Quickstart: Cursor

This page takes you from an empty folder to a component published to Staging, with Cursor doing the writing. It assumes the CLI quickstart step 1: you have taggor.mjs in the project folder, with its hash checked. Creating the account and approving the sign-in are yours; the agent never creates an account, accepts terms or types a secret.

1. Install the agent’s instructions

node taggor.mjs init --agent cursor
node taggor.mjs agent install --agent cursor   # or: add it to a project that has taggor.json

This writes the rules to .cursor/rules/taggor.mdc, with alwaysApply: true so every chat in the project gets them. The text is the same block the other agents get. Use Cursor’s agent mode, which can run terminal commands: the quickstart needs it to run check and publish.

2. Sign in

With a browser, you run it yourself, or the agent asks you to:

node taggor.mjs login
node taggor.mjs whoami

login prints a code and a URL; open the URL, check that the code matches, approve. Where there is no browser (a remote machine, a container, a CI job), use an API key instead: in the portal, open Settings, then API keys, create a key with the scopes the work needs, and set it as TAGGOR_API_KEY in the environment that starts the agent. When the variable is set, it wins over any login and whoami shows what the key can do. The key is shown once, to you: never paste it into the chat, into a file of the project or into a prompt, and delete it when the work is done.

3. Ask for a component

Open the project folder in Cursor, start a chat in agent mode and paste the prompt below. Cursor asks before it runs a command; read each one, and approve it only if it is one of the commands in the steps below.

Create a quote request form for my test site and publish it to staging.
Then give me the preview link and the script tag to paste on my site.

4. What you should see

Whatever the agent, the work goes in the same order. Each step says what you should see, so you can tell when the agent is off the path.

1. It checks who it is signed in as. It runs node taggor.mjs whoami --json. If it gets exit code 3 (not signed in), it stops and asks you to run node taggor.mjs login and approve the sign-in in the browser. It never asks you for a password or a token.
2. It learns your site before it writes. It reads the colors, the font and the tone of your project or page, so the component looks like it belongs there.
3. It writes one component in components/<tag>/src/index.tsx: a single defineTaggorComponent call (the contract). For a form that collects personal data it asks you for the privacy policy URL instead of inventing one.
4. It checks, and repeats until the check passes. It runs node taggor.mjs check --json. Exit code 0 means the gates passed. Exit code 6 means one failed: the JSON has error.errors[], and each entry has a code, a hint, a file and a line. The agent fixes that file and runs check again. Every code has a page, build diagnostics, with what it means and what to change. The agent never skips a check and never edits the check’s rules.
5. It publishes to Staging. It runs node taggor.mjs publish --json, without --channel. The answer holds the build, the version, a preview_url, the sites the component is attached to (sites) and the ones it is not (sites_not_attached, each with the command that attaches it: a production site that only came from taggor.json is never attached by a publish). A build that fails comes back as exit code 6 with the same error.errors[]: back to step 4.
6. It hands over to you. It gives you the preview link, and node taggor.mjs sites snippet gives the script tag. Pasting the snippet on your site is yours: see Install the snippet and the platform guides.

What the agent does not do unless you asked for exactly that in the conversation: move a component to Production (promote), pass --yes, roll back, detach a component, delete anything, or choose a site or a channel for you. If you see it reach for one of those, stop it.

5. When something goes wrong

• Exit code 3: not signed in. Run login (or set TAGGOR_API_KEY) and ask the agent to continue.
• Exit code 6, build or gate: the agent reads error.errors[] and fixes the file. If it fails the same code three times, open the code’s page under build diagnostics and tell the agent what it says.
• Any other failure: the message ends with a code and a link to its page in API and CLI errors. Give the agent the request_id if the message has one.
• The agent wants to skip a check, use --yes or paste a token into a file: say no. Those are never the fix; the rules the installer wrote tell it so.

Last real run of this prompt

Not run yet with Cursor. The date, the Cursor version and the result of the first real run on staging go here when it is done. Until then, this page is written from the installed rules and the CLI’s behavior, not from a recorded session.