Skip to content
NZT
ES · Leer en español

Usage

Getting started

How to use NZT once it is installed. Open your agent, ask for what you want to build, approve the plan and move forward one unit at a time. No new commands.

Once it is installed, there are no new commands to learn. Open your agent (claude or codex) in any project and ask for what you need in plain language: the kernel is already loaded and routes the request to the right phase and skill.

Step by step

  1. Open your agent. Go to the project folder and run claude or codex. NZT loads on its own from the global instructions file.
  2. Ask for what you want to build. NZT classifies the request. If something is not defined yet, it asks about the rules before choosing technologies.
  3. Review and approve the plan. It proposes units, artifacts, omissions and stopping points, and saves Plan/state.json with approved: false. It writes no code until you approve.
  4. Move forward one unit at a time. At the end of each unit it updates the state, reports and waits. You can authorize a batch if you prefer more autonomy.
  5. Resume whenever you want. In a new session, NZT reads Plan/state.json and continues from the next entry point.

How to start

Try it on a small, real task, in guided mode. Compare the process and the evidence, then decide which brakes your team adopts.

Not installed yet? Read the guide for Claude Code or for Codex CLI.

See also