taskflow

Long-running detached tasks maintain state by using a managed flow model with a flowId, currentStep, stateJson, and waitJson to ensure continuity and traceability across revisions and restarts. This approach keeps a single ownership context for multi-step jobs.

Subagent workflows require orchestration under a single owner context using revision-tracked mutations and waiting support. This ensures deterministic state updates and conflict safety when subagents run in parallel or wait for external input.

Multi-step workflows can wait for external input or parallel tasks and then resume using deterministic mutations. The flow model tracks the currentStep and stateJson, allowing the job to resume or finish safely when the required input arrives.

Plugin orchestration supports conflict-safe state mutations by applying revision-tracked mutations to a managed flow. This ensures that state changes are deterministic and traceable even when multiple steps or tasks update the same flow context.

A managed flow model is needed for long-running data processing pipelines when tasks span multiple prompts or detached sub-tasks. It provides durable flow identity, ownership tracking, and revision control to maintain continuity throughout the pipeline.

Multi-step tasks lose ownership context after a restart without a durable flow model. Using a managed flow with a flowId and stateJson ensures that the ownership context and progress state survive restarts and revisions.