# Turn a skill or spec into an agent

> Turns a Claude skill, a prompt file, a spec or an MCP server you already have into a Mindset agent and the resources it needs.

**Skill for your coding agent:** [`mindset-build-agent/SKILL.md`](https://docs.mindset.ai/skills/mindset-build-agent/SKILL.md)

Hand your coding agent what you already have, in full. It builds the agent and its resources, and keeps them as files in your repo as [Keep your agents in your code](https://docs.mindset.ai/docs/keep-your-agents-in-your-code) describes.

## Ask your coding agent

```
Turn the invoice exceptions skill below, which our finance team wrote, into a Mindset agent kept under agents/. Reuse any connection that already reaches our finance system, and list what won't come across before you build anything.
```

Skill: [mindset-build-agent](https://docs.mindset.ai/skills/mindset-build-agent/SKILL.md)

## What gets built

| What you have                                          | What it becomes                                                                                                                                                              |
| ------------------------------------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| A Claude skill, a prompt file, a spec or a description | Prompt text on an [agent](https://docs.mindset.ai/docs/agents), plus the connections, functions and widgets it needs. A [script](https://docs.mindset.ai/docs/scripts) only for the part that must wait or happen in order |
| An MCP server                                          | A [connection](https://docs.mindset.ai/docs/connections-and-operations), with each of its tools as an operation                                                                                     |

Each prompt field holds 2,000 characters, so a long skill is condensed and your coding agent says what it shortened. Some things don't come across at all: a skill's bundled code or data files, steps that read files on someone's computer, and any system your org has no connection to.

## What you decide

- What to do about each part that won't come across, before anything is built.
- Whether to reuse an existing connection. Your coding agent judges a connection by what it reaches, not by its name.

## Done when

- The agent and its resources are files in your repo, applied to Sandbox.
- You've seen the list of what didn't come across, and agreed it.

## In the Studio

Admins can hand a skill or spec to **Orca**, in the workspace rail. It proposes a **Provisioning plan** with a row for each resource, marked **Build it** or **Use existing**, and builds nothing until you press **Approve plan**. Orca matches existing resources by name, so check each **Build it** row against what your **Connections** list already reaches. [Agents](https://docs.mindset.ai/docs/agents) explains Orca and the builders.

![Orca's starting cards: turn a Claude skill into an agent, provision an agent from a spec, or give an agent the resources it needs.](https://docs.mindset.ai/images/docs/studio/orca-start.png)