Consider using coroutines for top-level glue instead of DI

[effa]

Every time the core needs something from the state, it can just ask for it, by yielding an effect. After the intent is completed, new state and an action record is yielded. It could look something like this:

def perform_intent(state, intent):
    intent_performer = perform(state, intent)
    step = next(intent_performer)
    while step.effect:
        data = resolve_effect(step.effect)
        step = intent_performer.send(data)
    return step.state, step.action