JM

Justin McKelvey

Fractional CTO · 15 years, 50+ products shipped

AI for Business 10 min read

Claude Cowork Skills Explained: 15 Built-In + Building Custom

Claude Cowork ships with 15 reusable skills — structured prompts that act like slash commands inside Claude (e.g., /cs-draft, /business-pulse, /lead-triage). You can also build your own custom skills to encode any repeatable task. Skills are the building blocks of every Cowork workflow; without them, you're back to one-off prompting. As of June 2026, Anthropic ships skills inside the Claude for Small Business toggle on every Team plan and up.

If you've spent any time inside Claude Cowork, you've probably seen the word "skill" thrown around and assumed it meant the same thing as a workflow, an integration, or a custom GPT. It doesn't. And the conflation is the single biggest reason people set Cowork up, get mediocre output, and decide AI isn't ready for their business.

This post is the clean explainer. What a skill actually is, the 15 Anthropic ships with Cowork, which 5 to turn on first, and how to build a custom one without over-engineering it into a junk drawer.

Skill vs. workflow vs. connector — the three things people conflate

Before going any further, these three terms need to be pinned down. They are not interchangeable, and Cowork's documentation does a worse job than it should clarifying which is which.

  • A connector is a tool integration. Google Workspace, HubSpot, QuickBooks, Slack — the pipes Cowork uses to read and write data. There are 8 native ones. A connector by itself doesn't do anything; it just exposes data and actions Cowork can use.
  • A skill is a reusable, structured prompt that Cowork treats like a callable command. /cs-draft is a skill. /business-pulse is a skill. Skills produce a specific kind of output (a customer-service reply, a weekly summary, a triaged lead). They run inside Cowork and can use any connected tool, but they're the unit of "ask the AI to do this specific job."
  • A workflow is a sequence — multiple skills, connector actions, and approval gates strung together. Example: "Every Monday at 8am, pull last week's revenue from QuickBooks, top 5 leads from HubSpot, and post a summary to #leadership Slack." That's a workflow. It uses 3 connectors and 2 or 3 skills.

The hierarchy: connectors are pipes, skills are commands, workflows are sequences. A workflow without skills is brittle one-off prompts. Skills without connectors are isolated text generators. You need all three for Cowork to actually save you time.

The 15 built-in skills that ship with Claude Cowork

Anthropic ships 15 first-party skills inside Cowork. They cover the operational jobs every small business does and never quite gets around to. Each has a slash command you can fire from any Cowork-enabled surface — Claude.ai, the desktop app, or Slack via the Claude bot.

Slash command What it does Connectors used
/business-pulse Weekly business health snapshot QuickBooks, HubSpot, Google/MS
/monday-brief Summary of the week ahead Google/MS Calendar, HubSpot
/cash-flow-snapshot Current cash position + 30-day forecast QuickBooks, PayPal
/invoice-chase Drafts follow-ups for overdue invoices QuickBooks, Google/MS Gmail
/lead-triage Scores and routes new leads HubSpot
/cs-draft Drafts a customer service reply Google/MS Gmail
/meeting-prep Generates prep notes for a meeting Calendar, HubSpot
/call-notes Turns raw call notes into branded follow-up Google/MS Docs, Gmail
/close-month Month-end close prep with flags QuickBooks, PayPal
/proposal-draft Generates a first-pass proposal HubSpot, Google/MS Docs
/contract-review Flags risks in inbound contracts DocuSign, Google/MS Docs
/slack-digest Summarizes channel activity Slack
/canva-onbrand Generates on-brand image from prompt Canva
/payment-recon Reconciles payments across PayPal + QB PayPal, QuickBooks
/pipeline-health Stalled-deal report with recommendations HubSpot

Anthropic refreshes this list roughly every quarter, so one or two slash commands may have been renamed by the time you read this. Open Settings → Skills to pull the current list — the categories above will still be accurate.

The 5 universal skills every business should turn on first

Don't turn on all 15. Pick the 5 that almost every business benefits from immediately, get them running well for a week, and then layer in the rest. The high-leverage starting set:

  1. /business-pulse — schedule it weekly. You'll catch revenue dips, stalled pipeline, and ops issues a week earlier than you would otherwise. The single best argument for paying for Cowork at all.
  2. /cs-draft — even if you have a CS team, having every reply drafted in your voice cuts response time by half and removes the worst part of inbox work.
  3. /meeting-prep — runs the night before any meeting on your calendar. You walk in knowing the contact's history, last touchpoint, and open deal. Borderline unfair.
  4. /invoice-chase — the AR work everyone hates and nobody does on time. Cowork drafts polite follow-ups; you click send.
  5. /lead-triage — scores new leads as they land and tells you which 3 to call today. Replaces a $40K/year SDR job for most solopreneurs.

