Difference to tool-calling agent or augmented LLM

[DiXue98]

An integral capability of ReAct Agent is to explicitly output the reasoning and make it part of its context, forcing the subsequent actions to stick to the reasoning. However, I do not see such component to induce the thought of LLM in this codebase. Please correct me if there is any thing I have missed or my understanding of ReAct is incomplete. Thanks.