Get started
Guides

Working as a team

Different people use different AI assistants. Here is how that stays tidy.

Everyone gets the rules, whatever they use

One file, AGENTS.md, carries your project’s rules. Codex, Cursor, OpenCode, Gemini and Grok all read it automatically. Claude Code reads it through a one-line CLAUDE.md.

So the baseline covers everybody with no per-person setup.

If someone wants the skill catalog inside their assistant rather than through init, they can install the plugin for Claude Code, Cursor, Codex, Grok, Gemini, OpenCode, or .agents. That is the procedures, not the ESLint guarantee — both can be installed.

Adding what a particular assistant needs

If a teammate uses something that needs its own small config:

npx @wiluszdamian/cue add cursor
npx @wiluszdamian/cue add claude-code

To see what is set up right now:

npx @wiluszdamian/cue list

To take one out again:

npx @wiluszdamian/cue remove cursor

remove uses a record of exactly what was installed, so it takes out what it put in and nothing else. Anything you edited by hand is kept and reported rather than deleted.

Nobody is forced

doctor notices an assistant somebody is using and mentions it:

[ warn ] Cursor detected but not configured
      Evidence: .cursor/ in this project.
      fix: cue add cursor

A warning, not an error. Detection is a suggestion, never a decision — a folder left over from an experiment is not consent to write files.

What ends up in git

Commit these. They are shared knowledge:

  • AGENTS.md and any assistant config
  • .agent-kb/ — what is known about your app
  • eslint.config.mjs, playwright.config.ts
  • .cue/install.json — the record of what was installed

Ignore these. Handled for you:

  • .auth/ — a live signed-in session
  • test-results/, playwright-report/
  • .env and other real environment files

Reviewing changes to the app notes

Treat .agent-kb/ like code. When a survey lands in a pull request, the diff shows exactly which elements appeared or vanished — which is often the clearest summary of a UI change anybody will get.

In CI

init writes two workflows.

Your tests, sharded and merged into one report.

Your guardrails:

- name: Constitution
  run: npx eslint . # the rules

- name: Environment
  run: npx @wiluszdamian/cue doctor --ci # is everything wired up

- name: Drift
  run: npx @wiluszdamian/cue sync --check # have the rules moved

Only genuine errors fail the build. Warnings never do — a team that has to suppress warnings to ship stops reading them, and then stops reading the errors too.