hitl_rerun

Human-in-the-Loop (single-node re-entry)

A Human-in-the-Loop workflow where one emitting FunctionNode both pauses for input and produces the final output. On resume the node is re-run from scratch (NodeConfig.RerunOnResume = &true).

• Concept: Single-node re-entry HITL with workflow.ResumeOrRequestInput.
• Needs LLM? No

For the two-node handoff variant, see ../hitl_simple.

Goal

Contrast two ways to do HITL. The two-node handoff variant (../hitl_simple) has one node ask and a separate node consume the reply. This sample collapses both phases into a single re-run node:

• workflow.ResumeOrRequestInput emits a RequestInput and returns ErrNodeInterrupted on the first pass (pause, no output);
• after the human replies, the node is re-run from the top, and the same call now returns the reply, which the body turns into the terminal output.

Workflow

graph LR
    User[User]
    subgraph "ADK Application Workflow"
        Start((Start)) --> G[Node: greet]
        G --> End((End))
    end
    User -- "1. hello" --> Start
    G -- "2. What's your name? (1st pass)" --> User
    User -- "3. Alice (re-run on resume)" --> G
    End -- "4. Hello, Alice!" --> User

Loading

The numbered edges are the user ↔ application exchange, in order. Because RerunOnResume = &true, the reply at step 3 re-enters greet (instead of flowing on to a successor, as it would in ../hitl_simple), so the same node executes twice: first it asks and pauses (step 2), then on resume it is re-run from the top and produces the greeting (step 4). The InterruptID embeds the invocation ID so the reply still correlates across the re-run within a single run, yet a later run re-prompts.

Running the sample

go run ./examples/workflow/hitl_rerun/ console

Example session

User -> hello
Agent -> What's your name?
User -> Alice
Agent -> Hello, Alice!