Freeze

This one pauses AI agent execution until you explicitly resume it, giving you a manual checkpoint in automated workflows. You'd use it when you want to review intermediate results before letting the agent continue, like checking data transformations before they get written to production or verifying API responses before the next chain of actions fires off. It's basically a human-in-the-loop pattern without having to rebuild your whole agent architecture. Simple concept, but genuinely useful when you're dealing with high-stakes operations or debugging agent behavior and need that moment to inspect state before things keep rolling.