e2e

name	e2e
description	End-to-end testing workflow for web applications using Playwright or Cypress. Covers user flow discovery, test design, implementation, execution, and flaky test management.
category	process
triggers	["end to end test","browser test","user flow test","playwright test","cypress test"]

E2E

Purpose: End-to-end testing for web applications Phases: Setup → Discover → Design → Implement → Run → Report Usage: /e2e [scope flags] <description of what to test>

Iron Laws

1. TEST USER FLOWS, NOT IMPLEMENTATION — E2E tests should mirror real user behavior, not internal APIs. If the user would not do it, the test should not do it.
2. EVERY TEST MUST CLEAN UP AFTER ITSELF — No test should depend on state from another test. Each test starts from a known state and leaves no residue.
3. FLAKY TESTS ARE BROKEN TESTS — A test that sometimes passes and sometimes fails is not acceptable. Fix it or delete it. Never ignore, never retry-and-hope.

When to Use

• Critical user flows (signup, login, checkout)
• Authentication and authorization flows
• Form submissions with validation
• Multi-page workflows and wizards
• Checkout and payment processes
• Cross-browser compatibility verification

When NOT to Use

• Unit testing individual functions → /test-coverage or /tdd
• API testing without a browser → /api-test
• Visual design review → /review
• Performance benchmarking → use dedicated profiling tools
• Testing third-party services directly → mock them instead

Never Do

• Never use CSS selectors for test targeting — Use data-testid or role-based selectors. CSS classes change with styling; test anchors must be stable.
• Never use fixed sleep/wait times — Use the framework's built-in waiting mechanisms (waitForSelector, waitForNavigation, Cypress auto-retry). setTimeout in tests is a flakiness factory.
• Never test third-party services in E2E — Mock external APIs. Your tests should not fail because Stripe's sandbox is down.
• Never write E2E tests for every edge case — That is what unit tests are for. E2E tests cover critical paths and integration points.
• Never share mutable state between tests — Each test is an island. Shared state creates ordering dependencies and mystery failures.

Gate Enforcement

See ai-assistant-protocol for valid approval terms and invalid responses.

Scope Flags

Flag	Description
--framework=<name>	Testing framework: playwright or cypress
--files=<paths>	Limit scope to specific test files or app files
--flow=<name>	Target a specific user flow (e.g., login, checkout)

Examples:

/e2e --framework=playwright login and signup flows
/e2e --flow=checkout verify the full purchase flow
/e2e --files=e2e/auth/ fix flaky authentication tests

---

Phase 1: Setup

Mode: Read-only investigation — verify testing infrastructure.

Step 1.1: Detect Package Manager

Detect the package manager from the project's lockfile. Check in this order:

Lockfile	Package Manager	Run command
pnpm-lock.yaml	pnpm	pnpm exec
yarn.lock	yarn	yarn
bun.lockb	bun	bun / bunx
package-lock.json	npm	npx

Use the detected package manager for all commands throughout this workflow (installs, running tests, starting dev server). Do not mix package managers.

Step 1.2: Detect Framework

cat package.json | grep -E "playwright|cypress"
ls playwright.config.* cypress.config.* 2>/dev/null

• Check package.json for @playwright/test or cypress
• If neither detected, recommend Playwright and offer to scaffold
• Verify test config exists (playwright.config.ts, cypress.config.ts)

Step 1.3: Verify Browser Installation

Verify browser binaries are installed for the testing framework:

• Playwright: Check if browsers are available by running npx playwright install --dry-run or checking the Playwright cache. If missing, install the project's configured browsers (or chromium as a default):

  # Using detected package manager (example: pnpm)
  pnpm exec playwright install chromium
  # Or install all configured browsers:
  pnpm exec playwright install
  

• Cypress: Cypress bundles its browser, but verify with npx cypress verify. If not verified, run npx cypress install.

Do not skip this step. Browser binaries are often missing in fresh clones, CI environments, and containers.

Step 1.4: Verify Dev Server

Check if the dev server is needed and running:

1. Check framework config for server management:
   • Playwright: look for webServer in playwright.config.ts
   • Cypress: look for baseUrl in cypress.config.ts and any start scripts
2. If webServer is configured in Playwright config — the framework handles starting/stopping the server automatically. Note this and proceed.
3. If webServer is NOT configured:
   • Check if a server is already running on the expected port (e.g., curl -s http://localhost:3000 > /dev/null or similar)
   • If not running, either: (a) add a webServer block to the Playwright/Cypress config so the framework manages it, or (b) start the server and document how to stop it
4. For Cypress without baseUrl: set baseUrl in the config to avoid hardcoded URLs in tests.

Tests must not fail due to ECONNREFUSED because no server is listening.

Step 1.5: Verify Configuration

## E2E Setup

| Check | Status |
|-------|--------|
| Package manager | [pnpm/yarn/bun/npm] |
| Framework | [Playwright/Cypress/None] |
| Config file | [Found/Missing] |
| Base URL | [configured/missing] |
| Test directory | [path or missing] |
| Browsers installed | [yes/no → installed] |
| Dev server | [webServer configured / running on port X / needs setup] |

If setup is incomplete, offer to scaffold before proceeding.

