Test Authoring Edho Ferdian

Test Authoring — Edho Ferdian Mode

You are helping someone write a unit or component test — choosing queries, structuring setup, mocking the network, testing a hook in isolation, asserting the right thing the first time. This skill is about craft at the point of authorship, not about judging a test after the fact.

---

The three-way boundary (read this before anything else)

This ecosystem has three other skills that sound adjacent to this one. They are not substitutes for each other — each answers a different question:

| Skill | Question it answers | When it runs | |---|---|---| | test-authoring-edho-ferdian (this skill) | How do I write this test well? Query choice, async handling, mocking, hook isolation, a11y assertions, snapshot discipline. | While the test is being written — during dev-kickoff's TEST stage, or any time someone is authoring a new unit/component test. | | code-review-edho-ferdian → references/test-quality-lens.md (TQ-01..06) | Is this already-written test actually good? Behavioral mapping, edge-case/error-path coverage, assertion strength, flakiness, isolation/naming, coverage-vs-behavior divergence. | After code exists, during review — judges tests that already exist, doesn't teach how to write new ones. | | dev-kickoff-edho-ferdian → references/test-design-checklist.md | Which edge-case categories and anti-patterns apply, regardless of stack? An 8-category checklist (null/undefined, empty, invalid types, boundary values, error paths, race conditions, large data, special characters) and 4 anti-patterns. Stack-agnostic — no React/Jest/pytest specifics. | Both by this skill (to decide which cases to author tests for) and by reviews (to check which cases got missed). It is the shared checklist, not owned by either side. | | e2e-testing-edho-ferdian | Does this full user journey work end-to-end in a real browser? Login → checkout, multi-page flows, anything touching money or auth. | Separate layer entirely — Playwright/Chrome-DevTools-driven, Page Object Model, journey mapping. Not unit or component level. |

In short:

• This skill = authoring. It teaches the mechanics of writing a good

test before it exists.

• test-quality-lens = reviewing. It judges a test that already exists,

with tool-backed evidence (run the suite, run coverage) before labeling anything High confidence.

• test-design-checklist = the shared checklist. Both authoring and

review should apply its 8 categories and 4 anti-patterns; it belongs to neither exclusively.

• e2e-testing = a different layer. Full user flows in a real browser,

not the unit/component tests this skill covers.

If someone asks "is this test any good?" about code that's already written, point them at code-review-edho-ferdian's test-quality-lens instead of this skill. If someone asks "does the whole login flow work?", point them at e2e-testing-edho-ferdian. This skill is for the moment of writing a new unit or component test.

> Evaluating an agent's reliability (pass@k / pass^k, model-based vs > code-based graders) is a different discipline from authoring software > tests and lives in gan-harness-edho-ferdian/references/evaluate-phase.md, > not here.

---

Core principle

Test what the user sees and does, not implementation details.

A good test:

• Renders the component with the same providers it has in production.
• Interacts via accessible queries (role, label) and userEvent.
• Asserts visible output and observable side effects (a callback fired, a

request was sent) — not internal state, not which hooks ran, not render count.

Apply the stack-agnostic edge-case categories from dev-kickoff-edho-ferdian/references/test-design-checklist.md when deciding which cases to write tests for — this skill covers how to write each one well once you've decided it belongs.

Baseline standards (stack-agnostic)

• references/baseline-testing-standards.md — the numbers and the loop:

the 80% coverage floor (and its three qualifications), the unit/ integration/E2E layer split, RED→GREEN→REFACTOR, Arrange-Act-Assert structure, behaviour-describing naming, the diagnose-in-order protocol for a failing test, and how to close a coverage gap deliberately rather than file-by-file. Read this before authoring in any stack — the stack-specific files below cover how, this file covers how much and in what order.

Stack-specific guidance

• React / Testing Library: references/react.md — query priority,

userEvent.setup() discipline, async assertions, MSW network mocking, renderHook for custom hooks (including the shared-QueryClient flake trap), jest-axe accessibility assertions, the RTL vs Playwright Component Testing vs full E2E decision boundary, and coverage expectations per layer.

• Regression tests (stack-agnostic): references/regression-testing.md —

writing the test a found bug earned, the four recurring regression patterns (dual-path shape drift, projection omission, error-state leakage, missing rollback), DB-free sandbox-mode testing, and why AI self-review is not evidence.

• Python/pytest: references/python-pytest.md — fixtures and scoping,

parametrize with ids, unittest.mock/pytest-mock (patch-site rule, autospec), pytest-asyncio async tests and AsyncMock, coverage configuration.

• Go: references/go.md — table-driven tests, t.Run subtests

(including the loop-variable capture trap), interface-based mocking, testify vs stdlib testing, benchmark tests.

• Vue: references/vue.md — Vue Test Utils mount strategies, testing

Composition API composables in isolation, Pinia store testing with createTestingPinia, Vitest async-update handling.

When a stack isn't covered above, apply the core principle and the stack-agnostic checklist above by hand, and note in your response that stack-specific guidance for that language isn't written yet — don't silently fall back to React-flavored advice for a non-React stack.

Language routing (fixed — see skill-authoring-edho-ferdian's canonical contract)

Communication to the user in Bahasa Indonesia; test code, test names, and assertions in English — fixed, never ask. Full contract: skill-authoring-edho-ferdian §7.

Source & license

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

• Author: edhoferdian
• Source: edhoferdian/EEF
• License: MIT

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