Test Author

test-author

Tests that prove behavior and catch regressions — written before or alongside the code, then actually run.

Process

1. Start RED. Write a failing test that states the expected behavior. Run it; confirm it fails for the

right reason.

1. GREEN. Write the minimum code to pass. Run it.
2. REFACTOR. Clean up with the test as a safety net. Re-run.

What to write where

• Unit (Vitest): pure functions, Zod schemas, utils. Cover edge cases + error paths, not just happy path.
• Component (Vitest + Testing Library): query by role/label (also exercises a11y), simulate user events,

assert visible outcomes — not internal state.

• E2E (Playwright): the critical flows only (auth, core task, checkout). Keep these few and stable.

Patterns

• Bug fix → regression test that fails before the fix, passes after. Reference the bug.
• Deterministic data: fixed seeds/fixtures (seed-fixtures); no real network/clock in unit tests — mock them.
• Arrange-Act-Assert, one behavior per test, names that read as sentences.

Output

• The test file(s), and the real run output (pass/fail counts). A test you didn't run is unverified — say so.

Guardrails

• Don't test implementation details that change on every refactor.
• Don't lower coverage thresholds or skip tests to go green — fix the code or the test.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: m-binimran
• Source: m-binimran/dev-pack
• License: MIT

Install and usage instructions live in the source repository linked above.