Turn a Claude skill into a $49 digital product

There are two markets for this, and confusing them is why most people's first skill sells nothing.

On PromptBase, I counted 52 priced listings on its Claude Skills page. The band ran $3.99 to $19.99, median $7.99, and the most common price was $9.99. Nothing on that page came near $49.

On Gumroad, searching the same category returns a different world: listings priced at $39, $47 and $49, and a long tail running well into the hundreds. Same file format. Roughly eight times the price.

The difference is not the file. It is what is wrapped around the file, and who the buyer thought they were buying from. This guide is how to build the version that belongs in the second market.

One thing up front about the money, because it matters. Every price in this guide is a public third-party listing or a platform's own published fee page, read on 12 September 2026 and linked so you can check it. None of them is a claim about what you will earn. Most listings on any marketplace sell very little, and the first one you publish is a lesson, not an income.

What you'll have when you're done

• A skill that installs and works on the surface your buyer actually uses
• A pack around it: licence, install sheet, worked example, reference file
• A price chosen against real listings rather than a feeling
• The net on a $49 sale, per marketplace, from each platform's own fee page
• A listing that names the job, not the technology
• [SCREENSHOT: the finished pack folder next to the live listing page]

Before you start

• A Claude plan you already use. Skills upload on claude.ai works on Pro, Max, Team and Enterprise with code execution enabled. Claude Code reads them off the filesystem.
• A job you have done more than five times. This is the whole prerequisite. A skill is a codified repetition. If you have not repeated it, you do not know what goes wrong.
• A payment platform account. Gumroad, Lemon Squeezy, Payhip and Whop all have no monthly fee at the entry tier. Pick after the pricing section, not before.
• A weekend. Two hours is the skill. The rest is the pack, and the pack is the product.

What a Claude skill actually is

Anthropic's own definition: Agent Skills are modular capabilities that extend Claude's functionality, and each Skill packages instructions, metadata and optional resources such as scripts and templates, which Claude uses automatically when relevant.

Concretely, it is a folder with a SKILL.md at the root:

invoice-reconciler/
  SKILL.md          main instructions
  REFERENCE.md      the fiddly details, loaded only when needed
  EXAMPLES.md       worked input and output
  scripts/
    match_rows.py   a utility the skill can run

The frontmatter rules are published and they are strict. Get these wrong and an upload fails:

That last row is the one that decides whether your product works for the person who bought it. Claude Code accepts extra fields that claude.ai and the Skills API do not, and Anthropic's docs are blunt about the result: including disallowed fields in uploads causes hard errors. So a skill written in Claude Code and shipped as-is can fail on the buyer's first attempt, on a surface you never tested.

The size model is worth understanding because it changes how you write:

Anthropic states there is no practical limit on bundled content because files do not consume context until accessed. Read that as a packaging instruction: the SKILL.md is the index, and everything long goes in a file next to it. A 20,000-word SKILL.md is not a bigger product, it is a broken one.

What the category actually charges

Two markets, two price bands. Here is what is publicly listed.

Two Gumroad listings I can quote at their stated USD price because they rendered natively in dollars: The Claude Code Playbook for Non-Developers at $49, and The 2026 AI Business Blueprint at $47.

The lesson is not "list on Gumroad". It is that price is set by the buyer's expectation on arrival, and the two marketplaces train opposite expectations. A PromptBase buyer is browsing prompts. A Gumroad buyer arrived from someone's newsletter or video expecting a thing that solves a problem. You can charge $49 to the second one, and the reason has almost nothing to do with the file.

What makes a skill worth paying for

Three tests. A skill that fails any one of them belongs in the $9 market, and no amount of pricing confidence changes that.

1. It does a job with a name. "Marketing prompts" is not a job. "Reconciles a Stripe payout CSV against an invoice list and flags the mismatches" is a job. If your listing cannot finish the sentence "this does ___ so you don't have to", the product is not finished.

2. It encodes something the buyer does not know. The first version of anything is a checklist. The version worth money contains the four things that went wrong the first four times, written as rules. That is what a buyer cannot generate themselves by asking Claude nicely, because they have not made the mistakes yet.

3. It works on the first try, on their machine. Most refunds on digital products are not "this was bad", they are "this did not run". See the surfaces section below, which is where this one is decided.

Step 1: Write the SKILL.md so it triggers

The description field is doing two jobs and most people only write for one. It is your listing copy, and it is the only text Claude sees when deciding whether this skill is relevant. Write it as a trigger.

