Skip to Content
Getting StartedFor AI-First Builders

Getting Started for AI-First Builders

Dot•requirements gives your AI assistant structured context about what you’re building, so it writes better code and tests that actually validate the right things.

What You’ll Get

  • AI gets it right the first time — clear requirements mean fewer rewrites and less debugging imprecise assumptions
  • Day-two features have a foundation — your AI assistant can read what came before
  • Tests stay honest — linking tests to requirements means you can trust they’re validating the behaviour you asked for
  • Build-first is fine — prototype first, capture requirements after; you get the same strong foundation either way

Set Up

One command:

npx @popoverai/dotrequirements@latest init

This will:

  • Create or connect to a project
  • Offer to set up your AI assistant (via dotreq ai-setup), installing the requirements-driven workflow guidance into its context file (CLAUDE.md / AGENTS.md)

That’s the whole integration: your assistant uses the dotreq CLI directly — there’s no server to install, configure, or restart. The workflow guidance tells it which commands to run and when.

A cloud account is optional, but it enables shared specs, hosted review, and coverage sync — the features that make AI-assisted development actually work well.

Supported assistants:

  • Claude Code
  • Cursor
  • OpenAI Codex
  • Google Antigravity
  • GitHub Copilot

Optional: To get the dotreq shorthand for future use, install globally:

npm install -g @popoverai/dotrequirements

The Workflow

Here’s how a typical feature development looks:

You describe what you want to build

“Add a login page with email and password. Users should see an error if credentials are wrong, and get redirected to the dashboard on success.”

AI captures requirements

Your assistant will ask: “Should we capture what this should do before we build it?”

If you say yes, it:

  1. Primes itself with your project’s format and style conventions (dotreq create-requirement-document)
  2. Writes a requirements document with clear, testable statements
  3. Has the draft style-reviewed: a fresh-eyes subagent judges it against your project’s style guide (dotreq style-check) and reports what to fix
  4. Validates the syntax (dotreq validate) and shows you the requirements for review

You review and approve

Check that the requirements capture your intent. Adjust if needed. The AI syncs approved requirements to the cloud with dotreq sync — your team sees them immediately.

AI implements and tests

Your assistant writes the code, then writes tests that reference the requirements.

After the tests pass, the AI has them reviewed the same way: a subagent judges the tests against the requirements they claim to validate (dotreq review-test) — catching tests that pass without actually checking the specified behaviour.

You accept the result

Run the tests, verify the implementation works, and move on. Coverage is tracked automatically.

Why the review steps use a subagent

The composing assistant grades its own homework badly — it reviews with its rationalizations still in context. Dispatching a separate agent that sees only the style guide and the content gives you genuinely fresh eyes, on your existing AI subscription. Assistants without subagent support run the same commands and review in-conversation.

Next Steps

Last updated on
For Product ManagersHow It Works