Join the open beta
SkillsFree while the beta runs

Write it down once.Agents follow it.

A Skill is a written procedure in your Library: the inputs, the steps and what a good result looks like. Publish it, and the next piece of work starts from it.

You explained it well once.Keep the explanation.

Morrow writes product copy the same way every time: facts from the product sheet, the studio's tone, a short first line. Written down, that is a Skill, and nobody has to retype it into a chat again.

  1. Write

    Describe the method

    One SKILL.md file with the inputs, the steps and the checks a good result must pass. Reference files sit in the same folder.

  2. Publish

    Publish a revision

    A draft stays a draft while you work on it. Publishing saves that exact version, and agents read the published one.

  3. Use

    Name it in the request

    Mention the agent and say which Skill to follow. It reads the published version before it starts the work.

  4. Review

    Check what came out

    Read the result where you asked for it. When the method was wrong, fix the Skill and publish the next revision.

One Skill

The procedure, as written.And as an agent is given it.

A Skill has its own workspace: the conversation that shaped it on the left, the card an agent is handed on the right. Each Run then records which revision it was given, under Run evidence.

A Skill describes how

It never gives an agent access. Every read and every change is still checked against the agent's own permissions.

Each publish is a numbered revision

Publishing shows the files that changed since the last one, and the conversation gets a line: published revision 3.

Publishing is a permission

People and agents who may publish Skills in a Space can do it. Everyone else asks for approval.

Use it inside Autopilot.Or from your own AI.

Skills live in the Library, next to the Knowledge they refer to. The same published procedure reaches your agents and your own chat client.

In a message or a Task

Point an agent at the Skill

Mention the agent, name the Skill and say what you need. The method does not have to fit in the message.

# Task MORR-24 in Shop > @Shop Builder write the Dune vase page following the product-copy Skill read skills/product-copy, revision 3 read Tone of voice
From your chat

The same Skills in Claude

Connect Autopilot as an MCP server and the published Skills you can read appear as prompts in clients that support them. More in Bring your AI.

# in Claude, connected to Morrow > /product-copy Aurora lamp prompt product-copy, from Shop autopilot.knowledge.search "Aurora"

Start with the voice of your brand.

See a voice guide, a facts sheet and a reviewed draft come together in one workflow.

See brand work

Skills, answered.

How is a Skill different from a document in the Library?

A Knowledge page holds facts: the brand guide, the price list, a decision. A Skill holds a method: what to read, which steps to take and what the result must satisfy. A Skill usually points at Knowledge pages and tells the agent how to use them.

What happens when I edit a published Skill?

Your edits are a draft. Agents keep reading the published revision until you publish again, so a half-finished change never reaches real work.

Can a Skill give an agent access to something?

No. A Skill is text. If the method says to read a file the agent cannot read, the agent is refused there, exactly as it would be without the Skill.

Do we have to write every Skill ourselves?

No. You can import a public Skill from a GitHub or skills.sh link. It arrives as a draft with its source files unchanged, and you review it before you publish. If you would rather hand the writing over, QUESTPIE Works turns your processes into Skills with you.

Does a Skill guarantee a good result?

It makes the work repeatable. The result still deserves a reader, which is why it lands in a Task or a thread where someone can ask for changes.

Stop explaining it twice.

Write one procedure this week and let the next piece of work start from it. Free during the open beta.