PocketFlow 的编排能力进一步增强了思考能力

PocketFlow's design concept is excellent. In the current Agent application scenario, most of them are based on workflow patterns, so using the Flow approach can indeed bring significant results. After in-depth use, I think it is possible to further optimize the component design and raise the level of atomization. Ideally, developers only need to write simple workflow orchestration code to complete complex tasks, such as `QuestionNode >> LLMNode >> RAGNode`. More specifically, if the atomization capability is strong enough, even low-code orchestration processes such as Dify can be converted to high-code implementations in PocketFlow. I have verified the feasibility of this direction, and the main challenge we face at present is the data sharing mechanism. Currently, PocketFlow shares data between nodes through the `shared` object, and each node can freely define the data to be output to shared. However, in actual applications, due to the different input requirements of each node, it is often necessary to re-configure the input and output settings of each node during workflow orchestration, which causes many inconveniences for users.