> ## 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 skill improvement suggestions

> Check a skill's recent execution evidence, review proposed instruction changes, and save selected suggestions as a new version.

Atlan can review a skill's recent recorded runs and propose changes to its `SKILL.md`. Editors review the proposed changes and choose which ones to save as a new version.

## Before you begin

* Skill improvement must be configured for your account, and you need edit access to the skill.
* The skill needs enough recent invocations across distinct recorded sessions. The app shows the required counts and the version they apply to.
* Trace content must be available for the check. Run counts alone do not provide enough evidence to judge instructions.

The suggestions card is visible only to editors. Availability depends on the service's readiness checks; having a **Usage** tab does not establish that suggestions are ready.

## Run the check

1. Open the skill and select **Usage**, or find **Improve the skill** on **Overview**.
2. Read the run and session counts. If the card says **Suggestions not available yet**, collect the missing evidence on the indicated version.
3. When the card says **Ready for suggestions**, select **Get suggestions**.
4. Follow the progress in **Usage** while Atlan reads, judges, and reviews the recorded work.

A failed check leaves the skill unchanged. If the runs do not contain enough readable content, the card explains why it could not produce suggestions.

## Review and apply changes

1. Expand each suggestion and read its reason and before-and-after diff. A proposal can rewrite the description, replace a section, add a section, or remove one.
2. Check its supporting evidence. The app links traces you created and counts other supporting runs without exposing their trace content.
3. Clear the checkboxes for changes you do not want. All suggestions start selected.
4. Select **Apply suggestions**. The button names the number selected when applicable; at most six can be accepted in one decision.
5. Select **View changes** in the success message, then confirm the new version in **Source**.

Applying suggestions saves a new version. Its security checks and workspace release policy still determine when people receive it. See [Versions and releases](/registry/skills/concepts/versions).

If the skill changed after the check or another editor already decided on the proposal, the service can refuse the apply. Reload the latest state and run another check instead of applying an old diff by hand.

## Read skill quality

When a judged check is available, **Skill quality** summarizes the runs it reviewed:

| Measure | What the check judged |
| - | - |
| **Reached the goal** | Whether the run did what the person asked |
| **Used at the right time** | Whether this skill fit the request |
| **Needed correction** | Whether a later reply corrected, undid, or complained about the result |

The correction rate uses only runs with a later user reply. If none had a reply, there is nothing to count. Select a version to compare the evidence recorded for that version; these judgments describe the reviewed sample, not every execution of the skill.


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