Which Claude Skills Should a Small Business Install First?

Coding Liquids tutorial cover featuring Sagnik Bhattacharya for Which Claude Skills Should a Small Business Install First?
Coding Liquids tutorial cover featuring Sagnik Bhattacharya for Which Claude Skills Should a Small Business Install First?

A small business should install Anthropic's own Excel, Word, PDF and PowerPoint skills first (built in once code execution is on), then doc-coauthoring for proposals, internal-comms for team updates and skill-creator to build your own. Turn them on in that order from Customize, then Skills, and read any third-party skill's files before switching it on.

The Complete Excel Course with AI Integration

From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.

By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner

Buy now

The order matters because the risk rises as you go. The first four are created and maintained by Anthropic and switch on automatically when a task needs them. The next two are Anthropic examples you'll probably adapt. Skill-creator is where your own skills begin, and skills from anyone else come last, after a safety check, because Anthropic's help centre names prompt injection and data leaks through malicious package code as the main risks.

What is a Claude skill, and how is it different from connectors and projects?

A Claude skill is a folder of instructions, and optionally scripts and reference files, that Claude loads when a task matches its description. Claude doesn't read every skill at the start of a chat: it sees each skill's name and one-line description, loads the full instructions only when a request fits, and opens any extra files only if those instructions point to them. That's why a well-written description is most of what makes a skill work.

The other ways to customise Claude do different jobs. A connector lets Claude reach another app, such as your inbox or accounts. A project holds background knowledge that's always loaded in that project's chats. A plugin packages skills, connectors and commands as one install. Skills are the "how we do this task" layer. If you haven't set up connectors yet, the companion tutorial on nine Claude connectors worth switching on covers that side.

Anthropic's help centre lists skills on every plan, Free included, as long as code execution is on. Pro is $17 a month billed annually or $20 monthly; Team is $20 per seat billed annually or $25 monthly, with a two-seat minimum. One of Anthropic's developer pages lists only the paid plans for skills, so if you're on Free and can't see them, that's the likely reason.

Why do Claude skills need code execution switched on?

Claude skills need code execution switched on, because they run in Claude's code sandbox and nothing below works until it is:

  • Free, Pro, Max and Ultra: open Settings, then Capabilities, and turn on "Code execution and file creation" (called "Create and edit files" in Anthropic's updated settings layout).
  • Team: skills are on by default at the organisation level; members turn individual skills on in Customize, then Skills.
  • Enterprise: an owner checks that both code execution and Skills are enabled under Organization settings, then Plugins and skills, on the Policy tab.

Then open Customize, then Skills. The Your skills tab groups skills by where they came from (created by you, from your organisation, shared with you, and from Anthropic and partners), and the Discover tab lists skills you can add.

Which seven Claude skills should a small business install first?

A small business should install Anthropic's Excel, Word, PDF and PowerPoint skills first, then doc-coauthoring, internal-comms and skill-creator, in that order.

The worked examples come from an illustrative five-person insurance broker: two account handlers, a claims handler, an office manager and the owner. Its documents are what you'd expect: renewal letters, quote comparison spreadsheets, insurer PDFs, proposal forms and the odd presentation for a larger commercial client.

1. Excel (xlsx): spreadsheets with working formulas

What it does. Anthropic's Excel skill handles any task where a spreadsheet is the main input or output: opening, editing and fixing .xlsx and .csv files, adding columns and formulas, formatting, charts, cleaning messy data and converting between table formats. It's one of the built-in skills that Claude uses automatically when relevant.

A broker job. Three insurers have quoted for a client's commercial combined policy, each in a different PDF layout. The handler uploads the three quotes and asks:

Build a comparison spreadsheet from these three quotes: one row per
cover section, columns for each insurer's limit, excess and premium,
a total premium row with formulas, and a Notes column flagging any
section one insurer excludes. Highlight the lowest premium in each row.

An illustrative result: a workbook with 11 rows, live SUM formulas and a note that one insurer excludes theft from the yard. What the handler fixed: the third insurer's premium included taxes and fees in the PDF and the others didn't, so the totals weren't like for like. The fix was a line in the prompt ("Show premiums before tax, with tax in a separate column"), which later went into the broker's own skill.

Safety check. None needed beyond normal care: it's Anthropic's skill. Do check every formula on the first few workbooks, especially totals.

2. Word (docx): letters and reports in your format

What it does. The Word skill creates, reads and edits .docx files and Word templates, including tables of contents, headings, page numbers and letterheads, find-and-replace, and tracked changes and comments.

