DESIGN.md, BRAND.md and skills

The MCP tools give an agent your brand on request. The files on this page make sure it asks, and give it the brand where it can't connect: a front page that routes each task, skills that teach the habit, a folder in your repository, and plain-text guides for agents that haven't connected yet.

BRAND.md#

BRAND.md is your brand's own front page, written by your team in Knowledge under Start here. It says what the company is, the rules every task follows, and which items each kind of work needs.

It opens the brief, the response an agent gets when it calls get_brand with no arguments. References in it, written as [[kind:Name]] (such as [[template:Instagram Story]]), reach the agent as the exact call that fetches each item. Until someone edits the starter file, it isn't served, so an agent never reads a skeleton as your brand's word.

See Knowledge for how to write it.

Agent skills#

A skill teaches an agent when to reach for the tools and what counts as done. Archetype publishes four. They hold no brand data, so they are public, and one install works for every brand you connect.

Each one is a SKILL.md file at https://www.heyarchetype.com/skills/<name>/SKILL.md, so it works in any agent that supports the convention. Settings → Agents, MCP tab, step 2 (Add the skills, optional) gives the install commands for Claude Code and Codex, and links to the files for other clients. For Claude Code:

mkdir -p ~/.claude/skills/brand-build && curl -sL https://www.heyarchetype.com/skills/brand-build/SKILL.md -o ~/.claude/skills/brand-build/SKILL.md

For Codex, use ~/.agents/skills/ instead. Connect the MCP server first: the skills call its tools, so without it they have nothing to do.

The brand as files in a repository#

An agent working in a repository reads its files whether or not it calls a tool. The archetype/ folder puts the brand there:

The CLI writes and refreshes it with npx archetype-os pull, and points your root AGENTS.md at archetype/AGENTS.md. Without the CLI, download the folder as a zip: in Settings → Agents, open the Repository tab, choose Do it by hand instead, and use the download under No CLI? Unzip this at your repo root. (It shows when Tokens is picked and Automatic updates is off.)

Scripts can fetch the same files over HTTP with an API key from Settings → Agents → API, sent as Authorization: Bearer <key>:

llms.txt and llms-full.txt#

Two public plain-text files describe Archetype to agents that haven't connected yet:

• https://www.heyarchetype.com/llms.txt is the index: what Archetype serves, every page of these docs as markdown, the skills, and the MCP address.
• https://www.heyarchetype.com/llms-full.txt is the full integration guide in one fetch: setup for every client, every tool with its arguments, the get_brand domains in full, how to write with update_brand, the HTTP endpoints, the CLI, and the CI check.

Both are generated from the same source as the server, so the tool reference in them always matches what the server answers. Every docs page is also served as markdown at its address plus .md, such as /docs/agents/tools.md.

DESIGN.md#

DESIGN.md is an open format from Google Labs, still marked alpha: one file at the root of a repository, with design tokens in YAML at the top and prose under them on how to use them. Agents such as Claude Code, Cursor and Codex read it when AGENTS.md or CLAUDE.md points at it.

Archetype's DESIGN.md generator turns any website into one, free and with no account. It reads one page and its stylesheets, calls no AI model, and only writes values the site declares.

A single file holds values, not parts: your real components, logos, templates and voice aren't in it, and nothing checks what the agent built. For a brand you keep in Archetype, the archetype/ folder and the MCP server cover that ground and stay current when the brand changes.

Next steps#