taskflow

To manage workflow state for long-running tasks, you need a durable flow with a single owner context that persists stateJson between steps and provides a stable return context after detached runs finish.

You can resume a multi-step orchestration job after restarts by using lifecycle methods like resume and setWaiting, which restore the previous execution context using stored waitJson and linked child task references.

Revision tracking is a mechanism for conflict-safe mutations that ensures state consistency across detached tasks. You need it when multiple steps attempt concurrent updates to the same stateJson within a single workflow.

You coordinate detached tasks under a single owner session by calling createManaged to establish a flow identity, then using runTask to execute linked child tasks that report back to the parent flow id.

Yes, you can cancel or fail a managed workflow midway by calling the cancel or finish lifecycle methods, which update the flow state to blocked, failed, or cancelled while preserving the current step and revision history.

You should avoid stateful orchestration for task execution when jobs are short-lived, complete within a single prompt, or do not require persistent stateJson and revision-tracked mutations across multiple detached steps.