Agently TriggerFlow Config

This skill covers TriggerFlow definition-level export, import, copy, and inspection. It focuses on blueprint copy, flow-config roundtrip, JSON or YAML flow files, handler registration for restored flows, and Mermaid visualization. It does not cover execution save/load, pause-and-resume runtime state persistence, or provider-specific model configuration.

Prerequisite: Agently >= 4.0.8.5.

Scope

Use this skill for:

• save_blue_print() and load_blue_print()
• get_flow_config()
• get_json_flow() and get_yaml_flow()
• load_flow_config(), load_json_flow(), and load_yaml_flow()
• to_mermaid(mode="simplified" | "detailed")
• exported TriggerFlow contract metadata in flow config and Mermaid
• flow-definition roundtrip across processes or repositories
• understanding what is serializable in a TriggerFlow definition and what must be re-registered or reinjected at runtime

Do not use this skill for:

• execution save() / load() after a workflow has already started running
• interrupt persistence and resume-after-restart mechanics
• model provider setup or output-schema design
• runtime-stream lifecycle as the primary topic

Workflow

1. Start with references/definition-surfaces.md to choose between blueprint copy, config export, and Mermaid inspection.
2. If the task is about JSON or YAML roundtrip, read references/export-import-roundtrip.md.
3. If the task is about what must be registered again after loading, read references/handler-registration.md.
4. If the task is about contract metadata in exported config or Mermaid, read references/contract-metadata.md.
5. If the task is about diagrams and inspection, read references/mermaid-usage.md.
6. If behavior still looks wrong, use references/troubleshooting.md.

Core Mental Model

TriggerFlow config APIs work on workflow definitions, not execution instances.

The main surfaces are:

• blueprint copy for in-process definition reuse
• flow config for serializable definition export and import
• Mermaid for inspection and communication

Use them like this:

• same process, same Python runtime, no file roundtrip required -> blueprint copy
• transport, repository storage, or human-editable artifacts -> JSON or YAML flow config
• architecture review or visual debugging -> Mermaid

Selection Rules

• in-memory reusable definition object -> save_blue_print() / load_blue_print()
• serializable dictionary form -> get_flow_config()
• share or store the definition as a file -> get_json_flow() or get_yaml_flow()
• restore a definition from stored JSON or YAML -> load_json_flow() or load_yaml_flow()
• inspect the structure visually with grouped nodes -> to_mermaid(mode="simplified")
• inspect internal nodes and callable labels in more detail -> to_mermaid(mode="detailed")
• exported artifacts should preserve contract metadata for inspection or schema discussion -> flow config and Mermaid
• restored flow needs chunk handlers or condition handlers -> register them before loading the config
• restored flow needs runtime resources -> inject them again at runtime after loading
• workflow has already started and should resume later -> use a separate execution-state skill, not this one

Important Boundaries

• flow config serializes the definition, not a running execution
• flow config preserves exported contract metadata, schema labels, and system interrupt metadata
• loading config restores contract metadata for inspection and re-export, but it does not reconstruct live runtime validators from the original Python contract types
• runtime resources are not serialized into flow config
• non-serializable handlers such as anonymous lambdas may still appear in Mermaid, but they are not valid export targets for get_flow_config()
• loading a config rebuilds the flow definition and then recompiles handlers against the currently registered callable registry

References

• references/source-map.md
• references/definition-surfaces.md
• references/export-import-roundtrip.md
• references/handler-registration.md
• references/contract-metadata.md
• references/mermaid-usage.md
• references/troubleshooting.md