Start here
Generative Specification is a way of working with an AI coding assistant. You write down what the system must be, the assistant derives the code, and something outside the model checks the result. This site teaches it, free and open.
Pick the time you have
| You have | Do this |
|---|---|
| Ten minutes | Read the two-page introduction: the problem, the idea and the method |
| An hour | Follow your first hour on a project of your own: a spec, a constitution file for the assistant, and quality gates. No tools required beyond what you already use |
| About half an hour of video | The free course: thirteen short lessons in English and Spanish. The videos are being published; the lesson list is final |
| An afternoon | Learn by goal or role, then Practice for the guide that fits your project |
Five words you will meet
- Spec: a written description of what the system must do, precise enough to check.
- Stateless reader: an assistant that starts every session with no memory of your project.
- Sentinel: a small root file that tells the assistant what the system is and what to read next.
- Gate: an automated check outside the model that can refuse a build.
- Ratchet: every miss leaves a permanent rule, test or gate behind.
Not sure where you fit?
- By goal or role: one row for each situation, with the pages to read in order.
- By property: study one idea at a time, from the seven properties of a derivable project.
- By course sequence: the thirteen lessons in order.
Want a quick read of your own project?
The free audit prompt at pragmaworks.dev/audit grades a repository you point it at. It is the first step of lesson 13 in the course.