AI Prompt & Agent-Rules Pack

AI prompts for QA testers, plus agent rules for Cursor, Claude Code and Copilot

121 copy-paste prompts written for real QA tasks, from test case design and bug reports to Playwright code and flaky test triage. Plus 16 agent-rules files so the AI in your IDE writes tests the way a senior SDET would.

Buy on Gumroad, $12 See a free sample

$12 one-time · instant download · secure checkout through Gumroad

At a glance

AI Prompt & Agent-Rules Pack for QA Engineers

121 prompts
15 chapters
16 agent-rules files
66-page PDF

Test case design, edge cases, bug reports, test plans and API tests
Playwright, Selenium and Cypress code, debugging and flaky test triage
16 agent-rules files for Cursor, Claude Code, GitHub Copilot and AGENTS.md
66-page PDF guide, plus Markdown, TXT and a CSV prompt library

$12 one-time · instant download

Buy on Gumroad, $12

Secure checkout through Gumroad · TestLinkPro never sees your card details · instant download · no account needed. How buying works

What’s inside

121 AI prompts for QA testers, in 15 chapters

A vague prompt gets a generic answer: five happy-path test cases and an expected result that says “should work correctly”. Each prompt in this pack tells the AI which test design technique to apply, what format to return and what it must not invent, so you get output you can paste into your test management tool, bug tracker or codebase.

Start Here: Context & Session Setup (5 prompts)Set up the AI once with your product, stack and conventions, and anonymise logs before you share them.
Test Case Design from Requirements & User Stories (12 prompts)User story → test cases, Gherkin, equivalence partitions and boundaries, decision tables, state transitions, pairwise, traceability gaps, exploratory charters.
Edge, Negative & Boundary Cases (10 prompts)Forms, limits, dates and time zones, concurrency, Unicode, third-party failures, mobile interruptions.
Bug Reports (8 prompts)Rough notes → complete report, severity/priority, log analysis, duplicates, intermittent bugs.
Test Plans & Strategy (9 prompts)One-page plans, risk-based strategy, entry/exit criteria, estimation, UAT, status reports.
API Testing (11 prompts)OpenAPI → test cases, Playwright API tests, REST Assured, pytest, Postman/Bruno, contract testing, GraphQL, auth.
UI Automation: Playwright, Selenium & Cypress (12 prompts)Manual steps → code, page objects, fixtures and auth, framework migration, visual and accessibility checks, AI agents with guardrails.
Debugging Failing Scripts (6 prompts)Error diagnosis, broken locators, traces, CI-only failures, timeouts, test code review.
Flaky Test Triage (6 prompts)Root-cause classification, CI history analysis, removing hard waits, parallel data collisions, quarantine policy.
Test Data Generation (8 prompts)Realistic data sets, invalid data, SQL setup/cleanup and validation, factories, LLM evaluation data, mocks.
Regression Scoping (7 prompts)Scope from a change or a git diff, suite design, upgrade impact, smoke suites.
Release Notes, Risk Review & Go/No-Go (7 prompts)Release notes, risk assessment, go/no-go, known issues, monitoring, escaped-defect retros.
Accessibility Checks (7 prompts)WCAG 2.2 AA checklists, keyboard and screen reader scripts, axe results → bugs, contrast.
Security Checks for QA (7 prompts)OWASP Top 10:2025-aligned test ideas, access control matrix, sessions, safe input checks, file uploads.
Maintaining Automation & CI (6 prompts)Refactoring, GitHub Actions pipelines, speeding up suites, framework docs, what to automate next.

Every prompt has a “use when” line and clear {{PLACEHOLDERS}}. 24 prompts include an example output so you know what good looks like.

16 agent-rules files: Cursor rules, CLAUDE.md, Copilot instructions and AGENTS.md

Drop these into your repository so the AI in your IDE writes test automation the way a senior SDET would: stable locators, no hard waits, independent tests and assertions that actually fail when the feature breaks.

Cursor (7 rules)

Project rules in .cursor/rules/*.mdc: core standards (always on), Playwright TypeScript, Cypress, Selenium (Java/Python), API tests, flaky-test triage, and a manual @test-review rule.

Claude Code (4 files)

A CLAUDE.md template plus 3 skills that add the /new-test, /triage-failure and /review-tests commands.

GitHub Copilot (4 files)

copilot-instructions.md, 2 path-specific .instructions.md files (Playwright tests and API tests) and a reusable prompt file for triaging a failing test.

AGENTS.md (1 file)

For other coding agents that read AGENTS.md, plus an install README for all the rule files.

Contents page of the AI Prompt and Agent-Rules Pack PDF listing the 15 chapters — The contents page of the 66-page PDF guide.

Preview of the agent-rules files for Cursor, Claude Code, GitHub Copilot and AGENTS.md — The agent-rules files, ready to copy into your repository.

Formats: a 66-page PDF guide, Markdown files (one per chapter plus all-in-one), a plain TXT file, a CSV prompt library you can import into Google Sheets, Excel or Notion, and a Quick Start PDF.

Free sample

Try one prompt and one rule for free

Below is prompt TCD-01 from chapter 2, exactly as it appears in the pack, with its example output. Copy it into ChatGPT, Claude, Gemini or Copilot and replace the placeholders.

TCD-01 · User story → structured test cases

Use when: you have a user story with acceptance criteria and need a first set of test cases quickly.

Prompt

Write test cases for the user story below.

