Test Design Skill#

A structured approach to designing tests that give confidence without excessive maintenance burden.

Core Philosophy#

Test behavior, not implementation. Tests that couple to internal implementation details become fragile and refactor-resistant. Aim to test what the code does, not how it does it.

The best test is the one that fails when something breaks, and passes when everything works. Perfect coverage means nothing if tests don't catch real bugs.

Testing Pyramid#

text

        ┌─────────┐
        │   E2E   │  ← Few, slow, high confidence
       ┌┴─────────┴┐
       │ Integration│ ← Some, medium speed
      ┌┴───────────┴┐
      │    Unit     │ ← Many, fast, granular
      └─────────────┘

Unit tests form the base — fast, isolated, cover individual functions. Integration tests verify components work together. E2E tests validate complete user journeys.

Test Case Design#

Arrange-Act-Assert (AAA) Pattern#

python

def test_withdraw_amount_exceeds_balance():
    # Arrange
    account = Account(balance=Decimal("100.00"))

    # Act
    result = account.withdraw(Decimal("150.00"))

    # Assert
    assert result.is_failed()
    assert account.balance == Decimal("100.00")  # unchanged

Given-When-Then (GWT) Pattern#

gherkin

Feature: Account Withdrawal

  Scenario: Withdrawal amount exceeds balance
    Given an account with balance $100
    When I withdraw $150
    Then the withdrawal is rejected
    And the balance remains $100

Test Categories#

🔴 Happy Path Tests#

• Verify the main success scenario works
• At least one per feature/function

🟡 Edge Case Tests#

• Boundary values (0, -1, max integer)
• Empty/null inputs
• Rare but possible conditions

🟡 Error Case Tests#

• Invalid inputs
• Missing dependencies
• Permission denied scenarios

🟢 Edge/Extreme Cases#

• Very large inputs
• Very small inputs
• Unicode/special characters
• Concurrent access

Test Doubles (Mocks/Stubs/Fakes)#

TypePurposeUse WhenDummyFill parameter listsNever actually usedStubProvide predetermined responsesTest requires specific dataSpyRecord how something was calledVerify interactionsMockPre-programmed expectationsVerify behavior + interactionsFakeWorking implementation (simplified)Real impl too slow/complex

Rule: Prefer real objects over mocks when practical. Mocks hide integration problems.

Test Naming Convention#

text

test_<unit>_<scenario>_<expected_result>

Examples:

• test_user_create_with_valid_input_succeeds
• test_user_create_with_duplicate_email_fails
• test_order_cancel_after_shipment_refunds_full_amount

Coverage Targets#

LayerTargetPurposeUnit70-80%Core business logicIntegration40-60%Key flows work togetherE2ECritical paths onlyUser journeys

Note: Coverage is a guide, not a goal. 100% coverage with shallow tests is worse than 70% with meaningful assertions.

Test Execution#

Unit Tests#

Run frequently during development (every save).

bash

# Go
go test ./...

# Python
pytest tests/unit/ -v

# TypeScript
npm test -- --coverage

Integration Tests#

Run before PR, after unit tests pass.

bash

# Start dependencies
docker-compose up -d

# Run integration suite
pytest tests/integration/ -v

E2E Tests#

Run on CI against deployed environment.

bash

# CI/CD pipeline
npm run test:e2e -- --specs=critical-paths

Test Data Management#

Strategies#

1. Factories/Fixtures — Create test data on demand
2. Seeded Data — Deterministic test datasets
3. Randomized Data — Fuzz testing with random inputs
4. Snapshot Testing — Verify serializable outputs

Best Practices#

• Tests should be independent (no shared state)
• Each test should clean up after itself
• Use meaningful data, not magic values
• Avoid test interdependencies

File Structure#

text

test-design/
├── SKILL.md
└── references/
    ├── test-case-templates.md
    ├── testing-pyramid.md
    ├── mocking-strategies.md
    ├── test-naming-conventions.md
    └── per-type/
        ├── unit-test-guide.md
        ├── integration-test-guide.md
        └── e2e-test-guide.md