Files
nick-doc/09 - Audits/Activity Log.md
2026-06-05 07:14:48 +03:30

124 KiB
Raw Blame History

title, tags, created
title tags created
Activity Log
audit
log
append-only
2026-05-28

Activity Log

Append-only log of every git push from backend and frontend. Newest entries on top. Maintained by agents per the rule in ../AGENTS.md.


2026-06-05 — backend@v2.8.84, frontend@v2.8.101 — Notifications wired end-to-end (id-seam normalize + Mini App joins user room)

Commits: backend v2.8.84, frontend v2.8.101 Touched: backend services/notification/NotificationService.ts; frontend sections/telegram/hooks/use-telegram-notifications.ts Why: No notifications reached the Mini App (in-app, real-time, or Telegram push). Two root causes. (1) Id seam: notifications were created with whatever id the event carried — usually a Postgres uuid (sellerId/buyerId) — but everything that consumes a notification keys on the user's session legacy ObjectId: the in-app fetch (req.user.id), the user-<id> socket room, and TelegramLink.userId. So a notification stored under a uuid was invisible to fetch, never reached the room, and sendTelegramNotificationToUser(uuid) found no_link. Fix: NotificationService.createNotification/createNotificationsBulk now normalise userId via toCanonicalUserId (uuid → users.legacy_object_id) before persist + real-time emit + Telegram forward, so all three line up. (2) Mini App never joined its room: the socket connected but use-telegram-notifications only registered listeners — it never emitted join-user-room, so the backend's targeted new-notification emits had no subscriber. Added a joinUserRoom(userId) effect that (re)joins on every connect. selfId is the session ObjectId, matching the socket's authed id and the now-normalised emit target. Verification: backend npx tsc --noEmit; frontend npx tsc --noEmit --ignoreDeprecations 6.0 + eslint — clean. After deploy: a marketplace event (offer received/accepted, payment, delivery) creates a notification that (a) shows in the Mini App bell list, (b) bumps the unread badge live, and (c) arrives as a Telegram bot message for linked users.


2026-06-05 — backend@v2.8.83, frontend@v2.8.100 — Select-offer 403 (id seam) + offer delivery-time [object Object]

Commits: backend v2.8.83, frontend v2.8.100 Touched: backend services/marketplace/marketplaceController.ts (selectOffer); frontend sections/telegram/view/telegram-request-detail-view.tsx Why: Testing the new Mini App «انتخاب و پرداخت» surfaced two bugs. (1) 403 on select-offer: selectOffer gated buyer ownership with toIdString(purchaseRequest.buyerId) !== buyerId — the session legacy ObjectId vs the PG-uuid buyerId (the recurring seam) → the real buyer got 403. Switched to !(await sameUser(buyerId, toIdString(purchaseRequest.buyerId))). (2) Delivery time rendered [object Object]: the offer card assumed offer.deliveryTime was a number, but it can be an object ({amount, unit}). Now extracts dt.amount ?? dt.value ?? offer.deliveryTimeAmount when it isn't a plain number. Verification: backend npx tsc --noEmit; frontend npx tsc --noEmit --ignoreDeprecations 6.0 + eslint — clean. After deploy: buyer taps «انتخاب و پرداخت» → offer accepted (no 403) → in-shell payment opens; delivery time shows a number of days, not [object Object].


2026-06-05 — frontend@v2.8.99 — Mini App buyer: see received offers + select & pay in-shell

Commits: frontend only (v2.8.99) Touched: sections/telegram/view/telegram-request-detail-view.tsx, sections/telegram/locales/{fa,en,types}.ts (offers_title, offer_select_pay, offer_delivery_days, offer_accepting) Why: When the buyer's stepper reached «انتخاب و پرداخت» (step 3, status received_offers) the Mini App had no way to act — the pay CTA only showed for pending_payment. Now, for a buyer on an offer-selection status, the detail view fetches and lists the received offers (seller, price, delivery days, description) and each has a «انتخاب و پرداخت» button → acceptOffer(requestId, offerId) (sets selectedOfferId, moves to pending_payment) → refresh() → opens the in-shell direct-transfer payment screen via onPay. Reused the existing offers SWR (broadened from review-only to also fire on received_offers/offers_received/in_negotiation). Verification: frontend npx tsc --noEmit --ignoreDeprecations 6.0 + eslint clean. After deploy: buyer opens a request with an offer → sees the offer card → «انتخاب و پرداخت» → in-shell payment opens; paying advances the stepper.