These five cover sales, finance, ops, and customer service. Get them dialed in for 2 weeks before adding anything else.

Industry-specific skill suggestions

Once the universal 5 are humming, the next layer depends on what kind of business you run.

B2B services (consultants, agencies, fractional execs). Add /proposal-draft, /call-notes, and /pipeline-health. The proposal-and-follow-up loop is where B2B services either close deals or lose them, and these three skills automate the parts everyone procrastinates on.

DTC / e-commerce. Add /cs-draft (more volume than B2B), /canva-onbrand, and /payment-recon. Customer-service volume and creative iteration are the time sinks; reconciliation is the back-office sink. Note: if you're on Shopify or WooCommerce, you'll need a Zapier bridge to feed order data into Cowork — that's a real limitation worth knowing.

Professional services (law, accounting, real estate). Add /contract-review, /close-month, and /meeting-prep. The compliance, financial close, and client-intake cycles are where time leaks. /contract-review is a 10x time saver — but always read the AI's flags yourself. The skill highlights risks; it doesn't replace your judgment.

How to build a custom skill (step-by-step)

The 15 built-ins cover the common ground. But every business has at least 2 or 3 jobs that are uniquely yours — and that's where custom skills earn their keep. Here's how to build one without over-engineering it.

I'll walk through a real example: /discovery-recap — a skill that takes raw notes from a discovery call and turns them into a branded follow-up email with a proposed next step.

Step 1: Pick a job you do at least weekly. Custom skills are worth building when you do something 4+ times a month. Below that, the time to author the skill exceeds the time it saves. The discovery-call follow-up qualifies — I do 6 to 10 a month.

Step 2: Write the skill as if you're training a new hire. Open Settings → Skills → New Skill. Cowork prompts you for a name, a slash command, and the instructions. Most people screw the instructions up by being either too vague ("write a follow-up email") or too rigid ("use exactly this template every time"). The sweet spot is structured but flexible.

Here's roughly what I wrote for /discovery-recap:

"You're drafting a follow-up email after a discovery call. Input: raw notes from the call (bullets, partial sentences, whatever the user pastes). Output: a 4-paragraph email in my voice that (1) thanks them for the time, (2) reflects back the 2 specific pain points they raised, (3) proposes a concrete next step based on the call, (4) closes with a soft CTA. Use the voice and offers in my Business Brain. If the next step isn't clear from the notes, default to: 'I'll send a short scoping doc by end of week.' Never invent details that weren't in the notes."

Step 3: Add 2-3 examples. Cowork lets you attach example inputs and outputs to a skill. This is where the magic happens — examples train tone better than any instruction. Paste 2-3 real past follow-ups with the original raw notes alongside them. Don't skip this. A skill without examples is just a verbose prompt.

Step 4: Add an escape valve. Every custom skill should have a line like: "If the input doesn't contain enough information to do this job well, say so and ask for the missing piece. Don't fabricate." This single sentence prevents the most common custom-skill failure mode — hallucinated specifics that look right and read wrong.

Step 5: Test on 3 real inputs. Don't ship a skill on day one. Run it against 3 real past examples, compare the output to what you would have written, and iterate the instructions or examples until the gap closes. This usually takes 20 minutes. Worth every second.

Step 6: Save, test the slash command, and document it. Hit save, then trigger /discovery-recap from a real conversation. If the output is good, document the skill in a shared doc so your team (or future-you) knows it exists and what it expects as input.

That's it. From idea to working custom skill in under an hour. The skill now compounds — every time you run it, you save the 10 to 15 minutes it would've taken to draft from scratch.

Skills + Business Brain — why skills need Brain context to sound like you

Skills tell Cowork how to do a job. The Business Brain tells Cowork who is doing it. Without the Brain loaded, even a perfectly written skill produces generic AI output — competent, useful, and instantly recognizable as not-you.

The Brain is where you load brand voice (with real writing samples), your offers and pricing, your ICP, and the way your sales motion actually runs. Every skill — built-in or custom — references the Brain when generating output. Run /cs-draft with no Brain, get a generic customer service reply. Run the same skill with a fully loaded Brain, get a reply that sounds like you wrote it at your desk.

If you're going to invest the time to set up Cowork, the order is: connectors first, Brain second, skills third. Skip the Brain and the skills are a polished version of generic.

Common custom-skill mistakes

I've helped clients build a few hundred custom skills at this point. The same three mistakes show up over and over.

