Backend deploy target is self-hosted Docker (VPS / Kubernetes / Docker
Engine), not Cloud Run. Add a multi-stage Dockerfile (Node 20, bcrypt
compiled in build stage, non-root runtime), .dockerignore, a staging
docker-compose, and DEPLOY.md covering install, build, migrate, run, and
log mapping/rotation. Pin engines.node>=20. Update deployment.md runbook
and backend/CLAUDE.md infra line off Cloud Run.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
setup-test-mitra-otp.mjs adds a phone+mitra-scoped entry to the
app_config.test_otp_bypass allowlist and ensures an ACTIVE mitra row
(createMitra defaults inactive -> 403). Dev/QA convenience; the bypass is
checked before Fazpass in requestOtp so it short-circuits even when
FAZPASS_ENABLED=true. Idempotent.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Belt-and-suspenders, not a bug fix: storage (timestamptz) and timer math are already tz-independent. Add SERVER_TZ env (default UTC) via getServerTimezone(); db/client.js pins the DB session timezone (reads env directly to avoid an import cycle); server.js pins process.env.TZ and asserts at boot that the DB session matches (logs [tz] or a loud warning). Keeps any future date_trunc/::date reporting deterministic and surfaces a misconfigured server early. Documented in backend/CLAUDE.md + .env.example.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
TransactionType.FIRST_SESSION_DISCOUNT ('first_session_discount', 22 chars) overflowed the VARCHAR(20) column, throwing in acceptPairingRequest AFTER the session was flipped to ACTIVE but before startSessionTimer/startSessionListener/PAIRED-notify ran. Every first-session-discount pairing thus half-completed: lost transaction row, no server-side timer, and a 500 to the mitra so its app never opened the chat. Widen the column (CREATE TABLE + idempotent ALTER). Deferred hardening (bookkeeping INSERT in the critical path) logged in TECH_DEBT.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- Test-OTP bypass allowlist for Apple reviewers / QA: phone-scoped static OTPs
managed in CC (Settings → Test OTP Bypass), bcrypt-hashed on save, kill-switch
toggle, per-entry expires_at. New `otp_requests` columns (is_bypass, code_hash)
+ DB CHECK enforcing bypass-row shape.
- Hash-at-rest for stub OTPs: replaced plaintext `<ref>:<code>` storage with
bcrypt(code_hash); reference goes to fazpass_reference alone. Verify routes on
sovereign is_bypass flag, defers code_hash-NULL rows to Fazpass.
- Fazpass integration (gated by FAZPASS_ENABLED env, default off): new
fazpass.service.js calling /v1/otp/{request,verify}; distinct errors for wrong
OTP (CODE_MISMATCH 401) vs provider outage (OTP_PROVIDER_FAILED 502).
- Removed redundant Free Trial CC section (was a back-compat shim for the same
pricing_promotions row as "Diskon Sesi Pertama") + unused alias in
pricing.service.js.
208 tests green (34 new for OTP + Fazpass). Fazpass API + dashboard PDFs added
at project root for reference (docs are auth-gated).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Customer end-of-session (figma §6):
- PricingBottomSheet: ghost "cukup, akhiri sesi" CTA + dedup divider
- chat_screen._runEndSessionFlow chains ConfirmEndStep1 → ConfirmEndStep2
→ ClosingMessageSheet (or "lewati saja" → close + /home). The four
popup/sheet widgets already existed; this commit just wires them
- showModalBottomSheet: showDragHandle=false to suppress the Material 3
auto-injected handle that was stacking with our own pill
Notification sound on API 33+:
- Bump channel halobestie_chat_v1 → halobestie_chat_v2, created from
native Kotlin in MainActivity.kt with AudioAttributes contentType
CONTENT_TYPE_SONIFICATION. flutter_local_notifications' default of
CONTENT_TYPE_UNKNOWN was causing Android 13 to silently drop audio
focus while the notification still posted (isNoisy=true). Both apps
- Backend FCM payload channelId updated to v2
- AndroidManifest meta-data: default_notification_icon + color → brand
silhouette tinted pink instead of generic Android bell. Both apps
Customer pairing reliability:
- pairing_notifier: applyPairedFromPush({sessionId, mitraName}) unsticks
searching screen when WS push failed and FCM/active-session-poll is
the first signal. Idempotent across PairingSearchingData,
PairingTargetedWaitingData, PairingErrorData (covers ALREADY_ACTIVE)
- notification_service: dispatches every FCM data payload to an
onDataMessage callback (foreground + tap + cold-start). main.dart
wires that to applyPairedFromPush on type=='paired'. Foreground
'paired' no longer renders a local banner — screen self-advances
- main.dart activeSession listener also calls applyPairedFromPush when
a session appears server-side while pairing is in a waiting state.
Covers stale ALREADY_ACTIVE recovery without a full page refresh
Auth refresh token race:
- auth_notifier._refreshFromStorage shares a single in-flight Future
across all callers (Auth.build + 401-retry path). Backend rotates
refresh tokens, so concurrent callers using the same stored token
would race → loser 401s → catch wipes flutter_secure_storage → user
appears logged out after kill+reopen
Polish:
- method_pick_screen: resizeToAvoidBottomInset=false — prevents the
one-frame overflow when entering with the previous screen's keyboard
still animating out
- bestie_history: BestieHistoryItem now carries `status` (backend
already returns it). Removed _rawHistoryProvider that fetched the
same endpoint just to read status; the two providers could go out
of sync mid-rebuild and throw RangeError(length) on indexing
Xendit Stage 8 (carried from WIP):
- xendit_checkout_screen: embedded webview hosting Xendit's invoice
page (intercepts halobestie:// deeplink + return-page URLs for
deterministic pop)
- waiting_payment_screen: auto-pushes the webview when the backend
payload includes xendit_invoice_url; spinner card + "Buka ulang
halaman pembayaran" CTA for the QR-fallback path
- pubspec: webview_flutter ^4.13.0
Maestro infra:
- subflows/onboarding_returning_user: drop the "Mulai" carousel wait
(splash auto-advances since 2026-05-26); tap phone-field hint
instead of point; drop hideKeyboard (sends BACK → /home when the
IME isn't actually up)
- New flow ts-customer-06-01-end_session_via_timeup_sheet: drives
the full path to the chat-expired banner. Last step blocked by a
Maestro+Flutter gesture quirk on the perpanjang ElevatedButton
(raw `adb input tap` works at the same coords). Documented in
memory; deeplink fixture or manual verify recommended
- ChatExpiredBanner button wrapped with Semantics(identifier:
'chat_extend_button', button: true, onTap: …) — good hygiene for
future tests even though it doesn't fix the dadb tap issue
.dev/: tracked wsl_emulator_bridge.ps1 + wsl_tcp_relay.py for
Maestro-on-WSL setup (Windows-side netsh portproxy + WSL-side
loopback relays). Both referenced from existing CLAUDE.md notes.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Backend wraps idn-finlogos npm at /assets/payment-icons/<slug>.svg with
1y immutable cache. Mobile drops bundled SVGs (only placeholder remains)
and fetches via flutter_cache_manager. payment_methods.icon is now a
CSV of slugs; catalog emits icon_urls[]. CARDS tile renders Visa + MC +
JCB side by side.
- Per-method min/max amount bounds (BIGINT, nullable). Picker greys out
out-of-range tiles with subtitle; backend gates with INVALID_PAYMENT_AMOUNT
(422). Defense in depth against stale-catalog clients.
- Xendit channel codes corrected from authoritative docs
(BCA_VA -> BCA_VIRTUAL_ACCOUNT, CREDIT_CARD -> CARDS, ovo -> ovo-new,
shopeepay -> shopee-pay, ...). 18 methods x 5 groups seeded with
Xendit-published per-channel min/max.
- Re-runnable seed (ON CONFLICT DO NOTHING on payment_code + new unique
index on group name). Operator CC edits never clobbered across re-runs.
One-shot reset + inspect scripts under backend/.dev/.
- Customer redirect HTML pages at /payment/return/{success,failure},
brand-styled with "Buka HaloBestie" CTA firing halobestie:// deeplink.
URL scheme registered on Android (intent-filter w/ BROWSABLE on
MainActivity) and iOS (CFBundleURLTypes). Waiting-payment poller still
owns confirmation; deeplink just brings the activity to foreground.
- Control center payment-catalog page: min/max inputs + columns. Other
CC pages restyled with new theme tokens (separate work, bundled here).
169/169 backend tests pass. See requirement/phase5-payment-revamp-2026-05-27.md
for the full revamp doc. Stage 8 (E2E) still pending: webhook URL routing
decision + two client_app follow-ups (legacy /chat/request removal,
extension Custom Tab).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Payment catalog (Phase 5.x — see requirement/phase5-payment-catalog-plan.md):
- New tables payment_method_groups + payment_methods with seed (3 groups,
10 methods; GoPay seeded inactive pending Xendit channel confirmation).
- payment-catalog.service.js with two-layer cache (60s in-process + 1h
Valkey) and config:invalidate pub/sub fanout. Mutator API + casing-
tolerant findActiveMethodByCode for downstream validation.
- App-facing GET /api/client/payment-methods returns pre-grouped JSON,
active-only, empty groups dropped server-side.
- POST /api/client/payment-requests now validates `method` against the
catalog (INVALID_PAYMENT_METHOD 422) and stamps
product_metadata.preferred_payment_code (upper-cased).
- Control-center /internal/payment-{groups,methods}{,/:id,/reorder}
endpoints (full CRUD + idempotent reorder). New Payment Catalog page
wired into the CC nav.
- Customer app renders the catalog as collapsible groups (first expanded)
via paymentCatalogProvider; QRIS-only hardcoded fallback on 5xx so
checkout never hard-fails. Replaces the hardcoded _PayMethod enum.
- 10 brand SVGs (~63KB) bundled in client_app/assets/payment_icons/ from
github.com/hafidznoor/idn-finlogos. Xendit's per-channel media-asset
pages were planned but found decommissioned during implementation —
switched to idn-finlogos with the standard "channels-we-accept"
trademark posture. See assets/payment_icons/README.md for the workflow
to add new methods.
- 16 vitest cases covering the service + cache; full backend suite green
(162/162).
Customer-app splash + register polish:
- Splash rewritten per figma S1: warm vertical gradient, two ImageFiltered
radial orbs, 96×96 rounded-square logo tile, "HaloBestie" + "kamu gak
harus ngerasain ini sendirian." Self-driving navigation via context.go
after a 2.5s post-frame timer (native Android splash burns ~1-1.5s
before Flutter paints — 1s timer yielded near-zero visible duration).
Router early-returns null for isSplash so it never moves us off /splash
on its own.
- 3-page onboarding carousel removed: user clarified the new splash
REPLACES that carousel. Dropped /onboarding route, OnboardingScreen,
onboardingDoneProvider + gating, dead splash_{1,2,3}.png + the
splash_chat_hebat.png Flutter asset. Phase 4 /onboarding/* subroutes
untouched; Android-native launch_background drawable left alone.
- Register screen (login-by-phone) polished: circular pink back button +
72×72 logo badge (same brandLogoBg pink as splash, Transform.scale 1.4
to fill the tile). Step-dots indicator removed.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
When the operator sets require_mitra_ping=false, the auto-offline sweep
early-returns (by design — "don't gate online status on heartbeat
freshness"). The three Valkey read paths still gated on heartbeat
freshness anyway, which trapped the system: sweep won't remove the
mitra from mitras:online, but readers reject them as stale. The customer
CTA stayed permanently disabled with no recovery.
Fix all three to skip the heartbeat-freshness check when require_ping
is off, matching the sweep's contract:
- computeAvailabilityFromValkey (customer beacon)
- isMitraReachable (extension service)
- findAvailableMitrasFromValkey (pairing candidate finder)
The Postgres fallbacks already did the right thing (is_online only,
no heartbeat compare); this aligns the Valkey hot path.
Also: PATCH /internal/config/mitra-ping now publishes config:invalidate
for require_mitra_ping and mitra_stale_after_seconds, and the subscriber
in mitra-status.service was widened to listen for both. Flipping the
toggle in CC now busts the 10s availability snapshot immediately instead
of waiting out the TTL.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Every Xendit invoice now carries metadata: { app: 'halobestie_v2' } so an
external webhook router (no DB access) can fan out v1/v2 traffic purely off
the echoed payload.
Every inbound webhook lands in a new webhook_logs table BEFORE auth or
business logic, so a forensic row survives 401/409/unknown/exception paths.
Primary fields are parsed as columns; raw_body keeps the full payload
verbatim. The handler captures outcome in closure-scoped vars and stamps
http_status/processing_result/processing_error in a single update before
the lone reply.send() — Fastify flushes reply.send() immediately, which
defeated the original finally-block stamp.
A non-UUID external_id no longer crashes the Postgres cast; it ACKs with
ignored_non_uuid_external_id so Xendit stops retrying legacy old-app IDs.
When the DB log itself fails, an optional rolling JSONL file sink absorbs
the event. Disabled by default — opt in via XENDIT_WEBHOOK_FALLBACK_ENABLED.
Naming: <NAME>-YYYY-MM-DD.jsonl in XENDIT_WEBHOOK_FALLBACK_DIR (default
./logs), basename XENDIT_WEBHOOK_FALLBACK_NAME (default
xendit-webhook-fallback). No stdout fallback by design.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Backend
- payment_sessions → payment_requests rename across DB schema + 29 files
- payment.service.js becomes product-agnostic owner: EventEmitter +
Xendit wrapper + requestPayment / confirmPayment public API; legacy
aliases retained for existing chat callers
- Webhook handler at POST /api/shared/payment/webhooks/xendit, with
constant-time token verification (8 vitest cases)
- Server-driven pairing: payment.service emits
payment_request.confirmed → pairing subscriber starts the blast.
Legacy POST /chat/request still works during the cutover.
- Reconciliation sweeper extended (re-emits events for confirmed rows
with no chat session)
- SIGTERM drain + startup reconciliation pass in server.js
Customer app
- waiting_payment_screen opens xendit_invoice_url via
LaunchMode.inAppBrowserView
- searching / no-bestie / targeted-waiting / pairing-notifier updated
to consume the new payment_request_id contract
- pending_payments_provider + bestie-unavailable dialog migrated
Dev / testing
- XENDIT_ENABLED=false is the safe default; .env.example documents the
four new vars
- backend/.dev/xendit-fake-webhook.sh exercises the handler without
ngrok
- 90/92 backend tests pass (two pre-existing session-timer flakes,
unrelated); client_app analyzer clean
- requirement/phase5-xendit-plan.md is the canonical reference
Stage 8 (live E2E) blocked on Xendit test-mode keys. The dashboard's
single-webhook-URL constraint will be worked around via a self-poll
script next session.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- mitra_app/CLAUDE.md: pitfall entry for the InputDecorationTheme
min-height collision that broke chat-input centering. Walks through
the working recipe (constraints: BoxConstraints(), Material +
StadiumBorder + Center wrapper). Points at chat_screen.dart::_InputBar
in both apps as the source of truth.
- backend/CLAUDE.md: two new convention sections.
- Config-source: when to use DB-stored (operator-tunable via CC) vs
env-driven (deploy-fixed). Codifies the pattern shipped today for
MITRA_HEARTBEAT_CADENCE_SECONDS so Xendit credentials / callback
tokens follow the same shape tomorrow.
- FCM channel: single shared `halobestie_chat_v1` channel for both
apps, target via android.notification.channelId. Bump the channel
ID when introducing a new sound (Android API 26+ binds sound at
channel-create time).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Splits the single mitra_ping_interval_seconds config (which conflated
"how often the app pings" with "how long until offline" through a
hidden ×3 multiplier) into two orthogonal knobs:
- mitra_stale_after_seconds (CC-tunable, app_config DB row): the
operator-facing offline threshold. What you set is what you get —
no multiplier. Default 45s (preserves today's effective grace at
the legacy 15s ping default).
- MITRA_HEARTBEAT_CADENCE_SECONDS (env var, default 30s): how often
the mitra app sends a heartbeat. Backend-fixed per deployment;
surfaced to the mitra app via /api/mitra/status.
Backend:
- config.service: getMitraPingConfig returns the new tuple
{require_ping, stale_after_seconds, heartbeat_cadence_seconds}.
Env parser handles blank/non-numeric → 30 fallback.
- mitra-status.service::autoOfflineStaleMitras drops the *3 and uses
stale_after_seconds directly.
- mitra-status.service::getStatus returns heartbeat_cadence_seconds
instead of ping_interval_seconds.
- /internal/config/mitra-ping PATCH validates
stale_after_seconds >= cadence, returns 422 with a clear message
("stale_after_seconds must be a number >= heartbeat cadence (30s)").
- migrate.js: adds mitra_stale_after_seconds default 45. The old
mitra_ping_interval_seconds key is left in place (vestigial) —
no live code reads it; safe to drop after one release.
Mitra app:
- status_notifier reads heartbeat_cadence_seconds, uses it directly
as the Timer.periodic interval. Defaults to 30s if missing (older
backend safety).
Control center:
- SettingsPage: renames "Interval Ping" → "Ambang offline", input
min={heartbeat_cadence_seconds}, shows the cadence as a read-only
value with explanation that it's env-controlled.
Verified end-to-end on dev backend:
- GET /api/mitra/status returns {…, heartbeat_cadence_seconds: 30}
- GET /internal/config/mitra-ping returns {require_ping,
stale_after_seconds: 45, heartbeat_cadence_seconds: 30}
- PATCH with stale_after_seconds=20 → 422 with cadence message
- PATCH with stale_after_seconds=120 → 200, persisted
- Env override (=60, blank, "foo") parses correctly.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The customer app now uses the same halobestie_notif.ogg as the mitra
app (shipped in the previous commit). Channel ID unified across both
apps so backend FCM stops branching per recipient.
- client_app: same channel bump (chat_messages → halobestie_chat_v1)
+ RawResourceAndroidNotificationSound binding, both at channel-
create time and per-notification details. .ogg copied to
client_app/android/app/src/main/res/raw/halobestie_notif.ogg
(same 32 KB asset, identical file).
- Backend: drop the per-recipientType channel ID branch; everyone
targets halobestie_chat_v1 now.
Verified on emulator-5554 (customer): dumpsys shows the channel
bound to android.resource://com.mybestie/raw/halobestie_notif.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Drops the system notification ding for incoming mitra FCM (Curhat Baru,
Permintaan Perpanjang) and plays the HaloBestie audio mark instead.
Source: a 2.8s mono AAC inside a 3GPP container the user supplied;
converted to 32 KB OGG (Vorbis q5) for Android since the channel-sound
API needs `res/raw/<name>.<ext>` and OGG is the smallest universally
supported short-sound format on Android 5+.
- mitra_app: bump notification channel ID from `chat_messages` to
`halobestie_chat_v1` (Android binds channel sound at create time
on API 26+, so existing installs with the old channel need a fresh
ID to pick up the new sound — can't mutate in place). Bind
RawResourceAndroidNotificationSound('halobestie_notif') at both
channel-create time and per-notification details (latter covers
API 24/25 where channels don't exist).
- Backend: branch FCM `android.notification.channelId` by recipient
type — mitras → `halobestie_chat_v1`, customers → `chat_messages`
(unchanged). Customer app keeps system sound until/unless we ship
a customer-side sound too.
Verified on emulator-5556 via `adb shell dumpsys notification` — the
new channel resolves to
`android.resource://com.mybestie.mitra/raw/halobestie_notif`. The OGG
ships inside the APK (32092 bytes, confirmed via `unzip -l`).
Follow-up (iOS): bundle the same sound as `.caf` under ios/Runner +
register as a Runner-target resource in pbxproj + reference filename
in the APS payload. Deferred until iOS testing comes back into scope.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Replaces the placeholder "Hubungi Koordinator" row with two real
contacts pulled from backend config (support_handles_json), and drops
the "Hapus Akun" CTA. Mirrors the figma BestieProfile design but uses
the same WA/TG channel as the customer Tanya Admin sheet — business
decided the same ops team triages both audiences.
Backend:
- Promote support-handles route from /api/client to /api/shared
(renamed file + export). Both apps now consume the same endpoint;
hitting /api/client/* from mitra would violate the per-app
convention in mitra_app/CLAUDE.md.
- client_app provider updated to /api/shared/support-handles.
Mitra app:
- New support_handles_provider mirroring the client_app one. Adds a
`displayHandle` getter that strips the URL scheme for the subtitle
("https://wa.me/X" → "wa.me/X", "https://t.me/Y" → "t.me/Y") so the
row looks like the figma without exposing raw URLs.
- Profil screen now lists: Chat WhatsApp Kami, Chat Telegram Kami,
Syarat & Ketentuan, Kebijakan Privasi. Danger zone simplified to
Keluar only — mitras request account deletion through the same
WA/TG channels (no separate self-service path).
- url_launcher added as a runtime dep, launches deeplinks in
externalApplication mode with graceful snackbar fallback when
parsing or launching fails.
