メニュー
How to write and run tests in this repo — unit vs integration vs e2e, the build tags and env vars each tier needs, running a single test, and the pre-commit gate. Use before writing a test, running the suite, or claiming work is done.
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/fjaeckel/ninerlog-api --skill testing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。
The OpenAPI-first workflow for this repo. Use whenever adding, removing, or changing an HTTP endpoint, request/response field, status code, query parameter, or auth requirement — and before touching anything under internal/api/generated/ or api-spec/openapi.yaml.
Database schema changes in this repo — creating golang-migrate migration pairs, naming and numbering rules, the reversible-down requirement, the test-DB init script, and repository/model follow-up. Use whenever adding a table, column, index, or constraint.
Keeping everything a user owns inside the JSON export, the JSON import and cloud backups. Use whenever adding a table that accumulates user rows, adding a field to one, or touching GET /exports/json, POST /imports/json or internal/service/cloudbackup — and when the classification test fails on a new table.
SKILL.md を表示中
| name | testing |
| description | How to write and run tests in this repo — unit vs integration vs e2e, the build tags and env vars each tier needs, running a single test, and the pre-commit gate. Use before writing a test, running the suite, or claiming work is done. |
All code is expected to be tested (target: 90% coverage). Tests are table-driven with t.Run
and testify assertions; repositories are mocked at the interface level for unit tests.
| Tier | Where | Needs | Command |
|---|---|---|---|
| Unit | next to the code | nothing | make test (go test -v -short -cover ./...) |
| Repository | internal/repository/postgres | Docker Postgres on :5433 | make test-integration |
| Tagged integration | *_integration_test.go | -tags=integration + TEST_DB_* | see below |
| E2E | test/e2e/ | -tags=e2e + test DB | make test-e2e |
| Full-stack E2E | test/e2e/ | full Docker stack on :3333 | make test-e2e-full |
| Performance | test/performance/ (k6) | k6 + Docker | make test-perf, make bench |
Build tags matter. Files under test/e2e/ start with //go:build e2e and are invisible
without -tags=e2e. Likewise internal/repository/postgres/*_integration_test.go carry
//go:build integration — and neither make test-integration nor CI passes that tag, so those
files only run when you add it yourself:
export TEST_DB_HOST=localhost TEST_DB_PORT=5433 TEST_DB_USER=testuser \
TEST_DB_PASSWORD=testpass TEST_DB_NAME=ninerlog_test
docker compose -f docker-compose.test.yaml up -d
go test -tags=integration ./internal/repository/postgres/...
go test -short -run TestFlightService_Create ./internal/service/
go test -tags=e2e -run TestE2E_Flights ./test/e2e/...
bash scripts/run-e2e-tests.sh TestE2E_Flights # full stack, single test
make test-e2e-full / scripts/run-e2e-tests.sh brings up docker-compose.e2e.yaml
(API + Postgres + MailPit for email + S3/SFTP/WebDAV for backup tests), waits for
http://localhost:3333, runs the suite, then tears down. --keep leaves the stack running.
Helpers live in internal/testutil (SetupTestDB, TeardownTestDB, fixtures, API client).
Business logic and validation, error paths, authn/authz, ownership enforcement, regulatory
rules (EASA/FAA currency), edge/boundary values, and that responses match the OpenAPI spec.
Do not test third-party libraries or generated code (internal/api/generated/).
make fmt
make lint
make test
make route-check
bash scripts/run-e2e-tests.sh
Add make migrate-check when db/migrations/ changed and make dashboard-check when a metric
or dashboard did.
All must be green. If you find a regression, do not work around it in the test — fix it, or file a GitHub issue documenting it. Never mark work complete without a green run.
When a behaviour change makes an e2e test wrong, or a red test might be stale rather than a
regression, see .claude/skills/e2e-sync/SKILL.md.