waterball-software-academy/aixbdd

AIBDD is a structured workflow for building software with AI coding agents in a disciplined, specification first way. The problem it addresses is that most AI assisted coding starts too fast: the agent begins writing code before the requirements are actually clear, which leads to wasted work, missed edge cases, and tests that pass without really validating anything. The workflow moves a product idea through a defined series of steps: clarifying exactly what needs to be built, discovering the boundary rules of the feature, generating a technical plan, producing executable acceptance criteria (the conditions that must be true for the feature to count as done), breaking the work into tasks, then executing those tasks through a Red, Green, Refactor cycle. In software development, Red means writing a failing test first, Green means writing the code that makes it pass, and Refactor means cleaning up the code afterward without breaking the tests. AIBDD treats each of these as a separate, checked step rather than letting the agent jump straight to claiming something is done. When requirements change partway through, a reconcile command is meant to cascade the correction through every downstream artifact starting from the earliest affected step, instead of requiring someone to manually patch things throughout the project by hand. There is also a rewind option to roll the workflow back to an earlier known stage. The project is delivered as a set of commands and skills meant to be used inside an AI coding agent, aimed at founders, tech leads, and product people who want the agent to build against a clear, agreed specification rather than a vague description. It is written in Python. The README notes that the project is still in active development and testing, with functionality that is not yet complete, and it is written partly in Traditional Chinese.