  • Over-engineering. The skill becomes a 600-word specification with nested conditionals and a 12-step process. Cowork chokes on it and the output gets worse, not better. Rule of thumb: if the instructions are longer than the typical output, simplify.
  • No examples. People skip the examples step because it feels like extra work. It's not extra — it's the most important step. A 200-word instruction with 3 examples beats a 600-word instruction with none, every time.
  • No escape valve. Without an explicit "ask if you don't know" instruction, the AI will invent. Always include the escape valve. It's the difference between a skill that produces drafts you trust and one that produces drafts you have to fact-check line by line.

FAQ

Can I edit Anthropic's built-in skills?
You can't edit the underlying skill, but you can clone it. In Settings → Skills, hit "Duplicate" on any built-in, rename it (e.g., /cs-draft-mine), and modify the clone. The original stays intact. Most teams end up running cloned versions of the built-ins because the defaults are generic by design.
Can I share skills with my team?
Yes. Skills authored at the workspace level are available to every Cowork user in your Team or Enterprise plan. You can also keep a skill private to your own account. The toggle is in the skill's settings — "Workspace" vs "Personal."
Can skills run on a schedule?
Skills don't run on schedules directly — workflows do. To run a skill on a schedule, wrap it in a workflow (Settings → Workflows → New) and set the trigger to a time or cron expression. That workflow can then call the skill weekly, daily, or on any cadence.
Do skills count as separate Cowork actions for billing?
Each skill invocation is an action, and Anthropic counts Cowork actions against your plan's monthly limit. As of June 2026, Team plans include enough actions that a typical small business won't hit the cap. If you do, you'll see a warning at 80% usage and an option to upgrade.
What's the limit on custom skills?
There's a soft cap of 50 custom skills per workspace on the Team plan and 200 on Enterprise. Most businesses use 8 to 15 — beyond that, the skill list becomes a junk drawer and people stop using them. If you're approaching the cap, audit and consolidate, don't ask for more.
Can I version control my skills?
Cowork keeps an internal version history per skill — you can roll back to any prior version inside Settings. Full external version control (Git-style) isn't supported yet. For now, the workaround is exporting skill definitions as JSON (Settings → Export) and storing them in your own repo. Anthropic has hinted at native Git integration on the roadmap.

The bottom line

Skills are the unit of leverage inside Claude Cowork. The 15 built-ins cover most of the operational jobs every small business does — start with the universal 5, layer in industry-specifics, and build custom skills for the work that's uniquely yours. Skip the Business Brain and the output stays generic. Build skills without examples and the output stays generic. Get both right and Cowork starts saving you 5-10 hours a week.

If you want a ready-to-go library of skill recipes built for your business type, my Claude for Small Business install includes a starter set of 8 custom skills tuned to your Business Brain — done in 2 weeks instead of 2 months of trial and error. Details at /claude-for-small-business.

Frequently Asked Questions

Can I edit Anthropic's built-in skills?
You can't edit the underlying skill, but you can clone it. In Settings → Skills, hit 'Duplicate' on any built-in, rename it (e.g., /cs-draft-mine), and modify the clone. The original stays intact. Most teams end up running cloned versions of the built-ins because the defaults are generic by design.
Can I share skills with my team?
Yes. Skills authored at the workspace level are available to every Cowork user in your Team or Enterprise plan. You can also keep a skill private to your own account. The toggle is in the skill's settings — 'Workspace' vs 'Personal.'
Can skills run on a schedule?
Skills don't run on schedules directly — workflows do. To run a skill on a schedule, wrap it in a workflow (Settings → Workflows → New) and set the trigger to a time or cron expression. That workflow can then call the skill weekly, daily, or on any cadence.
Do skills count as separate Cowork actions for billing?
Each skill invocation is an action, and Anthropic counts Cowork actions against your plan's monthly limit. As of June 2026, Team plans include enough actions that a typical small business won't hit the cap. If you do, you'll see a warning at 80% usage and an option to upgrade.
What's the limit on custom skills?
There's a soft cap of 50 custom skills per workspace on the Team plan and 200 on Enterprise. Most businesses use 8 to 15 — beyond that, the skill list becomes a junk drawer and people stop using them. If you're approaching the cap, audit and consolidate, don't ask for more.
Can I version control my skills?
Cowork keeps an internal version history per skill — you can roll back to any prior version inside Settings. Full external version control (Git-style) isn't supported yet. For now, the workaround is exporting skill definitions as JSON (Settings → Export) and storing them in your own repo. Anthropic has hinted at native Git integration on the roadmap.

If this was useful, here are two ways I can help: