사용 방법
- 설치 명령을 복사하고 자신의 에이전트 환경에 설치하기 전에 소스 패키지를 확인하세요.
- 필요한 도구, 인증 정보, 지원 환경은 원본 문서에서 확인하세요.
- 입력과 원하는 산출물을 설명하고 출처와 결과를 대조하세요.
npx skills add https://github.com/satanlittlehelper/ai-dev-kit --skill testing-changes시작 전 확인
- 설치에는 모델 구독, API 한도, 파일 접근 권한이 포함되지 않습니다.
- 별점 수와 설치 수는 활동 지표이며 독립적인 품질·보안 검증이 아닙니다.
원본 저장소의 제목, 예시, 코드, 표, 링크, 이미지를 유지해 표시합니다.
Testing Changes
Overview
Manual end-to-end test run of the current branch: analyze the changes, write a test plan, start every app, run the tests, show a report, stop what this run started once the report is approved, delete the test plan and the report. The test plan is written and saved before anything starts, but it is not sent for approval; nothing this run started outlives the run.
Phases
| # | Phase | Read |
|---|---|---|
| 0 | Preconditions: infrastructure is up, no stale run | references/environment.md § Preconditions |
| 1-2 | Analyze changes and task documents, write and save the test plan | references/test-plan.md |
| 3 | Check Playwright MCP and Max login, start all apps, write the state file | references/environment.md |
| 4 | Run the tests, record actual vs expected | references/execution.md |
| 5 | Write the report, show it through Plannotator; the apps keep running while the user reviews | references/report.md |
| 6 | Stop every app this run started, once the report is approved (or the review is dismissed, or the run is aborted). On approval also delete the test plan and the report | references/environment.md § Teardown, § Cleanup of documents |
Read each reference when you reach its phase, not upfront.
Hard rules
- The test plan and the report are written in Russian (code, commands, identifiers stay as-is).
- The test plan is not sent for approval: no Plan Mode, no waiting. Write it, save it, run it.
- Never start infrastructure. If it is down, stop and say what is down.
- API →
curl. Frontend → Playwright (the backend is covered through the page's own requests). Max bot or Max mini app → the web version of Max only. Existing e2e specs that cover a changed behavior join the plan. - Start the whole monorepo: every app, including those the tests do not touch, without asking which. Use only the project's own commands (
testing.apps[].startorstart:*scripts). Never override env vars or addNODE_TLS_REJECT_UNAUTHORIZED=0. - Record every app in
tmp/test-run-state.jsonthe moment it starts. Stop everyowned: trueentry in Phase 6. An app that already answers before you start it is not yours: do not start or stop it. - On report approval only: delete the test plan file and the run's temp files (app logs), and the report with its evidence in
tmp/reports/<run>/, without asking. Dismissed review or abort: delete nothing. - Never fix code, commit, or push during a test run. Report failures; fix only on request.
- Playwright runs headless; open a visible window only for the Max login (
references/environment.md). Logging in is the user's job: never type a phone number or a code. - Bash: no
cd, nosleepchains (use Monitor), one check per call.
Project config
Project values live in .claude/dev-conventions.json under testing (schema in references/environment.md). No such block: discover start:* scripts in package.json, show them in the plan, offer to save them to the config.
Common mistakes
| Mistake | Fix |
|---|---|
| Starting apps before the plan is written and saved | Phase 3 begins only after the plan file is saved |
Killing only the parent npm/nx process | Kill the process group (kill -- -<pgid>) and verify the ports are free |
| Reporting "passed" without actual results | Every test gets expected vs actual |