pavel-molyanov/molyanov-ai-dev

README preview

# AI-First Development Framework

**English** | [Русский](README.ru.md)

A framework for AI-First development with [Claude Code](https://docs.anthropic.com/en/docs/claude-code) and [Codex](https://github.com/openai/codex). One methodology, two compatible runtimes.

At its core is a spec-driven pipeline: first we plan the work in detail through an interview with the user and research of the codebase, agree on the specifications, and only then write code. Every stage is checked by specialized validator agents. Code is written via TDD.

The framework is multilingual: it talks to you, runs interviews, and writes user-facing specs in the language you write in, while keeping technical documents (tech-spec, tasks, code) in English.

You set your language in one place — the instruction file: `~/.claude/CLAUDE.md` for Claude and `~/.codex/AGENTS.md` for Codex (or a project's `CLAUDE.md` / `AGENTS.md`). In the `## Language` section, declare your language, e.g. `This user writes in English.` Skills and agents read the language from there, so even subagents that never see the chat produce output in your language.

## Two runtimes: Claude and Codex

The methodology supports both agents at once.

- **Claude is the source of truth.** You edit files in `~/.claude/`, `CLAUDE.md`, and projects' `.claude/**`.
- **Codex is the generated runtime.** `~/.codex/`, `AGENTS.md`, and `.codex/**` are created from the Claude sources by `scripts/sync-to-codex.py`. Do not edit them by hand.

The commands (`/new-user-spec`, `/do-task`, `/done`, etc.), skills, and agents are the same. Only the interpreter changes.

View full repository on GitHub →