[Torgh]

Adds Playwright end-to-end tests. Until now, UI changes such as #21 (dragging, dialogs, maps) and the budget were only checked by hand in a browser.

What is tested (11 tests, about 30 s locally)

Desktop app (production build, next start):

1. Demo trip: 7 days, “📝 2 to sort out”, “💰 €1,307 of €1,500”, a timing warning, the map's pins.
2. Dragging: a card dragged with the mouse to another day stays there after a reload.
3. Card editor:
   • adding a card;
   • the price hint warns about the missing currency in “1500” and reads “€12.50”;
   • Paid, saving (the header total goes up), deleting (it goes back).
4. To sort out: the to-do list shows what is waiting and what is still to book.
5. Budget dialog:
   • a Paid tick moves the totals;
   • switching to PLN converts the total and the limit, using the fixed test rates.
6. Trip settings: a later last day adds a day.
7. Map: a pin highlights its card.
8. Export and print: the JSON export parses back with parseTripFile, and the print page shows the bookings and the total.
9. Saving: “Saved on this PC”, and the trip file lands in the data folder.

Phone app (Vite build, Pixel 7 viewport):

10. Days, the day's spend, “✓ paid”, swiping between days, the Budget tab, and a budget row leading to its day.
11. Importing a .tripdeck.json file.

How it stays reliable

• Offline, no keys. Requests leaving the machine are blocked in the browser. The map gets a plain style; the pins still show. Exchange rates are fixed (1 EUR = 4.25 PLN). The app's /api/route, /api/places, /api/elevation, /api/timezone and /api/ai answer 503, as they do offline, so the board shows straight-line estimates.
• No keys in the builds. The apps are started with empty Google and Anthropic variables, which win over .env.local.
• Separate data. The desktop app gets a temporary TRIPDECK_HOME set to “This PC” storage, so your ~/.tripdeck and OneDrive are never touched.
• Errors fail tests. Any uncaught error or console.error in a page fails the test; blocked requests aside.
• Service workers are blocked, so nothing slips past the stubs.

Checked

• 11/11 pass locally (Edge on Windows), and 55/55 with --repeat-each=5 in parallel.
• The tests catch real breakage. I broke two things on purpose: dropped the limit from the budget in the header, and broke swiping on the phone. 3 tests failed, as they should; the code was restored afterwards.
• Typecheck (TS 7), lint and the 288 unit tests pass.

CI

• New job “End-to-end tests (Playwright)”: it installs Chromium, runs bun run test:e2e and, on failure, uploads the report with screenshots and traces as an artifact (kept 7 days).
• Once it has proven stable, it can be added to the required checks of protect-main.

Locally

bun run test:e2e. On Windows it uses the installed Edge. Elsewhere, run bunx playwright install chromium once.

🤖 Generated with Claude Code

[Torgh]

Dropping the end-to-end test idea for now.