Claude Skills for Small Business: A Practical Guide
Claude skills are reusable instruction packages that handle payroll planning, month-end close, and client follow-up. What they are and how to use them.
Payroll planning eats your Tuesday afternoon. Client follow-up lives in a draft folder nobody opens. Both jobs have repeatable steps.
Claude skills are reusable instruction packages that teach Claude to do a specific job, like planning payroll or drafting a quote. Anthropic ships 15 small-business skills inside Claude Cowork, connected to QuickBooks, PayPal, HubSpot, and Canva. You can also write your own as a simple folder with a SKILL.md file. You approve everything before it sends or posts.
That short list is the whole product in one breath. The two pages that rank first for this question are Anthropic’s launch materials, and both describe what shipped.
Neither shows the file inside the skills, and neither says which jobs to automate first. This page adds the part launch pages skip: what the files look like and where they pay for themselves.
TL;DR
- A Claude skill is a folder with a SKILL.md file. YAML frontmatter holds the name and the description, and the description decides when Claude loads the skill.
- Claude for Small Business ships 15 workflows and 15 skills inside Claude Cowork. They read your connected tools and stage the work. Nothing goes out until you approve.
- You can build your own skill in minutes. On Claude for Mac you can record the task and let Claude write most of the file.
- Ready-made skills arrive with Claude Code and sync from claude.ai. Public collections hold more. Deleting a folder removes one.
What are Claude skills for small business?
A Claude skill is a small instruction file that teaches Claude one job, from planning payroll to drafting a quote. It loads only when the job matches, because the description at the top is its trigger. Skills repeat your exact steps every time.
That same format sits under every skill Anthropic ships, whether Anthropic wrote it or you did. It follows the open Agent Skills specification. A folder you build for one tool can move to another that adopts it, per the Anthropic help center on custom skills.
Skills appear in two places. Claude Cowork ships the ready-made suite, and Claude Code ships bundled skills and the tools to write your own.
The format and the trigger are the same. The what are Claude Code skills post walks through it in plain language.
| Prompt | Slash command | Skill | |
|---|---|---|---|
| What it is | One instruction you type once | A skill you trigger by name | A folder that teaches one workflow |
| When it loads | The second you hit send | When you type /name |
When the task matches the description |
| What it remembers | Nothing after the reply | Whatever the file says | The same steps on every run |
| Who writes it | You, every time | You, once | You, once. Claude repeats it |
A skill fires on its description. Wrong words in the prompt, and the file never loads.
What comes with Claude for Small Business?
Claude for Small Business is a toggle inside Claude Cowork. It ships 15 ready-to-run agentic workflows and 15 skills that cover the jobs owners do weekly. The skills plug into the tools you already use.
Small businesses account for 44 percent of U.S. GDP and employ nearly half the private-sector workforce, says the Claude for Small Business announcement, which draws on SBA data. Their AI use often stops at the chat window. The suite is Anthropic’s answer to that gap.
Headline jobs from the same announcement:
- Payroll planning with a 30-day cash forecast
- Month-end close with a plain-English P&L
- The Monday business pulse, ranked by what needs you
- Campaign planning with Canva assets and a staged send
- Invoice chasing with reminders matched to each customer
The list also includes a margin analyzer. The full inventory lives on the solutions page. Each connected tool handles a specific job, the announcement spells out:
| Connector | What it handles inside Claude |
|---|---|
| Intuit QuickBooks | Payroll planning, the monthly close, cash flow, tax-season prep, reconciliation |
| PayPal | Settlements, invoicing, disputes, refunds |
| HubSpot | Lead triage, customer pulse, campaign attribution |
| Canva | Content generation, team editing, publishing, performance tracking |
| DocuSign | Contract signatures, status tracking, executed-copy filing |
Google Workspace and Microsoft 365 round out the list. The announcement names all seven tools the suite works inside.
The trust model is the approval gate. Half the owners in Anthropic’s survey named data security as their single biggest hesitation. The product answers it three ways:
- You approve the plan first, then approve again before anything sends, posts, or pays, the approval-first design.
- Existing permissions hold. If an employee cannot see it in QuickBooks or Drive, Claude cannot see it either.
- Anthropic does not train on your business data by default on Team and Enterprise plans.
Claude proposes and stages. You hold every approve button.
The jobs that earn a skill file
I run three tests before a job earns a skill file in my own business. These are the tests:
- Repeats weekly. A monthly job stays a prompt.
- Has steps you can write down. If the order changes every time, no file can hold it.
- Produces output you can judge. You need to spot a wrong result at a glance.
The first files in my business came from the jobs with the longest paper trail:
- Client follow-up
- Proposal and quote writing
- The standard operating procedures behind my lead-gen
My outreach system sends about 200 form submissions a day, and the follow-up cadence runs as a skill file. The proposal a prospect reads comes out of the Proposal & Quote Writer file. The SOP Builder runs the same way.
The same risk-reversal runs through my agency. Prospects get a working website before they pay anything.
Anyone selling a decade of agent experience is selling you something. Anthropic shipped this suite in May 2026, per the launch announcement, and the category is young.
The ceiling is what you can write down. I built the automated team that wrote this page, and every stage runs off a skill file.
A job earns a file when it repeats weekly and the output is checkable at a glance. Everything else stays a prompt.
How do you use Claude skills in your business?
You start one with a plain prompt. “Get me ready for payroll on the 15th” fires the payroll skill. Claude reads your connected tools and drafts the forecast. The reminders wait for your approval.
The Claude Academy small business tutorial runs the four starter skills end to end, and two details stand out. Claude notes which tools it read from, so you can verify any number at its source.
