google--adk-python/contributing/samples/workflows/loop_config/README.md

Workflow Loop Config Sample

Overview

This sample demonstrates how to define a workflow with a feedback loop using a YAML configuration file. It mirrors the workflow_samples/loop sample, but uses YAML to define the workflow structure instead of Python.

Sample Inputs

• Python programming

• Baking cookies

Graph

graph TD
    START --> process_input[process_input]
    process_input --> generate_headline[generate_headline.yaml]
    generate_headline --> evaluate_headline[evaluate_headline.yaml]
    evaluate_headline --> route_headline[route_headline]
    route_headline -->|unrelated| generate_headline

How To

This sample uses some special syntax in root_agent.yaml to support dynamic resolution and graph construction:

1. _code Suffix

Fields ending with _code (like output_schema_code in evaluate_headline.yaml) tell the ADK YAML mapper to resolve the value as a Python code reference rather than treating it as a plain string.

• If it starts with ., it resolves relative to the current agent directory's Python package path.
• Example: output_schema_code: .agent.Feedback resolves to the Feedback Pydantic model in agent.py in the same directory.

2. Function References in Edges

If a string in the edge list does not end with .yaml and is not 'START', it is treated as a function reference.

• If it starts with ., it resolves relative to the current agent directory's Python package path.
• Example: .agent.process_input resolves to the process_input function in agent.py.
• It automatically creates a FunctionNode with the function's name as the node name.

3. External Agent Files

Agents can be defined in their own YAML files and referenced by filename in the edges list.

• Example: generate_headline.yaml references the agent defined in that file.
• The mapper caches resolved nodes by their string value, so using the same filename in multiple edges correctly reuses the same agent instance, preserving the graph structure (e.g. for loops).