---
title: "Agent Skills"
url: "https://adaptocms.com/docs/agent-skills/"
source: adaptocms.com
---

# Agent Skills

Adapto Agent Skills is a Claude Code plugin that lets your coding agent operate Adapto CMS end to end. Install it once, and your agent can check your setup, scaffold a site, design and apply a schema, and research, write, translate, and publish content. Every step runs locally first and waits for your approval before anything reaches the CMS or goes live.

It sits one level above the [CLI](https://github.com/adaptocms/adapto-cms-cli). The CLI gives an agent the raw commands; the skills give it a governed workflow around them, with a project brain that keeps the writing on-brand and a review gate at every point content could ship.

> **Status: early access.** The pipeline runs end to end today. Expect rough edges, and tell us what breaks. It is a Claude Code plugin; other agents use the CLI and REST API directly.

## Requirements

-   **Claude Code.** The skills run as a Claude Code plugin.
-   **An Adapto project.** The free tier is enough to run the whole loop. Grab an API key from your project's developer tools.
-   **The `adapto` CLI.** The skills drive it. The `install` skill sets it up for you, or install it yourself:

```
curl -sSL https://raw.githubusercontent.com/adaptocms/adapto-cms-cli/main/scripts/install.sh | bash
```

## Install

In Claude Code, add the marketplace and install the plugin:

```
/plugin marketplace add adaptocms/adapto-cms-agent-skills
/plugin install adapto@adaptocms
```

To update or remove it later:

```
/plugin update adapto@adaptocms        # update to the latest version
/plugin uninstall adapto@adaptocms     # remove the plugin
```

Once installed, ask your agent to run `adapto:doctor` to confirm the environment is ready.

## The studio workflow

The skills form one pipeline. You run it once to set up, then repeat the content loop each cycle. Three steps stop for your approval.

1.  **Brain.** `adapto:project-define` builds a local knowledge base at `.adapto/project/`: your identity, voice, audience, pillars, and glossary. Every other skill reads it. *(approve)*
2.  **Research.** `adapto:content-research` gathers web, competitor, and keyword findings into a dated dossier. Drop your own Search Console data into `.adapto/sources/` and it folds them in.
3.  **Plan.** `adapto:content-plan` turns the dossier into a slate of per-piece briefs. *(approve)*
4.  **Write.** `adapto:content-create` writes each brief into an on-brand Markdown draft with SEO metadata, drafted and critiqued by subagents.
5.  **Review.** You read the drafts as files, then `adapto:content-upload` pushes the approved ones to Adapto as drafts. *(approve)*
6.  **Publish.** `adapto:publish` takes the reviewed drafts live. *(approve)*

No step touches your live site on its own. Content lands as a draft in your repo and in the CMS; publishing is a separate step you run yourself.

## What's included

Sixteen skills and three subagents, grouped by job:

-   **Setup and health.** `doctor` (read-only environment check), `install` (the CLI), `scaffold` (a Next, Astro, or SvelteKit app plus the `.adapto/` studio).
-   **Project brain.** `project-define` (build it), `project-learn` (fold in what the agent learns; you review the diff).
-   **Schema.** `schema-design` (propose a schema to a plan file), `schema-apply` (create the collections and categories, idempotent).
-   **Content pipeline.** `content-research`, `content-plan`, `content-create`, `content-upload`, plus `content-seed` (a condensed express lane) and `seo-wire` (render meta, OG, and JSON-LD, and generate `llms.txt`).
-   **Localize, publish, UI copy.** `translate` (localization with a structural-parity gate), `publish` (go live or archive back), `microcopy` (seed or extract UI strings).
-   **Subagents.** `adapto-researcher`, `adapto-writer`, and `adapto-editor`, dispatched by the skills above.

## How it stays safe

-   **Plan-then-apply.** Every skill that changes anything prints a plan and waits for your approval before it acts.
-   **Draft-first.** Content is written to `.adapto/drafts/` for you to read as files, then uploaded as drafts. Nothing publishes without a separate, deliberate step.
-   **Anti-slop.** Every draft passes a prose gate that strips AI-tells, then an editor subagent checks voice and completeness before it reaches you.
-   **It never edits your read-client.** The skills write content and their own studio files. They do not touch the SDK or the code that fetches your content.

## Your first run

For a brand-new site:

```
adapto:scaffold          # create the app + the .adapto/ studio
adapto:project-define    # teach it your brand and voice
adapto:content-seed      # research, write, and upload a few starter drafts
adapto:publish           # take the ones you approve live
```

For an existing Adapto project, skip `scaffold` and start with `adapto:doctor`, then `adapto:project-define`.

## Related

-   [SDK Reference](/docs/sdk-reference), the data layer your site uses to read content.
-   [CLI & AI Agents](/docs/cli-ai-agents), the raw `adapto` CLI the skills are built on.
-   [GitHub](https://github.com/adaptocms/adapto-cms-agent-skills), where the skills live and the changelog lands.