# Use skills in chat (/docs/ai/skills/use-in-chat)









Use a skill to apply a review checklist, writing style, or another repeatable method in chat. You keep the selected model or Agent.

## Before you begin [#before-you-begin]

You need the following access:

* a signed-in Portal session in the correct organization
* permission to read skills and use chat
* an available chat model or Agent

Skills work with a model or Agent in ordinary text chat. They are not available in **Deep Research**, live **Voicebot**, or the OpenAI Responses and Anthropic-compatible API paths.

If **Skills** is unavailable, ask your tenant administrator to check your access.

## Inspect a skill [#inspect-a-skill]

1. In **Chat & Work**, open **Skills**.
2. To find a skill shared by another member, select **Discover**.
3. Search by name or description.
4. Open the skill.
5. Read its description, activation rule, and warnings in **Overview**.
6. Select **Contents** to inspect `SKILL.md` and the supporting files.

<img alt="Skill contents with SKILL.md, a supporting reference file, and Preview and Source controls." src="__img0" />

For Markdown files, **Preview** shows formatted text. **Source** shows the file text. Images in skill previews appear as placeholders rather than loading image URLs.

## Add a skill to Yours [#add-a-skill-to-yours]

1. On the skill page, select **Add to Yours**.
2. Return to **Skills**.
3. Check that the skill appears under **Yours**.

**Discover** keeps the row and marks the skill **Added**. Adding a skill selects the original. It does not create a copy.

To narrow **Yours**, use **Filter** to choose **Uploaded by me** or **From your organization**. To change the order, choose **Recently updated** or **Name**.

## Mention a skill in chat [#mention-a-skill-in-chat]

1. Open a conversation or select **New chat**.
2. In the composer, type `/`.
3. Pick a skill from the suggestions.
4. Write the task.
5. Include the input the skill needs.
6. Send the message.
7. Check that chat shows **Loaded** with the skill name and revision.

<img alt="Chat composer with the slash picker showing a skill from Yours." src="__img1" />

The picker inserts `/skill:<name>`. For example, a skill named `release-summary` uses this reference:

```text
/skill:release-summary Summarize these changes: added a CSV export, fixed a duplicate notification.
```

You can also type the reference directly or open &#x2A;*+** → **Skills**. The picker lists skills in **Yours**, not every shared skill.

You can mention more than one skill in a message. All explicit references must resolve before generation begins; repeating the same skill does not load it twice.

If two selected skills have the same name, use the picker to identify the intended skill. It binds the selection to that skill's identity. Copied or manually typed reference text does not preserve that binding and can be ambiguous; choose again from the picker. Typing `/name` alone is not the canonical `/skill:name` reference.

References inside code blocks, inline code, or after a backslash remain literal text. They do not load a skill.

<img alt="Conversation with a loaded skill, a supporting resource read, and an active skill chip above the composer." src="__img2" />

## Check automatic activation [#check-automatic-activation]

1. Open the skill's **Overview**.
2. Check the activation rule under **How to use**.
3. If the assistant can load the skill, send a relevant task without a reference.
4. Check chat for **Loaded** before relying on the skill's instructions.

Selection does not force activation. The assistant chooses whether to load a relevant skill. **Loads only when you mention it** means an explicit reference is required.

If chat says a skill was not offered to the assistant, mention that skill with `/`.

## Inspect the revision used by a conversation [#inspect-the-revision-used-by-a-conversation]

1. Expand the **Loaded** row in chat.
2. Select **Open skill**.
3. Check the revision number on the skill page.
4. To inspect a newer revision, select **View current**.

A conversation keeps the revision from its first activation. Later edits do not replace that revision in the existing conversation. A new conversation uses the currently published revision when the skill activates; selecting a shared skill is not a permanent freeze of its package. Inspect updates when their instructions matter to your workflow.

## Stop using a skill [#stop-using-a-skill]

1. Above the composer, find the active skill's chip.
2. Select its &#x2A;*×** control.
3. Check that the skill no longer appears among the active skills.

The chat history keeps the instructions and resources it already received. **Stop using** excludes the skill's instructions from later model requests. It also prevents further resource reads and automatic activation in that conversation.

To resume the skill, mention it again. **Use again** on the stopped skill's chat row inserts the reference, which you then send. The conversation reuses its pinned revision. To use a newer revision, start a new conversation.

## Remove a skill from Yours [#remove-a-skill-from-yours]

1. Open the skill from **Yours**.
2. Open the **In Yours** menu.
3. Select **Remove from Yours**.

The skill leaves your chat picker. Removal does not delete the package or retract content from existing conversations.

## Resolve a refused reference [#resolve-a-refused-reference]

If chat cannot resolve a reference, no answer starts and the message remains in the composer.

* If the skill is not selected, add it to **Yours**.
* If the name is ambiguous, choose the intended skill from the picker.
* If the skill is no longer shared or is deleted, remove the reference or choose another skill.

Then send the message again.

If the instructions do not fit the model's context, use a shorter skill or a model with a larger context.

## Related pages [#related-pages]

* [Upload and manage skills](/docs/ai/skills/manage-skills)
* [Skill package reference](/docs/reference/skill-packages)
* [Chat & Work](/docs/use/portal/chat-and-work)
* [Use an MCP Server in chat](/docs/ai/mcp/use-in-chat)