2026-06-05 — frontend@v2.8.98 — Mini App seller shop: product-type filter + show controls for small shops

Commits: frontend only (v2.8.98) Touched: sections/telegram/view/telegram-seller-shop-view.tsx Why: The seller shop (per-seller template list) already had search + sort but (1) no filter, and (2) the controls bar was gated on allTemplates.length > 3, so a typical 3-template shop showed no controls at all. Added a product-type filter (chips built only for the types the seller actually sells — physical/digital/service/consultation, plus «همه») wired through TelegramListControls, and lowered the controls threshold to > 1 so shops with 23 templates still get search/filter/sort. Verification: frontend npx tsc --noEmit --ignoreDeprecations 6.0 + eslint clean. After deploy: opening a seller shop with ≥2 templates shows the controls bar; the filter chips reflect the seller's product types and narrow the list.


2026-06-04 — backend@v2.8.82, frontend@v2.8.97 — Stepper advances: offer→step, seller offer detection, real-time on code-verify

Commits: backend v2.8.82, frontend v2.8.97 Touched: backend services/marketplace/marketplaceController.ts (getOffersForRequest sellerId filter, verifyDeliveryCode socket emit); frontend sections/request/request-config.tsx (buyer received_offers→3), actions/marketplace.ts (getSellerOfferForRequest fallback) Why: Three related stepper bugs surfaced while testing an offer end-to-end. (1) Buyer stepper stuck at "awaiting offers" (2) after an offer arrived: determineBuyerStep('received_offers') returned offersCount>0 ? 3 : 2, but callers that don't thread offersCount (the Mini App) got 2. The status itself means an offer arrived on the buyer's own request, so it now returns 3 unconditionally. (2) Seller stuck on "send offer" (1) after submitting: getSellerOfferForRequest matched offers by o.sellerId === user._id, but o.sellerId is a PG uuid and user._id is the legacy ObjectId (the seam) → no match → sellerOfferStatus undefined → step 1. Backend getOffersForRequest now filters to the seller's own offers via sameUser when ?sellerId= is passed, and the client returns data[0] as a fallback. (3) Seller needed a manual refresh to go from "awaiting buyer" (4) to "receive funds" (5): the seller's code entry (verifyDeliveryCode) replaced the buyer's confirmDelivery in v2.8.95 but never emitted a socket event, so no live update. Added the purchase-request-update / status-changed emit on successful verify (matching what confirmDelivery did). Verification: backend npx tsc --noEmit; frontend npx tsc --noEmit --ignoreDeprecations 6.0 + eslint — all clean. After deploy: buyer sees step 3 once an offer arrives; seller sees step 2 right after sending an offer; seller's stepper advances to "receive funds" without a manual refresh after entering the code.


2026-06-04 — frontend@v2.8.96 — Mini App account: don't show "email verified" when the user has no email

Commits: frontend only (v2.8.96) Touched: sections/telegram/view/telegram-account-view.tsx, sections/telegram/locales/{fa,en,types}.ts (email_not_set) Why: A Telegram sign-up has isEmailVerified=true but no email address, so the account header showed a green «ایمیل تأیید شده» badge even though the email field was empty. Gated the verified badge on user?.email && user?.isEmailVerified. Added a third state for the no-email case: a «ایمیل ثبت نشده» badge that opens the in-shell settings (onOpenSettings) to add an email. The not-verified (has-email-but-unverified) branch is unchanged. Verification: frontend npx tsc --noEmit --ignoreDeprecations 6.0 + eslint clean. After deploy: a Telegram user with no email sees «ایمیل ثبت نشده» (→ settings), not a false «ایمیل تأیید شده».


