Claude Code for QA engineers — 5 practical workflows

Claude Code is a CLI environment that gives the AI assistant context of your entire repository. For a QA engineer it's the most efficient way to generate tests, fix flaky scenarios and keep the framework consistent — without opening a browser.

This article is a practical guide with 5 workflows we actually use at our clients.

1. Setup: Claude Code for a QA project

Start by installing the official CLI from claude.com/claude-code. After claude login open the repo and run claude. On the first run, configure CLAUDE.md at the root with your project conventions:

# Test Automation — konvencie
- Framework: Cypress 14 + TypeScript
- Page Object pattern v cypress/support/pages/
- Žiadne hardcoded waits — používaj cy.intercept() alebo .should()
- data-qa atribúty pre selektory
- Test data cez factories (cypress/support/factories/)
- Mailgun pre email flows

Claude reads this file with every instruction. You save hours of explanation.

2. Generating a test from a user story

Classic case: analyst writes a user story in Jira, you need a test. In the Claude Code environment:

> Napíš Cypress test pre túto user story:
"Ako používateľ chcem pridať položku do košíka, prejsť na checkout
a dokončiť platbu cez Stripe. Po úspešnej platbe chcem vidieť
potvrdenie objednávky s číslom."

Pozri si existujúci paymentPage v cypress/support/pages/ a použij
konvencie ako v ostatných testoch v cypress/e2e/checkout/.

Claude reads existing files, understands the style guide, and generates a test that compiles and runs — not just looks right.

3. Refactoring flaky tests

Got 15 flaky tests and a week-long deadline? Point Claude at the list:

> Zanalyzuj cypress/e2e/checkout/payment.spec.ts.
Identifikuj anti-flaky patterns (hardcoded timeouts, chýbajúce
intercepty, DOM race conditions) a oprav ich. Zachovaj business
logiku nezmenenú.

In ~70% of cases the output is immediately usable. For the remaining 30%, a review round pays off — Claude sometimes replaces cy.wait() with cy.intercept() that catches the wrong request.

4. Pull request review

Add a sub-agent (file .claude/agents/qa-reviewer.md) with a system prompt:

Si QA reviewer. Skontroluj PR a upozorni na:
- duplikáty existujúcich testov
- hardcoded timeouts
- chýbajúce data-qa selektory
- testy bez negative scenárov
- porušenia Page Object patternu

Invoke it via /agents qa-reviewer on every PR. You get a comment the dev team drops straight into the review.

5. Auto-documentation of the test suite

Documentation is what QA teams love most — reading it, not writing it. Claude does it for you:

> Zgeneruj README.md pre cypress/e2e/checkout/.
Pre každý spec súbor vypíš: čo testuje, aké use-case-y pokrýva,
aké pre-conditions vyžaduje. Formát: markdown tabuľka.

The output is readable documentation you can show business stakeholders without them having to look at code.

What Claude Code doesn't solve

• It doesn't understand your business — you have to give it context via CLAUDE.md and examples.
• It can't actually run tests through your CI — but it can write a Jenkinsfile or GitHub Action.
• Occasionally hallucinates a selector — always validate that it runs.

Measurable gains

• Test-writing time from 3 h → ~50 min.
• PR reviews with the QA AI agent catch 40–60% of issues before the human reviewer.
• Junior onboarding to the framework: from 3 weeks down to 5 days (CLAUDE.md is the study material).

---

Want the same approach at your company? Get in touch — dohodneme 30-minute discovery call.