Updates [[feedback-mitra-internal-audience]] — pre-login rule still
holds (no admin CTAs on S3a/S3b/AccountInactive), but the post-login
Profil tab now does surface WA/TG. Overrides decided 2026-05-21.
Verified on emulator-5556: Profil tab renders both rows with handles
from `wa.me/6285173310010` + `t.me/halobestie`, Keluar present, no
Hapus Akun button.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Today the customer's "Perpanjang" only reaches the mitra via session-
scoped WS. If the mitra is on Home/Undangan, in a different session, or
backgrounded, the WS send no-ops and the 10s safeguard timeout fires
auto-reject (or auto-approve if the mitra happens to also have an
active general WS, depending on config) — either way the mitra never
saw the request.
Backend:
- extension.service.js::requestExtension now falls back to FCM via
notification.service when the mitra isn't on the session WS. Mirrors
the pairing notifyMitra pattern (Curhat Baru). Customer display name
is pulled into the session lookup for the FCM body.
- shared.chat.routes.js: /chat/:sessionId/info now returns
pending_extension (extension_id, duration_minutes, price,
requested_at, expires_at, timeout_seconds) so the chat screen can
rehydrate the accept/reject UI after a cold-start FCM tap. expires_at
is derived from requested_at + extension_timeout_seconds config.
Mitra app:
- mitra_chat_notifier.dart::connect parses pending_extension from /info
and seeds MitraChatConnectedData.extensionRequest — the existing
_buildExtensionView renders unchanged.
- notification_service.dart::_navigateFromMessage handles
type=extension_request → pushes /chat/session/<id>. Composes with
the new /info pending_extension to bring the mitra straight into the
accept/reject view.
Verified end-to-end on dev backend (FCM call returned sent=true; /info
returns pending_extension when within timeout window). Visual delivery
on emulator-5556 deferred — API 24 AVD queues FCM 5-30 min per
feedback-emulator-avd-versions.
Out of scope (follow-ups):
- Customer-side FCM for EXTENSION_RESPONSE (accepted/rejected/timeout)
- Perpanjang tab list endpoint + Flutter provider + UI
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Brings the mitra app to figma-bestie parity for Home (§1), Undangan
inbox with Curhat Baru + Perpanjang tabs (§2), and the incoming-popup
+ active-chat flow (§3). Home now lives inside a StatefulShellRoute
with BestieTabBar so Profil + Undangan + Home share one shell.
- Shell: features/shell/ (StatefulShellRoute, BestieTabBar, 3 branches)
- Undangan: features/undangan/ — Curhat Baru reads
chatRequestProvider.pendingInvites; row Terima delegates accept to
the notifier and ChatRequestOverlay owns nav (no double-push).
Perpanjang tab stubbed (empty state) until backend exposes
pendingExtensionsProvider.
- Profil: features/profile/ — Bestie-styled stub
- Home: refactored to body-only (shell owns chrome)
- Popup: chat_request_overlay + chat_request_notifier updated to
serve the list rows, not just the modal
- Chat: mitra_chat_screen polish
- Theme: accentAmber tokens for the Perpanjang tab + halo_orb widget
(loading spinner used by undangan list states)
- Login: replace broken GoRouterState location guard with
_expectOtpPush flag — was stacking duplicate /otp pages on OTP
resend (see project-otp-nav-bug-fixed-2026-05-21)
Maestro:
- 17 new flows under .maestro/flows/ts-mitra-{1,2,3}-* covering home
online/offline variants, undangan empty/populated/tolak states,
popup curhat-baru → accept → chat → ended banner, plus popup
dismiss/expire/cancelled edge cases
- 4 new §A OTP flows (07/08/09/10) for invalid/mismatch/expired/cooldown
- Helper scripts: force_mitra_online/offline, force_pairing_timeout,
force_session_expires_at, delete_mitra_status_row,
customer_blast_now (js), customer_cancel_latest_blast
- Backend: POST /internal/_test/delete-mitra-status-row supports the
"fresh mitra with no status row" test setup
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
§1 notif banner: permission_handler v11 returns granted unconditionally
for Permission.notification on Android <13 because POST_NOTIFICATIONS
didn't exist as a runtime permission. Result: SHome1st amber "notifikasi
off" banner never showed on API 24-32 even when the user toggled
notifications off in Settings → Apps. Add a
NotificationManagerCompat.areNotificationsEnabled() pre-check via
flutter_local_notifications (works from API 19+) so the banner reflects
the real OS state on older Android.
§5 chat delivery: the contract is "WS when foreground, FCM when
background", but the previous build only honoured (1) — Android keeps
the TCP socket alive after the Dart isolate is paused, so backend's
`socket.readyState === 1` check returned true and FCM never fired.
Fix has five parts (all required together):
1. Customer-side lifecycle observer in client_app/main.dart closes
chatProvider's WS on paused/detached, reconnects on resumed.
2. `_appPaused` gate in main.dart suppresses the activeSessionProvider
listener's auto-reconnect (15s poll in active_session_notifier
would otherwise re-open the WS the next tick after the observer
closed it — defeating the fallback).
3. Mitra-side lifecycle observer in mitra_app/main.dart stashes
`_pausedChatSessionId`, calls mitraChatProvider.disconnect(), and
re-issues connect(saved) on resumed.
4. MitraChat gains a `_connectedSessionId` field + getter so the
observer in step 3 can read it back across disconnect (disconnect
clears it; the next connect overwrites it).
5. SearchingScreen resets pairingProvider when entering with a new
draft.paymentId — previously it retained PairingActiveData with
the *old* sessionId after a session ended, and the next pairing
flow navigated straight to that completed session showing
"Sesi sudah berakhir".
Backend additions under /internal/_test/* for assertion harness:
inspectSessionWsState + GET /ws-connection-state,
POST /send-chat-message-as-mitra (with delivered_via),
POST /send-chat-message-as-customer (with delivered_via),
POST /send-fcm-chat-message (raw FCM dispatch).
Maestro coverage:
- ts-customer-05-01: mitra → customer message when customer is
backgrounded → delivered_via=fcm.
- ts-customer-05-02: customer → mitra message when mitra is
backgrounded → delivered_via=fcm.
- ts-customer-01-01: §1 notif-denied banner on home. Documented
precondition: mitra must be force-stopped or backgrounded on the
chat screen before 05-02 runs (Maestro can only drive one --udid
per run; mitra-side lifecycle observer end-to-end is deferred).
Helper scripts under client_app/.maestro/scripts/:
inspect_ws_state.js, assert_ws_state.js,
send_chat_message_as_mitra.js, assert_delivered_via.js (takes
SENDER=mitra|customer to route to the matching backend endpoint).
README_section_05.md documents the test plan, helper scripts, and the
deferred mitra-side maestro driving. Both apps tested manually on
API 28 AVDs where FCM delivery is sub-second; API 24 has 5-30 min
heartbeats that make it impractical for FCM-related testing.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Inverse coverage for the auth path: TS-01..TS-06 all wipe the customer
row (drop_customer=true) so every OTP path lands on the new-user
set-name branch. TS-07 instead seeds an existing identified customer
(phone + display_name + is_anonymous=false) and verifies the OTP
sign-in returns the existing row unchanged via
resolveCustomerForIdentity branch 1, so /auth/set-name is never shown.
