, and slash commands	Distributing a set of related skills to a team
Projects	Static background knowledge always loaded in specific chats	Giving Claude persistent context about a project (docs, code, brand guide)
MCP	Connects Claude to external services via the Model Context Protocol	Letting Claude query your database, search your wiki, call your API
Custom Instructions	Broad preferences applied to all conversations	"Always respond in X tone" or "Use metric units"

Source: claude.com/docs/skills/overview

The key distinction: Skills load on-demand (progressive disclosure), Projects are always loaded, Custom Instructions are global, and MCP handles external connections. Use Skills when you have a repeatable task that doesn't need to be in context all the time.

The Agent Skills open standard: why it matters

The SKILL.md format isn't proprietary to Anthropic. It's an open standard called the Agent Skills specification, maintained at agentskills.io. The specification was originally developed by Anthropic and released as open source in late 2025. As of 2026, it has been adopted by multiple agent platforms — meaning a skill you write for Claude can also work in other tools that support the standard (agentskills.io).

The specification defines:

• Required frontmatter fields: name (1-64 chars, lowercase alphanumeric + hyphens, must match the folder name) and description (1-1024 chars)
• Optional fields: license, compatibility (max 500 chars, environment requirements), metadata (arbitrary key-value map), allowed-tools (experimental, pre-approved tools)
• The scripts/, references/, and assets/ directory conventions
• Progressive disclosure as the loading architecture
• Validation via skills-ref validate ./my-skill

If you're building skills you want to share or use across tools, validate them against the spec. This is also how you know your skill will work when Claude Code, Codex CLI, Cursor, or Gemini CLI load it.

What this means for you

If you're using AI for your work or small business, skills are the difference between an assistant and an employee. Here's the progression:

1. Today: Install one built-in skill (the PowerPoint or PDF skill). Use it. See how the output changes.
2. This week: Modify a built-in skill for your brand. Upload your brand colors, fonts, and tone. Save it.
3. This month: Connect one external tool (Slack, Canva, or Notion) as a connector. Run a workflow end-to-end inside Claude.
4. Going forward: Create one custom skill for the workflow you repeat most. Write a detailed description, document both what to do and what to avoid, iterate with feedback, and save it.

Skills are learned by reps, not by reading. Pick one level to try this week and do it. The compound payoff — Claude getting better at your specific work every time you add a skill — is why this is the most underused power feature in the entire Claude ecosystem.

For a deeper look at the framework behind building skills (as opposed to installing them), see our guide to building AI agent skills instead of collecting them. If you're hitting Claude Code token limits while writing skills, our free open-source tools to cut Claude Code tokens can help. And if you want persistent memory that complements skills (they handle task workflows, not long-term recall), see our Claude + Obsidian memory setup guide.

FAQ

Q: What is a Claude Skill?

A: A Claude Skill is a folder containing a SKILL.md file with YAML frontmatter and markdown instructions that teaches Claude how to handle a specific task. Claude loads the skill automatically when your request matches its description, and follows the instructions to produce consistent output every time.

Q: Do Claude Skills work on the free plan?

A: No. Skills require a Claude Pro ($20/month), Max ($100-$200/month), Team ($25/seat/month), or Enterprise plan as of mid-2026. They also require code execution to be enabled in your settings (claude.com/pricing, claude.com/docs/skills/overview).

Q: What is the difference between a skill and a plugin?

A: A skill is a single SKILL.md folder that defines one task workflow. A plugin is a shareable package that can bundle multiple skills, connectors, slash commands, and sub-agents together. Plugins are the distribution mechanism; skills are the content (claude.com/docs/skills/overview).

Q: How do I make my Claude Skill trigger automatically?

A: Write a specific, keyword-rich description in the YAML frontmatter. List every word someone might use when asking for the task your skill handles (e.g., "slides, deck, presentation, pptx, PowerPoint"). The description — not the body — is what Claude reads at startup to decide whether to activate the skill (agentskills.io/specification).

Q: Can I use Claude Skills in other AI tools?

A: Yes. The SKILL.md format is an open standard (the Agent Skills specification at agentskills.io). Skills you create can work in any tool that implements the standard, including Claude Code, OpenAI Codex CLI, Cursor, Gemini CLI, and GitHub Copilot.

Q: How many skills can I install at once?

A: Anthropic has not published a hard limit. Because skills use progressive disclosure (only ~100 tokens of metadata load at startup), you can keep many installed without overwhelming Claude's context. In practice, 3-10 skills covering your main workflows is the sweet spot; too many with overlapping scope can cause inconsistent behavior.

Q: Where can I find pre-built skills to install?

A: Anthropic's official skills are in the anthropics/skills repository on GitHub (17 skills as of July 2026, Apache 2.0 licensed). You can also browse the built-in Skills browser in Claude.ai settings, and partner skills from companies like Canva, Notion, Figma, and Atlassian are available in the Plugins tab.

Sources

1. Claude Skills — Official Page — Anthropic's official Skills landing page with overview, use cases, and partner integrations.
2. Skills Overview — Claude.ai Documentation — Official documentation covering availability, progressive disclosure, types of skills, and feature comparisons.
3. Agent Skills Specification — The open standard specification for the SKILL.md format, including frontmatter schema, directory structure, and validation.
4. anthropics/skills — GitHub Repository — Anthropic's official open-source skills repository (17 skills, Apache 2.0).
5. Claude Pricing Plans — Official pricing for Free, Pro, Max, Team, and Enterprise plans.
6. Skill Creator — SKILL.md — The Skill Creator meta-skill from Anthropic's repo, showing how skills are structured and iterated.

Updates & Corrections

• 2026-08-01 — Initial publication. All facts verified against primary sources (Anthropic official docs, GitHub repo, agentskills.io specification). Pricing and plan availability verified against claude.com/pricing as of August 2026.

---