ADK Developer

The ADK Developer skill provides a structured framework for engineers and AI developers to design, implement, and maintain intelligent agents using the Python-based Google Agent Development Kit. This skill covers the full lifecycle of agent development, from foundational primitives like LlmAgent and BaseAgent to complex multi-agent orchestration and deployment strategies. It is designed for developers who need to integrate LLMs—specifically using Vertex AI—into robust, scalable workflows with clear observability, state management, and debugging capabilities. Users can expect to learn how to structure projects, define custom tools, manage session memory, and ensure secure communication between agents.

• Full support for ADK primitives including Session, State, Memory, and Event management for building stateful conversational systems.

• Comprehensive workflow orchestration guidance, covering both simple LlmAgent patterns and complex multi-agent system architectures.

• Deep integration with Vertex AI for secure model execution using high-performance models like gemini-2.5-flash.

• Standardized project layout recommendations to enable native CLI discovery and maintainable codebases.

• Advanced capabilities for streaming, performance optimization, and observability via event logging and tracing tools.

• Always initialize sessions by checking for repository updates via provided internal scripts to ensure synchronization with current API references.

• Mandatory configuration protocol requires users to explicitly define GOOGLE_CLOUD_PROJECT and GOOGLE_CLOUD_LOCATION; implicit defaults are strictly prohibited for security reasons.

• Prioritize simplicity by starting with single LlmAgent designs before escalating to multi-agent or WorkflowAgent architectures.

• Enforce strict structural boundaries: all agent code must reside in dedicated subdirectories rather than the root directory to support the ADK discovery mechanism.

• Design artifacts should be presented for validation before any implementation code is generated to ensure architectural alignment.

• Utilize standard tools including FunctionTool and OpenAPIToolset, ensuring all tool-based authentication is handled according to defined security best practices.