A broker job. Renewal letters follow a set layout: the broker's letterhead, a summary of changes in cover, the premium, and a reminder of the duty to disclose material changes. Upload last year's approved letter as a .docx and ask Claude to produce this year's version for a named client from the renewal terms, keeping the layout. The handler reviews it with tracked changes on, so every edit the office manager makes later is visible.

What to watch. The skill makes a good document; it doesn't know your regulatory wording. Keep the compliance paragraphs as fixed text you paste in, not text Claude writes.

3. PDF: forms, merges and scanned documents

What it does. According to its description, the PDF skill reads and extracts text and tables, merges and splits files, rotates pages, adds watermarks, creates new PDFs, fills PDF forms, encrypts and decrypts files, extracts images and runs OCR on scanned PDFs so they become searchable. It's also one of Anthropic's examples of a skill that carries scripts, which it runs to fill forms and extract tables.

A broker job. A new client's previous broker sent a 40-page scanned bundle. Ask Claude to make it searchable, extract the schedule of insured items into a table, and split out the claims history pages into their own file. Ten minutes of checking replaces an hour of retyping. A second job: filling an insurer's PDF proposal form from the client's fact-find, with every filled field listed afterwards so the handler can check it against the source.

What to watch. OCR on poor scans misreads numbers. Check sums insured and dates against the original page before anything goes to an insurer.

4. PowerPoint (pptx): decks from notes and templates

What it does. The PowerPoint skill creates, reads and edits slide decks, works with templates (.potx), layouts and speaker notes, and combines or splits slide files.

A broker job. Once a year the owner presents a renewal review to the broker's largest commercial client. Upload the broker's slide template and the comparison spreadsheet from skill 1, and ask for an eight-slide deck: claims over the year, market conditions in plain words, three options with premiums, and a recommendation slide left blank for the owner to write. Leaving the recommendation to a person is deliberate.

Why fourth. Most small firms use it least, so it can wait a week. It's still Anthropic's and still low risk.

5. doc-coauthoring: proposals that hold up

What it does. This Anthropic example skill guides you through writing a structured document, such as a proposal, a decision document or a spec, in three stages: gathering context, refining structure, and testing the draft on a reader. Its value is the process: it asks the questions you'd otherwise skip.

A broker job. A proposal to manage a local group of shops' insurance as one programme. Instead of "write me a proposal", the skill takes the owner through who the reader is, what decision they need to make and what evidence they'll want, then drafts and asks how a sceptical reader would respond. An illustrative exchange from the reader test: "The client's finance director will ask how the combined premium compares with what each shop pays now. The draft doesn't say." That gap would have cost the meeting.

Where to find it. Anthropic's skills repository says its example skills are already available on paid Claude plans. Look in Customize, then Skills, under the Anthropic group.

6. internal-comms: weekly updates in one format

What it does. This example skill covers internal communications: status reports, leadership updates, "3P" updates (progress, plans, problems), company newsletters, FAQs and incident reports. It works by loading a guideline file for each type of message.

A broker job. A Friday update to the team: renewals won and lost, claims opened and closed, anything a handler is stuck on. Paste the week's numbers and notes and ask for a 3P update. An illustrative result: "Progress: 14 renewals completed, 12 retained. Plans: three large renewals due next week. Problems: one insurer has stopped writing fleet business, affecting two clients." Three lines everyone reads.

Make it yours. The guideline files reflect Anthropic's formats. The skill's source is public in Anthropic's repository, so copy its structure, replace the examples with your own (your Friday update, your client newsletter) and upload it as a custom skill. That's the step skill-creator makes easier.

7. skill-creator: building the broker's own skills

What it does. Anthropic's skill-creator helps you create new skills, improve existing ones, run test prompts and sharpen a skill's description so it triggers at the right time. Anthropic's docs say that on claude.ai you turn it on under Customize, then Skills, where it's listed as from Anthropic, and that in chat it works through your test prompts one at a time and shows the results.

A broker job. The first custom skill the owner built was "renewal-letter": the house layout, the fixed compliance wording as a reference file, the tax line from skill 1's lesson, and three approved letters as examples. Describe what you want and skill-creator produces the folder, writes the SKILL.md file with its required name and description, and bundles the files. A good second skill is your brand voice; the tutorial on building one brand-voice skill every staff member uses walks through it.

The rules a custom skill must follow. Anthropic's docs set two required frontmatter fields: a name of lowercase letters, numbers and hyphens, up to 64 characters, matching the folder name; and a description of up to 1,024 characters saying what the skill does and when to use it. You upload a skill as a ZIP file whose top level is the skill's folder, through the + button in Customize, then Skills, then Create skill, then Upload a skill. A ZIP with SKILL.md loose at its root isn't recognised.

