TriggerFlow

Suggested reading order:

1. Overview: boundaries, mental model, and a minimal flow.
2. Lifecycle: open / sealed / closed, and start / close entrypoints.
3. State and Resources: choosing state, flow_data, and runtime_resources.
4. Events and Streams: emit / when and runtime stream.
5. Patterns: branching, match, batch, for_each, event-driven loops.
6. Sub-Flow: parent-child flow composition.
7. Persistence and Blueprint: save/load and definition export.
8. Pause and Resume: human intervention and external resume.
9. Distributed Pause and Resume Boundaries: host-managed recovery, resource ownership, and production boundaries.
10. Runtime Intervention: adding supplemental context without pausing or mutating the graph.
11. Model Integration: calling agents / requests inside chunks.
12. Execution Result: reading snapshots, state, compatibility results, interventions, and metadata.
13. Compatibility: migrating old .end(), set_result(), and runtime_data.

Dynamic Task is documented separately as an application-level facade: Dynamic Task. It uses TriggerFlow as the execution substrate, but ordinary users should start from Agently.create_dynamic_task(...).