Taskflow

When you need a background job to survive across multiple detached tasks, waits, and restarts without losing track of who started it or where you left off, this is the coordination layer. It gives you durable flow identity, revision-safe state mutations, parent-child task linking, and explicit waiting semantics for external events or human replies. The API is verbose by design: you create a managed flow, run linked child tasks, set waiting status with structured wait metadata, resume when unblocked, and finish or fail. It's not branching logic or a workflow engine. Think of it as the bookkeeping runtime that lets your orchestration code in Lobster or a plugin survive process boundaries and come back to the right step with the right context intact.