Google ADK Python Skill

Code-first AI agent framework with tool integration, multi-agent coordination, and workflow orchestration.

When to Use

• Building multi-agent systems with hierarchical coordination (researcher → writer → editor)
• Creating workflow pipelines (sequential/parallel execution, loop patterns)
• Integrating LLMs with tools (Google Search, Code Execution, custom functions)
• Deploying production agents to Vertex AI, Cloud Run, or custom infrastructure

Quick Start

pip install google-adk

from google.adk.agents import LlmAgent
from google.adk.tools import google_search

# Single agent with tools
agent = LlmAgent(
    name="search_assistant",
    model="gemini-2.5-flash",
    instruction="You are a helpful assistant that searches the web.",
    tools=[google_search]
)

# Multi-agent system
researcher = LlmAgent(name="Researcher", tools=[google_search])
writer = LlmAgent(name="Writer")
coordinator = LlmAgent(
    name="Coordinator",
    sub_agents=[researcher, writer]
)

Common Use Cases

Research Pipeline

Who: Content team lead

"Build a research assistant that searches tech news, summarizes findings,
and generates a weekly report. Need human approval before publishing."

Code Analysis System

Who: Engineering manager

"Create an agent that reviews PR code, runs tests, checks documentation,
and suggests improvements. Parallel execution for speed."

Customer Support Router

Who: Support operations lead

"Build a support agent that classifies tickets, searches knowledge base,
drafts responses, and escalates complex issues to specialists."

Data Processing Workflow

Who: Analytics engineer

"Create a sequential pipeline: fetch data from APIs, clean it,
run analysis, generate visualizations. Loop through multiple datasets."

Key Differences

Workflow Types:

• SequentialAgent: Execute agents in order (research → summarize → write)
• ParallelAgent: Run concurrently (web + papers + experts search)
• LoopAgent: Repeat with iteration logic (process each dataset)

Quick Reference

Agent Creation

# Basic agent
agent = LlmAgent(
    name="agent_name",
    model="gemini-2.5-flash",
    instruction="Your instructions here",
    description="Agent description",
    tools=[tool1, tool2]
)

# Multi-agent coordinator
coordinator = LlmAgent(
    name="coordinator",
    model="gemini-2.5-flash",
    sub_agents=[agent1, agent2, agent3]
)

Custom Tools

from google.adk.tools import Tool

def calculate_sum(a: int, b: int) -> int:
    """Calculate sum of two numbers."""
    return a + b

sum_tool = Tool.from_function(calculate_sum)

Workflows

from google.adk.agents import SequentialAgent, ParallelAgent, LoopAgent

# Sequential
seq = SequentialAgent(name="pipeline", agents=[a1, a2, a3])

# Parallel
par = ParallelAgent(name="concurrent", agents=[a1, a2, a3])

# Loop
loop = LoopAgent(name="iterator", agent=processor)

Human-in-the-Loop

agent = LlmAgent(
    name="careful_agent",
    tools=[google_search],
    tool_confirmation=True  # Requires approval
)

Supported Models

• gemini-2.5-flash (recommended for speed)
• gemini-2.5-pro (complex tasks)
• gemini-1.5-flash, gemini-1.5-pro
• Model-agnostic (supports other LLM providers)

Pro Tips

• Not activating? Say: “Use google-adk-python skill to build a multi-agent research system”
• Use SequentialAgent for dependent tasks, ParallelAgent for independent ones
• Custom tools = any Python function with type hints + docstring
• Enable tool_confirmation=True for sensitive operations (data deletion, API calls)
• Hierarchical structure: coordinator → specialized agents → tools
• Deploy to Cloud Run for containerized hosting, Vertex AI for managed scaling
• Development version: pip install git+https://github.com/google/adk-python.git@main

Related Skills

• AI Multimodal - Gemini API integration
• Planning - Agent workflow design
• Backend Development - API integration

Key Takeaway

Google ADK enables code-first AI agent development with tool integration, multi-agent coordination, and workflow orchestration. Optimized for Gemini, deployable to Cloud Run/Vertex AI, model-agnostic for flexibility.