Agentic

By Sander Janssen

Make your site safely editable by an AI content agent. Guardrails, validation, and a reviewed publish flow: bulk content work that can’t break the site.

Let your client edit their Statamic site by talking to an AI agent, without letting it break anything.

Agentic is a Statamic starter kit. Your client describes changes in plain language to Claude Code, the agent edits the content, and a preview URL updates. When they say "publish", you get one pull request to approve.

It asks one thing of your site: pages are built with a replicator page builder, the block-based setup many Statamic sites already use. That is what keeps the agent safe. It can only compose pages from the blocks you defined and fill in their fields, so it never writes markup or invents a layout, and every edit can be checked against your blueprints. Everything else is up to you: which blocks you make, how they look, Antlers or Blade, your CSS and front-end tooling.

What that looks like

Client: "Here's an overview doc for our six services. Set up a page for each one, in our tone."

The agent looks up which content blocks the site actually has, builds the six pages out of them, writes the copy in the client's own voice (it keeps a notes file on their tone and preferences), checks its work, and pushes to the preview site. The client reviews all six at once, says "looks good, publish it", and one pull request lands on your desk.

For a one-line tweak, the Control Panel is already fine and Agentic isn't trying to replace it. Where it earns its keep is volume and structure: redoing a whole page, drafting a batch of posts, a site-wide tone pass, "here are three new team members, add them to the team page in our usual style".

Why it's safe to hand over

A man in a 90s office, on the phone, looking worried

Your client, ten minutes after you gave them repo access and their agent broke the website. Agentic exists so this call never happens.

  • The agent can only touch content. Code, config, templates and CI are blocked, both on the editor's machine and on GitHub's side. It's default-deny, not a polite request.
  • Broken content can't go live. Every change is validated against your blueprints, using the same rules a Control Panel save enforces, before it's committed and again on every push.
  • The agent knows your site. A catalogue of every available block and field is generated from your own fieldsets, so it can't invent a block or field that doesn't exist.
  • You own production. The client's agent only pushes to the preview branch. Nothing reaches the live site except through a pull request you merge.

Getting started

statamic new my-site sanderjn/statamic-agentic

…or in an existing Statamic project:

php please starter-kit:install sanderjn/statamic-agentic

Then follow SETUP.md in your new site: one setup command, a repo with two branches, your host, and one email to hand over to the editor.

  • Statamic v6, PHP 8.3+, single locale (multilingual is on the roadmap).
  • Deployment is your choice: Ploi, Forge, Vapor, SSG. Agentic sets up the repo, branches and CI, not the host.
  • Your editor needs a GitHub account and their own paid Claude Code account. Flag that cost up front.

For developers

The kit has exactly one opinion: pages are built from blocks, the sets of a single replicator fieldset (resources/fieldsets/page_builder.yaml). Add a set, add a matching resources/views/blocks/<handle>.antlers.html (or .blade.php), regenerate the catalogue. The agent's reference, the validation, the Control Panel and the front-end rendering all derive from that one file: no switch statement, no second registry. The kit ships one example block and a minimal layout for you to replace.

Antlers or Blade. The agent only edits content, so the kit doesn't care how you render it. The shipped views are Antlers; if you work in Blade, replace the five small views in resources/views/ and loop the page builder like this:

@foreach ($page_builder as $set)
    @include('blocks.' . $set->type)
@endforeach

Each block partial then reads its fields from $set (e.g. {!! $set->body !!}).

The edit guard also applies to your own Claude Code sessions. Set AGENTIC_DEVELOPER=1 (via env in an untracked .claude/settings.local.json) to lift it while you build.

How it works → What the agent reads, how the guardrails and the validator work, and how changes go live.

License

MIT: use it, strip it down, rebuild it however you like. If you improve it, or learn something worth sharing from running it for a client, open an issue or PR.