Adds:
* /internal/_test/seed-customer endpoint — upserts a customer with
phone + display_name + is_anonymous=false.
* client_app/.maestro/scripts/seed_customer.js helper.
* client_app/.maestro/flows/ts-07_returning_existing_name_skips_setname.yaml.
* TS-07 scenario doc + coverage-map row in
requirement/phase4-customer-flow.md.
The flow asserts the "halo, <name>" greeting on the returning-user home
variant (identified users always land on _SHomeReturningView regardless
of chat history) plus an explicit notVisible on "Siapa namamu" as a
belt-and-braces check.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Stages 5.1, 5.3, 5.4 of the returning-user flow rework. All three §4
entry paths now require payment BEFORE pairing, matching the updated
mermaid spec.
* Spec (requirement/flow_customer.mermaid.md §4): payment block converges
three call-sites (bestie-yang-udah-kenal-online, bestie-baru,
offline-popup → cari bestie lain). PairRoute dispatches lama → targeted
pair, baru/cari-lain → §3 blast. §3 retains its post-payment-shared
contract.
* Stage 5.1 (client_app): PaymentDraft carries targetedMitraId +
topicSensitivity. bestie_history_list seeds the draft + pushes
/payment/entry (was legacy /payment). searching_screen branches on
draft.targetedMitraId for blast-vs-targeted dispatch.
payment_entry uses resetExceptTarget(); bestie_choice_sheet + home
_onCurhatBestieBaruPressed call explicit reset() before push so
the keepAlive draft can't leak stale targeting into a blast.
* Stage 5.3 (client_app): new BestieOfflineVariant.prePayReturning.
Bestie-history-list _BestieRow splits tappable from dim so offline
rows render dimmed but route taps into the popup. CTA "cari bestie
lain" resets the draft + pushes /payment/entry.
* Stage 5.4 (client_app): deleted legacy /payment route,
payment_screen.dart, payment_notifier.dart(+.g.dart). router cleaned.
* Tests (requirement/phase4-customer-flow.md + client_app/.maestro/):
six Maestro flows TS-01..TS-06 covering every §4 branching point,
all passing end-to-end. Shared onboarding prelude under
.maestro/subflows/. New helper scripts: accept_latest_pending,
force_mitra_offline, force_other_mitra_online,
reset_all_mitras_online, mitra_accept_latest_internal. New backend
_test endpoints to match. /reset-phone now cascade-deletes
customer_transactions (FK was blocking). /force-pairing-timeout
branches targeted (RETURNING_CHAT_TIMEOUT via
expireTargetedPairingRequest, now exported) vs blast (PAIRING_FAILED).
seed_history_session also outputs MITRA_NAME_RE (regex-escaped) for
reliable selectors against display names containing regex specials.
* mitra_app: dispose-during-deactivate guardrail for back-press on the
mitra chat screen after the customer's goodbye message. Pending real
emulator repro verification (carried over from 2026-05-15).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Replaces the two `pricing_*_tiers_json` blobs and five `first_session_discount_*`
keys in app_config with dedicated `pricing_tiers` and `pricing_promotions`
tables plus matching `_history` audit tables. UUID PKs, UNIQUE(mode, minutes)
natural-key constraint, optimistic-lock via `updated_at` token returning 409
STALE_WRITE on conflicts. Every mutation writes a history row capturing the
operator (changed_by from request.auth.userId) and change_kind.
CC SettingsPage replaces the JSON-textarea editors with per-row tables —
add / edit / soft-delete / reactivate / reorder, plus a buffered first-session
discount form with the same optimistic-lock contract. `minutes` and `mode` are
read-only on edit since they form the natural key; operators soft-delete and
recreate to change duration.
Stage 5 fixes a latent leak: `client.payment.routes.js` had its own local
`readDiscountConfig` that still read from app_config — would have silently
fallen to hardcoded defaults once the legacy rows were deleted. Now reads from
pricing_promotions via the shared service helper, so CC edits to the first-
session discount affect actual payment pricing on the next request.
Customer-facing GET /api/client/chat/pricing shape unchanged (id values are
now UUIDs instead of "5"/"12"/"60" but lookups happen by (mode, minutes), so
no app changes needed). 27 new backend tests, all green.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Chat-screen performance (customer + mitra):
- Parent screens have zero `ref.watch` — only `ref.listen` for side effects
- Body extracted into its own `ConsumerStatefulWidget`; AppBar parts split
into narrow `.select` consumers (mode, sensitivity, timer)
- Per-second timer ticks routed to dedicated providers
(`chatRemainingSecondsProvider` + new `mitraChatRemainingSecondsProvider`)
so WS `session_tick` frames don't invalidate the rest of the chat state
Dispose-in-ref bug fix:
- `home_screen.dart`, `payment_screen.dart`, `mitra_chat_screen.dart` —
ref-using cleanup moved from `dispose()` to `deactivate()`. Modern
Riverpod invalidates `ref` the moment `dispose()` runs; the resulting
silent error corrupts the widget-tree finalize and the next screen
appears frozen
- `halo_lints` package added at repo root with `no_ref_in_dispose` rule
to catch this pattern in CI / IDE analysis
- `custom_lint` activated in both apps' `analysis_options.yaml`
(was installed but never wired in — also brings `riverpod_lint`'s
`avoid_ref_inside_state_dispose` online)
- CLAUDE.md Pitfalls section added to client_app + mitra_app
Phase 4 §3 retryable blast-failure (Option A):
- Backend `expirePairingRequest` + all-rejected use
`recordIntermediateFailure` instead of `failPaymentSession` so the
payment session stays `confirmed` for re-blast
- WS `pairing_failed` payload carries `is_terminal: false` on the
retryable paths; client parses the flag and exposes `retryBlast()`
- "Coba cari lagi" CTA on S7 Timeout now re-blasts on the same payment
- Pairing service test updated to reflect the new semantics
Customer waiting-payment screen navigation patch:
- `_navigateTerminal` uses `Future.microtask` + `addPostFrameCallback`
redundancy after a release-mode bug where polling stopped but
`context.go` never fired, leaving the screen visually stuck on
"menunggu pembayaran"
See requirement/resume-2026-05-15.md for next-day pickup checklist
(mitra release rebuild + S21 Ultra install + retest is the gating item).
Bundles unrelated in-flight Phase 4 §2.x work that was already on disk
(ESP screen removal, USP one-time gate scaffolding, bestie-availability
public route, OTP service edits, Maestro flow tweaks) — kept together
to avoid a partial-rebase mess.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds `customers.account_belongs_to UUID NULL` and refactors customer
sign-in (phone/Google/Apple) so an anon row that re-verifies into an
existing customer no longer 409s. Instead the anon row stays intact
with a breadcrumb pointing at the real customer; tokens are issued
for the existing user. Actual data reconciliation onto the existing
row (chat_sessions, customer_transactions, payment_sessions,
pairing_failures) is deferred.
Backend
- migrate.js: ADD COLUMN account_belongs_to UUID REFERENCES customers(id)
ON DELETE SET NULL.
