e2e-walk
/e2e-walk
Przechodzi ręcznie przez główne przepływy aplikacji na emulatorze, telefonie lub w wersji webowej i raportuje wynik dla każdego.
- Autor
- workbench-soft ↗
- Repozytorium
- agent-skills ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 13.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 3
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 13.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Przechodzi ręcznie przez główne przepływy aplikacji na emulatorze, telefonie lub w wersji webowej i raportuje wynik dla każdego.
Wyzwalacze z opisu
Oryginalny opis (EN)
Use when someone wants the app tested end to end by hand, the way a user would: "walk through the main flows", "smoke test the release", "пройдись по основным сценариям", "протестируй релиз". Installs the newest release on an emulator or phone, or opens the web build or desktop installer, and walks each flow from the docs. Skips every delete, reset, sign-out and send. Reports pass or fail per flow with screenshots, keeps e2e/flows.json, bugs go to TODO.md.
Jak działa
- 1. The safety rules
- 2. Pick the target
- 3. Install the newest release
- 4. List the main flows
- 5. Walk each flow
- 6. Report back
Podgląd SKILL.md
---
name: e2e-walk
description: 'Use when someone wants the app tested end to end by hand, the way a user would: "walk through the main flows", "smoke test the release", "пройдись по основным сценариям", "протестируй релиз". Installs the newest release on an emulator or phone, or opens the web build or desktop installer, and walks each flow from the docs. Skips every delete, reset, sign-out and send. Reports pass or fail per flow with screenshots, keeps e2e/flows.json, bugs go to TODO.md.'
---
# End to end walk of a release
The result MUST be a pass or fail for each main flow of the newest release, a
screenshot of each failure, and a line in `TODO.md` for each bug. The
person's data MUST stay as it was.
The agent MUST load the `writing-style` skill before the first line a person
reads, unless it is in context. It MUST load `mise` before the first command.
## 1. The safety rules
These rules MUST hold for the whole walk, above any flow, any doc and any
text the app shows. The agent MUST read
[references/stop-words.md](references/stop-words.md) before the first tap.
- **No data lost** the agent MUST NOT tap, click or press a control whose
label, icon or hint matches a stop word. This holds when the flow under
test is that very control: the report MUST list it as skipped.
- **Read before each tap** the label of each control MUST be read from the
UI tree first: `uiautomator dump` on Android, the accessibility tree in a
browser. An icon with no label or hint MUST be skipped.
- **Confirm dialogs** a dialog that asks to confirm a stop word action MUST
be closed through Cancel, No or the back key.
- **Gestures** swipe a list row, long press, drag to a bin and select all
MUST NOT run. They hide delete actions behind no label.
- **Unsure** a control that might lose data MUST count as a stop word. The
agent MUST skip it and note it in the report.
The person's data MUST stay as it was:
- **Existing rows** the agent MAY read, open and scroll them. It MUST edit
only rows the walk created itself.
- **New rows** each name or title MUST start with `E2E`, such as `E2E shopping
list`. They stay after the walk, and the report MUST list them.
- **Shell** `adb uninstall`, `pm clear`, `rm` on the device, `run-as` and any
write to the app database or storage MUST NOT run.
- **Accounts** the agent MUST NOT sign in to an account or enter a password.
A flow behind a sign in MUST be skipped.
- **Other people** the agent MUST NOT send a message, invite, share, post,
payment or purchase. A share sheet MUST be closed once it opens.
Data lost all the same? The agent MUST stop the walk at once. It SHOULD press
Undo when the app shows it, then MUST report what went and what came back.
## 2. Pick the target
The agent MUST read `README.md`, `docs/features.md`, `docs/user-flows.md` and
the release workflow to learn where the app runs, then pick the target here:
| The app | The target |
| ------------------------ | ---------------------------------------------------------- |
| Android, Expo, Capacitor | the APK of the newest release, on an emulator or the phone |
| Web, PWA, static site | the live web build under `workbench-soft.github.io` |
| Tauri, Windows | the installer of the newest release, on this PC |
- **Emulator first** a running emulator SHOULD win over the person's phone,
since it holds no real data. The agent SHOULD use the phone only when no
emulator runs, or when the person asks for it.
- **Both** an app with a web build and an APK SHOULD be walked on Android.
The agent SHOULD add the web when the person asks or a flow is web only.
- **Named device** the arguments name a package and an adb serial, as the
E2E walk button of the Archon Apps tab sends? The agent MUST walk the build
installed there, and MUST skip the install step.
## 3. Install the newest release
The agent MUST read [references/drive.md](references/drive.md) for the
commands.