2026-06-04 — frontend@v2.8.95 — Delivery confirmation: code is entered by the seller, buyer's «تایید دریافت کالا» removed

Commits: frontend only (v2.8.95) Touched: sections/telegram/view/telegram-request-detail-view.tsx (Mini App buyer: drop confirm button), sections/request/components/buyer-steps/step-5-receive-goods.tsx (web buyer: drop confirm button), sections/request/components/seller-steps/step-4-waiting-for-confirmation.tsx (web seller: code display → code input) Why: The handover flow was wrong/duplicated: the buyer had a «تایید دریافت کالا» button (confirmDelivery) AND the seller was shown the expected code. Per the user, the single correct mechanism is: the buyer gives the 6-digit code, the seller enters it to confirm — and that 403'd anyway (confirmDelivery compared session ObjectId !== PG-uuid buyerId — same id seam). Changes: (1) Mini App buyer (status delivery) now only displays the code to hand over — the confirmDelivery button/handler/import removed. (2) Web buyer step-5: removed the «تایید دریافت کالا» button (kept the code display + an explanatory note). (3) Web seller step-4: replaced the code display (getDeliveryCode, which leaked the code to the seller and 403s post-v2.8.79) with a 6-digit input + «تایید تحویل کالا» calling verifyDeliveryCode(id, code). Both confirmDelivery and verifyDeliveryCode reach the same delivered state, so no escrow-release regression. (Mini App seller already had the input from v2.8.94.) Verification: frontend npx tsc --noEmit --ignoreDeprecations 6.0 + eslint clean. After deploy: buyer sees only the code (no confirm button); seller enters the buyer's code → status → delivered; no 403.


2026-06-04 — backend@v2.8.81 — Public shop settings: resolve uuid→legacy ObjectId (name/avatar were blank publicly)

Commits: backend v2.8.81 Touched: services/marketplace/shopSettingsStore.ts (getSellerShopSettings + new resolveSellerLegacyId) Why: Found while setting shop name/avatar/cover for the three seed sellers. After PUT-ing settings, the seller's OWN GET /marketplace/shop/settings returned them correctly, but the PUBLIC GET /marketplace/shop/settings/:sellerId (what the shop page shows buyers) returned data:null — blank name, no avatar/cover. Root cause: the recurring Mongo↔PG id seam. shop_settings rows are keyed by seller_legacy_object_id (the seller's legacy Mongo ObjectId, which is what req.user.id is on the own-settings path), but the public sellers list returns each seller by their PG uuid, so getSellerShopSettings(uuid) queried where seller_legacy_object_id = <uuid> and matched nothing. Fix: when the direct lookup misses, resolve the incoming id via select legacy_object_id from users where id=$1 (a 24-hex id is treated as already-legacy) and retry. Mirrors the existing reverse resolveSellerUuid helper. Verification: backend npx tsc --noEmit clean. After deploy: GET /marketplace/shop/settings/<pg-uuid> returns the saved name/avatar/cover for all three sellers (was null). Seeded: seller/seller1/seller2 @amn.gg each have a shop name, description, avatar, cover image, social links, and 3 active products.


2026-06-04 — backend@v2.8.80 — Shop sellers list cache: invalidate global templates:list on per-seller change

Commits: backend v2.8.80 Touched: services/redis/cacheService.ts (invalidateTemplatesCache) Why: Found while seeding shops for three sellers via API: after creating templates for seller1@amn.gg/seller2@amn.gg, the public shop endpoint (GET /marketplace/request-templates/sellers) still returned only the first seller. Root cause — the global sellers list is cached under Redis key templates:list, but RequestTemplateService create/update/delete call invalidateTemplatesCache(sellerId) which only deleted templates:seller:<id>, never templates:list. So a newly-onboarded seller didn't appear in /shop until the 5-minute TTL elapsed. Fix: when invalidating with a sellerId, also del('templates:list') since the aggregated list changes whenever any seller adds/edits/removes a template. Verification: backend npx tsc --noEmit clean. After deploy: a brand-new seller's first template appears in GET …/request-templates/sellers (and dev.amn.gg/shop) immediately, not after TTL. (Seeded data: seller/seller1/seller2 @amn.gg each have 3 active templates — two at 0.01 USDT, one at 0.05 — with images, BSC USDT/USDC payment config.)


