Claude Skills: What They Are and How to Build One for Your Business

Most people who use Claude re-explain themselves every single session.

Same brand voice instructions. Same output format requirements. Same tone guidelines. Same list of words they never want Claude to use. The same context, pasted in fresh at the start of every conversation, producing slightly different results every time because the model is starting from scratch each time.

Claude Skills fix this. Not partially completely.

A Skill is a saved, reusable instruction pack that Claude loads automatically when it recognises a relevant task. You teach Claude once your voice, your format, your workflow, your quality bar and it applies that knowledge consistently across every session, every output, every time. No re-explaining. No copy-pasting the same system prompt. No prompting lottery where the output quality depends on how well you worded the instruction that morning.

This guide explains exactly what Skills are, how they work technically (without requiring you to be technical), the two types that matter for business use, and the step-by-step process for building your first one this afternoon.

What a Skill Actually Is

The official definition from Anthropic's API documentation: Skills are reusable, filesystem-based resources that provide Claude with domain-specific expertise workflows, context, and best practices that transform general-purpose agents into specialists.

The plain-English version: a Skill is like an onboarding document for a new team member, except the team member is Claude, and the onboarding happens in milliseconds.

When you hire a freelance writer, you send them your brand guide, your style rules, your audience description, your formatting preferences. You do that once. After that, they know how to work with you. Skills do the same thing for Claude except the knowledge persists across every session, automatically, without you ever sending the onboarding document again.

Every Skill lives in a folder on your computer or in Claude's settings, and at its core is a file called SKILL.md. This file tells Claude three things: what the Skill is called, when to use it (the trigger conditions), and what to do when it activates (the instructions).

The architecture is deliberately lightweight. Claude scans each Skill's name and description using roughly 100 tokens per Skill when a session starts. The full instructions which can run up to 5,000 tokens only load when Claude determines the Skill is relevant to the current task. Supporting files, scripts, and templates load only when explicitly needed. This means you can have dozens of Skills installed without any of them slowing down conversations where they're irrelevant.

The Two Types of Skills And Why the Distinction Matters

Understanding this split changes how you think about which Skills to build and which to install.

Type 1: Capability Uplift Skills

These give Claude new abilities it does not have by default access to web data, document processing, browser control, API integrations.

Examples of Capability Uplift Skills:

Firecrawl gives Claude reliable access to live web data during a session. When Claude needs to pull information from a URL, most default tools fail on JavaScript-heavy sites. The Firecrawl Skill gives Claude a reliable extraction mechanism that handles modern web pages correctly.

File/Document Processing gives Claude direct access to PDFs, spreadsheets, and CSVs for parsing, cleaning, validating, and converting into pipeline-ready formats. Without this Skill, document processing requires manual uploads for each file.

webapp-testing lets Claude test local web applications using Playwright for UI verification and debugging. Opens up QA workflows that weren't possible before.

agent-browser lets Claude control any web interface through stable element references. Full support for clicks, form fills, screenshots, and parallel browser sessions.

These Skills expand what Claude can do. They are the difference between Claude knowing about a capability and Claude actually having it available in your session.

Type 2: Encoded Preference Skills

These teach Claude how you want things done your standards, your voice, your format, your quality bar. They don't add new capabilities. They make Claude's existing capabilities consistent with your specific requirements.

Examples of Encoded Preference Skills:

brand-voice your tone guidelines, vocabulary preferences, sentence rhythm, the words you never want used, the structural conventions your content follows. Every piece of writing Claude produces in sessions where this Skill is active matches your brand standards without you mentioning them.

content-format your blog post structure, heading hierarchy, section lengths, FAQ format, image prompt style. Claude produces content in exactly this structure every time, without you specifying it.

client-report how your client-facing documents look, what sections they always include, how data is presented, what the executive summary covers. Every report Claude produces matches the template your clients expect.

email-style how you write professional emails, your sign-off conventions, your formality level by recipient type, the phrases you use and avoid.

These Skills are the ones nobody publishes because they encode your taste, not a generalised best practice. The most valuable Skill in your library three months from now will be one you built yourself, tuned to your specific way of working, refined through dozens of real uses until it consistently produces output you would have written yourself.

