Large language model agents have made significant strides in code generation, but most existing systems assume a predefined repository architecture. This assumption fails in zero-to-all code generation, where an agent must build an entire software project from natural-language requirements while maintaining a modular structure throughout development.
To address this, researchers at Hugging Face present Repo0, a continuous structural evolution framework. Repo0 maintains an explicit architectural state instantiated as a Dual-Directed-Acyclic-Graph (Dual-DAG), comprising a requirement-level DAG, a component-level DAG, and their alignment. Starting from natural-language requirements, it iteratively evolves component boundaries through structural actions guided by modularity metrics until structural convergence, after which the converged architecture guides test-driven development code generation.
Evaluated on six real-world repositories from RepoCraft using GPT-5 mini and DeepSeek V3.2, Repo0 achieves the highest Functionality Coverage and Pass Rate across all settings. Compared with RPG, the strongest repository-planning baseline, Repo0 improves Functionality Coverage by up to 20.08 percentage points and Pass Rate by up to 29.74 percentage points.
Ablation and structural-evolution analyses further demonstrate the importance of the Dual-DAG architectural state, modularity-guided structural evolution, and explicit structural convergence. The framework not only generates code but also builds the architecture as it goes, continuously reshaping the repository with cohesion and coupling until the structure makes sense.