2026-06-04 — frontend@v2.8.94 — Telegram Mini App: seller delivery-code entry field

Commits: frontend only (v2.8.94); backend stays v2.8.79 Touched: telegram-request-detail-view.tsx (seller code-entry card), locales/{fa,en,types}.ts (seller_code_* strings) Why: The escrow handover is two-sided: the buyer generates a 6-digit delivery code (shown in v2.8.92) and the seller enters it to confirm delivery. The Mini App seller had no field for this — at status delivery the seller only saw «منتظر تایید خریدار» with nowhere to type the code. Added a code-entry card for role === 'seller' && status === 'delivery': a numeric 6-digit input (digit-filtered, monospace, centred) + «تایید تحویل کالا» button calling verifyDeliveryCode(requestId, code) then refresh() so the status advances to delivered. Backend verifyDeliveryCode already uses the sameUser seller gate (v2.8.77), so this composes with the id-seam fix. Verification: frontend npx tsc --noEmit --ignoreDeprecations 6.0 + eslint clean. Admin verifies after deploy: buyer opens delivery request → sees code; seller opens same request in Mini App → enters that code → «تایید تحویل کالا» → status → delivered.


2026-06-04 — backend@v2.8.79 — Delivery code: apply sameUser to buyer auth gates (Mini App showed "—")

Commits: backend v2.8.79 (frontend stays v2.8.93) Touched: services/marketplace/marketplaceController.ts (generateDeliveryCode, getDeliveryCode) Why: In the Mini App receive-goods step the delivery code rendered as «—»: both getDeliveryCode and generateDeliveryCode returned 403. Root cause is the recurring Mongo↔PG id seam — these two handlers still used the strict request.buyerId !== userId comparison (session legacy ObjectId vs PG-uuid buyerId), unlike verifyDeliveryCode which was already moved to the async sameUser helper in v2.8.77. Replaced both buyer checks with !(await sameUser(userId, toIdString(request.buyerId))). Added an explicit !userId unauthorized guard in getDeliveryCode to restore the string narrowing the removed !== previously provided (so getDeliveryCode(id, userId) still type-checks). The repo layer (getDeliveryCodeForUser) already resolves the seam via resolveEntityId, so once the code exists it loads without regenerating. Verification: backend npx tsc --noEmit clean. Admin verifies after deploy: buyer opens a delivery-status request in the Mini App → a 6-digit delivery code appears (no «—», no 403); seller can verify it.


2026-06-04 — backend@v2.8.78, frontend@v2.8.93 — Gate shop orders on payment: abandon-cancel + 30-min TTL sweep

Commits: backend v2.8.78, frontend v2.8.93 Touched: backend services/admin/dataCleanupService.ts (new cancelStaleUnpaidRequests), services/admin/ttlCleanupJob.ts (schedule it every 5 min); frontend telegram-payment-view.tsx (abandon-cancel on unmount) Why: Shop checkout creates the purchase request in pending_payment before the buyer pays — unavoidable because the direct-balance deposit address is derived per-request, so the request must exist for the AMN scanner to credit it (the payment-intent route requires purchaseRequestId/sellerOfferId/sellerId). That left unpaid orders lingering. Now, per the user's choice (both mechanisms, 30-min window, only cancel orders with no deposit): (1) Frontend abandon-cancel — when the buyer leaves the in-shell checkout payment screen without completing, the just-created order is set to cancelled via updateRequestStatus; two refs (paidRef, depositRef) suppress the cancel if the deposit was confirmed or a partial deposit was detected, so funds are never orphaned. (2) Backend TTL sweepcancelStaleUnpaidRequests runs every 5 min, cancels pending_payment requests older than 30 minutes that have no active payment (findActivePaymentForRequest === null); any active/slow on-chain payment is left untouched. The sweep is the backstop for app-closed cases the frontend unmount can't catch. Verification: backend npx tsc --noEmit clean; frontend npx tsc --noEmit --ignoreDeprecations 6.0 + eslint clean. Admin verifies after deploy: shop checkout → reach payment step → press back without paying → order shows cancelled (not lingering pending_payment); a paid/underpaid order is NOT cancelled.