---
name: stripe-payout-reconciler
description: Reconciles a Stripe payout CSV against an invoice list, matching on
  amount and date within a tolerance, and produces a mismatch report. Use when the
  user mentions payout reconciliation, unmatched invoices, Stripe settlement files,
  or asks why a payout total does not equal the sum of its invoices.
license: See LICENSE.md
---

Note the second sentence. Naming the phrases a user would actually type is what makes the skill fire without being asked for by name, and a skill that has to be invoked manually is a document with extra steps.

Step 2: Build the pack around it

The file is not the product. This is:

The install sheet matters more than it sounds, because Anthropic is explicit that custom Skills do not sync across surfaces. A skill uploaded to claude.ai has to be uploaded separately to the API, API skills are not available on claude.ai, and Claude Code skills are filesystem-based and separate from both.

That API row is a product decision, not a footnote. If your skill fetches anything, it does not work on the API, and your listing has to say so before someone pays.

Step 3: Write the licence

You are selling a file that copies perfectly. The licence is the only thing that says what happens next, and it does not need a lawyer to be useful.

Every marketplace already enforces roughly the same four-part shape. PromptBase's terms are the clearest public statement of it, and they are worth reading as a template:

• A non-exclusive, worldwide, perpetual licence to use the item for any purpose
• No use for harmful, illegal or malicious activities
• No direct resale, redistribution or transfer without the creator's written consent
• The creator keeps the intellectual property; buying does not transfer ownership

Write those four in plain English, put them in LICENSE.md, and put a one-line summary in the listing. Two additions worth making explicit, because buyers ask:

• Commercial use. Say yes or no in one sentence. Most buyers of a working skill want to use it at their job, and silence here costs you sales.
• Team use. Per person or per company. Ambiguity here is how a $49 product gets installed by fourteen people.

One thing no marketplace gives you: none of these platforms supplies a licence agreement template. Gumroad offers licence keys, which is DRM and not a licence. PromptBase applies its own marketplace terms to every sale automatically. Everyone else expects you to bring your own text. Write it once and reuse it.

Step 4: Pick where to list, on the published fees

Every one of these publishes its rate. No monthly fee at the entry tier on any of them.

Read the two Gumroad rows together, because that gap is the real decision. Gumroad's own wording: 10% plus $0.50 per transaction for sales through your profile or direct links to your customers, and 30% per transaction when new customers find and buy through the Discover marketplace. You are not choosing a platform, you are choosing whether you are paying for traffic. Nine dollars and thirty cents a sale, in Gumroad's case.

Payhip publishes paid tiers that buy the fee down: $29 a month drops it to 2%, $99 a month removes it. The arithmetic is simple and worth doing once. Against the free tier's 5%, the $29 plan only pays for itself past roughly $970 a month in sales. Do not buy the volume plan before the volume exists.

Step 5: The listing

A listing for a working tool is not a feature list. Five blocks, in this order:

1. The job, in one line. What it does and for whom. No tool names in the headline; the buyer has a problem, not a stack.
2. The before and after. The input they have now, the output they get. If you can show a real one, show it. This block sells the product.
3. What is in the zip. The six files from Step 2, named. This is the visible difference between a file and a product, and it is what justifies the price band.
4. Which surfaces it works on. Claude Code, claude.ai, the API, with the network-access caveat stated. Honest here is cheaper than a refund later.
5. The licence, in one sentence. Commercial use yes or no, resale no.

The naming trap

This one catches a surprising number of live listings, and it is two separate rules stacked.

The technical rule: a skill's name field cannot contain the reserved words anthropic or claude. That is a validation constraint, and it will simply reject your folder.

The brand rule is broader. Anthropic's guidance says you can accurately state, in plain text, that your product has Claude Code preinstalled or that it runs Claude Code, but you cannot use the Claude Code or Anthropic names or logos as part of your own product, feature or company name, in your own logo, or in a way that suggests Anthropic built, endorses or is partnered with your product.

So:

The descriptive sentence is allowed and it is also better copy. Naming the job beats borrowing a brand.

What I would do differently at the start

Build the install sheet before the skill. Not after. Writing down how a stranger installs this on three surfaces forces the scope decision, and the scope decision is the product decision.

Test on the surface you do not use. Whatever you built it in, ship the first test on another. Almost every one-star review of a digital tool is an install problem wearing a quality complaint.

Price against the market you are selling into, not the file. The same folder is a $9 item where buyers browse prompts and a $49 item where buyers arrive from someone they trust. If you are bringing your own audience, start at the higher band and pay 10% instead of 30%.

Ship version 1.1 within a month. A dated changelog is the cheapest trust signal in this category, and it is the thing the $9 listings never have.