<story>
{{USER_STORY_AND_ACCEPTANCE_CRITERIA}}
</story>

Requirements:
- Cover every acceptance criterion with at least one positive and one negative test.
- Add edge cases the criteria don't mention but a real user would hit.
- Output a table with columns: ID (TC-{{MODULE}}-001...), Title, Type (Positive/Negative/Edge), Preconditions, Steps (numbered), Test Data, Expected Result, Priority (Critical/High/Medium/Low), AC covered.
- Expected results must be observable: exact messages, URLs, status codes, DB values.
- After the table, list any acceptance criteria that are ambiguous or untestable and suggest a rewrite.

Example output (excerpt)

| ID | Title | Type | Preconditions | Steps | Test Data | Expected Result | Priority | AC |
|---|---|---|---|---|---|---|---|---|
| TC-RESET-001 | Reset link sent to registered email | Positive | User exists | 1. Open /forgot-password 2. Enter email 3. Click "Send link" | ana@example.com | Message "If an account exists, we've sent a link" shown; email received within 2 min | High | AC1 |
| TC-RESET-004 | Unregistered email gets the same message | Negative | No account for email | Same as TC-RESET-001 | nobody@example.com | Same generic message as TC-RESET-001; no email sent | High | AC1 |
| TC-RESET-007 | Reset link expires after 30 min | Edge | Link generated 31 min ago | Open the link | - | "This link has expired" page with option to request a new one | High | AC3 |

Ambiguous: AC2 "link should be secure" → Suggested rewrite: "Link contains a single-use random token, expires after 30 minutes and becomes invalid after a successful reset."

Example outputs use fictional data. Your results will differ, so always review AI output before you use it.

Cursor rules for Playwright, Cypress and Selenium: the golden rules

From qa-automation-core.mdc, the always-on Cursor rule that the framework-specific rules build on. The same standards are in the Claude Code, Copilot and AGENTS.md files.

A test must fail when the feature is broken. Never weaken, delete or skip an assertion just to make a test pass.
If a failure looks like a real product bug, say so and stop. Do not "fix" the test to match broken behaviour.
No hard waits (waitForTimeout, sleep, Thread.sleep, cy.wait(<number>)). Wait for a condition instead.
Every test is independent: it creates the data it needs and does not rely on other tests or on execution order.
No secrets, tokens, passwords or real personal data in code. Read them from environment variables or the config layer.
Keep tests deterministic: control time, randomness and external services where needed.

Sample prompt cards from the AI Prompt Pack PDF with use-when lines, placeholders and example output — Prompt cards in the PDF guide.

How it works

How to use the prompts

Set the context once
Start a new chat with a prompt from chapter 1 (for example CTX-01), filled in with your product and stack. Every later answer gets more specific.
Pick a task prompt and fill it in
Copy it from the PDF, Markdown or CSV library and replace every {{PLACEHOLDER}}. Paste long material between the tags provided.
Review, then iterate
Treat the output as a strong first draft. Check it, run generated code, and keep a human in charge of Pass/Fail and release decisions.

Who it’s for

For testers, SDETs and developers who use AI

Manual testers who want to design better test cases faster.
Automation engineers and SDETs using Playwright, Selenium or Cypress who want AI-generated code they don’t have to rewrite.
QA leads who write plans, strategies, status reports and go/no-go recommendations.
Developers who write their own tests and want their AI assistant to follow good testing practice.

Built on real test design

The prompts use real techniques (boundary value analysis, decision tables, state transitions, pairwise and risk-based testing) and include responsible-use guidance: anonymise data first, verify outputs, and never let AI “fix” a test by hiding a real bug.

Current for 2026 tools and standards: Playwright test agents, Cypress cy.prompt, WCAG 2.2 and OWASP Top 10:2025.

Not affiliated with OpenAI, Anthropic, Google, Microsoft, GitHub, Cursor or any tool vendor.

Built by Kamran Ghaffar, a Quality Assurance professional with 10+ years of experience in software quality, test automation and QA leadership.

FAQ

Questions about the AI Prompt & Agent-Rules Pack for QA Engineers

Which AI tools does it work with?

Any capable assistant: ChatGPT, Claude, Gemini, Microsoft Copilot, or the chat panel in VS Code, Cursor or JetBrains IDEs. The agent-rules files are for Cursor, Claude Code, GitHub Copilot and tools that read AGENTS.md.

Do I need a paid AI subscription?

No. The prompts work on free plans too. Long inputs (big requirement documents, large code files) work better with models that have larger context windows, which are often on paid plans.

I only use Selenium (or Cypress). Is it still useful?

Yes. The UI automation and debugging prompts cover Playwright, Selenium (Java and Python) and Cypress, and there are separate Cursor rules for each. Most of the other chapters (test design, bugs, plans, API, data, regression, accessibility, security) don’t depend on any framework.

Is it safe to paste company data into AI tools?

Follow your employer’s AI policy. The pack includes a prompt to anonymise logs and tickets before you share them, and guidance on what never to paste (credentials, customer data, confidential code in unapproved tools).

Will AI replace my testing?

No, and the prompts don’t pretend it will. They speed up the drafting so you have more time for exploration and risk judgement. Always review the output.

Can I use it at work or for clients?

Yes. Use the prompts and add the rule files to your company’s or clients’ repositories. You can’t resell or redistribute the pack itself (see LICENSE.txt in the download).

Do I get updates?

Major updates or new versions of the pack may be released as a separate paid upgrade.