Separation of planning concerns in ReAct-style LLM agents. Planner fine-tuning on synthetic trajectories. - View it on GitHub