Capstone: a five-test suite, your quiz, and your certificate

Twenty days ago your suite did not exist; today you finish it, prove it, and decide where it goes next. Two artefacts first, then the assessment.

The standards file

Everything this course taught you works only if every future test follows it, including tests written by teammates and by AI agents. The way that happens is a standards file at the root of the test repo, named AGENTS.md or CLAUDE.md, that every agent reads before writing a line. The core of mine, genericised from a real suite:

- Specs never call raw framework commands. Every
  interaction goes through a named page-object action.
- All locators live in page objects, never in specs.
- Stable locators only: role, label, data-testid.
  Never positional XPath or nth-child chains.
- No fixed-time sleeps, ever. Wait on a request,
  a visible element, or a content assertion.
- Every test stands alone and passes in any order.
- Known product bugs are quarantined with a ticket,
  never deleted and never "fixed" in the test.
- Test names are sentences: verify that <condition>
  <expected result>.

Create AGENTS.md in your project and put those rules in it, edited to your taste. When a new test arrives in review, the question is simple: does the diff break any line in this file. My full, downloadable standards files for Playwright, Cypress, and WebdriverIO live at AI Test Automation Standards.

The capstone

Five tests against Tumble Kitchen, in your existing project, held to the standard above:

1. The sign-in boundary loop from Day 10. One table, six generated tests, both edges of the 8 to 64 rule.
2. Search with an exact price from Day 8. Count, name, price, and the store’s own status line.
3. The cart count test from Day 12. With its quarantined sibling still carrying the ticket note.
4. The API price match from Day 11. The feed and the card agreeing on the plantain plate.
5. The discount boundary red from Day 16. Still red, still telling the truth. On a team you would quarantine it with a ticket like the cart test; here it stays red so anyone reading your work sees a suite that caught a real bug.

The rubric

Score your own work before you call it done. Every line is something we built together:

• Every locator lives in the page object. Zero locators in spec files.
• Every interaction is a named user action: addToOrder, applyDiscount, searchAndWaitForResults.
• Not one fixed sleep anywhere. Anything that waits, waits on a request or an element.
• Every test name is a sentence starting with “verify that”, readable by a non-coder.
• Every test passes alone, in any order, twice in a row with the same answer.
• The suite is honest: greens that prove promises, one truthful red pointing at a real bug, one quarantined test carrying its ticket.
• AGENTS.md sits at the repo root, and the suite passes its own rules.

If all seven hold, you have a portfolio piece, and you can talk through every line of it because you typed every line of it.

The assessment and reviewed certificate

Your final email contains a private code tied to this course and email address. The assessment has 30 scenario questions, with six in each competency section. You need 24 correct overall and at least four correct in every section.

• The questions are timed: 45 minutes, with the countdown on screen.
• Expect more than one format: pick one, pick two, put steps in order, and read the evidence before you answer.
• The practical work is untimed. Prepare it before you start the questions.

You also complete three applied responses and submit the five-test Playwright practical, run evidence, failure diagnosis, CI result, and automation strategy. I review the work against the published rubric. The certificate is issued only after approval.

Unlock the assessment with your final-email code

If a question trips you, use the objective-level review link to revisit the relevant lesson before the next attempt opens.

Where this goes next

AI agents can now draft tests like the ones you just wrote, in seconds, and most of what they draft should not be merged. Knowing the difference is the next skill, and everything this course trained you to check is exactly what you will be checking in an agent’s work. My AI in QA course picks up right there, with the same store and your same suite.

Thank you for spending these three weeks with me. If you or your team wants help doing this on real work, my work with me page is open. Otherwise, my newsletter goes out every week, and it is where the practical stuff lands first. See you there.