2026-06-04 — frontend@v2.8.92 — Telegram Mini App: buyer receive-goods/confirm + checkout stepper line fix

Commits: frontend only (v2.8.92); backend stays at v2.8.77 Touched: telegram-request-detail-view.tsx (buyer receive-goods flow), components/telegram-request-stepper.tsx (row width cap/centre), locales/{fa,en,types}.ts (delivery-code + confirm-receipt strings) Why: (1) The buyer had no way to confirm receipt inside the Mini App — once the seller shipped (status delivery, fixed in backend v2.8.77's sameUser patch), the buyer was stuck at step 5 «دریافت کالا» with no action. Mirrored the web step-5-receive-goods: on status === 'delivery' for a buyer, auto-load/generate the delivery code (getDeliveryCode → fallback generateDeliveryCode) shown dashed/monospace for the buyer to hand the seller, plus a «تایید دریافت کالا» button calling confirmDelivery(requestId) then refresh() so the stepper advances to «تایید نهایی». (2) The 3-step checkout stepper stretched each cell across the full Telegram width, leaving a long connector stub past the last dot — capped the steps row at steps.length * 96 and centred it (margin: 0 auto), so both the 3-step checkout and 6-step request steppers render tight, even connectors. Verification: frontend npx tsc --noEmit --ignoreDeprecations 6.0 clean. Admin verifies after deploy: buyer opens a delivery-status request → sees delivery code + «تایید دریافت کالا» → tap → request advances to تایید نهایی; checkout stepper line is clean (no stub).


2026-06-03 — scanner@ccd96e8, backend@22ae0bd, frontend@10c4292 — direct scanner balance checks and balance watches

Commits: scanner ccd96e8 (tag v0.1.8), backend 22ae0bd (tag v2.8.60), frontend 10c4292 (tag v2.8.60) Touched: scanner: balance.go, balance_watch.go, balance_test.go, api.go, config.go, intent.go, main.go, README.md, VERSION; backend: src/services/payment/adapters/amnPayAdapter.ts, src/routes/amnScannerWebhookRoutes.ts, package.json, package-lock.json; frontend: package.json version metadata only. Why: Add scanner primitives for non-smart-contract/direct-address token payments: synchronous ERC-20 balance checks, persisted balance watches with 5/10/20/40-minute cadence decay, 7-day expiry, signed balance_changed callbacks, and backend adapter/webhook plumbing. Backend still needs the product decision layer that turns validated balance deltas into funded payments; documented in the new direct-address PRD. Verification: scanner GOCACHE=/private/tmp/codex-go-cache go test -count=1 ./...; backend npm run typecheck; frontend not built because only version metadata changed. Linked docs updated: Scanner API, Scanner Architecture, Payment Flow - Scanner, ScannerBalanceWatch, Scanner Operations, Environment Variables, PRD - Direct Address Token Payments via Scanner Balance Watches


2026-06-03 — frontend@9bafbbb — Telegram Mini App: full in-shell shop, account tab parity, and shopping cart (v2.8.57v2.8.59)

