Project2Task: Graph-Guided Project-Level Planning for Autonomous Research

Project2Task addresses the challenge that research agents can execute individual tasks like literature search, hypothesis generation, code execution, and manuscript drafting, but lack the ability to plan and manage entire projects. The framework uses a graph representation to model a research project's agenda, breaking it into bounded subtasks with clear dependencies and parallel alternatives. This enables more autonomous and structured execution of complex research workflows. The approach is expected to improve how AI systems handle multi-stage research by supporting dependency-aware sequencing and concurrent exploration. Future work may focus on evaluating Project2Task across diverse research domains and integrating it with existing agent frameworks.