Feature Implementation Plan: customer_support_agent

📋 Todo Checklist

• Create the agent directory structure.
• Create the in-memory ticket database.
• Implement the ticket lookup tool.
• Implement the ticket summarization tool.
• Define the main customer support agent.
• Final Review and Testing.

🔍 Analysis & Investigation

Codebase Structure

The current project is a clean slate. Based on the AGENTS.md documentation, ADK agents should be organized into their own directories. The convention is to have a directory (e.g., customer_support_agent/) containing an agent.py file (which defines the root_agent) and an __init__.py file (which exports the agent module). Other related files, like tools and data, can be co-located in this directory.

Current Architecture

There is no existing architecture. The plan proposes a simple, single-agent architecture. A single LlmAgent will be equipped with tools to interact with an in-memory ticket database. This is a straightforward approach that aligns well with the capabilities of the ADK framework as described in the documentation.

Dependencies & Integration Points

The agent will depend on the google-adk library. The core integration points are:

• An in-memory data source for tickets.
• Tools (Python functions) that the agent can call.
• The LlmAgent itself, which orchestrates the use of tools based on user input.

Considerations & Challenges

• Summarization Strategy: The summarization tool can be implemented in two ways: either as a simple function that calls another LLM for summarization or as a dedicated LlmAgent. For this plan, we will use a separate LlmAgent for summarization to demonstrate a multi-agent (in a sense) setup, where one agent uses another as a tool.
• Error Handling: The tools should handle cases where a ticket ID is not found.
• Scalability: An in-memory database is suitable for the requested 20 sample tickets but will not scale for a production system. For a real-world application, this would be replaced with a connection to a proper database or API.

📝 Implementation Plan

Prerequisites

• google-adk installed (pip install google-adk).
• A configured Gemini API key (typically in a .env file).

Step-by-Step Implementation

1. Create Agent Directory Structure:

   • Files to create:
     • customer_support_agent/
     • customer_support_agent/__init__.py
   • Changes needed:
     • In customer_support_agent/__init__.py, add the line: from . import agent

2. Create In-Memory Ticket Database:

   • File to create: customer_support_agent/tickets.py
   • Changes needed:
     • Define a list of 20 dictionaries, where each dictionary represents a ticket with id, title, and description keys.
     • The id should be a unique string.
     • The title and description should be descriptive strings.

3. Implement Agent Tools:

   • File to create: customer_support_agent/tools.py
   • Changes needed:
     • Import the ticket data from customer_support_agent.tickets.
     • Create a function get_ticket_details(ticket_id: str) -> str that takes a ticket ID and returns the ticket details as a formatted string or a JSON string. It should handle cases where the ticket is not found.
     • Import LlmAgent from google.adk.agents.
     • Define a summarizer_agent using LlmAgent. This agent's instruction will be to summarize the provided text.
     • Create a function summarize_ticket(ticket_id: str) -> str. This function will first call get_ticket_details to get the ticket's description, and then it will use the summarizer_agent to summarize the description.

4. Define the Main Customer Support Agent:

   • File to create: customer_support_agent/agent.py
   • Changes needed:
     • Import LlmAgent from google.adk.agents.
     • Import the get_ticket_details and summarize_ticket functions from customer_support_agent.tools.
     • Define the root_agent as an LlmAgent.
     • Set the name to customer_support_agent.
     • Use the gemini-2.5-flash model.
     • Write a clear instruction that tells the agent it is a customer support assistant and explains how to use the get_ticket_details and summarize_ticket tools.
     • Provide the get_ticket_details and summarize_ticket functions in the tools list.

🎯 Testing Strategy

The user will run adk web . from the root of the project. This will start the ADK web server. The user can then interact with the customer_support_agent through the web UI to test its ability to retrieve and summarize tickets.

✅ Success Criteria

• The agent can successfully find and return the full details of a ticket when given a valid ID.
• The agent provides a helpful message when a ticket ID is not found.
• The agent can provide a concise summary of a ticket when requested.
• The agent correctly uses the tools to perform these actions.