Automation

Tells the model when the context passes 250k tokens (heads-up) and 300k (hand off), so it finishes the task it's on and then hands off to a fresh session instead of starting another.