Commits: a8ae1e3 (v2.8.57), 6dc3918 (v2.8.58), 9bafbbb (v2.8.59) — frontend only; backend stays at v2.8.56 Touched: telegram-mini-app-view.tsx (shell nav: openSellerId, overlayScreen='cart', BackButton dismissal chain), telegram-shop-view.tsx (cart badge header button, TelegramShopRow converted from <a href> to in-shell onOpen), telegram-seller-shop-view.tsx (new — seller header, active templates with budget/usage, add/remove-to-cart buttons, floating cart CTA), telegram-cart-view.tsx (new — qty stepper, remove, USDT total, "Continue to payment" → web checkout), telegram-account-view.tsx (new — profile header, preferences section, help section, sign-out with bottom-sheet confirm), hooks/use-telegram-seller-shop.ts (new — SWR over getSellerWithTemplates), hooks/use-telegram-cart.ts (new — localStorage app-request-template-checkout, custom tg-cart-changed event), hooks/use-telegram-shops.ts (new — SWR over getTemplateSellers), telegram locales (fa/en/types) Why: Three sequential buyer-parity milestones shipped in one session. (1) v2.8.57: tapping a seller in فروشگاه previously opened the web dashboard inside the webview; the seller store now renders entirely in-shell with TelegramSellerShopView — seller header, active templates, budget/usage count — and template ordering hands off to the web from-template checkout where the wallet stack lives. (2) v2.8.58: the web account menu (تنظیمات عمومی، اعلان‌ها، کیف پول، آدرس‌های تحویل، Passkey) had no Mini App counterpart; all five are now accessible from the account tab — notifications open the existing in-shell overlay, the other four deep-link to web dashboard pages (labeled «در داشبورد وب باز می‌شود») because passkey and wallet require browser-context APIs unavailable in Telegram. (3) v2.8.59: phase 1 of full buyer parity — new useTelegramCart hook writes to the same localStorage key (app-request-template-checkout) that the web checkout provider reads, so the Mini App cart IS the web cart with no sync step; TelegramCartView adds qty controls, remove, and a "Continue to payment" link that hands off seamlessly. Remaining buyer-parity roadmap: offers view/accept on requests, delivery confirmation, payments list, points/referral, addresses CRUD, TON Connect payments (Telegram Wallet is TON-native; backend already has tonProofService). Verification: tsc + eslint clean across all three commits. Admin verifies after deploy: (shop) فروشگاه → tap seller → in-shell shop → سفارش این قالب → web checkout; (cart) افزودن به سبد → cart badge → سبد overlay → ادامه و پرداخت → web checkout shows same items; (account) account tab → preferences rows → notifications overlay in-shell, remaining rows open web dashboard. Linked docs updated: 04 - Flows/Telegram Mini App.md (major update — navigation model, all view files, shop/cart/checkout flow, account tab, SDK surfaces, API call table), 01 - Architecture/Frontend Architecture.md (updated — Telegram Mini App section)


2026-06-02 — backend@cf59726, frontend@a2b972b — normalize Postgres repository store modes

Commits: backend cf59726 (version 2.8.37), frontend a2b972b (version 2.8.37) Touched:

  • Backend: src/db/repositories/factory.ts, src/services/health/healthCheckService.ts, __tests__/repository-factory-modes.test.ts, __tests__/health-check-service.test.ts, package.json, package-lock.json
  • Frontend: package.json version metadata only. Why: Fix migration-control drift found after enabling notification PG mode in dev. Repository factory flags now treat postgres as an alias for pg, so NOTIFICATION_STORE=postgres actually resolves to the Drizzle notification repo. Release-hold mode is now independent from dispute mode, so DISPUTE_STORE=postgres no longer makes health or repository selection falsely report release holds as cut over. Verification: Backend npm test -- --runTestsByPath __tests__/repository-factory-modes.test.ts __tests__/health-check-service.test.ts __tests__/notification-service-repo.test.ts --runInBand; backend npm run typecheck; backend/frontend git diff --check. Linked docs updated: Postgres Runtime Cutover Status

2026-06-02 — backend@882096f, deployment@8764fdf — enable notification Postgres dev cutover

Commits: backend b64995a and 882096f (version 2.8.36), frontend 28ad8e6 (version 2.8.36, already on remote), deployment 8764fdf Touched:

  • Backend: src/services/marketplace/index.ts, __tests__/marketplace-runtime-import-surface.test.ts, package.json, package-lock.json
  • Deployment: docker-compose.yml, gatus/config.yaml Why: Continue MongoDB removal by detaching the unmounted legacy marketplace router export that still pulled in top-level Mongoose/model imports, and move notifications from PG-capable/operator-ready to PG-backed in the dev runtime. Dev Gatus now requires eight enabled Postgres stores and explicitly asserts storeModes.notification == "postgres". Verification: Backend npm run typecheck; backend npm test -- --runTestsByPath __tests__/marketplace-runtime-import-surface.test.ts __tests__/blog-service-repo.test.ts __tests__/health-check-service.test.ts --runInBand; backend active-surface static scan for non-type top-level mongoose/models/* imports returned no matches; backend/frontend/deployment git diff --check. Frontend push was not needed because remote frontend was already at 2.8.36. Linked docs updated: Postgres Runtime Cutover Status

