The hours went in. The revenue never did.

Your skills live on GitHub. Your margin lives somewhere else.

If you shipped a reusable skill in the last year — a Claude agent, a prompt pack, a dataset, an orchestration kit — your work is already being consumed. The part that pays for it isn't you.

Agents don't pay because there's nothing to pay to.

Open source taught the last decade how to share code. It didn't teach the next one how to price an agent-consumable skill. So the default is still: publish, get scraped, watch it compound into someone else's product.

A model can't negotiate with a README. It needs terms it can parse, request, and obey. Until that layer exists, every skill you ship is subsidy.

A decade on the licensing problem. Then the consumer changed.

I spent most of the last decade on licensing and IP — co-building license.rocks, and then working alongside enterprises and creators whose output was the IP. Music and art especially: world-class craft, almost no licensing scaffolding around it. For a long time, the job was building that scaffolding for humans to read.

Then the consumer changed.

A model doesn't read a PDF. It doesn't care what's in your footer. An agent deciding whether to use your skill runs at thousands of requests per minute. Either the terms are in a format it can parse, or the terms might as well not exist. Every document built for human readers goes invisible the moment an agent is in the loop.

Look at the licensing surfaces that exist today — all built for human readers. A LICENSE.md in the repo: models ignore it. A clickwrap on a distribution page: agents don't click. A platform-specific agreement: solves one surface, leaves every other exposed.

The durable fix had to be a neutral protocol. Not a platform. Not a storefront. A file format and a status lifecycle any skill can ship with, any agent can parse, and any chain can verify.

That's ProofMeta Protocol.

A manifest. A request. A grant.

ProofMeta wraps a skill in a machine-readable license. Agents discover it, request access, get a decision, and act inside the terms you set. The wrapper is the surface. The skill stays yours.

# proofmeta.manifest.yaml
item:     prompt-chain-kit
license:
  type:     commercial-use
  price:    "€0.02 / call"
  duration: 30d
  chain:    any
status:   OPEN

Every state is machine-readable. Every transition is auditable. Every revoke is enforceable.

Four parts. All open. Shipping in the open — what's ready and what isn't.

The protocol is a specification, not a product. Everything below is covered under Apache 2.0, versioned in public, and shippable without asking. Status badges mark what's written down, what's in progress, and what's planned — no vaporware.

Honest disclosure: pre-launch.

ProofMeta Protocol is pre-launch. There are no customer quotes yet — and we won't invent them. The placeholders below will fill as builders wrap their first skills. If you want to be one of them, the spec is below.

The honest objections, answered honestly.

One person. A decade on the licensing problem.

I'm Daud. I've been in software since the late 1990s and spent most of the last decade on licensing and IP — co-building license.rocks, then working alongside enterprises and, more often, creators whose output was the IP but who had no licensing layer to hold it. Music and art especially: the craft was world-class, the rights scaffolding around it was almost always missing.

That's the pattern ProofMeta is built on. Agents are the next consumer, and they need the same thing those creators needed — a wrapper the surface can actually read. Not a leap into a new category. The compounding of a decade of work into the era where models, not lawyers, are the parsers.

ProofMeta is a one-person protocol project today. Apache 2.0, written in public, open to anyone who wants to build on it.