Part 2 Exercises: Practice Building Agents¶

Complete these exercises to reinforce your understanding of agent creation. Remember to use the WXO Agent Architect mode to create your agents.

Exercise 1: Personality Agent (Easy)¶

Goal: Create an agent with a unique personality

Task: Create a pirate-themed customer service agent that helps users but speaks like a pirate.

Steps Needed:
1. Create agent yaml file
2. Define the agent with appropriate instructions
3. Import and test the agent

Ask Bob for help:

Bob, create a YAML file for a pirate-themed customer service agent that speaks like a pirate but is still helpful

Test prompts:
- "Hello, I need help with my order"
- "What can you do?"
- "Thank you for your help"

Reminder: Use orchestrate agents import -f <your_agent_name>.yaml to import your agent and orchestrate chat ask --agent-name <your_agent_name> to test it.

Success criteria:
- Agent responds in pirate speak
- Agent is still helpful and professional
- Agent stays in character

Exercise 2: Domain Expert Agent (Medium)¶

Goal: Create an agent that's an expert in a specific domain

Task: Create a Python programming expert agent that can answer coding questions.

Requirements:
- Agent should introduce itself as a Python expert
- Should provide code examples when relevant
- Should explain concepts clearly
- Should admit when questions are outside Python scope

Ask Bob for help:

Bob, create an agent that's a Python programming expert. It should provide code examples and explain concepts clearly.

Test prompts:
- "What's the difference between a list and a tuple?"
- "Show me how to read a CSV file"
- "Explain decorators in Python"
- "How do I deploy a Java application?" (out of scope)

Success criteria:
- Provides accurate Python information
- Includes code examples
- Handles out-of-scope questions gracefully

Exercise 3: Multi-lingual Agent (Medium)¶

Goal: Create an agent that can respond in multiple languages

Task: Modify the hello-agent to support English, Spanish, and French.

Requirements:
- Agent should detect the user's language
- Should respond in the same language
- Should be able to switch languages mid-conversation

Ask Bob for help:

Bob, modify hello-agent.yaml to support English, Spanish, and French. The agent should detect and respond in the user's language.

Test prompts:
- "Hello, how are you?"
- "Hola, ¿cómo estás?"
- "Bonjour, comment allez-vous?"
- "Can you switch to Spanish?" (then continue in Spanish)

Success criteria:
- Responds correctly in all three languages
- Maintains context when switching languages
- Provides natural translations

Exercise 4: Structured Output Agent (Advanced)¶

Goal: Create an agent that returns responses in a specific format

Task: Create a "Product Recommender" agent that always returns recommendations in a structured format.

Requirements:
- Agent should ask clarifying questions about user preferences
- Should return recommendations in this format:

Product: [name]
Price: [price]
Rating: [rating]
Why: [reason]

- Should recommend 2-3 products

Ask Bob for help:

Bob, create an agent that recommends products and always formats responses with Product, Price, Rating, and Why fields

Test prompts:
- "I need a laptop for programming"
- "Recommend a good book for learning Python"
- "What's a good gift for a tech enthusiast?"

Success criteria:
- Always uses the specified format
- Provides relevant recommendations
- Explains reasoning clearly

Exercise 5: Conversational Flow Agent (Advanced)¶

Goal: Create an agent that guides users through a multi-step process

Task: Create a "Survey Agent" that collects user feedback through a series of questions.

Requirements:
- Agent should introduce itself and explain the survey
- Should ask questions one at a time
- Should remember previous answers
- Should thank the user at the end

Questions to ask:
1. "How would you rate your experience? (1-5)"
2. "What did you like most?"
3. "What could be improved?"
4. "Would you recommend us to others? (Yes/No)"

Ask Bob for help:

Bob, create a survey agent that asks users 4 questions one at a time and remembers their answers

Test prompts:
- Start the conversation and go through all questions
- Try to skip questions
- Try to go back and change an answer

Success criteria:
- Asks questions in order
- Remembers all answers
- Handles edge cases (skipping, changing answers)
- Provides a summary at the end

Bonus Challenge: Debug This Agent¶

Goal: Practice debugging agent issues

Task: This agent has several problems. Find and fix them!

kind: native
name: broken-agent
description: An agent that doesn't work properly
instructions: |
  You are a helpful assistant.

llm: invalid-model-name
config:
  hidden: true

Problems to find:
1. Instructions are too vague
2. Invalid model name
3. Agent is hidden (won't show in UI)
4. Missing description details

Ask Bob for help:

Bob, review this agent YAML and identify all the problems: [paste YAML]

Fix the agent and test it!

Reflection Questions¶

After completing the exercises, consider:

What makes good agent instructions?
Specific role definition
Clear capabilities and limitations
Behavior guidelines
Output format requirements
How do you test agents effectively?
Use diverse test prompts
Test edge cases
Verify consistency
Check error handling
When should you create a new agent vs. modifying existing ones?
New domain/purpose → new agent
Refinement/improvement → modify existing
Different personality → could be either
How can Bob help you build better agents?
Generate initial YAML
Review and improve instructions
Debug issues
Suggest test cases

Next Steps¶

Once you've completed these exercises, you're ready to learn how to customize Bob's behavior with custom rules!

Continue to Part 2b: Bob Custom Rules →