Agent Skills: Visual Testing Philosophy

Visual Testing Philosophy

When helping users add or improve visual tests, follow these principles and best practices.

Core Philosophy

Visual tests capture the REAL thing.

Unlike component libraries that render in isolation, Vizzly captures screenshots from your actual functional tests - the same tests that verify your app works. This means:

• You're testing what users actually see
• Integration issues are caught (not just isolated components)
• Visual tests are a byproduct of existing tests, not extra work

When to Add Visual Tests

High-value screenshot opportunities:

1. After page navigations - Full page renders after goto() or routing changes
2. After user interactions - State changes from clicks, form submissions, etc.
3. Critical user flows - Checkout, login, onboarding, settings
4. Different viewport sizes - Desktop, tablet, mobile breakpoints
5. Different user states - Logged in/out, empty/full data, error states

Lower value (consider skipping):

• Loading spinners (inherently unstable)
• Animations mid-transition
• Highly dynamic content (dates, random data)
• Third-party content you don't control

Screenshot Naming Conventions

Good names are descriptive, consistent, and sortable.

Pattern: <feature>-<state>-<variant>

Good examples:

homepage-desktop
homepage-mobile
login-form-empty
login-form-filled
login-form-error
checkout-cart-with-items
checkout-cart-empty
user-profile-settings-tab

Avoid:

test1                    # Not descriptive
page                     # Too generic
myComponent              # CamelCase inconsistent
home_page_after_login    # Underscores (use hyphens)

Include viewport when testing responsive:

homepage-1920x1080
homepage-768x1024
homepage-375x667

Best Practices for Stable Screenshots

1. Wait for Stability

Capture screenshots after the page is fully loaded:

// Wait for network idle
await page.waitForLoadState('networkidle');

// Or wait for specific element
await page.waitForSelector('.main-content');

// Then screenshot
await vizzlyScreenshot('homepage', await page.screenshot());

2. Handle Dynamic Content

Mock or stabilize dynamic data:

// Mock dates
await page.evaluate(() => {
  Date.now = () => new Date('2024-01-01').getTime();
});

// Hide dynamic elements
await page.evaluate(() => {
  document.querySelector('.timestamp')?.remove();
});

3. Disable Animations

Prevent animation-related flakiness:

await page.addStyleTag({
  content: `
    *, *::before, *::after {
      animation-duration: 0s !important;
      transition-duration: 0s !important;
    }
  `
});

4. Use Meaningful Thresholds

Configure thresholds based on your needs:

// In vizzly.config.js
export default defineConfig({
  threshold: 0.1,  // 0.1% - strict, catches minor changes
  // threshold: 1.0,  // 1% - lenient, ignores anti-aliasing
  // threshold: 0,    // 0% - pixel-perfect (may be flaky)
});

Threshold guidelines:

• 0.05% - 0.1% - Most projects, catches real changes, ignores rendering noise
• 0.5% - 1% - If you have font rendering issues across environments
• 2%+ - Only for very dynamic UIs or when stability is a problem

What Makes a Good Visual Test

A good visual test:

• Captures a stable, representative state
• Has a descriptive name that explains what's being tested
• Runs consistently across environments
• Catches real visual regressions
• Doesn't flake on unimportant changes

A bad visual test:

• Captures loading states or animations
• Has vague naming ("test1", "page")
• Includes timestamps or random data
• Breaks on every run due to dynamic content
• Tests third-party widgets you don't control

Adding Screenshots to Existing Tests

When adding visual tests to an existing test suite:

1. Identify key states - What are the important visual states in this test?
2. Add screenshots after assertions - The test already waits for the right state
3. Use descriptive names - Match the test description
4. Run and review - Check the first screenshots look correct
5. Accept baselines - These become your reference

Example integration:

test('user can log in', async ({ page }) => {
  await page.goto('/login');
  await vizzlyScreenshot('login-form-empty', await page.screenshot());

  await page.fill('#email', 'user@example.com');
  await page.fill('#password', 'password');
  await vizzlyScreenshot('login-form-filled', await page.screenshot());

  await page.click('button[type="submit"]');
  await page.waitForURL('/dashboard');
  await vizzlyScreenshot('dashboard-after-login', await page.screenshot());
});

TDD Workflow

The visual TDD workflow:

1. Start TDD server: vizzly tdd start
2. Run tests in watch mode: Your test runner with --watch
3. Make changes - Code updates trigger test reruns
4. Review in dashboard - See visual changes at http://localhost:47392
5. Accept or fix - Accept intentional changes, fix regressions
6. Repeat - Iterate until visual tests pass

Common Pitfalls

Flaky tests:

• Usually caused by animations, loading states, or dynamic content
• Solution: Wait for stability, mock data, disable animations

Too many screenshots:

• Don't screenshot everything - focus on critical states
• Quality over quantity

Brittle baselines:

• If baselines break constantly, your thresholds may be too strict
• Or you're capturing unstable states

Ignoring failures:

• Don't blindly accept all changes
• Each failure is either a regression to fix or a change to consciously accept