メニュー
Write tests that verify observable behavior (contract), not implementation details. Auto-invoked when writing or reviewing tests.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/growilabs/growi --skill essential-test-designコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
Organize and clean up specification documents after implementation completion. Removes implementation details while preserving essential context for future refactoring.
Review a task implementation against approved specs, task boundaries, and verification evidence. Use after an implementer finishes a task, after remediation, or before accepting a task as complete.
Validate feature-level integration after all tasks are implemented. Checks cross-task consistency, full test suite, and overall spec coverage.
SKILL.md を表示中
SOC 職業分類に基づく
| name | essential-test-design |
| description | Write tests that verify observable behavior (contract), not implementation details. Auto-invoked when writing or reviewing tests. |
Tests that are tightly coupled to implementation details cause two failures:
setTimeout was called)setTimeout to a delay() utility, spy breaks)Both undermine the purpose of testing: detecting regressions in behavior.
A test is "essential" when it:
Ask: "What does the caller of this function experience?" — test that.
// BAD: Tests implementation, not behavior
// Breaks if implementation changes from setTimeout to any other delay mechanism
const spy = vi.spyOn(global, 'setTimeout');
await exponentialBackoff(1);
expect(spy).(expect.(), );
// BAD: The "arrange" is set up only to make the "assert" trivially pass
// This is a self-fulfilling prophecy, not a meaningful test
vi.advanceTimersByTime(1000);
await promise;
// No assertion — "it didn't throw" is not a valuable test
// GOOD: Tests the observable contract
// "Does not resolve before the expected delay, resolves at the expected delay"
let resolved = false;
mailService.exponentialBackoff(1).then(() => { resolved = true });
await vi.advanceTimersByTimeAsync(999);
expect(resolved).toBe(false); // Catches: delay too short
await vi.advanceTimersByTimeAsync(1);
expect(resolved).toBe(true); // Catches: delay too long or hangs
When writing a test, ask these questions in order:
setTimeout to Bun.sleep() shouldn't break the testexpect(resolved).toBe(false) at 999ms would catch itUse fake timers + boundary assertions (as shown above).
Assert on output shape/values, not on which internal helper was called.
// BAD
const spy = vi.spyOn(utils, 'formatDate');
transform(input);
expect(spy).toHaveBeenCalled();
// GOOD
const result = transform(input);
expect(result.date).toBe('2026-01-01');
Mocking the boundary (API/DB) is acceptable — that IS the observable behavior.
// OK: The contract IS "sends an email via mailer"
expect(mockMailer.sendMail).toHaveBeenCalledWith(
expect.objectContaining({ to: 'user@example.com' })
);
Test the number of attempts and the final outcome, not the internal flow.
// GOOD: Contract = "retries N times, then fails with specific error"
mockMailer.sendMail.mockRejectedValue(new Error('fail'));
await expect(sendWithRetry(config, 3)).rejects.toThrow('failed after 3 attempts');
expect(mockMailer.sendMail).toHaveBeenCalledTimes(3);
A spec that asserts a codebase invariant (e.g. "no static import chain from a boot entrypoint reaches a heavy package") can rot silently: if its walk starts from a wrong or renamed root it traces nothing and passes vacuously — green forever, guarding nothing.
Real case: the no-eager-*-imports.spec.ts drift specs in apps/app each
shipped with mutation evidence; boot-rooted walks caught two real leak paths
(an admin route, a group-sync service) that module-rooted walks could not see.