Lo-fi editorial illustration split between a professional repeatedly re-explaining context to Claude versus the same professional using a skill for instant consistent output

How Skills Work: The Technical Picture (Without the Technical Pain)

You do not need to write code to build a Skill. But understanding the architecture helps you build better ones.

Every Skill lives in a folder with this structure:

1your-skill-name/
2├── SKILL.md          ← the main file Claude reads
3├── scripts/          ← optional: executable scripts Claude can run
4 │   └── helper.py
5└── resources/        ← optional: templates, reference files, examples
6    └── template.json

The SKILL.md file has two parts. The frontmatter the metadata at the top tells Claude what the Skill is called and when to activate it. The body contains the actual instructions Claude follows when the Skill is active.

Here is what a minimal SKILL.md looks like for a brand voice Skill:

1---
2name: brand-voice
3description: |
4  Use this skill when writing any client-facing content —
5  blog posts, emails, social media, reports, proposals.
6  Applies brand tone, vocabulary standards, and formatting rules.
7---
8
9# Brand Voice Skill
10
11## Tone Guidelines
12- Direct and commercial — no filler phrases, no buzzwords
13- Sentences vary in length — mix short punchy lines with longer explanations
14- First person where appropriate — write as if the founder is speaking
15- Never use: "leverage", "comprehensive", "seamless", "holistic", "actionable"
16
17## Formatting Rules
18- Blog posts: H2 headers only (no H3 unless a list requires it)
19- Paragraphs: maximum 4 sentences. Mix 1-sentence paragraphs deliberately.
20- Every major claim cites a specific named source
21- FAQ sections at the end of every blog post — minimum 6 questions
22
23## Voice Examples
24[Paste 2-3 paragraphs of your best existing content here]

The description field in the frontmatter is the most important line in the entire file. It is what Claude reads to decide whether to activate the Skill. If the description is vague, the Skill never fires. If it is specific naming the exact situations where it applies Claude loads it reliably every time.

How to Install Skills

Skills are available across three surfaces, and the installation process differs slightly for each.

Claude.ai (web and desktop): Go to Settings → Features → Custom Skills. Upload your Skill folder as a zip file. Available on Pro, Max, Team, and Enterprise plans with code execution enabled.

Claude Code (terminal): Place the Skill folder in ~/.claude/skills/your-skill-name/. Claude Code discovers it automatically in the next session. No restart required.

Via the API: Upload Skills programmatically using the API's skill management endpoints. For teams running Claude in production workflows, this is how Skills are deployed and updated without manual intervention.

Pre-built Skills from the marketplace: In Claude Desktop, go to Customize → Personal Plugins → Browse Plugins. The marketplace contains Skills and plugin bundles from Anthropic and the community frontend design, document processing, internal communications, MCP builders, and dozens of others. Install any of them with one click.

One important note: Custom Skills uploaded to Claude.ai are individual to each user. They are not automatically shared across an organisation. Team and Enterprise plan admins can deploy Skills workspace-wide through the admin panel this is how consistent brand standards and workflow Skills get applied to an entire team without each member building them individually.

Building Your First Skill: Three Paths

Path 1: The Skill Creator (Recommended for Non-Technical Users)

The Skill Creator is a built-in tool inside Claude Cowork that builds your Skill interactively no SKILL.md editing required.

Step 1: Open Claude Cowork on your desktop. Type: "Use the skill-creator to help me build a Skill for [your task]."

Step 2: The Skill Creator runs an interview. It asks specific questions about your task what triggers the Skill, what the output should look like, what rules Claude should follow, what examples you want it to reference. Answer specifically. The more concrete your answers, the more precisely the Skill will match your expectations. Select Claude's pre-made answers if they fit, or type your own your own answers produce better Skills.

Step 3: Claude generates the complete Skill folder correctly named, with the SKILL.md file, frontmatter, and instructions written based on your interview answers.

