NZT · A method for coding agents
From idea
to product.
Every day, with discipline.
NZT gives your coding agent —Claude Code or Codex CLI— method, memory and brakes. From a one-line fix to a whole product: the same rhythm, at the right depth.
Version 0.1.0The problem
An agent can get to the wrong place very fast.
- Without NZT
Assumes and writes.
It writes code before anyone has defined what it should do.
With NZTAsks before it assumes.
Functional analysis comes first, and anything undefined becomes a question, not an assumption.
- Without NZT
Never stops.
It starts and does not stop until something breaks.
With NZTStops where you can review.
It cuts the work into reviewable units and stops to wait for you at the end of each one.
- Without NZT
Loses the thread.
It runs out of context and you lose the thread.
With NZTPicks up without re-explaining.
The plan and the state live in Plan/state.json: you resume without explaining it all again.
- Without NZT
Tells you it works.
Without showing what it tested, skipped or assumed.
With NZTShows you the evidence.
It reports what happened: the test that failed, the step it skipped, the assumption it made.
What you gain is not just better code: it is being able to review what it did, pick up where you left off, and not find out at the end that it understood something else.
The cycle
One rhythm, scaled to the request.
Understand
What, and what for.
Propose
Units and scope.
Execute
One unit.
Verify
Real evidence.
Record
Persistent state.
A one-line change does not trigger the whole method: the cycle stays, the depth changes.
First step
Try it on today's task.
A small, real task, in guided mode. Everything else —install, phases, brakes and CLI— lives in the docs.