Session 03 Lab — Prompt Engineering Fundamentals

Duration: 2 hours
Difficulty: Beginner
Prerequisites: Sessions 01–02 completed
Deliverable: A prompt engineering cheat sheet + a fully prompted REST API endpoint

---

Lab Overview

Write structured prompts, build a REST API from comments, see how context affects suggestions, and create a cheat sheet for later sessions.

Exercise	Topic	Time
1	Prompt Challenge Rounds	40 min
2	Comment-Driven API Development	40 min
3	Context Manipulation Experiments	20 min
4	Prompt Cheat Sheet	20 min

Before you start

Read the course safety baseline. Confirm that Copilot Chat and inline suggestions are available.

If access is unavailable

Write the prompts in the supplied templates, implement the API manually, and have a peer assess each prompt against the acceptance criteria.

---

Exercise 1: Prompt Challenge Rounds (40 min)

Objective

Practice writing effective prompts under time pressure. For each challenge, write the best prompt you can, then compare with reference prompts.

Steps

1. Open lab/starter/prompt-challenges.md. The file has five timed challenges.

2. Set a timer for each challenge (8 minutes each). For each one:

3. Read the challenge description
4. Write a prompt in Copilot Chat (or as a code comment)
5. Evaluate the output against the requirements.
6. Iterate on your prompt if the first output isn't right

7. Record your prompt and the quality of the output (1–5)

8. Work through all 5 challenges:

#	Challenge	Time
1	Data validation function	8 min
2	Error handling wrapper	8 min
3	Unit test generation	8 min
4	Documentation generation	8 min
5	Algorithm implementation	8 min

1. After each challenge, reflect:
2. Was my first prompt good enough, or did I need to iterate?
3. What information was missing from my initial prompt?

4. How did I improve the prompt on subsequent attempts?

5. Compare with lab/solution/prompt-challenges-answers.md to see example prompts that meet the requirements.

Prompt Engineering Tips (apply these!)

Technique	Example
Be specific	"validate email format" NOT "check input"
Provide constraints	"max 20 lines, use regex, return bool"
Give examples	"e.g., 'user@test.com' → True, 'invalid' → False"
State the context	"This is a Flask route handler for a REST API"
Specify output format	"Return a dict with 'valid' (bool) and 'errors' (list)"

Troubleshooting

Problem	Solution
Time pressure feels too tight	Focus on getting a working prompt first, then refine. 8 minutes is generous once you practice.
Output doesn't match requirements	Your prompt is missing constraints. Re-read the challenge, identify what you left out.
Copilot generates a different approach	Evaluate whether it meets the requirements. The approach may differ from yours.

---

Exercise 2: Comment-Driven API Development (40 min)

Objective

Build a complete CRUD REST API for a todo list using ONLY comment-driven prompting. You'll write comments describing what you want, then let Copilot generate the code.

Steps

1. Open the starter project at lab/starter/api-starter/. Review the files:
2. package.json — project dependencies
3. server.js — minimal Express setup

4. routes/todos.js — this is where you'll work — it has comment prompts but no code

5. Install dependencies:

cd sessions/session-03-prompt-engineering/lab/starter/api-starter
npm install

1. Open routes/todos.js. You'll see comment blocks describing each endpoint. Your job: position your cursor after each comment block and let Copilot generate the implementation.

2. Build the API endpoint by endpoint. Follow this workflow for each one:

Step A: Read the comment prompt already in the file
Step B: Place your cursor on the blank line after the comment
Step C: Press Enter and wait for Copilot's suggestion
Step D: Review the suggestion — does it match the comment's requirements?
Step E: Accept (Tab) or refine the comment and try again

1. Endpoints to build (comments are already in the starter file):

#	Method	Route	Description
1	GET	/	List all todos, optional ?status= filter
2	GET	/:id	Get a single todo by ID
3	POST	/	Create a new todo
4	PUT	/:id	Update a todo
5	DELETE	/:id	Delete a todo

1. Test your API:

npm start

Then in another terminal:

# Create a todo
curl -X POST http://localhost:3000/todos \
  -H "Content-Type: application/json" \
  -d '{"title": "Learn prompting", "status": "pending"}'

# List all todos
curl http://localhost:3000/todos

# Get specific todo
curl http://localhost:3000/todos/1

# Update a todo
curl -X PUT http://localhost:3000/todos/1 \
  -H "Content-Type: application/json" \
  -d '{"status": "completed"}'

# Delete a todo
curl -X DELETE http://localhost:3000/todos/1

Expected: Each endpoint returns JSON responses with appropriate status codes (200, 201, 404, 400).

