Tool: Browser manual smoke

개요: 에이전트가 크롬을 직접 열어 화면을 조작하고 지금 이 순간 제대로 뜨는지 확인합니다. 빠르지만 반복할 수 없으므로 라벨을 분명히 붙입니다.

Proves

• The page renders, the element exists, the click does what was asked — right now
• Layout / copy changes look as intended (with a screenshot)

Does not prove: it will keep working. This is a smoke check, not a regression test.

Use when

• Frontend change where a human would normally "open it and click around"
• Verifying a UI fix before deciding whether a Playwright spec is worth writing
• No E2E setup exists and the orchestration allows manual smoke

Do not use when

• The orchestration requires repeatable evidence (→ playwright-e2e)
• The action is destructive or irreversible on real data

Inputs

Need	Source
Running app	dev server URL from the mission
Browser control	product capability — see ../entrypoints/
What to check	the mission's acceptance list; do not wander

Procedure

1. Navigate to the target URL
2. Snapshot (accessibility tree) → confirm expected elements exist
3. Perform the mission's actions (click, type, submit)
4. Snapshot / screenshot → confirm expected result
5. Record each step as evidence

Stop and report after four failed attempts at the same step. Do not loop.

Evidence format

Under ## Artifacts, labelled clearly as manual:

kind:    manual smoke (not repeatable)
url:     http://127.0.0.1:3000/items
steps:   open → click "New item" → fill name → submit
observed: list shows the new row; no console errors
screenshot: /path/to/shot.png

QA treats manual smoke as weaker evidence than a passing spec and may ask for playwright-e2e when the orchestration requires it.

BLOCKED conditions

• Login / captcha / passkey needs a human
• Browser control is not available in the current product
• Page depends on data that does not exist locally

Current vs extension

Item	Status
Agent-driven browser smoke	current when the product offers browser control
Promoting a smoke path to a Playwright spec	recommended follow-up, not automatic