docs: record 'Oops' crash fix (intl + PWA middleware) verified via Playwright

docs/ROO-HANDOVER.md

@@ -6,6 +6,56 @@
 
 ---
 
+## ✅ "OOPS" CRASH FIXED — 2026-06-13 (update 3) — verified in a real browser
+
+The previous "RESOLVED" was wrong: login was verified with **curl only**, which cannot
+execute client-side JS. In a real browser **every page (including `/login`) crashed on
+hydration** with the "Oops! Something went wrong" boundary. Diagnosed by installing a
+**Playwright browser probe** (`scripts/debug/dashboard-probe.mjs`) that logs in and
+captures real console/network errors + a screenshot. Three root causes, all fixed in
+commit **`4be9c4c`** (frontend rebuilt + redeployed):
+
+8. **App-wide intl hydration crash ("Oops" on every page).** Root `app/layout.tsx`
+   renders global client components (`PwaInstallPrompt` → `useTranslations`, `Toaster`,
+   `Sonner`) as siblings of `{children}` inside `<Providers>`, but only each route-group
+   layout wrapped *its own* children in `NextIntlClientProvider`. Those global components
+   mounted with **no intl context** → "No intl context found" → hydration crash.
+   **Fix:** root layout is now `async` and wraps the body in `NextIntlClientProvider`
+   via `getMessages()`. Nested route-group providers remain valid (next-intl nests).
+
+9. **PWA assets intercepted by auth middleware** (manifest.json syntax error + stale cache).
+   The middleware `matcher` did not exclude `/manifest.json` or `/sw.js`, so unauthenticated
+   browsers got **307 → /login (HTML)** for both → browser parsed HTML as JSON
+   (`manifest.json:1 Syntax error`) and an HTML/old service worker kept serving **stale
+   bundles** ("website hasn't changed after redeploy"). **Fix:** matcher now excludes
+   `manifest.json`, `sw.js`, `icons`, `offline`. Verified: `manifest.json` → 200
+   `application/json`, `sw.js` → 200 `application/javascript`.
+
+10. **Service-worker stale cache.** Bumped `CACHE_NAME` `v1`→`v2` in `public/sw.js` so the
+    `activate` handler purges old cached bundles from clients that loaded the broken build.
+
+**Browser verification (Playwright):** login `admin@test.de`/`test123` → lands on
+`/dashboard`, **no `pageerror`**, dashboard renders. The "Oops" is gone.
+
+### ⚠️ Known follow-up (non-fatal — dashboard renders via mock fallback)
+Dashboard API endpoints are **mis-wired**: frontend calls `/api/backend/dashboard/stats`,
+`/distributions/recent`, `/consent/check` → proxy rewrites to `/api/v1/dashboard/stats`
+etc., but the backend has **no such routes** (real stats route is `/api/v1/clubs/me/stats`
+in `ClubController`; `/dashboard` only exists in `PortalController`) → HTTP 500/404.
+The dashboard falls back to `mockClubStats`/`mockRecentDistributions` so it still renders.
+Frontend `ClubStats` type also differs from backend `ClubStatsResponse`. Fix = align
+service paths + DTO shape, or add `/api/v1/dashboard/*` endpoints to the backend.
+
+### 🔧 New tooling installed on the Fedora 44 workstation
+- **Node.js 22.22.2 + npm 10.9.7** via `sudo dnf install -y nodejs npm` (was MISSING).
+- **`@playwright/mcp` v0.0.76** (run via `npx @playwright/mcp@latest`) + Chromium
+  headless-shell (`npx playwright install chromium`). Wired into Roo `mcp_settings.json`
+  (server name `playwright`). Chromium launches fine on Fedora 44 despite the
+  "OS not officially supported" warning.
+- **Probe:** `scripts/debug/dashboard-probe.mjs` — reusable client-side debugger.
+
+---
+
 ## ✅ LOGIN WORKING — 2026-06-13 (update 2) — full auth flow verified
 
 After deploy, login showed a client-side "Oops! Something went wrong" error boundary.