In your paper and the provided JSON description, you outline a set of spatial operators that the LLM can call.I understand that the function names and descriptions are packaged in a prompt or a JSON schema for the LLM. Could you please tell me how to use the chosen local methods, and how is its result fed back to the LLM to formulate the final natural language answer?