> ## Documentation Index
> Fetch the complete documentation index at: https://platform.atlan.com/llms.txt
> Use this file to discover all available pages before exploring further.

> ## Agent Instructions
> To act on Atlan objects, use the Atlan MCP server at https://api.atlan.com/mcp or the SDK. To install or publish skills from a terminal, use the `aireg` CLI; `aireg --map json` prints its command map. The `atlanai` CLI is deprecated. Run a read-only identity check before any write.
> The docs MCP server at /mcp searches these docs only. It cannot read or change Atlan objects.
> SDK packages: Python `atlanai` (PyPI) and TypeScript `@atlanai/sdk` (npm). Show Python first, then TypeScript.
> These docs cover Atlan AI Registry, Atlan's hosted product. The AI registry specification and aireg CLI are maintained separately. /open-source lists publicly reachable signed installers and explains that the source repository and hosted CLI reference currently require GitHub access. Do not claim the repository or its reference docs are publicly accessible. Use these docs for Atlan-specific setup such as the adapter, sign-in and workspaces.

# Review agent-drafted definitions

> Let an agent propose ontology definitions, then review the preview in Studio and approve saving them as drafts yourself.

<Badge color="orange">Preview</Badge>

An agent can draft new ontology definitions for you, but only a person can save them. The agent prepares an import document and checks it; you review the preview in Studio and approve it. Nothing is published until you publish it yourself.

```mermaid theme={null}
flowchart LR
  A["Agent drafts a<br/>native JSON document"] --> P["Agent checks it with<br/>get_import_preview"]
  P --> Y["You preview it in<br/>Studio Import & export"]
  Y -->|approve| S["Save approved drafts"]
  Y -->|don't approve| N["Nothing is saved"]
  S --> D["New drafts<br/>no release"]
```

## What agents can and cannot do

| An agent can | An agent cannot |
| - | - |
| Draft a native JSON, RDF, OWL or SKOS document. | Save drafts, apply an import, or publish a release through the Ontology MCP tools. All six are read-only. |
| Call `extension__ontology__get_import_preview` to see what the document would create and whether it validates. | Approve its own proposal. Studio's final save checks a fresh, direct human sign-in and refuses agents, delegated users and machine clients. |
| Hand you the document to review. | Edit or overwrite existing definitions through this flow. A save only creates new definitions. |

Binding mappings and knowledge records an agent suggests stay as text for you to enter separately. This flow creates logical definitions only.

## Before you begin

* You can create in the ontology's workspace, and you sign in to Studio as yourself.
* The agent has given you its document, for example a native JSON file that adds a `Return` object type and a `returned_in` link to the retail model.

<Steps>
  <Step title="Open the import preview">
    In the ontology, open **Definitions** and choose **Import file**. **Importing into** shows the destination ontology; check it is the right one.

    Drop the file, choose **Choose file**, or paste the document. Pick the **Format**: **Studio JSON**, **RDF · Turtle**, **OWL** or **SKOS**. For RDF, OWL and SKOS, also set the **Base IRI**. Then choose **Preview import**.
  </Step>

  <Step title="Review the proposal">
    Studio shows **Preview only.** and **Would create** with the number of definitions. Nothing has been saved yet.

    * Open each proposed definition and read its properties and links.
    * Read the **Import report**. **nothing lost** means the document mapped fully; otherwise it counts approximations, omissions and unsupported parts.
    * If Studio shows **Import validation errors**, the proposal can't be saved. Ask the agent to fix the named definitions, then preview the new document.
  </Step>

  <Step title="Approve or reject">
    To **approve**, select **I approve creating these definitions as drafts in the selected ontology. Nothing will be published.**, then choose **Save approved drafts**.

    To **reject**, leave the box clear and close the page, or choose **Choose another**. Nothing is written.

    The approval covers this exact preview. If the document, the format options or the destination ontology changes, Studio needs a fresh preview and a fresh approval.
  </Step>

  <Step title="Publish when ready">
    The new definitions are ordinary drafts. Edit them like any other, then validate and publish on the **Publish** page when you are ready.
  </Step>
</Steps>

## Verify

* Studio shows **Drafts saved.** and **Saved *N* definitions as drafts atomically. Review and publish separately.** Choose **View definitions** to see them.
* **Definitions** lists the new definitions as drafts, and the toolbar shows them as unpublished.
* The live release is unchanged until you publish.

If Studio reports **Save outcome unknown.**, check **Definitions** before trying again: a failed response doesn't prove the save failed, and the same preview can't be submitted twice.

## Limits

* One save creates at most 100 definitions, all or none. A larger valid proposal shows **This proposal cannot be applied synchronously. Reduce its size and preview it again.**
* A document can be at most 8 MiB.
* Approval needs a direct human sign-in. Without one, the save fails with "direct human draft approval required"; see [Draft save returns 403](/cookbooks/troubleshooting/ontology#draft-save-returns-403).

## Related

* [Preview ontology import](/tools/mcp/ontology/references/ontology-get-import-preview): the tool an agent uses to check its document.
* [Publish an ontology through the API](/cookbooks/how-tos/ontology): import, validate and publish from scripts with your own credentials.
* [Ontology concepts](/ontology/concepts): drafts, releases and the definition kinds.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.