Step 4: Claude runs an evaluation automatically. This step is critical and most people skip it because they are in a hurry to use the Skill. The evaluation tests the Skill against sample inputs and shows you where the outputs fall short of your expectations. Fix the gaps in the SKILL.md before deploying, not after.

Step 5: Click "Always Allow" when prompted, and the Skill is active in your current session. Upload it to your Settings to make it permanent.

Path 2: Write It From Scratch (For Specific, Complex Workflows)

For business owners who know exactly what they want and want full control over the instruction logic, writing directly is faster than the Skill Creator interview for complex Skills.

The minimal viable Skill takes 30 minutes to write and test:

1. Create a folder named your-skill-name (lowercase, hyphens, no spaces)
2. Inside it, create SKILL.md
3. Write the frontmatter: name and description
4. Write the instructions: trigger conditions, step-by-step process, output format, quality standards, examples
5. Upload to Claude.ai Settings or place in ~/.claude/skills/
6. Test by triggering the Skill with a real task and reviewing the output against your expectations

The most common mistake: writing instructions that are too broad. A Skill called "writing" that covers blogs, emails, social posts, and reports produces inconsistent results because different tasks require different conventions. Build one Skill per task type. A blog-post Skill. An email-outreach Skill. A client-report Skill. Narrow focus produces reliable activation and consistent output.

Lo-fi editorial illustration of business owner completing the Claude Skill Creator interview on laptop with handwritten workflow notes on notepad beside keyboard

Path 3: MakeMy Skill (Fastest for Simple Skills)

For straightforward Skills where you know the task but don't want to go through the interview process, makemyskill.com lets you describe the Skill in plain language and generates the SKILL.md file automatically.

Describe the Skill in as much detail as possible the longer and more specific, the better the output. The tool searches the web for relevant context before writing the instructions, which is useful when you are not sure what best practices exist for a task. Download the generated Skill, review the instructions, upload to Claude, and test.

This is the fastest path for simple Skills. For complex Skills that encode your specific voice or multi-step workflow logic, the Skill Creator interview produces better results because it extracts nuance that a one-paragraph description misses.

The Business Skills Worth Building First

These are the highest-ROI Skills for most business owners the ones that produce the most time savings relative to the effort of building them:

Brand Voice Skill the single most valuable Skill for any business producing written content. Encodes tone, vocabulary rules, structural conventions, and voice examples. Every piece of content Claude produces in sessions where this Skill is active is on-brand without you mentioning it.

Weekly Report Skill your report structure, the metrics it always covers, how data is presented, the narrative format your stakeholders expect. Build once, deploy every reporting cycle. What used to take two hours to draft takes 20 minutes to review.

Email Outreach Skill your cold outreach framework, your tone by recipient type, the structural rules that produce replies (under 100 words, one question, specific hook). Every outreach draft reflects the approach that actually works for you.

Content Brief Skill how you brief content the sections you always include, the questions you always ask, the research inputs you require before writing starts. Every brief Claude produces follows your exact methodology.

Client Summary Skill how you communicate project status to clients. The language level, the structure, what you highlight, what you omit, how you frame problems. Every client communication is consistent with the relationship standard you want to maintain.

SOX/Compliance Documentation Skill for finance and legal teams, a Skill that encodes the specific document format, required sections, and language standards your auditors expect. Every draft starts from the correct baseline without anyone remembering to specify it.

The Mistakes That Break Skills

Vague descriptions: If the description does not tell Claude exactly when to use the Skill, it won't activate reliably. "Use this skill for writing" is too broad. "Use this skill when writing blog posts, LinkedIn articles, or website copy in the brand voice" is specific enough to trigger consistently.

Instructions that try to do too much: A Skill that covers six different task types produces six inconsistent output styles. One Skill, one task type. If you find yourself writing "if the task is X, do Y, but if the task is Z, do W" that is two Skills, not one.

Skipping the evaluation: The Skill Creator builds an evaluation automatically. Most people skip reviewing it. The evaluation is the difference between a Skill that works 80% of the time and one that works 95% of the time. Review every evaluation result before deploying a Skill in production use.

