Understand the SR Method

The SR Method, for Specification Runtime, is a method for building complex applications with a code AI without letting it improvise the product.

It starts from a user intent, turns it into usable documentation, then asks Codex to work through short, verified and documented lots.

Promise

Codex should work like a careful developer:

• understand the need before acting;
• read verifiable sources;
• separate facts, assumptions, and decisions;
• split work into short lots;
• change a limited surface;
• test what it delivers;
• document how to resume.

Core rule

The AI is free during exploration, but constrained during execution.

Codex may analyze, compare, diagnose, and recommend. In a project with strict human validation, it must not edit files, push to GitHub, install dependencies, migrate a database, or change configuration until the user validates the scope.

Operating loop

Understand -> verify -> split -> modify -> test -> document -> prepare resume

What the method prevents

The SR Method reduces several common drifts:

• Codex codes without reading sources;
• scope expands without validation;
• business decisions are invented;
• a lot is marked done while real user testing is still missing;
• a long conversation loses context;
• resume depends on oral memory;
• runtime AI agents are designed as simple prompts.

Experimental and contributive method

The SR Method comes from real AI-assisted development practice. It is experimental, improvable and open to contributions.

It does not replace user expertise. It provides a frame to make Codex work controllable, auditable and resumable.