Part 2 — Optional Exercises¶
Complete these only after hello_world_agent passes the main checkpoint.
Exercise 1: Change the persona¶
Ask Bob to create agents/workshop_guide.yaml for an agent that:
- welcomes participants
- explains the nine mandatory parts
- answers in no more than four sentences
- says when it does not know an answer
Ask Bob to validate and import the file, then suggest prompts that test every requirement. Manual fallback:
orchestrate agents import -f agents/workshop_guide.yaml
orchestrate chat ask --agent-name workshop_guide
Success means the name uses snake_case, the YAML validates, and the response
follows the length and uncertainty instructions.
Exercise 2: Repair an agent¶
Ask Bob to identify the problems before changing anything:
spec_version: v1
kind: native
name: broken_agent
description: An agent that does not work properly
llm: invalid-model-name
instructions: |
Be helpful.
hidden: true
Expected observations:
- the model is not one of the workshop's available models
- the instructions do not define a useful role or boundaries
hidden: trueprevents normal discovery in the UI- the description is too vague for collaborator routing
After reviewing the diagnosis, ask Bob to fix the file, validate and import it, suggest one test for each change, and explain why the change was needed.
Exercise 3: Add starter prompts¶
Ask Bob to add two starter prompts to workshop_guide. Each prompt needs a
unique id, title, subtitle, and prompt:
starter_prompts:
prompts:
- id: explain-workshop
title: Explain the workshop
subtitle: See the learning path
prompt: What will I build in this workshop?
Ask Bob to create the second prompt, validate and re-import the agent, generate draft webchat configuration, and help you confirm that both prompts appear.