Claude Skills: A Practical Guide for Business Owners

A Claude Skill is a folder containing a SKILL.md file with instructions that teach Claude how to do a specific job. It works across Claude.ai, Claude Code, and the API without modification.

Think of it this way: without a skill, Claude is a brilliant generalist that guesses what you want. With a skill, Claude follows the exact method you’ve defined — your frameworks, your voice, your standards. Same AI, completely different output.

Most tutorials focus on skills for developers. This guide is for business owners who want to use Claude as a genuine thinking partner and production tool — not just a code generator.

Every skill is a folder with a SKILL.md file inside it. The file has two parts:

YAML frontmatter (between --- markers) that tells Claude the skill’s name and when to use it. The name uses kebab-case (e.g. sparring-partner). The description is critical — Claude uses it to decide whether to activate the skill for the current task.

Markdown instructions below the frontmatter that tell Claude what to do when the skill is invoked. This is where your method, framework, and quality standards live.

You can also add reference files (like REFERENCE.md) for supplemental information that Claude pulls in when needed.

Where skills live:

• Claude.ai: Upload as a ZIP file via Customize > Skills. Click the “+” button, then “+ Create skill.”
• Claude Code: Place in ~/.claude/skills/ and Claude reads them automatically.
• API: Same SKILL.md format, loaded programmatically.

The same skill works everywhere. Write once, use across all Claude surfaces.

This is the question everyone asks. The official answer from Anthropic is clear:

Projects give Claude persistent context about a specific body of work. Your company, a client, a research initiative. Everything in the project’s context window is loaded into every conversation.

Skills teach Claude how to do something. A project might contain all the background on your product launch. A skill teaches Claude your writing standards or your code review process.

The practical test: if you find yourself copying the same instructions across multiple projects, that’s a signal to create a skill instead.

I use both together. My Claude Projects have Soul (who I am) and Context (what’s happening). My skills have methods (how to do specific tasks). The project tells Claude who it’s working with. The skill tells it how to work.

Every skill I build follows the same structure. The frontmatter handles identity; the body handles method:

The body of the SKILL.md typically includes:

Anthropic’s advice on descriptions: make them “a little bit pushy.” Claude has a tendency to under-trigger skills, so a clear, specific description helps it recognise when to load yours.

Most skill tutorials show developer examples. Here are skills built for business work — consulting, content creation, and operations:

The thinking skills (sparring partner, first principles) are deeply personal — built from how I actually make decisions. Nobody else’s version would work the same way for me.

The production skills (LinkedIn writer, copywriting fundamentals) are more transferable. The GTM system skills (ICP extractor, pain language miner) were built for a specific client project but the patterns apply to any B2B operation.

This is one of the things I do for clients. I sit with them, extract their expertise through conversation, and build skills that capture how they actually work. Their objection handling. Their pricing logic. Their voice. Packaged as skills that any AI tool can use.

Pick something you do repeatedly and explain well. The task you find yourself describing to every new team member, every new client, every new project.

Write out how you do it: the steps, the rules, the examples, the things to avoid. Put it in the SKILL.md format with a clear name and description. Upload it to Claude.ai or drop it in your skills folder.

Start with one skill. Use it for a week. Notice what’s missing — Claude will show you by getting things wrong in specific, fixable ways. Update the instructions. The first version is never the best, but it’s infinitely better than no version.

Your skill library grows naturally from your work. Every project, every client, every repeated task is a candidate. Over time, you stop building from scratch and start assembling from parts.

That’s when AI stops feeling like a tool and starts feeling like a team.

The March 2026 update added three significant capabilities:

• Built-in evals. Test your skill’s output quality against specific criteria. This means you can measure whether v2 of a skill actually performs better than v1.
• A/B testing. Run two versions of a skill side by side and compare results. Useful for iterating on content creation or analysis skills.
• Trigger optimisation. Improved how Claude decides when to activate a skill. The progressive disclosure system is now smarter about matching tasks to skills.

For business users, the most impactful change is progressive disclosure. You can now maintain a large library of skills without worrying about context window limits. Claude reads the skill descriptions, decides which ones are relevant, and loads only those. Dozens of skills, zero overhead.

This is part of a series on building AI systems for business. Start with Meta-Prompting 101 for the method, then Claude Projects for the environment. Skills are the tools.

The full picture