AI

Repo0: Design-Driven Zero-to-All Code Generation

Repo0 is a framework for generating code from natural-language requirements without assuming a predefined repository architecture. It uses a Dual-Directed-Acyclic-Graph (Dual-DAG) to maintain an explicit architectural state and iteratively evolves component boundaries through structural actions guided by modularity metrics until convergence. Repo0 was evaluated on six real-world repositories using GPT-5 mini and DeepSeek V3.2, achieving the highest Functionality Coverage and
Repo0 is a framework for generating code from natural-language requirements without assuming a predefined repository architecture. It uses a Dual-Directed-Acyclic-Graph (Dual-DAG) to maintain an explicit architectural state and iteratively evolves component boundaries through structural actions guided by modularity metrics until convergence. Repo0 was evaluated on six real-world repositories using GPT-5 mini and DeepSeek V3.2, achieving the highest Functionality Coverage and Pass Rate compared to a baseline repository-planning method. --- Why it matters: This matters because it addresses the limitation of existing code generation systems that assume a predefined repository architecture, allowing for more flexible and modular software development. Source: https://arxiv.org/abs/2608.19854

This article was originally published at: https://arxiv.org/abs/2608.19854