- customer.service.js: stampAccountBelongsTo helper; account_belongs_to
exposed in CUSTOMER_SELECT.
- auth.service.js: new shared resolveCustomerForIdentity (4-case logic);
normalizeIdentityConflict + IDENTITY_ALREADY_LINKED 409 deleted;
completeCustomerPhoneSignIn / signInWithGoogle / signInWithApple all
route through the shared helper.
- client.auth.routes.js: new resolveAnonymousCustomerId picks the anon
prefix ONLY from a verified Bearer JWT — closes the UUID-leak attack
where a tamper-able body field could mis-route someone else's
transactions. /otp/verify, /google, /apple all use it; the body field
`anonymous_customer_id` is no longer accepted on any of them.
- test/services/auth.service.test.js: 9 Vitest cases covering phone +
Google + Apple, all 4 logic cases + multi-merge accumulation.
Customer app
- auth_notifier.dart::verifyOtp: drop `skipAuth: true` and the dead
body field so ApiClient auto-attaches the anon's Bearer from
AuthBridge. Survives the AuthOtpSentData state transition (the
earlier `_currentAnonymousCustomerId()` state-drop bug is bypassed by
sourcing the id from the bridge instead of state).
- Google + Apple client paths remain unchanged (gated on provider
creds; mirror this fix when wiring lands).
Docs
- flow_customer.mermaid.md: new §2.1 sub-section with the merge
diagram, schema note, replaces-current-behaviour paragraph, and
Bearer-only security callout.
- phase3.4-testing.md: §1.5 line 76 simplified (no more per-path
split); new §1.5.1 with the 5-step operator scenario + DB invariants
+ curl recipe + Vitest pointer; new §1.5.2 covering Google/Apple
parity (deferred client work flagged).
Verification (against live dev backend, before this commit):
- Vitest: 9/9 in auth.service.test.js; 49/51 overall (2 unrelated
pre-existing failures in session-timer.service.test.js).
- Operator Node smoke: 14/14 in the §1.5.1 scenario; 11/11 in the
Bearer-precedence cases.
- Real-device UI walkthrough on SM-A530F still pending — see resume
memory `project_phase4_2_1_resume_test`.
Sister WIP bundled in migrate.js + customer.service.js: `usp_seen`
column + `markCustomerUspSeen` helper (Phase 4 USP one-time gate, was
already uncommitted in the working tree).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Closes the Stage 10 acceptance criterion §10.11 #13 (Maestro coverage).
- New dev-only `POST /internal/_test/seed-pending-payment` — inserts a
payment_sessions row in `pending` status with expires_at 20m out, so
the Pembayaran sub-tab has a deterministic row to render. Body
accepts { phone, isExtension?, amount?, durationMinutes?, mode? }.
Gated on NODE_ENV != 'production' like the other test routes.
- New Maestro helper script `seed_pending_payment.js` mirrors the
existing seed_history_session pattern.
- New flow `09_chat_tab.yaml`:
cold-start onboarding → home (returning view) →
seed completed session + seed pending payment →
tap "💬 chat" bottom-nav → lands on /chat/aktif via redirect →
assert "aktif" / "pembayaran" / "selesai" pills + empty-state copy →
tap pembayaran → assert "menunggu pembayaran sesi" + "bayar Rp..." →
tap selesai → assert "X menit" duration row → tap row → assert
"Transkrip Chat" appbar → back → still on /chat/selesai.
Maestro parsed the YAML cleanly and started executing against the
device; full run requires backend + online mitra in dev DB (same
pre-reqs as flows 03/05/06/08).
- TECH_DEBT entry: Stage 10 retired the standalone bestie-history list
screen, which means (a) the "curhat lagi" targeted-payment entry
point has no UI affordance anywhere in the app — its plumbing in
payment_notifier / payment_screen is now orphaned, and (b) the
Stage 8 flow `08_returning_targeted.yaml` will fail at
`assertVisible: "Riwayat Chat"` because it expects the deleted
screen. Three fix paths listed in the entry for product to pick.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Backend half of Stage 10 — the new Chat tab in the customer app that
replaces /chat/history with a 3-sub-tab list (Aktif / Pembayaran /
Selesai).
- New GET /api/client/payment-sessions/pending — returns the customer's
pending initial + extension payment sessions. Filter is status='pending'
AND expires_at > NOW(). Mitra info comes from session_extensions →
chat_sessions for extension rows, payment_sessions.targeted_mitra_id
for targeted-curhat-lagi initial rows. TTL reuses the existing
payment_session_timeout_minutes app_config row (default 20m) — no new
config row needed since payment is still mocked.
- getCustomerHistory migrated from offset (page/limit) to cursor
pagination. Cursor is base64url(`<endedAtIso>|<id>`) with id-tiebreak
in ORDER BY so rows with identical timestamps don't duplicate or skip
across pages. SELECT now JOINs payment_sessions to surface `mode`
(chat/call) for the Selesai-row voice-call pill.
- requirement/flow_customer.mermaid.md: new §7 Chat Tab subgraph + Figma
cross-ref entry for SChatList.
- requirement/phase4-customer-flow-plan.md: Stage 10 plan section. Also
carries forward earlier uncommitted "Post-Stage-8 corrections" notes
from the Stage 9 sweep (boot path / SHome1st / onboarding fixes).
Tests: +7 for getCustomerPendingPayments (initial null mitra,
targeted-mitra fill, extension-via-session JOIN, mixed-newest-first,
expired excluded, non-pending excluded, customer scoping). +10 for
cursor history (empty, exact-fit, multi-page walk, same-timestamp
tiebreak, limit clamp, customer scoping, CLOSING+COMPLETED only).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Stage 9 sweep on Client_Phone AVD + physical mitra phone:
- 01_smoke ✅
- 02_onboarding_verified ✅
- 03_onboarding_anon ✅
- 04_payment_expired ✅
- 05_searching_timeout: in progress when wrap-up began
- 06–08: not yet attempted
## Real shipping bugs fixed (would have hit prod)
1. **Router carve-out too narrow** (router.dart). The AuthAnonymousData
carve-out only protected /auth/display-name. On refreshListenable
notify after loginAnonymous resolves, GoRouter re-evaluates the
*bottom* of the navigation stack (/welcome — also an auth route),
and the AuthAnonymousData fallback redirected to /home, tearing down
the verif sheet before it could open. Loosened to allow any auth
route under AuthAnonymousData.
2. **Phase 4 multi-screen payment never called startSearch**
(searching_screen.dart). The legacy single-screen /payment did
`pairing.startSearch()` on confirm. The Phase 4 flow is
waiting → notif-gate → /chat/searching with no intermediate that
owned the call — customers would land on the searching screen with
no pairing in flight and never get matched. Added the kickoff to
searching_screen::initState when state is PairingInitialData and
paymentDraft.paymentId is set.
## Test infrastructure
- Self-contained Maestro flows 04 + 05 with inline verified-onboarding
prelude, distinct test phones per flow, robust waits.
- 02 + 03 fixed: malformed `extendedWaitUntil` (visible: + notVisible:
true → Maestro parsed as compound predicate); now use proper
notVisible: block.