1. Improve your comments if needed. If Copilot's first suggestion isn't right:
2. Add more detail to the comment (constraints, edge cases, response format)
3. Add an example: // Example: GET /todos?status=completed → [{id: 1, ...}]

4. Specify error handling: // Return 404 with {error: "Todo not found"} if ID doesn't exist

5. Compare with lab/solution/api-complete/ to see the reference implementation.

Troubleshooting

Problem	Solution
Copilot doesn't generate anything	Make sure the comment is descriptive. Try: // GET / - Return all todos from the array as JSON
Generated code has errors	Check for missing imports or syntax issues. Copilot sometimes forgets middleware.
Port 3000 already in use	Change PORT in server.js, or kill the other process: lsof -ti:3000 \| xargs kill

---

Exercise 3: Context Manipulation Experiments (20 min)

Objective

Discover how opening and closing files in VS Code changes Copilot's suggestions. Copilot uses "neighboring tabs" as context.

Steps

1. Open lab/starter/context-experiment/models/user.js in VS Code. Read it — it defines a User model with specific field names and validation rules.

2. Open lab/starter/context-experiment/routes/users.js in a second tab. This file is nearly empty — just comment prompts.

3. Experiment 1: With context. With models/user.js open in a neighboring tab, place your cursor in routes/users.js after the first comment and let Copilot suggest code.

Expected: Copilot generates route handlers that reference the User model's fields (name, email, age) by name, and possibly imports the model file.

1. Close models/user.js (close the tab, don't just switch away).

2. Experiment 2: Without context. Delete the code Copilot generated. Position your cursor again and let Copilot suggest new code.

Expected: Without the model file open, Copilot generates more generic code — different field names, possibly a different data structure.

1. Document the difference. Note:
2. Did field names change?
3. Did the import/require change?

4. Was the validation logic different?

5. Experiment 3: Add middleware context. Now open middleware/auth.js alongside routes/users.js. Let Copilot suggest code for the protected routes.

Expected: With the auth middleware visible, Copilot is more likely to add authentication checks to routes.

1. Complete all route handlers with both models/user.js and middleware/auth.js open. Compare with lab/solution/context-experiment/.

What to remember

The files you have open can influence Copilot's suggestions. This is the "neighboring tabs" effect. Teach trainees to open relevant files before asking Copilot to generate code.

Troubleshooting

Problem	Solution
Can't see a difference	Try a more complex prompt. The effect is stronger when the model file has unique field names or patterns.
Copilot ignores open files	Copilot prioritizes the current file. Make sure the model has distinctive patterns (unique field names, custom validation).

---

Exercise 4: Prompt Cheat Sheet (20 min)

Objective

Create a personal prompt engineering cheat sheet for later labs.

Steps

1. Open lab/starter/cheat-sheet-template.md in VS Code.

2. Fill in each section based on what you learned in Exercises 1–3. The template covers:

Section	What to document
Inline suggestions	Best comment formats for triggering good completions
Chat prompts	Prompt structures that worked well in Session 02
Comment-driven dev	Patterns from Exercise 2 that produced clean code
Context tricks	What you learned in Exercise 3 about open files
Anti-patterns	Prompts that consistently produce bad results
Favorite prompts	Your top 5 go-to prompts

1. Include at least 3 concrete examples for each section. Write the actual prompt you would use instead of generic advice.

2. Add a "Trainer Tips" section with advice you'd give trainees about prompting.

3. Save your cheat sheet. You'll update this in future sessions as you learn more.

By the end: A filled-in cheat sheet with real examples. Keep it as a reference and share it with trainees if useful.

Troubleshooting

Problem	Solution
Don't know what to write	Go back to Exercise 1 and review which prompts worked best. Pull examples from your Chat history.
Cheat sheet feels generic	Add specific code examples, not just descriptions. Show the prompt AND the expected output.

---

Wrap-up

Deliverables Checklist

• prompt-challenges.md — All 5 challenges attempted with your prompts recorded
• api-starter/routes/todos.js — All 5 CRUD endpoints built via comment-driven prompting
• Context experiment notes — documented how open files affected suggestions
• cheat-sheet-template.md — Completed prompt engineering cheat sheet

Trainer notes

1. Prompt structure matters more than length. State the task, context, and constraints.
2. Comments can guide inline suggestions. Write behavior and boundaries before the code.
3. Choose context deliberately. Open the files that show the pattern you need.
4. Refine when the output misses a requirement. Name the missing behavior, then evaluate the next response.
5. Update the cheat sheet with working examples.

Next session

In Session 05, you will use agent mode for multi-step coding work. The same clear prompts and constraints apply.