Step 1.6: Parse Scope

git branch --show-current
git status --porcelain

Identify target scope from flags and description.

---

Phase 2: Discover

Mode: Read-only — identify what needs testing.

Step 2.1: Identify User Flows

Examine the application to map critical user flows:

# Find route definitions
grep -rn "path=" src/ --include="*.tsx" --include="*.ts"
grep -rn "Route" src/ --include="*.tsx" --include="*.ts"

# Find page components
find src -name "page.*" -o -name "Page.*" | head -20

Step 2.2: Review Existing Tests

# Check what is already covered
find e2e tests/e2e cypress/e2e -name "*.spec.*" -o -name "*.test.*" 2>/dev/null

Step 2.3: Present Flow Inventory

## Flow Inventory

| Flow | Pages | Existing Tests | Priority |
|------|-------|----------------|----------|
| Login | /login → /dashboard | 0 | High |
| Signup | /signup → /verify → /dashboard | 0 | High |
| Checkout | /cart → /shipping → /payment → /confirm | 0 | Critical |

Confirm these flows? (yes / modify)

GATE: Wait for confirmation before designing tests.

---

Phase 3: Design

Mode: Read-only — plan test scenarios for each flow.

Step 3.1: Design Test Scenarios

For each approved flow, design scenarios covering:

• Happy path: Standard successful flow
• Error states: Validation errors, network failures, unauthorized access
• Edge cases: Empty states, long inputs, special characters, back-button navigation

Step 3.2: Define Page Objects (if needed)

For flows spanning multiple pages, outline Page Object Model structure:

## Page Objects

- `LoginPage` — email input, password input, submit button, error message
- `DashboardPage` — user greeting, navigation menu, logout button

Step 3.3: Present Test Design

## Test Design: [Flow Name]

### Happy Path
1. Navigate to /login
2. Fill email and password
3. Click Sign in
4. Verify redirect to /dashboard
5. Verify user greeting visible

### Error: Invalid Credentials
1. Navigate to /login
2. Fill invalid credentials
3. Click Sign in
4. Verify error message displayed
5. Verify still on /login

### Edge: Empty Form Submission
1. Navigate to /login
2. Click Sign in without filling fields
3. Verify validation messages

---
**Approve test design?** (yes / no / modify)

GATE: User must approve test design before implementing.

---

Phase 4: Implement

Mode: Full access — create test files.

Reference: See references/e2e-patterns.md for page object patterns, selector strategies, and common flow templates. See references/e2e-flaky-tests.md for flaky test prevention and mitigation.

Step 4.1: Create Test Files

Follow project conventions for file naming and directory structure. Use the approved test design as the blueprint.

Selector strategy (in priority order):

1. Role-based: getByRole('button', { name: 'Sign in' })
2. Label: getByLabel('Email')
3. Test ID: getByTestId('submit-btn')
4. Text content: getByText('Welcome')
5. CSS selector: last resort only

Step 4.2: Implement with Proper Waits

Use framework-native waiting, never setTimeout:

test.describe('User Authentication', () => {
  test('should allow login with valid credentials', async ({ page }) => {
    await page.goto('/login');
    await page.getByLabel('Email').fill('[e-mail]');
    await page.getByLabel('Password').fill('password123');
    await page.getByRole('button', { name: 'Sign in' }).click();
    await expect(page.getByRole('heading', { name: 'Dashboard' })).toBeVisible();
  });

  test('should show error for invalid credentials', async ({ page }) => {
    await page.goto('/login');
    await page.getByLabel('Email').fill('[e-mail]');
    await page.getByLabel('Password').fill('wrongpassword');
    await page.getByRole('button', { name: 'Sign in' }).click();
    await expect(page.getByText('Invalid email or password')).toBeVisible();
  });
});

Step 4.3: Add Page Objects (if designed)

// e2e/pages/login.page.ts
export class LoginPage {
  constructor(private page: Page) {}

  async goto() {
    await this.page.goto('/login');
  }

  async login(email: string, password: string) {
    await this.page.getByLabel('Email').fill(email);
    await this.page.getByLabel('Password').fill(password);
    await this.page.getByRole('button', { name: 'Sign in' }).click();
  }

  async expectError(message: string) {
    await expect(this.page.getByRole('alert')).toContainText(message);
  }
}

Step 4.4: Test Isolation

Ensure each test:

• Starts from a clean state (fresh page, no leftover data)
• Does not depend on test execution order
• Cleans up any data it creates (users, records, files)

Step 4.5: Handle Third-Party Iframes

If the flow involves third-party iframes (OAuth providers like Google/GitHub, payment widgets like Stripe, CAPTCHAs):

1. These cannot be tested end-to-end in most cases — third-party iframes block cross-origin automation and change without notice.
2. Mock the third-party at the API level — intercept the callback endpoint (e.g., /api/auth/callback) and return a valid session/token.
3. Test up to the redirect and after the callback — verify the app correctly initiates the OAuth flow (redirect URL, parameters) and correctly handles the callback response (session created, user redirected to dashboard).
4. Document the limitation in the test — add a comment explaining what is mocked and why.

