Skip to content

🧪 Practice lab: become a prompt engineer

Welcome to the control room of MagicFridge. You have learned the theory, now you must apply it. These exercises put you in the shoes of a quality engineer facing GUS (our AI).

Exercise 1: the anatomy of a prompt 🦴

(Objective: structure an effective prompt - LO 2.1.1)

Situation: a junior tester sends you this prompt she used to test the "Add to cart" function:

"Test if it works when I add apples."

Your mission: this prompt is mediocre. Identify 3 critical components that are most sorely missing to make it a pro-level prompt.🧐

Critical analysis

Select the missing elements (mentally), then reveal the analysis to compare.

Reveal the Senior QA's analysis

This prompt is unusable by the AI because it lacks:

  1. The role (persona): the AI does not know if it should act as an end-user or a technical tester.
  2. The context: "If it works" is too vague. Are we talking about performance? Functionality? Security?
  3. The output format: the AI will probably answer with a trivial sentence. We want a structured format (Gherkin or table).

Corrected version: "Act as a QA tester (role). For the 'Cart' function (context), generate 3 negative test cases for adding products (instruction). Present the result as a table (format)."

Exercise 2: which technique for which task? 🛠️

(Objective: choose the right prompting technique - LO 2.2.5)

Task: for each situation encountered at MagicFridge, determine, by clicking on the correct answer, the most suitable technique.

Your turn

1. You want the AI to generate 50 test cases in strict Gherkin format, without syntax errors.

●
●
●

2. You want the AI to analyze a complex user story, identify business rules, and then derive tests.

●
●
●

3. You need to test the API security but you are stuck and don't know what to ask the AI.

●
●

Exercise 3: system or user? 🤖

(Objective: distinguish prompt types - LO 2.1.3)

Context: you are configuring the MagicFridge automated test tool via its API. You have two text areas to fill.

  • Zone A: "You are a sarcastic QA assistant who always answers in rhymes."
  • Zone B: "Generate a test for the 'Login' button."
Which one is the System Prompt?

It is Zone A.

  • Why? It defines the immutable behavior and personality of the AI.
  • Zone B is the user prompt: it is the specific instruction at time T.

Exercise 4: the ruthless evaluation 📉

(Objective: use evaluation metrics - LO 2.3.1)

The test: you asked GUS to generate SQL scripts to insert test users into the database. You ask for 10 users.

The result: GUS successfully generates 10 SQL queries (INSERT INTO...). However, when executing them, 4 queries fail because the generated email addresses lack an @ (database constraint violation).

Which metric is impacted?

  1. Diversity
  2. Accuracy / success rate
  3. Recall
See the answer

Answer: 2. Accuracy (and success rate).

The AI produced code that looks syntactically correct, but is functionally wrong in 40% of cases.

  • Corrective action: use few-shot prompting by giving a valid email example, or add an explicit constraint: "Emails must respect the regex format *@*.*".


Have you validated this chapter?

If these exercises helped you see things more clearly, a small coffee for the coach would be greatly appreciated! ☕