Claude Skills: Teach Your AI Once, Use It Forever

Remember the scene in The Matrix where Neo gets kung fu uploaded directly into his brain? One moment, he knows nothing about martial arts. Next, he opens his eyes and says, "I know kung fu."

That's what Claude Skills do for your AI workflows.

When you want Claude to do something specific—follow your brand guidelines, format reports a certain way, apply your company's writing style—you usually have two options: prompt it every single time, or set up custom instructions in a project that only work when you're inside that project. Otherwise, you're back to copying and pasting the same detailed prompts over and over.

Skills change this completely. You create a knowledge package once—instructions, best practices, examples, specific guidance for a task—upload it to Claude, and you're done. From that point forward, whenever you mention anything relevant to that Skill (or even start a task it applies to), Claude automatically uses that knowledge. No manual triggering. No remembering which project has which instructions. It just works.

Anthropic launched Skills in October 2025, marking a meaningful shift in how we work with AI assistants. Instead of training yourself to write better prompts, you train Claude once and move on.

Stop using AI like a search box.

Join us live TODAY at 12:00 PM EST for All Things AI’s Lunch & Learn as Mark Hinkle shows you how to turn Claude, ChatGPT, and Gemini into real workflow engines—persistent systems that capture your expertise and automate repeatable work. This session breaks down how to build menu-driven AI projects that actually scale across your day-to-day tasks.

If you use AI at work, this will change how you use it.

Stop copying and pasting the same prompts. Skills let you upload expertise directly into Claude's brain.

The beauty of Skills is the "anywhere, anytime, automatically" part. You don't have to keep uploading prompts. You don't have to be in a specific project. Skills work across Claude.ai, the Claude desktop app, Claude Code, and the API—build once, use everywhere.

This guide shows you what Skills actually are, how to enable them, and how to create your first custom Skill—whether you want Claude to do it conversationally or you prefer to build one manually—no coding experience required.

Skills require a paid Claude subscription:

All paid tiers also require Code Execution to be enabled, which is included in your subscription.

For Enterprise plans: Organization Owners must first enable both "Code execution and file creation" and "Skills" in Admin settings before individual members can access them.

For Team plans: Skills are enabled by default at the organization level.

For Pro and Max plans: You can enable Skills directly in your personal settings.

Pricing verified December 2025. Check Claude's pricing page for current rates.

Step 1: Navigate to Settings > Capabilities in Claude

Step 2: Ensure "Code execution and file creation" is toggled on

Step 3: Scroll to the Skills section

Step 4: Toggle on any Anthropic-provided Skills you want to use (Excel, Word, PowerPoint, PDF)

Step 5: To add custom Skills, click "Upload skill" and upload a ZIP file containing your Skill folder

Time: Under 2 minutes

Claude comes with several pre-built Skills that activate automatically when relevant:

You don't invoke these manually. Ask Claude to "create a PowerPoint about Q3 results," and it automatically loads the PowerPoint Skill and produces an actual .pptx file you can download. Ask it to "build a budget spreadsheet with formulas," and it uses the Excel Skill to generate a working .xlsx file.

Skills uses a clever system called "progressive disclosure" that keeps Claude fast while giving it access to specialized expertise.

At startup: Claude scans only the lightweight metadata for each Skill—just the name and a short description. This costs roughly 100 tokens per Skill, so you can have dozens installed without slowing anything down.

When you make a request: Claude matches your task against those descriptions. If there's a match, it loads the full Skill instructions into context (typically under 5,000 tokens).

If the Skill references other files: Claude loads only what it needs—a brand guide, a template, a script—on demand.

This is fundamentally different from Projects, which load all their context upfront and keep it loaded for the entire conversation. Skills are surgical: Claude pulls exactly what it needs, when it needs it.

You have two options: let Claude guide you through it conversationally, or build the Skill manually.

Option A: The Conversational Method (Easiest)

Claude includes a built-in "skill-creator" Skill that walks you through the process interactively. This is the fastest way to get started.

Step 1: Open a new Claude conversation

Step 2: Type something like:

For example: "Help me create a skill for writing customer support responses in our company's voice."

Step 3: Answer Claude's questions about your process

Claude will ask what the Skill should do, when it should activate, and what makes good output for this type of work. If you have relevant materials—templates, brand guidelines, examples of work you're proud of—upload them during this conversation.

Step 4: Review and download the Skill

Claude generates the properly structured Skill folder and provides it as a ZIP file for download.

Step 5: Upload the ZIP to Settings > Capabilities > Skills

Time: 10-30 minutes depending on complexity

Option B: The Manual Method

If you prefer more control, you can create a Skill folder yourself.

Step 1: Create a new folder on your computer (name it something descriptive like brand-guidelines or quarterly-report-format)

Step 2: Inside that folder, create a file named SKILL.md

Step 3: Add the required YAML frontmatter at the top:

The name field has a 64-character limit. The description is critical—Claude uses it to decide when to invoke your Skill, so be specific about what it does and when it applies. You have 200 characters to work with.

Step 4: Below the frontmatter, add your instructions in Markdown:

Step 5: ZIP the entire folder and upload it to Claude via Settings > Capabilities > Skills

Time: 15-45 minutes depending on how detailed your instructions are

Tip: If your Skill isn't activating when expected, check that your description clearly states both what the Skill does and when Claude should use it. Vague descriptions like "Help with documents" won't trigger reliably.

These three features solve related but different problems:

Custom Instructions (Settings > Profile) apply broadly to all your conversations. They're suitable for general preferences like "respond concisely" or "I work in healthcare."

Projects provide static background knowledge that's loaded whenever you start a chat within that project. Great for ongoing work where you need the same context every time—like a product launch or client engagement.

Skills provide specialized procedures that activate dynamically when needed and work everywhere across Claude. They're best for repeatable workflows you use across different contexts—formatting reports, applying brand guidelines, and analyzing specific types of data.

The key difference: Projects are always-on within their container. Skills are always available but only load when relevant.

No org-wide sharing: Custom Skills you upload are private to your individual account. If your team wants to use the same Skill, each person needs to upload it separately. Anthropic has indicated enterprise-wide deployment is coming, but no timeline has been announced.

No conversation memory: Skills don't carry state between sessions. Each conversation triggers Skills independently based on the current request.

Security considerations: Skills can include executable code, which makes them powerful but also means you should only install Skills from trusted sources. A malicious Skill could potentially exfiltrate data or execute unintended actions. Anthropic recommends auditing any Skill from an unfamiliar source before enabling it—read through the SKILL.md file and any bundled scripts to understand what they do.

Occasional misfires: The progressive disclosure system sometimes has minor hallucinations. Claude might claim a Skill exists that you never created, or fail to invoke one that clearly matches the context. If Claude isn't using a Skill you expect, try being more explicit: "Use my brand guidelines skill to create this presentation."

Skills aren't just a convenience feature. They represent a shift from "prompt engineering every time" to "teach once, use forever." The initial investment in creating a well-crafted Skill pays dividends through consistent execution and reduced cognitive load.

Your AI now knows kung fu—time to put it to work.

This is where AI gets real. No sales pitches—just 4,000 builders, engineers, operators, and execs sharing how AI actually works in practice, from hands-on workshops to real-world sessions, right in the heart of RTP and Duke’s AI ecosystem.

We want to make learning AI as practical, accessible, and immediately useful as possible—so you leave knowing what to build, what to deploy, and what actually drives results.

I appreciate your support.