2026-06-02 — backend@f1ba14b, frontend@b94d8a9 — add notification Postgres backfill tooling

Commits: backend f1ba14b (version 2.8.34), frontend b94d8a9 (version 2.8.34) Touched:

  • Backend: src/services/notification/notificationBackfill.ts, src/scripts/backfillNotificationPostgres.ts, src/db/backfill/backfill-notifications.ts, src/db/backfill/run-backfill.ts, scripts/smoke/notifications-postgres.sh, __tests__/notification-backfill.test.ts, package.json, package-lock.json
  • Frontend: package.json version metadata only. Why: Continue the MongoDB removal by making the already repo-backed notification domain operator-ready for Postgres cutover. Existing notification runtime paths use getNotificationRepo(); this adds Mongo→Postgres backfill, ordered-runner inclusion, dry-run support, and a focused PG smoke for notification create/list/read/delete behavior. Verification: Backend npm run typecheck; backend npm test -- --runTestsByPath __tests__/notification-service-repo.test.ts __tests__/notification-backfill.test.ts __tests__/health-check-service.test.ts --runInBand; backend bash scripts/smoke/marketplace-core-postgres-backfill.sh (static checks passed; optional live dry-run skipped because migration DSNs were not set); backend active-surface static scan for non-type top-level mongoose/models/* imports returned no matches; backend/frontend git diff --check. The new scripts/smoke/notifications-postgres.sh was added but not run locally because no PG_URL/DATABASE_URL/POSTGRES_URL is configured in the backend env files. Linked docs updated: Postgres Runtime Cutover Status

2026-06-02 — backend@10de752, frontend@3dfbac2 — defer legacy Mongo runtime imports

Commits: backend 10de752 (version 2.8.33), frontend 3dfbac2 (version 2.8.33) Touched:

  • Backend: src/infrastructure/database/connection.ts, src/services/health/healthCheckService.ts, src/services/admin/dataCleanupService.ts, src/services/payment/migration/reportService.ts, package.json, package-lock.json
  • Frontend: package.json version metadata only. Why: Continue removing MongoDB as a runtime dependency by deferring the remaining active startup/health/admin/report Mongoose/model loads. Legacy Mongo connection, health ping, admin cleanup, and SHKeeper migration report paths still work, but now load Mongoose/models only when those legacy actions actually run. Verification: Backend npm run typecheck; backend npm test -- --runTestsByPath __tests__/health-check-service.test.ts --runInBand; backend npm test -- --runTestsByPath __tests__/payment-migration.service.test.ts --runInBand; backend static scan for non-type top-level mongoose/models/* imports in active app/routes/services/infrastructure returned no matches; backend/frontend git diff --check; BASE_URL=https://dev.amn.gg bash scripts/smoke/backend-health.sh passed against dev before push. Linked docs updated: Postgres Runtime Cutover Status

2026-06-02 — backend@134d155, frontend@18af5dd — lazy-load PG-capable store Mongo fallbacks

Commits: backend 134d155 (version 2.8.32), frontend 18af5dd (version 2.8.32) Touched:

  • Backend: src/services/config/configStore.ts, src/services/marketplace/reviewStore.ts, src/services/marketplace/shopSettingsStore.ts, package.json, package-lock.json
  • Frontend: package.json version metadata only. Why: Continue removing MongoDB as a runtime dependency by making the already PG-capable config, review, and shop-settings stores avoid top-level Mongoose/model imports. Mongo fallback/backfill paths still lazy-load the legacy models when they are actually used, preserving rollback/mirror behavior. Verification: Backend npm run typecheck; backend bash scripts/smoke/marketplace-core-postgres-backfill.sh (static Jest checks passed; optional live dry-run skipped because MIGRATION_MONGO_URL/MIGRATION_PG_URL were not set); backend npx jest __tests__/health-check-service.test.ts __tests__/telegram-service.test.ts __tests__/telegram-auth.test.ts --runInBand rerun outside the sandbox after MongoMemoryServer hit local bind restrictions (16/16 passed); backend/frontend git diff --check. Linked docs updated: Postgres Runtime Cutover Status