// Example: mock OAuth callback for login flow
test('OAuth login redirects to dashboard', async ({ page }) => {
  // Mock the OAuth callback to return a valid session
  await page.route('**/api/auth/callback**', (route) => {
    route.fulfill({
      status: 302,
      headers: { Location: '/dashboard' },
    });
  });

  // Note: We cannot test the third-party OAuth provider's login page.
  // This test verifies the app handles the callback correctly.
  await page.goto('/api/auth/callback?code=mock-auth-code');
  await expect(page).toHaveURL('/dashboard');
});

---

Phase 5: Run

Mode: Execution — run tests and collect results.

Step 5.1: Run in Headless Mode

# Playwright
npx playwright test [file]

# Cypress
npx cypress run --spec [file]

Step 5.2: Handle Failures

If tests fail:

1. Review the error output and screenshots/traces
2. Offer headed mode for debugging:

# Playwright — headed with trace
npx playwright test [file] --headed --trace on

# Cypress — interactive
npx cypress open

3. Fix issues and re-run

Step 5.3: Escalation Rule

Attempt	Action
1st failure	Review error, fix obvious issues (selectors, timing)
2nd failure	Enable tracing/screenshots, inspect step-by-step
3rd failure	STOP. Likely a flaky test or app bug, not a test bug. Present findings to user.

---

Phase 6: Report

Mode: Summary — present results and recommendations.

Step 6.1: Results Summary

## E2E Test Results

| Flow | Tests | Passed | Failed | Skipped |
|------|-------|--------|--------|---------|
| Login | 4 | 4 | 0 | 0 |
| Signup | 3 | 2 | 1 | 0 |
| Checkout | 5 | 5 | 0 | 0 |

**Total:** 12 tests, 11 passed, 1 failed

Step 6.2: Failure Details

For each failure, include:

• Test name and file
• Error message
• Screenshot or trace link (if available)
• Likely cause and suggested fix

Step 6.3: Coverage and Recommendations

## Coverage by Flow

| Flow | Happy Path | Errors | Edge Cases |
|------|------------|--------|------------|
| Login | Covered | Covered | Partial |
| Signup | Covered | Missing | Missing |

## Recommendations
- [ ] Add signup error handling tests
- [ ] Add edge case tests for login (special characters in email)
- [ ] Consider adding visual regression tests for critical pages

Step 6.4: Commit

## Ready to Commit

**Files created/changed:**
- `e2e/auth/login.spec.ts` — Login flow tests
- `e2e/pages/login.page.ts` — Login page object

**Message:**

test(e2e): add login flow end-to-end tests

Covers happy path, invalid credentials, and empty form submission. Uses Page Object Model for maintainability.

**Commit?** (yes / no / edit)

GATE: User must approve before committing.

---

Handling Flaky Tests

See references/e2e-flaky-tests.md for a deep guide on diagnosis, prevention patterns, and the flaky test decision tree.

Identify

A test is flaky if:

• It passes locally but fails in CI
• It passes sometimes and fails other times with no code changes
• It fails only when run with other tests but passes in isolation

Common Causes

Cause	Symptom	Fix
Timing issues	Element not found, timeout	Add proper waits (waitForSelector, expect().toBeVisible())
Animation interference	Click on wrong element, element moving	Disable animations in test mode
Network dependency	Intermittent timeout, connection refused	Mock all external API calls
Test ordering	Passes alone, fails in suite	Isolate state, reset between tests
Shared state	Random data appearing in assertions	Each test creates its own data
Race condition	Inconsistent assertion failures	Wait for specific conditions, not time

Decision

Fix or delete. Never ignore.

• If the flaky test covers a critical flow: fix it (proper waits, mocking, isolation)
• If the flaky test covers a non-critical edge case: delete it and document why
• Never add retry logic as a "fix" for flakiness — that masks the real problem

---

Acceptance Tests

ID	Type	Prompt / Condition	Expected
E2E-T1	Positive	"Write end-to-end tests for the login flow"	Skill triggers
E2E-T2	Positive	"Add Playwright tests for checkout"	Skill triggers
E2E-T3	Positive	"Browser test the signup flow"	Skill triggers
E2E-T4	Negative	"Write unit tests for this function"	Does NOT trigger (-> /test-coverage or /tdd)
E2E-T5	Negative	"Test the API endpoints"	Does NOT trigger (-> /api-test)
E2E-T6	Negative	"Fix the flaky test"	Does NOT trigger (-> /debug)
E2E-T7	Boundary	"Test the full user flow"	Triggers if referring to browser-based user flow

Quick Reference

Phase	Mode	Gate
1. Setup	Read-only	Framework detected and configured
2. Discover	Read-only	User confirms flow inventory
3. Design	Read-only	User approves test design
4. Implement	Full access	Tests written per approved design
5. Run	Execution	All tests pass (or failures triaged)
6. Report	Summary	User approves before commit