The skills also learn from corrections. Tell /invoice-chase you chase at 15 days, and the change sticks.
| Skill | Prompt that starts it | What you get |
|---|---|---|
/monday-brief |
“Help me build a Monday morning brief every week in Slack.” | One page covering cash, settlements, pipeline, and calendar, ranked by what needs you |
/run-campaign |
“Find my weakest revenue month and plan a promo.” | Offer draft, Canva assets, a HubSpot segment, and a staged email |
/close-month |
“Close out March for me.” | Reconciled transactions, a flag list, and a plain-English P&L packet |
/plan-payroll |
“Get me ready for payroll on the 15th.” | A 30-day cash chart and a reminder per overdue customer |
Install is a toggle. Turn on the suite in Cowork and connect the tools you use. Then type the prompt.
The tutorial links a plugin install guide for setup details.
The starter skills are the fastest path in. One prompt, and the pattern shows up in the output.
How do you create a custom Claude skill?
Make a folder with a SKILL.md file. The file needs a name and a description. Save it where Claude looks, or record yourself doing the task in Claude for Mac and let Claude propose the skill. Then test it with real prompts and tighten the description until it fires.
For personal use, skills live in ~/.claude/skills/<skill-name>/. Project skills sit in .claude/skills/<skill-name>/ inside the repo. Claude Code reads the description and loads the folder when the task matches, per the Claude Code skills documentation.
The fastest route needs no typing. The Anthropic help center on custom skills walks the recording flow. It runs on Pro, Max, and Team plans in Cowork on Claude for Mac.
Record about 10 minutes of screen with narration on. Claude proposes a skill file from what it sees. You review the proposal before you save it.
Testing drives the whole build. Try several prompts that should trigger the skill, then watch whether Claude loads it.
If it stays quiet, change the description first. Write it in the words you would actually type.
A skill is a hiring decision for one task. The file is the job description.
What is the SKILL.md format?
SKILL.md is a text file with two parts. YAML frontmatter carries the name and the description, capped at 64 and 200 characters. The description is the trigger, so it states when Claude should load the file. Markdown below holds the instructions.
The Anthropic help center on custom skills publishes the field rules, and they are short:
| Part | Job | Limit |
|---|---|---|
name |
Human-friendly name for the skill | 64 characters |
description |
Tells Claude when to load the skill | 200 characters |
| Markdown body | The instructions Claude follows | One job per file |
| Reference files | Extra detail that loads only when needed | Any file in the folder |
| Scripts | Code Claude can run for the job | Review before enabling |
The description carries the load. The Claude Code skills documentation warns that the combined description and when_to_use text cuts off at 1,536 characters in the skill list. The trigger phrase has to lead.
A minimal file is a few lines. Open with ---. Inside sit the name: and description: lines.
Close with --- and write the instructions below. The step-by-step build, from blank folder to first test, lives in the how to create Claude Code skills post.
Name and description are the contract. One job per file keeps both honest.
Where can you find ready-made Claude skills?
Claude Code ships bundled skills like /doctor and /code-review, and claude.ai has more you can enable so they sync to your machines. GitHub holds public collections of skill files, and Anthropic’s help center links a starter repository. Owners also publish their own bundles.
Claude Code arrives with a set of bundled skills, per the bundled skills list:
/doctor/code-review/batch/debug/loop/claude-api
Skills you enable on claude.ai sync down to ~/.claude/skills/synced/ when sync is on, covered in the skills synced from claude.ai section. If I were starting fresh, the bundled set would be my first hour with the format.
Anthropic keeps a public repository of example skills on GitHub. The custom skills article links it so you can use the files as templates.
Plugins extend the same ecosystem. The skill-creator plugin from the official marketplace automates the comparison loop, the skills reference shows.
The format is identical everywhere, which is the point. I run my own skill files in the same spec.
The two that carry the most weight in my agency are the SOP Builder and the Proposal & Quote Writer. They are part of the skills bundle. Read the pattern there.
AutomateReal skills
A ready-made skill takes two minutes to try. Deleting the folder ends the experiment.
Fast answers to the questions owners ask
The follow-up questions, answered plainly.
How is a Claude skill different from a prompt or a slash command?
A prompt is one instruction you type once. A slash command is a skill you trigger by name.
A skill file loads on its own when the task matches its description, so it runs the same steps every time. Custom commands in Claude Code have merged into skills, the Claude Code skills documentation notes.
Where are Claude skills stored on disk?
Personal skills live in ~/.claude/skills/<skill-name>/SKILL.md, and project skills sit in .claude/skills/<skill-name>/SKILL.md inside the repo. Skills synced from claude.ai land in ~/.claude/skills/synced/.
To remove a skill, delete its folder. The Claude Code skills documentation shows the full location table.
Do skills work with Claude Code subagents?
Yes. Add context: fork to the frontmatter and the skill runs in its own subagent, with no access to your conversation history.
It executes in the background and delivers the result when it finishes. The subagent execution section of the Claude Code skills documentation spells it out.
How do I write a skill that calls a tool or a script?
Put the script in the skill folder and reference it from the SKILL.md file. Claude can install packages from PyPI or npm and run the code, the help center notes.
The allowed-tools frontmatter field pre-approves specific tools for the run, the skills reference explains. Never hardcode API keys or passwords in a skill file, the same article warns.
Do I need coding skills to create a Claude skill?
No. A skill is a folder and a text file with two fields. The recording feature on Claude for Mac builds one from your screen without any writing.
Scripts are optional extras. Skills run on Free through Enterprise plans, though code execution must be enabled, the help center notes.
Start with the job you do every week
Pick the weekly job that annoys you most. Run one starter workflow, or write your own description and see what fires. The first skill you keep pays for the setup in the same week.
If you want help finding that first workflow, a discovery call maps it in about 30 minutes.
AutomateReal services
For a managed version, see the skills bundle.