Loading too many Skills: Eight to twelve well-chosen Skills cover most of a senior professional's daily work. More than that and you start paying context tax every loaded Skill consumes tokens whether it helps or not. Audit your Skills monthly. Delete anything you have not triggered in 30 days.

Never updating Skills: A Skill built in January reflects your workflow in January. When your process changes, your brand evolves, or your output standards shift update the Skill. A Skill that no longer matches your current standards is worse than no Skill, because it produces output that looks like yours but isn't.

Skills vs. Projects vs. MCP: The Distinction That Matters

These three features get confused constantly. They are not the same, and they compose well together.

Projects are persistent workspaces with saved context the brief for a specific client, the details of an active listing, the scope of a running campaign. Projects hold the what the specific information relevant to one engagement.

Skills are reusable instruction packs that tell Claude how to work your brand voice, your report format, your quality standards. Skills travel across every Project because they define the way you always work, not the context of a specific project.

MCP (Model Context Protocol) gives Claude access to external tools and data sources your CRM, your inbox, your Google Drive, your project management platform. MCP handles the connections. Skills handle the behaviour.

A mature Claude setup uses all three: Projects hold engagement-specific context, Skills encode your working standards, and MCP connections give Claude access to the live data it needs to execute. The combination is what moves Claude from a general-purpose tool to operational infrastructure that reflects how your business actually works.

Lo-fi editorial illustration of business owner reviewing a whiteboard showing organised Claude Skills library alongside Projects and MCP connections in mature AI workflow setup

FAQ

What are Claude Skills and how are they different from prompts? Claude Skills are reusable instruction packs stored as SKILL.md files that Claude loads automatically when relevant tasks arise. Unlike prompts which you write fresh each session Skills persist across all sessions and activate contextually without you mentioning them. A Skill encodes your brand voice, workflow standards, or domain expertise once, and Claude applies them consistently every time.

Do I need to be technical to build a Claude Skill? No. The Skill Creator inside Claude Cowork builds Skills through a guided interview you answer questions about your workflow in plain English and Claude generates the SKILL.md file. makemyskill.com offers an even faster path for simple Skills. Writing SKILL.md directly from scratch is optional and only beneficial when you want precise control over complex instruction logic.

What plans include Claude Skills? Custom Skills are available on Pro, Max, Team, and Enterprise plans with code execution enabled. Pre-built Skills from the marketplace are available to all users. Team and Enterprise plan admins can deploy Skills workspace-wide through the admin panel. Custom Skills on Claude.ai are individual to each user they do not automatically sync across an organisation.

How many Skills should I install? Eight to twelve well-chosen Skills cover most of a professional's daily work. Beyond that, you start paying context tax each loaded Skill consumes tokens even when inactive. Install Skills that address your highest-frequency recurring tasks. Audit monthly and remove anything you have not used in 30 days.

What is the difference between a Capability Uplift Skill and an Encoded Preference Skill? Capability Uplift Skills give Claude new abilities web data access, document processing, browser control. Encoded Preference Skills teach Claude your specific standards your brand voice, your output format, your quality bar. Both are valuable. The distinction helps you understand which problem you are solving: expanding what Claude can do, or making what Claude already does consistent with how you work.

Can Claude Skills be used across Claude.ai, Claude Code, and the API? Yes. Skills are an open standard the same SKILL.md file works across Claude.ai, Claude Code, and the API without modification. Skills uploaded to Claude.ai must be separately uploaded to the API; they do not sync automatically between surfaces. Anthropic has published Skills as an open standard designed to be portable across AI platforms as the ecosystem matures.

What makes a Claude Skill stop working? Three common causes: a vague description that does not tell Claude when to activate the Skill, instructions that try to cover too many different task types, and Skills that no longer match your current workflow because they were never updated after your process changed. The Skill Creator's built-in evaluation catches most activation problems before deployment.

What is the most valuable Skill to build first? For most business owners producing written content, the brand voice Skill produces the most immediate ROI it eliminates the single most common repeated input across content, email, and report workflows. For operations and finance teams, the weekly report or compliance documentation Skill produces the most measurable time savings. Build the Skill that addresses your highest-frequency manual context-setting task first.