---
title: Save and run a Skill
description: Teach the agent a workflow once, save it as a Skill, then run it by name on any future session.
---

A **Skill** is a reusable ad-ops playbook. This recipe shows the full loop: build a workflow with the agent, save it as a Skill, then run it by name later — from any MCP client, by anyone on your team.

## 1. Save a workflow as a Skill

After the agent does something useful, ask it to capture the steps:

> Save what we just did as a skill called **"weekly meta audit"**: pull the last 14 days of Meta campaigns, flag any campaign whose CPA is above target, and end with a one-line summary. I want to run it every Monday.

The agent calls `skill_create` with a `name`, a `description` (so future agents know when to use it), and a `body` containing the steps. It returns the new Skill's slug, e.g. `weekly-meta-audit`.

## 2. Run it later, by name

In a fresh session — yours or a teammate's:

> Run my **weekly-meta-audit** skill.

The agent calls `skill_get({ skill_name: 'weekly-meta-audit' })`, reads the body as its instructions, and executes the steps against your AdCrunch tools (`list_entities`, `query_insights`, …).

## 3. Improve it in place

> Update the weekly-meta-audit skill to also compare against the prior 14 days.

The agent fetches the current revision with `skill_get`, then calls `skill_update` with `base_revision` so it can't clobber a concurrent edit. The next time anyone runs the skill, they get the improved version automatically.

:::tip[Discoverable by description]

Skills are discovered by their description, not their body — the agent runs `skill_list` and picks the right one. Write the description as "when to use this," and your agent will reach for the right playbook on its own.

:::

:::info[Also editable in the console]

Non-engineers can create and edit the same Skills from the console — see [Skills](/use/console/skills).

:::
