Skip to main content

Lab 4 Challenge: Transforming an Agent

Goal​

Your task is to create a new "Haiku Poet" agent by duplicating and modifying the "Echo" agent from the previous lab. This will demonstrate how to build upon existing agents.

Requirements​

  1. In your adk-training directory, duplicate the echo_agent directory and rename the copy to haiku_poet_agent.
  2. Navigate into the new haiku_poet_agent sub-directory.
  3. Follow the Python Approach below to modify the agent's behavior.
    • Change the name to haiku_poet_agent.
    • Update the description.
    • Craft a new instruction that gives the agent the persona of a wise poet who transforms the user's topic into a haiku.
    • Include at least two examples (few-shot prompts) in your instruction to guide the agent.
  4. Return to the parent adk-training directory.
  5. Run the new agent using the adk web haiku_poet_agent command.
  6. Interact with the agent in the Dev UI to verify that it correctly responds with a haiku.

Python Approach (Primary)​

Modify the agent.py file inside your new haiku_poet_agent directory. You will need to update the name, description, and instruction arguments in the LlmAgent constructor.

Alternative Approach: Using YAML Configuration​

If you are using a root_agent.yaml file:

  1. Modify the name, description, and instruction fields in your root_agent.yaml file.
  2. Ensure you do not have an agent.py file in the same directory, as the YAML file will take precedence.

Self-Reflection Questions​

  • How does providing examples (few-shot prompting) in the instruction help the LLM understand the task better than just describing it?
  • What would happen if you removed the constraint "Do not answer questions or have a conversation"? How would the agent's behavior change?
  • Experiment with creating a new persona for the agent (e.g., a Shakespearean poet, a futuristic robot). What are the key elements you need to include in the instruction to make the new persona convincing?

🕵️ Hidden Solution 🕵️​

Looking for the solution? Here's a hint (Base64 decode me): L2RvYy1hZGstdHJhaW5pbmcvbW9kdWxlMDQtbGxtYWdlbnQtZGVlcC1kaXZlL2xhYi1zb2x1dGlvbg==

The direct link is: Lab Solution