ADK Agent Architect (Python Edition)

Philosophy

Agents are modular execution units. Use the specialized references below to implement the specific architecture required for the task. All implementations must be Python 3.12+ compatible.

Logic Flow

1. •Analyze Complexity:
   • •For FSM-based topology (Markovian routing) -> Use the markov-topology skill.
   • •For dynamic reasoning/tool use -> LlmAgent.
   • •For fixed order/loop/parallel execution -> WorkflowAgent.
   • •For non-standard branching -> CustomAgent.
2. •Context Loading:
   • •If working on LLM-driven agents, read references/llm-agents.md.
   • •If working on Workflows (Sequential/Parallel/Loop), read references/workflow-agents.md.
   • •If working on Inheritance/Custom Logic, read references/custom-agents.md.
3. •Implementation: Generate Python 3.12+ code using google-adk.

Output Standards

• •Explicit InvocationContext handling.
• •Use of Jinja2 {var} templates in instructions.
• •Strict async implementation using AsyncGenerator[Event, None].
• •Typing: Use typing.Annotated, pydantic.BaseModel for schemas.