> For the complete documentation index, see [llms.txt](https://help.usejimo.com/docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.usejimo.com/docs/experiences/agent/skills/create-a-skill.md).

# Create a skill

{% hint style="warning" %}
**Beta** · This feature is currently in beta. Behavior and interface may still evolve.
{% endhint %}

A Skill runs on a page. So creating one starts by choosing **which page**, then teaching the Agent the procedure **on that page**, in your own product.

<figure><img src="/files/VCM2dhY02iXxwA5PUekF" alt=""><figcaption></figcaption></figure>

### Step 1: Pick the type

**Add Skill**, from the [Skills](/docs/experiences/agent/skills.md) page or from a scanned page's panel, offers the three [types](/docs/experiences/agent/skills.md#the-three-types-of-skills): Execute a flow, Guide user in a flow, Answer complex questions.

<figure><img src="/files/9SgIKqrOjIgA6iaYXXIs" alt=""><figcaption></figcaption></figure>

### Step 2: Pick the page

**Where should we create the skill?** lists your [scanned pages](/docs/experiences/agent/knowledge/interface/scan-a-page.md), each with its element and Skill counts. Search by name, or pick **Scan a new page** if the page has never been scanned.

<figure><img src="/files/7b2OsZ8RLTYrfeNgFGQR" alt=""><figcaption></figcaption></figure>

{% hint style="info" %}
The page you pick is where the Skill will run and which elements it can use. If the flow spans several pages, start from the one where the user begins.
{% endhint %}

Jimo then asks where to open it: pick the URL or environment, and **Open Page** launches your product with the Skill creation panel alongside.

<figure><img src="/files/7EL1vJoMy68IadtqIXB4" alt=""><figcaption></figcaption></figure>

### Step 3: Describe it, or show it

Two ways to teach the procedure.

<figure><img src="/files/V23m568ZSlg9Nv25ZdgM" alt=""><figcaption></figcaption></figure>

| Method          | How it works                                                                                                              | Best when                                                                    |
| --------------- | ------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------- |
| **Describe it** | Write what the Skill should do. The Agent asks follow-up questions, then drafts the steps from what it knows of the page. | The flow is simple, or the page is already well scanned.                     |
| **Show me**     | Do the task once on your product. The Agent captures and scans what you did.                                              | The flow has many steps or conditions, and is easier to do than to describe. |

<details>

<summary><strong>For an Explain Skill: pick the elements</strong></summary>

An **Answer complex questions** Skill starts from a different prompt (*What questions should your agent answer?*) and adds one step: after the clarifying questions, the Agent asks you to **Select Elements**, the parts of the page its answer should read as its source of truth.

<figure><img src="/files/ALjY59mEeifDAZicGRGl" alt=""><figcaption></figcaption></figure>

A floating toolbar opens on your page: click elements to select them, or **Select all relevant elements**, then **Done Selecting**.

<figure><img src="/files/y3GBhnwGWj6UIZjyc8ir" alt=""><figcaption></figcaption></figure>

The Agent **reviews your selection**. If you picked navigation aids where it needs data (a Settings link instead of the KPI cards), it says so and asks again. In the saved instructions, each selected element appears as a reference like `{{kpi.total-revenue-kpi}}`: the answer is built on them at runtime.

<figure><img src="/files/6ZQIq5Gz1vjCBSz8bPfG" alt=""><figcaption></figcaption></figure>

</details>

#### Describe it, in practice

The panel opens with a prompt matching your type, for example *What flow should your agent guide users through?* Write it in one sentence, in the words your users would use.

<figure><img src="/files/tR1xP7DLbUR0npt98ylt" alt=""><figcaption></figcaption></figure>

The Agent then **asks you what it needs to know**. It reads the page, states what it understood, and where the request is ambiguous it asks a question with ready-made options, plus **Something else...** to write your own. **Skip** lets it decide alone.

<figure><img src="/files/RARUro4x45YCdCM4yBAB" alt=""><figcaption></figcaption></figure>

Expect two or three rounds: scope first (which sections of the page), then depth (what each step should say). Each answer narrows the draft.

<mark style="color:purple;">\<IMG 11-08 · CAPTURÉE : captures/11-create-a-skill/11-08\_conversation-two-answers.png · LÉGENDE : Each answer narrows the draft></mark>

<figure><img src="/files/NRg4KhJQvxWHZLdIqb8r" alt=""><figcaption></figcaption></figure>

{% hint style="success" %}
**Answer the questions rather than skipping them.** The Agent uses your answers to fix scope. Skipping leaves it to guess, and a Skill built on the wrong scope is harder to fix than to redo.
{% endhint %}

### Step 4: Review the draft

The Agent proposes a card with everything it wrote:

* **Name** and type label
* **Description**: what the Skill covers and *when the Agent should use it*, phrased around the questions users ask
* **Instructions**: the Skill's goal, the **inputs needed from the user**, and the numbered **steps**

<figure><img src="/files/YQu6zIKfGP7iv97Mn9Wv" alt=""><figcaption></figcaption></figure>

Read the steps before saving: they are what the Agent will actually do. **Save skill** creates it.

### Step 5: Review, activate, test

The saved Skill opens in its own panel, with two tabs: **Settings**, everything below, and **Usage**, its statistics once it starts running. The menu next to the Active toggle holds **Delete**.

<figure><img src="/files/IULazhOCirgZHInBc3di" alt=""><figcaption></figcaption></figure>

| Element                        | What it does                                                                                                                                                                                                            |
| ------------------------------ | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Active** toggle              | *End users can trigger it.* Off keeps the Skill in draft, invisible to users.                                                                                                                                           |
| **Mode**                       | Switch between **Execute** and **Guidance** after creation: same structure, the only difference is whether the flow progresses automatically. An **Explain** Skill is a different construction and cannot be converted. |
| **Description for the agent**  | Editable. This is what decides *when* the Skill fires.                                                                                                                                                                  |
| **Instructions for the agent** | Editable. The goal, required inputs, and numbered steps.                                                                                                                                                                |
| **Try this skill**             | Runs it in a test chat, on the real page.                                                                                                                                                                               |

**Try this skill** opens a Test Chat with a suggested question. The Skill runs step by step: you see each step announced, and when it needs the user to act, the Agent waits with a **Waiting for your action... Skip** bar while pointing at the element on screen.

<figure><img src="/files/yRhV6fP4vY5K8m7Xpj6u" alt=""><figcaption></figcaption></figure>

The first step of any run is **Read page to get refs**: the Agent is inserting it itself, to line up the elements it is about to work with before doing anything. You do not write it, and it appears in every run.

An **Explain** Skill has no steps to watch: it answers straight away, in the structure its instructions describe, from the elements you selected.

<figure><img src="/files/vDcV3LCY01DHDGhntXRw" alt=""><figcaption></figcaption></figure>

{% hint style="warning" %}
**Test before activating.** A Skill that points at the wrong element or skips a step is worse than no Skill: the user follows it and gets stuck. Run it once end to end, on a real page, before turning Active on.
{% endhint %}

### Related pages

* [Skills](/docs/experiences/agent/skills.md): the three types, and what they replace
* [Interface](/docs/experiences/agent/knowledge/interface.md): scan a page before teaching a Skill on it
* [Page details](/docs/experiences/agent/knowledge/interface/page-details.md): the elements a Skill can use
* [Knowledge](/docs/experiences/agent/knowledge.md): the five kinds of Agent knowledge