SR Method in English

The SR Method helps teams develop with Codex without letting the model improvise the product. It turns a user intent into controlled work: context reading, explicit scope, human validation, short lots, tests, and resumable memory.

Recommended path

1. Understand the principles.
2. Start a Codex session with the right prompt.
3. Create or resume a project.
4. Work by lots.
5. Apply validation gates.
6. Close with tests, traces, and a recovery path.

Entry prompts

Use these prompts in Codex instead of running the pack scripts manually.

Install the SR Method in this project.
Read the SR Method documentation, inspect the repository, propose the exact scope, then wait for my validation before changing anything.

Strict SR resume.
Read AGENTS.md, SR_BOOTSTRAP, CURRENT_STATE and the latest NEXT_SESSION_PROMPT if it exists. Summarize the state and wait for validation before coding.

Runtime agent extension

The SR Agent Method is an optional extension for designing application runtime AI agents: agent contract, controlled runtime, authorized data, typed outputs, human validation, and logs.

New in SR 3.2.0

SR 3.2.0 adds two structuring concepts:

• SR Passes, to group several related lots into a bounded execution pass with order, preflight, human validations, and grouped E2E checks;
• a stronger SR Agent Method, to design framework-agnostic runtime AI agents from a bounded product action, stable internal representation, runtime contract, message builder, and strict validations.