google-adk

Conditional branching in a Google ADK SequentialAgent is implemented using the transfer_to_agent control flow mechanism. This allows your multi-agent system to dynamically route execution to different agents based on session state.

SequentialAgent executes agents in a linear order, LoopAgent repeats agent execution until a condition is met, and ParallelAgent runs multiple agents concurrently. These Google ADK patterns provide distinct orchestration strategies for complex multi-agent systems.

Data loss between agents often occurs due to a missing output_key. You must explicitly define output_key in your agent configuration to ensure data is correctly persisted in session.state and passed along the control flow.

Design a FunctionTool by defining a function that accepts the required parameters, and Google ADK will automatically inject the context. This enables your agents to securely access session state and interact with external tools during execution.

Resolve Google ADK context overflow errors by carefully managing session.state and limiting the data passed between agents. The skill provides debugging guidance to identify non-obvious behaviors causing excessive context accumulation in multi-agent systems.

Google ADK is designed for complex multi-agent system development, but building advanced control flows requires understanding state management and orchestration patterns. This skill addresses common implementation errors to help developers navigate non-obvious behaviors.