Agent Test Suite Architect: Targeting 85%+ Branch Coverage with Playwright & Vitest, Measured, Not Estimated (2026)

The Illusion of Green Tests: Why Happy-Path AI Code Breaks in Production

AI coding assistants are good at generating feature code and quick happy-path unit tests. Production software fails on edge cases, race conditions, null parameters, and third-party API timeouts, none of which a happy-path test ever exercises.

The more common failure, though, isn't missing tests. It's tests that pass and fail for reasons that have nothing to do with the code: a suite that stayed green for weeks and then broke five times in one afternoon because a CSS class got renamed for an unrelated reason, or because CI got a little slower and a fixed 3-second wait ran out before the page actually finished loading.

The principle of high-assurance QA: tests must be deterministic, isolated from external network dependencies, structured with Arrange-Act-Assert, and query the DOM through user-facing accessibility roles, not implementation details that are free to change.

The 4 Pillars of Modern Automated Testing

1. Playwright Accessibility-First Locators

Never target volatile CSS class names or XPath strings. Query in this priority order instead:

1. page.getByRole('button', { name: 'Submit' })
2. page.getByLabel('Email address')
3. page.getByPlaceholder('dev@company.com')
4. page.getByText('Payment successful')
5. page.getByTestId('checkout-card'), fallback for genuinely non-semantic widgets

2. Web-First Auto-Waiting Assertions

Eliminate hardcoded delays. Playwright's web-first assertions poll the DOM automatically until the condition is met or the timeout expires:

// Bad (flaky)
await page.waitForTimeout(3000);
expect(await page.isVisible('.success-badge')).toBe(true);

// Good (deterministic auto-waiting)
await expect(page.getByRole('alert')).toHaveText('Payment confirmed');
await expect(page.getByRole('button', { name: 'Download Invoice' })).toBeVisible();

3. Deterministic Third-Party API Mocking

Don't let unit or integration tests make live HTTP calls to payment gateways or cloud providers. Mock external services at the network layer:

vi.mock('stripe', () => ({
  default: vi.fn().mockImplementation(() => ({
    checkout: {
      sessions: {
        create: vi.fn().mockResolvedValue({ id: 'cs_test_123', url: 'https://pay.stripe.com' })
      }
    }
  }))
}));

4. Polyglot Table-Driven Tests (Go / Python / Vitest)

Test boundary values, zero, negative, overflow, null, with parameterized, table-driven tests instead of one-off assertions:

func TestCalculateDiscount(t *testing.T) {
    tests := []struct {
        name    string
        amount  float64
        want    float64
        wantErr bool
    }{
        {"zero amount", 0, 0, false},
        {"negative amount", -10, 0, true},
        {"tier 1 discount", 100, 90, false},
    }
    for _, tt := range tests {
        t.Run(tt.name, func(t *testing.T) {
            got, err := CalculateDiscount(tt.amount)
            if (err != nil) != tt.wantErr || got != tt.want {
                t.Errorf("got %v (err: %v), want %v", got, err, tt.want)
            }
        })
    }
}

Automating Test Suite Generation with Agent Skills

The Agent Test Suite Architect skill audits your application logic and generates high-coverage, zero-flake test suites with a full Test Gap Analysis matrix:

"Using the agent-test-suite-architect skill, audit my billing service and write comprehensive Vitest tests with Stripe API mocks, targeting 85%+ branch coverage and reporting the actually measured result."

Frequently Asked Questions