# Skills

What a skill is, how it differs from a prompt and a tool, how the library is organised, and what to do with it.

A skill is a set of reusable, versioned instructions that an agent loads for a specific task: a ticket triage routine, a comparables table, a one-page research brief. Write it once, attach it to as many agents as you like, and the same procedure behaves the same way everywhere.

**Who can use it:** the **Builder** role or higher, in a tenant licensed for the agent builder. Anyone in the workspace can read the library; creating, editing and deleting a skill are Builder rights.

Video: Skills (2:54), https://katonicdocsmedia.blob.core.windows.net/docs/skills/7fd98ecd471d/index.html

## Skills in four steps

The video above walks the library, a skill's page, the editor and the agent wizard; the pages under **What to do with it** go deeper on each part.

  
**Step 1. Read the library**

**Import** and **New skill** sit in the header, then the filters: source (**All**, **Yours**, **Platform**), vertical, role and a sort. **Your skills** are editable; **Platform built-ins** are read-only and can be forked. Each card carries the name, version, the description agents read to decide when to load it, tags and the last update. The card's menu opens the skill, attaches it to an agent, edits it or archives it.

  
**Step 2. Open a skill**

The header holds **Attach to agent**, **Export**, **Fork**, **Archive** and **Edit**; the trigger sits under **When agents use this**, then the instructions as written. **Version history** lists each saved version and **Used by** groups the agents on it as live, then drafts, with the version each is on. A built-in offers **Fork** in place of Edit and no Archive. **Import** takes a folder of skill files at once and says what it did with each.

  
**Step 3. Write one**

**New skill** opens the form. The first field is the trigger: describe the situation, not the task. **Name** is lowercase letters, digits and hyphens, fixed once the skill exists; **Version** is yours to set; **Vertical**, **Role** and **Tags** feed the filters. **Instructions** is the body, in Markdown, with a preview beside it; end with an output format. **Create skill** saves it under Your skills.

  
**Step 4. Attach it to an agent**

The wizard's **Skills** tab lists the library with a switch per row. The version pill beside an attached skill sets its policy: pinned, follow latest minor or follow latest, resolved when the agent is published. Expanding a row shows the instructions the agent will get. Publishing freezes that text into the version; an edit reaches the agent only when it is published again.

## What a skill is, against what it is not

| | What it is | Where it lives |
| --- | --- | --- |
| **A skill** | A procedure an agent follows, versioned and shared across agents | Skills |
| **A prompt** | The instructions that define one agent's behaviour | [Prompt Studio](/agents/prompt-studio) |
| **A tool** | Something the agent can call to act or fetch | [Tools](/tools/tool-catalog) |

A skill changes how the agent does something. A tool changes what it can reach. An agent's prompt is its own; a skill is borrowed by many.

## The library

![The Skills page: the description line with Import and New skill, filters for source, vertical, role and sort, the Your skills group with one card per skill showing its name, version, description, tags and last update, and the Platform built-ins group.](https://docs.katonic.ai/img/agents/skills/skills-library.png)
*Skills are versioned instructions an agent loads for a task. Your skills are editable; platform built-ins are read-only and can be forked.*

**Skills**, under **Develop** in Studio, is the library, with **Import** and **New skill** in the header. Filter by source (**All**, **Yours**, **Platform**), by vertical and by role, and sort by **Recently updated** or **Name A to Z**.

Two groups follow. **Your skills** are the ones your organisation wrote or imported, and are editable. **Platform built-ins** ship with the platform, are read-only, and can be forked.

Each card carries the skill's name, its version, its source, the description agents read to decide when to load it, its tags and when it was last updated. That description is the most important line on the card, because it is the one an agent reads.

## A skill's page

Selecting a skill opens it: **Export**, **Fork**, **Archive** and **Edit** in the header, then **When agents use this** with the trigger, the tags, the author and the last update, and the instructions themselves, rendered as the Markdown document their author wrote, tables and all. Below them, **Version history**, where each version carries a note saying what changed and who saved it; **Used by**, grouped as live agents then drafts, each with the version it is on and a note where it is behind the current one; and **Details**.

A platform built-in shows **Fork to edit** in place of Edit, and no Archive. That is enforced by the platform, not just hidden here.

## What to do with it

  - [Write a skill](https://docs.katonic.ai/agents/skills-author): Create one, version it, fork a built-in, export it.
  - [Attach a skill to an agent](https://docs.katonic.ai/agents/skills-attach): How a skill reaches an agent, and when an edit to it takes effect.
  - [Import skills in bulk](https://docs.katonic.ai/agents/skills-import): Bring in a folder of skill files from elsewhere.

## Limits and behaviour

- Platform built-ins cannot be edited or archived. Forking makes a copy you own, with no link back.
- A published agent keeps the skill text it was built with. Editing a skill reaches an agent the next time that agent is published, and not before.
- Exporting produces a Markdown file with front matter, which is what the importer reads, so a skill moves between tenants as a file.
- The card, the skill's own page and the agent builder all count the same way now, and label it, so a skill used by three agents reads **2 live · 1 draft** everywhere. Where only the published side can be counted it reads **Used by N agents**, and a skill nothing uses reads **Not attached yet**, which is a state rather than a count.

## Related

- [Write a skill](/agents/skills-author)
- [Attach a skill to an agent](/agents/skills-attach)
- [Prompt Studio](/agents/prompt-studio)

Source: https://docs.katonic.ai/agents/skills