- New dev-only POST /internal/_test/force-confirm-payment so flows can
advance past the waiting-payment screen without going through Xendit.
- /internal/_test/reset-phone now cascades through chat_messages →
chat_sessions → payment_sessions → auth_sessions before deleting the
customer row (FK 23503 was blocking re-runs).
- /internal/_test/force-pairing-timeout now accepts both
`searching` and `pending_acceptance` states (mitra-online dev means
the chat_session transitions through searching very quickly).
- mark_latest_payment_paid.js helper script for Stage 5+ flows.
## Maestro YAML quirks documented in flows
- text: matches anchored regex against the FULL content-desc — need .*
wildcards for substring, e.g. "mulai.*Rp.*" not "mulai".
- The middot `·` and other special unicode break naive matching;
always use .* anchors when the source string contains them.
- runFlow `when:` evaluates immediately; pair with waitForAnimationToEnd
or a preceding extendedWaitUntil before branching.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Bestie Choice Sheet on home Mulai Curhat CTA. When the user has at
least one prior session (bestieHistoryHasItemsProvider hits the chat-
sessions history endpoint), the CTA opens a HaloBottomSheet with two
cards: 'bestie yang udah kenal' -> /chat/history, 'bestie baru' ->
/payment/entry. Empty history -> direct to /payment/entry.
Bestie history list visual upgrade: HaloOrb (mitraId seed) + name +
last-session date + topic pills + sessions count + ONLINE pill.
Backend getCustomerHistory now returns topics, mitra_is_online,
sessions_count in a single payload (no per-row presence round-trip).
BestieOfflinePopup with two variants (returning | new_) replacing the
legacy BestieUnavailableDialog. tanya admin ghost CTA on both variants
opens the new TanyaAdminSheet. Stage 5's targeted-wait declined stub
+ Stage 7's chat-screen 409 stub + searching-screen call site all
migrated to the real component.
TanyaAdminSheet: HaloBottomSheet with WA + Telegram buttons, deeplinks
fetched via supportHandlesProvider (CC-config-driven). url_launcher
added to client_app; ios LSApplicationQueriesSchemes covers
https/http/whatsapp/tg.
Stage 2's OTP-blocked popup hubungi admin SnackBar stub also migrated
to TanyaAdminSheet.
Dev-only POST /internal/_test/seed-history-session lets Maestro 08
flow seed a history row before exercising the choice sheet.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Searching screen: soft-prompt card reskin, pulsing-dots panel replaces
the spinner, inline 5-min timeout panel with `coba cari lagi` (resets
pairing notifier + routes to /payment/entry for a fresh funnel — the
server-side payment is failed_pairing at that point so a stale retry
isn't valid) and `kembali ke home` ghost CTA.
Bestie-found screen: S9 Match-V4 reskin — HaloOrb + status dot +
'halo, aku bestie {name}' + `mulai sesi {N} menit →` with N pulled from
the active session's duration_minutes.
Targeted-wait overlay (new) at /chat/waiting-targeted/:mitraId. Three
sub-states from pairingProvider's PairingTargetedWaitingData:
waiting (20s countdown) / accepted (routes to chat) / declined (stubbed
BestieOfflinePopup with a TODO pointing to Stage 8). Reached via
payment_screen._routeToSearchOnConfirmed when the confirm carried a
targetedMitraId — keeps the mandatory payment-before-pairing invariant.
Dev-only POST /internal/_test/force-pairing-timeout drives the 5-min
timeout shortcut for the Maestro flow without waiting live.
Maestro 05_searching_timeout.yaml + force_pairing_timeout.js helper.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Six new screens under /payment/* + a paymentDraftProvider holding
mode/durationId/durationMinutes/priceIDR/paymentId/isFirstSessionDiscount
across the flow. PaymentEntryScreen handles the routing decision
(eligible+enabled -> /payment/discount-paywall, else /payment/method-pick)
and clears the draft on fresh entry.
Screens:
- discount_paywall_screen: S6 first-session discount with struck-through
gimmick price + actual price + 'mulai · Rp{actual}' CTA -> /payment/method
- method_pick_screen: chat vs call cards
- duration_pick_screen: tier list with chat|call mode toggle that resets
the selection on swap
- payment_method_screen: QRIS-first list, posts to existing
/api/client/payment-sessions with mode/duration/price/discount/method
- waiting_payment_screen: qr_flutter QR (encodes paymentId in mock mode),
20-min countdown header, 3s polling for status, pauses on background
via WidgetsBindingObserver
- payment_expired_screen: retry CTA -> /payment/method with draft retained
Status mapping: real payment_sessions.status uses 'confirmed'/'consumed'
for paid (not 'paid' as in plan) and 'expired'/'abandoned' as terminal.
home_screen 'Mulai Curhat' CTA now pushes /payment/entry.
Dev-only /internal/_test/force-expire-payment endpoint to drive Maestro
flow 04_payment_expired.yaml without waiting 20 minutes. Gated behind
NODE_ENV !== 'production'.
chat_opening_provider PricingData extended to carry Phase 4 chat/call
groups + firstSessionDiscount, back-compat with the Phase 3 shape.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Verif Choice Sheet on display_name_screen drives the user into either
the verified or anonymous onboarding sub-flow. ESP screen (12 chips,
multi-select, info-only) + USP screen are shared between both branches;
selections persist through to chat_sessions.topics on session start.
OTP-blocked popup (HaloPopup) listens for the four real OTP-rate-limit
error codes (OTP_RATE_LIMIT_PHONE, OTP_RATE_LIMIT_IP, OTP_COOLDOWN,
OTP_ATTEMPTS_EXCEEDED) and drops the user onto the anonymous path with
ESP/USP state preserved.
Auth-providers gating replaces the --dart-define=ENABLE_SOCIAL_AUTH
build flag with server-driven discovery. authProvidersProvider preloads
GET /api/shared/auth-providers at cold start; welcome/register/
force-register screens render Google/Apple buttons only when the
backend reports enabled:true. Falls back to phone-OTP-only when both
providers are off. social_auth_enabled.dart deleted; client_app/CLAUDE.md
updated to reflect the new gating contract.
Mitra app: chat screen renders an ESP chip strip above the first message
bubble when chat_sessions.topics is non-empty.
Backend session.service.js getSessionById SELECTs cs.topics so the mitra
side can read the customer's selected topics.
Maestro flows 02_onboarding_verified.yaml + 03_onboarding_anon.yaml.
Deviation from plan: plan referenced OTP error code 'otp_retry_exhausted';
real codes are OTP_RATE_LIMIT_*/OTP_COOLDOWN/OTP_ATTEMPTS_EXCEEDED -
popup listens for all four. Plan said 'has_paid_first_session'; live
endpoint returns 'has_consulted_before' - used the live field.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Dev-only /internal/_test/peek-otp + /internal/_test/reset-phone endpoints
gated by NODE_ENV !== 'production'. peek-otp reads the latest stub OTP
out of an in-memory map populated by otp.service.js fazpassSendStub;
reset-phone wipes otp_requests rows (and optionally the customers row)
so flows can re-run without tripping cooldowns.
