Task Planning

PlanMiddleware is a middleware used for structured decomposition and process management before executing complex tasks.

Additional Notes

Task planning is an efficient strategy for context engineering management. Before executing a task, the large language model first breaks down the overall task into multiple ordered subtasks, forming a task planning list (referred to as a plan in this library). It then executes each subtask sequentially, dynamically updating the task status after completing each step, until all subtasks are finished.

Parameter Description

Parameter	Description
system_prompt	System prompt. If None, the default prompt will be used. Type: str Required: No
custom_plan_tool_descriptions	Custom descriptions for planning-related tools. Type: dict Required: No
use_read_plan_tool	Whether to enable the read plan tool. Type: bool Required: No Default Value: True

The keys for the custom_plan_tool_descriptions dictionary can be any of the following three values:

Key	Description
write_plan	Description for the write plan tool
finish_sub_plan	Description for the finish sub-plan tool
read_plan	Description for the read plan tool

Usage Example

from langchain_dev_utils.agents.middleware import PlanMiddleware

agent = create_agent(
    model="openai:gpt-4o",
    middleware=[
        PlanMiddleware(
            custom_plan_tool_descriptions={
                "write_plan": "Used to write a plan, breaking down the task into multiple ordered subtasks.",
                "finish_sub_plan": "Used to complete a subtask, updating its status to finished.",
                "read_plan": "Used to query the current task planning list."
            },
            use_read_plan_tool=True,  # If you don't want to use the read plan tool, you can set this parameter to False
        )
    ],
)

response = agent.invoke(
    {"messages": [HumanMessage(content="I'm going to New York for a few days, help me plan the itinerary")]}
)
print(response)

Tool Description

PlanMiddleware requires the use of two mandatory tools: write_plan and finish_sub_plan, while the read_plan tool is enabled by default. If not needed, you can set the use_read_plan_tool parameter to False.

Comparison with the Official To-do List Middleware

This middleware is similar in function to the official To-do list middleware provided by LangChain, but differs in tool design:

Feature	Official To-do List Middleware	This Library's PlanMiddleware
Number of Tools	1 (write_todo)	3 (write_plan, finish_sub_plan, read_plan)
Functionality Focus	Oriented towards to-do lists	Specifically designed for planning lists
Operation Method	Adding and modifying are done through one tool	Writing, modifying, and querying are handled by separate tools

Whether referred to as todo or plan, the essence is the same concept. The key distinction of this middleware from the official one is that it provides three dedicated tools:

• write_plan: Used to write or update plan content
• finish_sub_plan: Used to update the status of a subtask after its completion
• read_plan: Used to query plan content