A minimal SKILL.md for the broker's letter skill:

---
name: renewal-letter
description: Draft a client renewal letter in our house layout from
  the insurer's renewal terms. Use whenever someone asks for a renewal
  letter, renewal summary or renewal invitation for a client.
---
1. Use the layout in examples/approved-letter-1.docx.
2. Summarise changes in cover in a table: section, last year, this year.
3. Show premium before tax, then tax, then total.
4. Paste references/compliance-wording.md exactly as written. Never
   rephrase it.
5. End with the handler's name and direct line, not the office number.

Testing the custom skill before the team relies on it

Anthropic's docs make a useful distinction: trying a few prompts tells you a skill loads, not whether the output is better. Skill-creator can run agreed test prompts through your skill and show the results, so give it three that cover the awkward cases. The broker's three, with illustrative results:

  • "Renewal letter for the café client from these terms." Loaded the skill, used the layout, kept the compliance text word for word. Pass.
  • "Write to the florist client: their insurer won't renew." The skill fired, but a non-renewal isn't a renewal letter, and the result read like good news. The description now ends "Do not use for non-renewals or cancellations."
  • "Renewal letter, premium up 18%, cover unchanged." The table showed no changes, which was right, but there was no sentence explaining the increase. Step 6 was added: "If the premium rises by more than 10%, add one plain sentence giving the insurer's stated reason, or flag that none was given."

Twenty minutes of testing caught two problems that would otherwise have reached clients. Re-run the same three prompts after every change to the skill.

One you might expect on the list: brand-guidelines

Anthropic's repository also has a brand-guidelines skill, and it's tempting. But it applies Anthropic's own colours and typography, so turning it on as-is would put Anthropic's look on your documents. Its value is as a model: Anthropic's docs point to it as a good example of a skill that is only instructions. Copy its shape, not its contents.

If the spreadsheet side is the slow part, try The Complete Excel Course with AI Integration on Coding Liquids.

How much time can the seven skills save in a first week?

In the illustrative broker's first week, the seven skills cut repeated jobs such as a renewal letter from 25 minutes to 10, checking included. The times are examples, not measurements.

An illustrative schedule and the numbers that came out of it, for the five-person broker on Claude Team (five seats at $25 a month billed monthly, $125 in total, or $100 on annual billing):

DaySkillJob triedTime beforeTime after, including checking
MondayExcelThree-insurer quote comparison45 min15 min
MondayWordRenewal letter25 min10 min
TuesdayPDFScanned bundle to searchable, schedule extracted60 min20 min
WednesdayPowerPointAnnual review deck2 hours50 min
Thursdaydoc-coauthoringProgramme proposalHalf a day2 hours, with a better draft
Fridayinternal-commsTeam update20 min5 min
Fridayskill-creatorBuild the renewal-letter skilln/a90 min, once

The times are illustrative; the pattern is typical. The biggest saving comes from the jobs the firm repeats weekly (letters and comparisons), and the custom skill is what makes those savings hold up, because the house rules stop living in one person's head.

How do you check a skill that isn't Anthropic's before turning it on?

Before turning on a skill that isn't Anthropic's, read every bundled file, because such skills aren't reviewed by Anthropic and their instructions and any scripts run as part of your conversation. Anthropic's advice is to install skills only from trusted sources and, for anything less trusted (including a colleague's), to read every bundled file first, paying attention to code dependencies, bundled scripts and images, and any instruction to connect to untrusted external network sources. Skills can also include, or tell Claude to install, third-party packages.

Here's that check filled in for an illustrative third-party "renewal-tracker" skill the office manager found on a public code site:

CheckWhat she foundVerdict
Who published it, and is there a real person or company behind it?An individual account, no other projectsCaution
Read SKILL.md in fullClear instructions for listing renewals from a spreadsheetFine
Does the description match what the instructions do?YesFine
Any scripts folder? Read each scriptscripts/fetch_rates.pyRead it
Does any script or instruction contact an outside address?The script sends the spreadsheet's contents to a web address to "fetch market rates"Stop: client data would leave the sandbox
Does it install packages?Installs one package not pinned to a versionStop
Hidden or unexplained files?An image file with no stated purposeCaution
Test with fake data before real dataNot reachedNot installed

The decision took ten minutes: don't install. The useful part, the instructions for listing renewals, went into the broker's own skill instead, without the script. If you plan to use third-party skills regularly, the fuller method is in a checklist to vet third-party Claude skills. Partner skills are a middle ground: Anthropic's help centre says its Skills Directory features skills from partners such as Notion, Figma and Atlassian, built to work with their own connectors.

