Does create_react_agent use tool call results from previous steps?

[wylswz]

I built an agent with create_react_agent with two tools, and I prompted it to execute one tool at a time, because there are dependencies. The result of first tool call is not recognized by the agent, so the second tool is never invoked.

Here's the tool definitions

    @tool(parse_docstring=True)
    def metabcr_data_preprocess(raw_data_path: str) -> str:
        """
        perform data preprocess for metabcr, and get final input file

        Args:
            raw_data_path: raw data path

        Returns:
            metabcr_input_file_path: metabcr input file path

        """
        return "/a/b/c/d"

    @tool(parse_docstring=True)
    def metabcr_predict(input_path: str) -> str:

        """
        Perform metabcr predict on BCR seq against antigen

        Args:
            input_path: input path

        Returns:
            metabcr_output_file_path: metabcr output file path
        """
        return "/c/d/e/f"

I added interrupts before tool execution, I'm not sure if that affects the result

[hinthornw]

It does if you are checkpointing

[wylswz]

It looks like there's something wrong with my model's context length.