google-adk-python_skill

Overview

This skill teaches how to build, evaluate, and deploy AI agents using Google ADK Python (Agent Development Kit). It focuses on code-first agent design, tool integration, multi-agent composition, workflow patterns, and deployment options like Vertex AI and Cloud Run. The guidance favors modular, testable implementations and human-in-the-loop safety controls.

How this skill works

The skill inspects how to define agents in Python (LlmAgent, BaseAgent) and compose them into workflow agents (SequentialAgent, ParallelAgent, LoopAgent). It shows how to attach pre-built tools (google_search, code_execution), convert Python functions into tools, and require confirmations for sensitive operations. It also covers multi-agent coordination, evaluation patterns, and deployment pipelines for Vertex AI, Cloud Run, or custom infrastructure.

When to use it

• You need tool-enabled LLM agents that call external APIs or execute code.
• You want predictable pipelines: sequential, parallel, or repeated tasks.
• You are building multi-agent systems with delegation and coordination.
• You require human-in-the-loop approval before tool execution.
• You plan to deploy agents at scale using Vertex AI or containerized services.

Best practices

• Adopt a code-first approach: define agents in Python for version control and testing.
• Design agents modularly: specialize agents for single responsibilities and compose them.
• Use pre-built tools and wrap custom Python functions as tools for consistency.
• Add tool confirmation for sensitive actions and integrate human-in-the-loop flows.
• Create evaluation test cases and iterate with the development UI for reproducible improvements.

Example use cases

• Research assistant: web search, summarization, and report generation via a SequentialAgent.
• Code assistant: run code, debug, and generate documentation with code_execution tools.
• Customer support: route queries to domain specialists and escalate using a coordinator agent.
• Content pipeline: parallel research agents feeding a writer agent to assemble drafts.
• Data workflows: fetch, process, and visualize data using workflow agents and custom tools.