On Enterprise plans there's an extra layer: organisations can turn on skill scanning, which checks each uploaded or edited third-party skill for malicious content before it can run, usually within one to two minutes. Small teams on Team or Pro don't have that, so the manual check above is the scan.

How do you share a Claude skill across a Team account?

A Team account can share a skill with specific people, publish it to the organisation's library, or have an owner provision it for everyone, each with a different level of control. Once the renewal-letter skill works, everyone should use the same version:

  • Share with specific people. From the skill's menu in Customize, then Skills, choose Share. Recipients get a view-only copy, turned off until they enable it; when you update the skill, they get the new version next time they use it. Useful, but it means an edit you make reaches everyone straight away, so test changes first.
  • Publish to the organisation. "Publish to org" submits the skill to your organisation's library. If your organisation requires review, an owner checks it before it's published and decides whether it's available to install, installed by default or required. Updates go through the same review, and users stay on the approved version until then.
  • Provision as an owner. Owners can provision skills for everyone; they appear in each person's list automatically.

Owners can also turn off user-created skills altogether. For a five-person broker, the sensible setting is: anyone can create skills, only the owner publishes to the organisation. Our comparison of ChatGPT, Claude and Gemini skills is worth a look if some of your team prefer another assistant. Skills you enable also follow you into the Claude add-ins for Excel, PowerPoint, Word and Outlook, so the renewal-letter skill works inside Word as well as in chat; see setting up Claude Projects as a shared team assistant for the background knowledge that pairs well with it.

If you want help fitting skills into how your business actually runs, I offer one-to-one AI consultations for small businesses.

How do you tell when a Claude skill isn't earning its place?

Skills are not earning their place when Claude skips them, applies them at the wrong time, or nobody uses them after a month. Check each skill after a month, using what Anthropic's troubleshooting guidance points to:

  • Claude doesn't use it when it should. Confirm it's toggled on, then rewrite the description so it says exactly when to use it. Being explicit in the request ("use my renewal-letter skill") works as a stopgap.
  • It fires when it shouldn't. The description is too broad. "Use for any letter" will grab complaint replies; "Use for renewal letters and renewal invitations" won't.
  • It has grown into three jobs. Anthropic's advice is focused skills, one per workflow; several small skills combine better than one large one.
  • Usage figures are flat. On Team and Enterprise, a skill's page shows adoption and activity figures. A skill nobody uses after a month either needs a better description or isn't needed.
  • Upload failures. The usual causes are a ZIP over the size limit, a folder name that doesn't match the skill name, a missing SKILL.md, or invalid characters in the name or description.

Seven skills, in that order, covers the documents most small firms produce every week. The point isn't the number; it's that each one is either maintained by Anthropic or read line by line by someone in your business before it touched client data.

Claude skills questions owners ask first

Which Claude plans include skills?

Skills are available on Free, Pro, Max, Ultra, Team and Enterprise, as long as code execution is switched on, according to Anthropic's help centre and pricing page. Pro costs $17 a month billed annually or $20 monthly; Team costs $20 per seat billed annually or $25 monthly, with a two-seat minimum. One of Anthropic's developer pages lists only the paid plans, so check your own Customize page.

Can my staff see skills I upload?

No, not until you share them. Anthropic says custom skills you upload aren't visible to colleagues until shared; admins can see their names and sharing status but not their files. On Team and Enterprise plans you can share a skill with specific people, who get a view-only copy, or publish it to the organisation's library, which can require an owner's review.

Are Anthropic's own skills safe to turn on?

They are the lowest-risk place to start, because Anthropic creates, tests and maintains them. Skills shared by colleagues or uploaded from elsewhere aren't reviewed by Anthropic, so read their files first. The risks Anthropic names are prompt injection and data leaks through malicious package code.

Do skills work in Excel and Word too?

Yes. Anthropic says skills you've enabled in Claude's settings are also available in the Claude add-ins for Excel, PowerPoint, Word and Outlook, where you can type / to pick one or describe the task and let Claude apply it. They also load in Claude Code when you sign in with the same account.

Further reads

Sources: Claude help centre articles on what skills are and using skills in Claude; Claude docs on skills, creating custom skills and connectors, skills and plugins; Anthropic's public skills repository and each skill's SKILL.md; claude.com skills and pricing pages (all checked 17 November 2026).

Have a specific process in mind?

Describe the job in your own words on a 1:1 call and I'll show you what AI you can use for it with the tools you already pay for.

Book a 1:1 call with me