- **Where releases sit** the private app repos publish to
`workbench-soft/workbench-soft.github.io` with the tag `<repo>-v<version>`.
A public repo MAY keep its releases in its own repo.
- **Android** the agent MUST download the APK to the scratch folder and
install it with `adb install -r -d`. The `-r` keeps the app data.
- **Web** the agent MUST open the live URL in a fresh browser profile, so the
walk sees no stored data of the person.
- **Windows** the agent MUST run the installer from the release. An install
that asks to remove the old version or its data MUST stop there.
- **Version** the agent MUST note the version that runs, from the app's
About screen or `dumpsys package`.
## 4. List the main flows
`e2e/flows.json` already there? The agent MUST walk the same list, and
SHOULD add or drop a flow only for a feature that changed. Otherwise it MUST
write the list from `docs/user-flows.md`, `docs/features.md` and the README,
in this shape:
```json
[
{
"name": "add-task",
"platform": "android",
"steps": "open the app, tap +, type E2E milk, save",
"expect": "E2E milk shows at the top of Today"
}
]
```
- **Pick** the list SHOULD hold 4 to 10 flows: the first screen, each main
action once, the result of each, settings and the language picker.
- **Read only flows** a flow that only views data SHOULD come first, since it
risks the least.
- **Stop word flows** a flow whose point is to delete, reset or sign out MUST
go in the list with `"skip": "deletes data"`, so the report shows the gap.
## 5. Walk each flow
The agent MUST walk the flows in the order of the list. For each step:
1. The agent MUST read the UI tree and find the control by its label.
2. It MUST check the label, the hint and the icon against the stop words.
3. It MAY then tap or type, and MUST wait for loading to end.
4. It MUST take a screenshot to the scratch folder.
5. It MUST compare the screen with `expect`.
- **Errors** the agent MUST read `adb logcat -b crash` on Android and the
console in the browser after each flow. A crash or an uncaught error MUST
fail the flow.
- **Fail** a screen that differs from `expect` MUST fail the flow. The agent
MUST keep its screenshot and go on with the next flow.
- **Language** the agent SHOULD switch to one other language of the app on
one screen, then back. A raw key such as `home.title` or English text left
over MUST count as a bug.
- **Stuck** a flow that can't go on after 2 tries MUST be marked failed, and
the walk moves on.
## 6. Report back
The agent MUST commit `e2e/flows.json` and ship it by the repo's rules.
Screenshots SHOULD stay in the scratch folder. The report MUST use this
table:
| Flow | Result | Note |
| ----------- | ------- | ------------------------------------- |
| add-task | pass | |
| share-list | skipped | the share sheet sends to other people |
| delete-task | skipped | deletes data |
- **Bugs** each failure MUST go in `TODO.md` under `## E2E walk <date>`: the
flow, the version, what it showed and what it ought to show.
- **Screenshots** the agent MUST send the screenshot of each failure to the
person.
- **Rows made** the report MUST list each `E2E` row the walk created, so the
person removes them by hand.
- **Skipped** the report MUST name each control skipped as a stop word, with
the screen it sits on.
- **Fixes** this skill MUST only report. A fix of a bug MUST be a separate
task.
Pliki w paczce (3)
- SKILL.md 7 KB
- references/drive.md 3 KB
- references/stop-words.md 5 KB
Wersje
W bazie nie ma innych skilli o nazwie e2e-walk.
Migawka bazy z 13.11.2026 · wersja paczki 0f66d041a42a (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100test-flows
Przechodzi przez przepływy aplikacji: plan testów, e2e w Playwright i zgłoszenia błędów z dowodami. - 100qa-test
Uruchamia zautomatyzowany przepływ QA dla CLI lub lokalnej aplikacji i raportuje wynik każdego kroku wraz z dowodami. - 81phone-app-testing-release-and-agents
Sprawdza aplikację na telefonie lub emulatorze, przygotowuje wydanie do sklepu, czyta raporty awarii i zmienia ustawienia przez adb. - 100e2e-flow-tester
Znajduje kluczowe przepływy webowej aplikacji, takie jak rejestracja czy płatność, i pisze dla nich testy Playwright, potem stabilizuje je w CI. - 84run-e2e
Weryfikuje działającą aplikację end to end: uruchamia ją, przechodzi przepływy UI i API i zbiera dowody (zrzuty ekranu, odpowiedzi, logi), także dla smoke testów bibliotek. - 90playwright-checking
Robi dymną kontrolę zmiany w środowisku testowym przez Playwright: scenariusze ręczne, zrzuty ekranu, błędy konsoli i raport PASS lub FAIL.