June 17, 2026 · 2 min read

Human-built vs auto-generated agent skills

Toby BanksFounder, Skillzy

The dispatch

New listings, creator interviews, the occasional discount. Every other Friday.

Anthropic, OpenAI, Google, and the rest are all shipping skill libraries. Most of what fills them is model-generated — plausible, generic, and never run against a real business on a bad day.

The gap auto-generation cannot close

A generated skill knows what a quote looks like. It does not know that this trade always gets burned on variations, or that this clinic must never confirm a booking without a deposit. That knowledge comes from doing the job, getting it wrong, and fixing it.

What human-built actually buys you

  • It has failed already — somewhere other than your business
  • It stops at the lines a person decided were dangerous
  • Someone can tell you why it works, not just that it does

Why this is the whole point

Every skill here was built by someone who needed it, ran it, and sells it because it held up. That is not a feature we added. It is the only reason to choose a marketplace over a generated list inside a model.

Featured listing

Patterns for prompts that hold up.

Prompt patterns that survive a model upgrade.

Agent prompt patterns · $24
Featured listing

Your first SKILL.md, the right way.

The short version of "how to publish a skill people will actually buy."

Your first SKILL.md, the right way · $9

Liked this? Get the next one.

Field Notes, every other Friday. No spam.