How to Use Claude Skills: The 20-Minute Setup That Stops You Re-Explaining Your Business (2026)

TL;DR: A Skill Is the Checklist You Keep Pasting Into Chat

Most of the "Claude Skills" content out there is written for developers building agent frameworks, and it makes a folder with one markdown file sound like infrastructure. It isn't. Anthropic's own docs describe the trigger plainly: create a skill when you keep pasting the same instructions or multi-step procedure into chat, or when a section of your always-on context has grown into a procedure rather than a fact. That's it. A Skill is a saved procedure with a routing rule on top. I run both of my businesses on a handful of them, and the ones that earn their keep are embarrassingly ordinary — how we write a proposal, how we triage an inbound lead, how we prep a Monday report. As of September 2026 they work the same way in Claude Code, the Claude.ai app, Cowork, and the API, because the format is an open standard.

What a Skill actually is (the definition, in one paragraph)

A Skill is a folder. Inside it is a file called SKILL.md with two parts: YAML frontmatter between --- markers, whose description tells Claude when to use the skill, and markdown content underneath, which is what Claude follows when it does. The folder can also hold supporting files — scripts, templates, reference documents — that load on demand. Claude reads only each skill's name and description up front; the full body loads when a task matches. So a 40-page reference document inside a skill costs you nothing until the moment it's needed. That's the difference between a Skill and stuffing everything into always-on context: Skills are lazy, in the good sense. (The Cowork-specific catalog — the 15 built-in skills and building custom ones inside Cowork — is its own post: Claude Cowork Skills explained. This one is the cross-surface mechanics.)

Step 1: Write the SKILL.md (10 minutes)

Here's the shape, using the first skill I'd tell any owner to build — the one that drafts replies to inbound leads in your voice, from your real facts:

---
description: Drafts a reply to an inbound lead or customer email in our voice. Use when the user pastes an email and asks for a reply, a draft, or "what should I say."
---

## How we reply
- Answer the actual question in the first two sentences.
- Quote prices only from business-facts.md in this folder. Never invent a number.
- One call to action, never two. If they already declined a call, no CTA.
- Sign off "Justin" — no "best regards."

## Before you draft
Read business-facts.md for current offers and policies.

Two things to notice. The description is written like a routing rule — specific enough that Claude activates it at exactly the right moment and not when you ask something unrelated. And the instructions point at a supporting file (business-facts.md) instead of repeating the facts, so the facts live in one place and the skill stays short. That two-file pattern — a lean SKILL.md plus a facts file — is the single most useful habit in this whole post.

Step 2: Put it where Claude looks (2 minutes)

Where a skill lives decides who can use it. In Claude Code, as of 2026:

• Personal — ~/.claude/skills/<skill-name>/SKILL.md. Available in every project you open. Start here.
• Project — .claude/skills/<skill-name>/SKILL.md inside a repository. Only that project sees it, and it travels with the repo when a teammate clones it.
• Plugin — bundled inside an installed plugin, namespaced as plugin-name:skill-name so it can't collide.
• Enterprise — pushed through managed settings to everyone in the organization.

When two skills share a name, enterprise beats personal and personal beats project — and any of them beats a bundled skill with the same name. Claude Code watches these folders live, so editing a SKILL.md takes effect in the current session without a restart. One gotcha worth knowing: Cowork and cloud sessions don't read the folders on your machine at all. They load the skills enabled on your claude.ai account, synced at session start, which you manage from Customize in the desktop sidebar or the skills settings on claude.ai. If a skill only exists in ~/.claude/skills/, a scheduled Cowork run will report it as not found. Enable it on the account and the problem goes away.

Step 3: Invoke it — or let Claude do it (1 minute)

Two ways. Ask something that matches the description ("here's an email from a lead, what should I say back?") and Claude loads the skill on its own. Or type / followed by the skill name to run it directly. Claude Code also lets you flip a skill to manual-only in the frontmatter, so Claude never triggers it unprompted — the right setting for anything that spends real time or money, like a full audit or a batch job. And yes: the old custom commands were merged into skills, so a .claude/commands/deploy.md and a .claude/skills/deploy/SKILL.md both create /deploy and behave the same.

Using Skills in Claude.ai, not just Claude Code

The same format runs in the app. Anthropic ships its own skills there — the document, spreadsheet, presentation, and PDF ones that make Claude produce real files — and you can upload a custom skill as a zip from the skills area in Settings. As reported for 2026, custom skills are available on Pro, Max, Team, and Enterprise with code execution turned on, and Team or Enterprise admins enable them for the whole organization first. Whatever you enable on the account is what Cowork picks up. For developers, the Skills API — reported generally available in August 2026 — lets the same SKILL.md ship inside a product. One file, four surfaces. That's the reason to learn the format once.

The three skills every business owner should build first

1. The reply skill (above). Inbound email in, draft in your voice out, prices only from your facts file. This is the one that pays daily.
2. The intake skill. "Use when the user pastes notes from a discovery call." It turns your messy call notes into the same five-section summary every time — situation, pain, budget signal, next step, risks — so your CRM stops being a graveyard of half-typed thoughts.
3. The Monday skill. "Use when the user asks for the weekly report." It reads the same three sources you check every Monday and produces the same one-page brief. Manual-only. Run it once a week and stop re-explaining what "the report" means.

Notice none of these are clever. That's the point — the value of a skill is repetition, not intelligence. The clever version is the one you'll never finish.

The mistakes I see in the first week

• Description written like a title. "Proposal skill" routes nothing. "Use when the user asks for a proposal, a scope, a quote, or an SOW" routes correctly.
• One skill that does everything. One clear job per skill. Two related jobs are two skills.
• Facts inside the skill. Prices, policies, and names change. Put them in a supporting file the skill reads, so you update one place.
• Skipping the account enable. Works on your laptop, fails in Cowork. See Step 2.

The micro-action for today

Open your chat history and find the instruction you've pasted most often in the last month. That's your first skill. Twenty minutes, one folder, one file. If you'd rather have the whole system installed — the skills, the facts file they read from, and the training so your team actually uses them — that's what the Claude for Small Business install is (from $4,500, about two weeks and three hours of your time). The do-it-yourself version with the templates is the CFSB Playbook ($397, 30-day money-back), and if you want to talk through which procedures are worth turning into skills before you build any, that's a free strategy call. Related reading: how to actually use Claude Code day to day and how to use Claude Projects, which is where the facts file often starts life.