Conditional Node

The Conditional Node enables dynamic branching logic within graph skills, allowing you to selectively trigger downstream nodes based on custom conditions.

Overview

The Conditional Node evaluates a user-defined function to determine which connected nodes should execute next. It operates based on the graph's order of execution (__max_connection__ edges).

Implementation

Function Signature

def run_next(skill_globals, graph_state) -> list[str]:
    """
    Determine which child nodes should execute next.
    Returns a list of node names to execute.
    """
    pass

Available Context

The function has access to two key objects:

1. skill_globals: Contains skill-level variables
2. graph_state: Provides a snapshot of the graph's execution state up to the current point
   • Structure: Nested dictionary using node names and output names as keys
   • Example: graph_state["node_name"]["output_name"]

Execution Rules

• Only immediate child nodes can be included in the return list
• Child nodes not included in the return list will be skipped
• Any downstream nodes that depend solely on skipped nodes will also be skipped

Example Usage

def run_next(skill_globals, graph_state) -> list[str]:
    # Check if previous node's confidence score is above threshold
    if graph_state["classification_node"]["confidence"] > 0.8:
        return ["high_confidence_path"]
    else:
        return ["low_confidence_path", "retry_path"]

In this example:

• If the confidence score from a previous classification is high, only the "high_confidence_path" node will execute
• Otherwise, both the "low_confidence_path" and "retry_path" nodes will execute
• Any other connected nodes will be skipped