Skip to content
Atomus

Claude Code, Cursor and Copilot

Three pieces make a coding agent reliable with Atomus. The skill gives it the knowledge, the AGENTS.md rules make it follow it in your repo, and the validator catches what slips through.

  1. Install the Atomus skill

    The skill follows the open Agent Skills format, so it works in Claude Code, Cursor, Copilot, Codex and other compatible agents.

    Terminal window
    npx skills add StanVisionAgency/atomus

    Or copy it by hand (Claude Code, all projects):

    Terminal window
    git clone --depth 1 https://github.com/StanVisionAgency/atomus.git /tmp/atomus
    mkdir -p ~/.claude/skills && cp -R /tmp/atomus/skills/atomus ~/.claude/skills/

    For one project, copy it to .claude/skills/atomus in the repo and commit it, so the whole team gets it.

  2. Add the Atomus rules to your repo

    Copy templates/consumer/AGENTS.atomus.md into your product repo. Append it to AGENTS.md, or save it next to it and link it. Each tool reads it from a different place:

    Claude Code reads CLAUDE.md. Point it at the shared file:

    See @AGENTS.md and @AGENTS.atomus.md.

  3. Give the agent the docs

    Point the agent at https://docs.atomus.io/llms.txt. Every page is also available as Markdown at …/index.md, for example https://docs.atomus.io/components/button/index.md. In Cursor, add https://docs.atomus.io/llms-full.txt under Docs and mention it with @Docs.

  4. Validate before you merge

    The skill ships a dependency-free validator. Ask the agent to run it, or add it to CI:

    Terminal window
    node ~/.claude/skills/atomus/scripts/validate.mjs src/
    src/Settings.tsx:12:7   error   invalid-value    <Button hierarchy="danger"> — allowed: "primary" | "secondary" | "outline" | "tertiary" | "link"
    src/Settings.tsx:18:22  error   raw-color        raw colour #18181b — same value as var(--color-text-primary) … (pick by intent)
    src/app.css:4:15        error   primitive-token  primitive --color-gray-200 — same value as var(--color-border-secondary) … (pick by intent)

What to ask for

Section titled “What to ask for”

Agents do best with the page type and the components named. For example:

Build a settings page with Atomus: a Page header, a Profile card with name, email and time zone fields and a Save button, and a Notifications card with toggles. Use only props from the Atomus skill and run the validator when you’re done.

The skill’s page patterns cover the app shell, dashboard, settings, auth, table view and website sections.

What the agent follows

Section titled “What the agent follows”
  • Rules for agents: the core rules and the forbidden list.
  • Choosing a component: the catalogue, other names (dialog → Modal) and decision trees.
  • The React API section at the end of every component page. The skill has the same tables.

Working on Atomus itself

Section titled “Working on Atomus itself”

The Atomus repo has its own AGENTS.md, with the repo map, build commands, which files are generated, and the AI policy: agents may not add tokens, disable lint rules or change thresholds without human review.