langchain-ai/langgraph ? reverse-engineered prompt

I want to build a Python library that lets developers create powerful, long-running AI agents. The core idea is that you can define an agent's logic as a graph, kind of like a flowchart, where each node is a specific step or a tool it can use. This will allow for building much more complex, multi-step agents.

These agents need to be truly stateful, so they can remember what happened in previous steps and maintain context. It's also really important that they're durable—they should be able to run for a long time without crashing, and if they do get interrupted, they can resume from where they left off. I also want a "human-in-the-loop" feature, so a person can step in at any point to review, approve, or change things before the agent continues. This should be a flexible, low-level framework for building really robust and observable agent systems.