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

# Skills

> Guide your Agent through recurring expert tasks with reusable, task-specific skills.

> [!WARNING]
>
> **Skills are currently in closed beta.** The feature is still being rolled out and may change. If you would like to try Skills for your Agents, please reach out to your Genow contact.

Skills let you capture how a specific, recurring task should be solved and hand that "recipe" to your Agent. Instead of relying on a generic search-and-answer for every question, the Agent can recognise a known task, load the matching skill, and follow its defined plan - using the files the skill provides.

***

## What Is a Skill?

A skill is a reusable, task-specific instruction set that an Agent Admin creates in Administration and assigns to one or more Agents. Each skill bundles:

* **A name** - a clear identifier for the task.
* **A description** - explains *when* the skill should be used. The Agent relies on this to decide whether a request matches the skill.
* **A body** - the plan, or step-by-step instructions, for how to solve the task (the **Instructions** field in the form).
* **References and assets** (optional) - files the skill reads or works with, for example a price list or an email template.

Skills follow the open [Agent Skills specification](https://agentskills.io/specification) (the `skill.md` format). You can fill in the fields directly in Administration or upload an existing `skill.md` file to pre-fill them.

> [!NOTE]
>
> Think of a skill as an "expert" for one job: one skill might look up delivery times, another might run a competitor comparison, and a third might assign the right sales region to a postal code.

***

## How Agents Use Skills

In plain terms: every skill has a short **name** and **description** that say what task it is for. When you ask the Agent something, it reads these descriptions and checks whether your request matches one of its skills.

* If a skill matches, the Agent "opens" it, reads its step-by-step instructions and attached files, and follows them to complete the task.
* If nothing matches, the Agent simply answers the normal way.

You never have to pick a skill yourself - the Agent decides automatically which one (if any) fits your request. This works like an orchestration layer: it first identifies which "expert" (skill) should handle the request, then follows that skill's plan instead of giving a generic answer.

***

## How Skills Fit into the Pipeline

Normally, Genow answers in two steps: it searches your connected knowledge (semantic search) and then generates a grounded answer. Skills add a task-specific layer on top of this:

1. When your request matches a skill, the Agent loads the skill's plan and any attached files.
2. It then follows that plan - for example, looking up a value in the skill's Excel file and returning it in a defined format - rather than producing a free-form answer.

This makes recurring expert tasks far more **reliable and repeatable**, because the Agent follows a fixed procedure and works only with the data the skill provides.

> [!NOTE]
>
> In the current beta, skills follow their written plan and work with their attached files. Running custom scripts or code is not yet supported.

***

## Skills vs. Agent Instructions

Skills complement the [Agent Context and Instructions](https://docs.genow.ai/en/use-case-management/settings/add-context.md):

| | Applies to | Purpose |
| :- | :- | :- |
| **Instructions and context fields** | *Every* question on the Agent | Always-on, general steering and context |
| **Skills** | Only when a request matches | On-demand recipes for specific, recurring tasks |

Use **Instructions** for rules that should always hold; use **Skills** for self-contained procedures that should only kick in for a particular kind of task.

> [!NOTE]
>
> Here, "Instructions" means the Agent-wide **Instructions** field in [Agent Context](https://docs.genow.ai/en/use-case-management/settings/add-context.md) - not the **Instructions** field inside a single skill, which holds that one skill's plan.

***

## Creating a Skill

You create and manage skills under <kbd>Administration</kbd> → <kbd>Skills</kbd> → <kbd>Create</kbd>.

The easiest workflow is to **write a `SKILL.md` file first and upload it** - Genow then fills in all the fields below automatically. You only review the result and, most importantly, assign the skill to an Agent.

**1. Upload your SKILL.md file (recommended)**

In the **Upload SKILL.md** area, select, drop, or paste your file (file type: MD, max 5 MB). Genow reads it and pre-fills every field below. You can also skip the upload and fill in the fields by hand.

**2. Check the fields**

Review the pre-filled fields and adjust them if needed (see the table below).

**3. Assign the skill to an Agent**

This is the key step. Under **Agents**, select the Agent (or Agents) that should use the skill - an Agent whose users actually need this task. A skill only takes effect on the Agents it is assigned to; everyone with access to that Agent can then use it.

### The Fields

| Field | Required | What it is for |
| :- | :- | :- |
| **Name** | Yes | A clear identifier for the task. |
| **Description** | Yes | Explains *when* the skill should be used. The Agent reads this to decide whether a request matches - so describe the task and the situations it covers. |
| **Agents** | Recommended | The Agent(s) the skill is assigned to. Without an assignment, no one can use the skill. |
| **License** | No | Optional licence information for the skill. |
| **Compatibility** | No | Optional note on what the skill is compatible with. |
| **Metadata** | No | Optional extra entries (add one row at a time). |
| **Instructions** | Core | The heart of the skill: the step-by-step plan telling the Agent how to carry out the task. |
| **References** | No | Files the Agent can read once it has loaded the skill. |
| **Assets** | No | Templates and other files the skill works with. |

> [!NOTE]
>
> **References vs. Assets.** References are documents the Agent *reads* for information - for example a product catalogue, a price list, or a guideline. Assets are files the skill *works with* or fills in - for example an email template, a quote or proposal template, or an Excel calculation sheet. Both accept file uploads of up to 5 MB each.

> [!NOTE]
>
> Agent Admins can manage the skills they created or that have been shared with them; Global Admins can manage all skills. Genow also provides a set of preset skills.

***

## Example Skills (Sales Context)

Skills are especially useful in sales, where the same expert tasks come up again and again:

* **Find the right product:** the customer describes their requirements, and the skill matches them to the best-fitting product from your catalogue.
* **Configure to the requirement:** the skill assembles the right configuration or options for the customer's needs, based on your product rules.
* **Draft a customer email:** the skill turns an answer into a ready-to-send email in your company's wording and format.
* **Look up a lead time:** the skill returns the delivery time for a product from a lead-time table.
* **Competitor comparison:** the skill compares your product with a competitor's on the parameters that matter.

### A Worked Example: "Product Finder"

| Element | Example |
| :- | :- |
| **Name** | Product Finder |
| **Description** | Use when a customer describes their requirements and needs a product recommendation from our catalogue. |
| **Instructions** | 1. Identify the key requirements; ask a clarifying question if something essential is missing. 2. Match the requirements against the catalogue. 3. Return the best one to three options, each with a short reason. 4. Follow the structure of the recommendation template. |
| **References** | The product catalogue and current price list - so the Agent reads accurate product data. |
| **Assets** | A recommendation template - so every answer follows the same structure. |

This shows a pattern you can reuse for other tasks: the **Instructions** define *how*, the **References** provide the *facts*, and the **Assets** define the *form* of the output.


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