Best practice to edit human message and rerun

[yf-yang]

I'd like to implement a fundamental human message edit functionality. Consider the following

async for mode, chunk in graph.astream(...):
  match mode:
    case 'update':
      print(graph.get_state(config=config).config['configurable']['checkpoint_id'])

With this method, I can get checkpoint id at ckpt 3/4/5 8/9/10 of the following

ckpt1 (no update yet, so cannot get ckpt id)
START
ckpt2 (no update yet, so cannot get ckpt id)
chatbot
ckpt3
tools
ckpt4
chatbot
ckpt5
None

ckpt6 (no update yet, so cannot get ckpt id)
START
ckpt7 (no update yet, so cannot get ckpt id)
chatbot
ckpt8
tools
ckpt9
chatbot
ckpt10
None

However, if I want to modify the second query's human message, I need to modify state from ckpt 6/7 and add_message reducer is not applicable if I update_state and resume from there, right? If update_state from ckpt 5, seems the graph will directly quit.

I am confused, what's the correct langgraph implementation to deal with that situation?