JS + shell helpers under .maestro/scripts/ wrap the endpoints for use
inside Maestro runScript steps. 01_smoke.yaml expanded from a launch-only
sanity check to a full cold-start onboarding -> force-register -> OTP ->
home walk.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Replaces the home-screen pending-requests banner with a "Riwayat
Permintaan" CTA that opens a list of the mitra's last 20 chat requests
(any status). Pending rows pin to the top; non-pending rows open a
read-only detail screen with a "Lihat percakapan" CTA on accepted rows.
Backend:
- New service `getRecentRequestsForMitra(mitraId, { limit })` capped at
20, pending pinned via `(response IS NULL AND status='pending_acceptance')
DESC`. Customer call_name returned verbatim, with `'Anonim'` only as
null-safety fallback (no anonymity-flag masking — see project memory).
- New route `GET /api/mitra/chat-requests/recent`. Strictly per-mitra
scoped via the existing `resolveMitra` preHandler.
Mitra app:
- New `RequestResponse` enum in core/constants.dart.
- New Riverpod notifier `requestHistoryProvider` (AsyncValue<List<...>>,
keepAlive) — pull-to-refresh + screen-mount fetch only, no WS.
- Two new screens (history list + detail) and two new GoRoutes.
- Home screen: `_PendingRequestsBanner` removed → `_RequestHistoryButton`
Card with red count badge. Live count comes from the existing
chatRequestProvider so nothing changes about the WS-driven badge math.
Plan + acceptance criteria in requirement/phase3.5-plan.md. flutter
analyze clean (zero new issues). Backend smoke-tested against real DB.
Real-device E2E pending.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Fastify public app now passes `trustProxy: true` so request.ip resolves to
the real client IP from X-Forwarded-For when behind Cloud Run / a load
balancer. Without this the per-IP rate limit was either useless or
collapsed all users into one shared LB IP.
- The `anonymity_enabled` config row + JS default + migration seed now
default to `false`. The flag is dead code (no business logic ever
consumed it) and the actual rule is simpler than the toggle implied:
mitras always see the customer's chosen call_name; only phone+email
are private. The whole feature is queued for rip-out as a separate
cleanup pass.
The per-IP OTP rate limit (10/hr) was also effectively disabled by
upserting `app_config.otp_max_per_ip_per_hour = 1000000` — a runtime
config change, not a code change. Per-phone (3/hr) + Fazpass cost
remains the real abuse gate.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
OtpError now carries an optional details object; rate-limit branches in
checkRateLimits compute retry_after_seconds (cooldown delta for OTP_COOLDOWN,
window-roll-out delta for OTP_RATE_LIMIT_PHONE / OTP_RATE_LIMIT_IP) so the
client can disable Kirim OTP / Kirim ulang CTAs with a real countdown.
All four sendAuthError helpers (client, mitra, shared, internal) now surface
err.details and log unhandled (no statusCode) errors at level 50.
New GET /api/shared/config/otp returns the resend cooldown so the OTP screen
can gate the resend countdown without hardcoding.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Drop NOT NULL on customers.display_name so phone-OTP and social signups can
land before the user picks a name; frontend then routes them to /auth/set-name.
Google sign-in no longer requests the name claim and Apple SDK scope is
trimmed to email only.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Customer chat refreshSessionStatus now clears sessionExpired carryover so the
goodbye composer renders correctly when re-opening a closing session from
history. Backend /api/shared/chat/:id/info returns goodbye_submitted_by_me;
both apps suppress the composer for the side that has already submitted and
render an awaiting-banner view instead.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Promotes the customer-side chat WebSocket to active-session-scoped (driven
by a new `activeSessionProvider`) so home reflects session state in real
time without a per-screen connection. Backend now auto-completes sessions
left in `closing` after a 5-minute grace window so abandoned goodbye flows
don't leave the customer's home permanently locked.
Customer:
- New `activeSessionProvider` (replaces `unread_notifier`) — single source
of truth for the active session + unread count; polled every 15s.
- Chat WS lifecycle moved to `main.dart` listener on activeSessionProvider.
Chat screen joins via `connectIfNotConnected`; the new
`refreshSessionStatus` reconciles flags from the server when re-entering
an already-connected session (covers missed `sessionClosing`/`sessionExpired`
WS events).
- Home filters `closing` from the "Sesi Aktif" CTA so a session pending
goodbye doesn't block "Mulai Curhat".
- Timer-expired UX is a non-dismissible modal (Tutup / Perpanjang) instead
of an inline bar.
- Early-end goodbye composer gets an amber "Sesi telah ditutup oleh Bestie"
banner. Goodbye TextEditingController lifted to state so focus changes
no longer wipe the message.
- Closure provider reset on chat_screen mount to avoid stale
`ClosureCompleteData` from a previous session leaking into a new view.
- Chat history now lists `closing` sessions with a "Belum ditutup" badge
that routes to the live chat (goodbye composer) instead of the transcript.
Mitra:
- Same goodbye-controller fix as customer.
- Same chat-history badge + routing for `closing` items.
Backend:
- New `EndedBy.SYSTEM_AUTO_CLOSE` constant.
- `startClosureGraceTimer` extracted in `session-timer.service.js`; wired
in from `closure.initiateEarlyEnd`, `extension.rejectExtension`, and
`extension.handleExtensionTimeout`. Cancelled when customer submits
goodbye.
- Restart recovery (`restoreActiveTimers`) re-arms grace timers and stamps
any orphaned `closing` rows with `system_auto_close`.
- `getCustomerHistory` / `getMitraHistory` include `closing` alongside
`completed`; ordering uses `COALESCE(ended_at, created_at)`.
Removed: dead `session_active_screen.dart` (no router entry).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Mitra auth: parse DioException response for proper error messages
(ACCOUNT_NOT_FOUND, ACCOUNT_INACTIVE) instead of generic "OTP invalid"
- Backend: add CORS to internal app (port 3001) for control center
- Control center: fix login race condition (wait for AuthContext verify
before navigating), fix MitraActivityPage fetching paginated data
- Stale session goodbye: both apps detect SESSION_NOT_ACTIVE/409 and
move to complete state instead of retrying endlessly
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Backend: new GET /api/mitra/chat-requests/pending endpoint
- Backend: getPendingRequestsForMitra() queries unresponded notifications
for sessions still in pending_acceptance status
- Mitra app: loadPendingRequests() fetches on screen load + status toggle
- Mitra app: activeRequestCount getter exposes queue size
- Mitra app: _PendingRequestsBanner widget shows count with tap-to-view
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- DB migration: add active_session_count column + mitra_notified index
- Constants: add MISSED to NotificationResponse
- Pairing service: record active_session_count on notification creation,
use MISSED (not IGNORED) when another mitra accepts first
- New mitra-activity.service.js: getMitraActivityLog (paginated),
getMitraActivitySummary (per-mitra aggregates with acceptance rate)
- New mitra-activity.routes.js: GET /internal/mitra-activity/log,
GET /internal/mitra-activity/summary
- Control center: new MitraActivityPage with summary table + detail log,
filters (mitra, date range), color-coded response types, pagination
- Register route in App.jsx, add "Aktivitas Mitra" nav link in Layout
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
