mcpField
V
verification-and-testing

verification-and-testing

github.com/alexbrunoその他09 回インストール今日更新
リポジトリ

verification-and-testing について

タスクの完了を宣言する前に、証拠に基づく検証と有意義なテストカバレッジを徹底します。テストの作成、`workflow/SPEC.md` の受け入れ基準の検証、コード変更後のビルド、リンター、実行時の動作確認に使います。

インストール

コマンドをコピーしてターミナルで実行します。

$npx skills add https://github.com/alexbruno/dev-skills --skill verification-and-testing

インストール後、Claude Code で/verification-and-testingを入力して読み込まれたか確認します。

README(原文)English

プロジェクトの README より

この README の日本語訳はまだありません。以下は英語の原文です。日本語の紹介はページ上部をご覧ください。

Verification & Testing

Never claim a change works, passes, or is complete without fresh, direct verification evidence collected after the final edit.

Core Rule

Evidence before completion claims. If you did not run a check after your last change and observe its output, do not report it as passing.

1. Align Verification with workflow/SPEC.md

  • Map every acceptance criterion in workflow/SPEC.md and the verification strategy in workflow/PLAN.md to a concrete check (automated test, build, lint, typecheck, or explicit runtime check).
  • Treat untested acceptance criteria as incomplete work.

2. Write Meaningful Tests

  • Test behavior, not implementation details: Assert observable outputs, state transitions, and side effects so tests survive safe refactoring (code-simplification).
  • Cover the edges: Include boundary values, empty/null inputs, invalid states, and error paths—not just the happy path.
  • Keep tests isolated and deterministic: Avoid shared mutable state, real network calls, or time-dependent flakiness unless explicitly testing integration boundaries.
  • Never weaken assertions: Do not relax, skip, or mock away the core behavior under test just to make a suite green.

3. Execution Gate After Final Edit

Before handing off work or moving to workflow/REVIEW.md:

  1. Run the relevant unit/integration tests, linter, typechecker, and build commands.
  2. Confirm all executed checks pass with zero skipped or relaxed assertions.
  3. Report the exact scope of what was run (e.g., full suite vs. targeted subset, any environment limitations).

コメント