2026-06-01 — backend@2c5c3c7, frontend@775a73b — route funds ledger through payment repo seam

Commits: backend 2c5c3c7, frontend 775a73b (backend 2.8.20, frontend 2.8.20) Touched:

  • Backend: src/services/payment/ledger/fundsLedgerService.ts, src/db/repositories/factory.ts, src/db/repositories/drizzle/DrizzlePaymentRepo.ts, src/db/repositories/mongo/MongoPaymentRepo.ts, __tests__/payment-ledger.service.test.ts, __tests__/mongo-payment-repo.test.ts, scripts/smoke/funds-ledger-repo.sh, package.json, package-lock.json
  • Frontend: package.json, Dockerfile version metadata only. Why: Start replacing the remaining PG-capable repository stores with a low-risk money-core slice. Funds ledger appends and balance reads now go through getPaymentRepo(), so REPO_PAYMENT=mongo|dual|pg can control the ledger path. The repo factory now lazy-loads PG/dual implementations so importing it in Mongo mode no longer requires PG_URL. Mongo/Drizzle payment repo stats were aligned with live behavior (buyerId, nested amount.amount, and completed counted as successful), and Drizzle ledger balance reads now match external/string refs as well as UUID refs. Verification: Backend git pull --rebase --autostash (already up to date); backend scripts/smoke/funds-ledger-repo.sh; backend npm test -- --runTestsByPath __tests__/payment-release-refund-orchestration.test.ts __tests__/money-safety.test.ts --runInBand (money-safety skipped by its own env guard); backend npm run typecheck -- --pretty false; backend npm run build:server; frontend git pull --rebase --autostash (already up to date); frontend npx tsc --noEmit --project tsconfig.json; frontend npm run build (passed with the existing non-fatal SSR getPosts fetch refusal during static page generation); backend/frontend git diff --check. Linked docs updated: Postgres Runtime Cutover Status, MongoDB to PostgreSQL Migration Plan (Drizzle), Payment

2026-06-01 — deployment@38cb75b — default PG-capable stores to Postgres in dev

Commits: deployment 38cb75b Touched:

  • Deployment: docker-compose.yml, gatus/config.yaml Why: Move the existing PG-capable runtime stores from opt-in env settings to the default dev runtime: AUTH_STORE, CONFIG_STORE, ADDRESS_STORE, CATEGORY_STORE, LEVEL_CONFIG_STORE, SHOP_SETTINGS_STORE, and REVIEW_STORE now default to postgres in compose. Gatus now verifies checks.postgres.enabledStoreCount >= 7 and asserts each of those store modes is postgres, so monitoring catches partial cutover drift. Verification: Deployment YAML parse via ruby -e 'require "yaml"; YAML.load_file("docker-compose.yml"); YAML.load_file("gatus/config.yaml"); puts "yaml ok"'; deployment docker compose config --quiet. Linked docs updated: Postgres Runtime Cutover Status, Monitoring, Gatus Monitoring - Proposed Config

2026-06-01 — backend@c5db471, frontend@f424a03 — add RequestTemplate Postgres backfill surface

Commits: backend c5db471, frontend f424a03 (backend 2.8.19, frontend 2.8.19) Touched:

  • Backend: src/db/schema/requestTemplate.ts, src/db/backfill/backfill-requestTemplates.ts, src/db/backfill/run-backfill.ts, src/db/backfill/_idMap.ts, src/db/schema/purchaseRequest.ts, src/db/migrations/0010_request_templates.sql, src/db/migrations/meta/_journal.json, __tests__/marketplace-core-backfill.test.ts, package.json, package-lock.json
  • Frontend: package.json, Dockerfile version metadata only. Why: Start the 28 migration pass with the lowest-risk marketplace-core gap. RequestTemplate now has a PG schema, id-map collection entry, ordered backfill step, and marketplace-core runner coverage. The same migration also adds the missing unique index for purchase_request_specifications (purchase_request_id, key), matching the existing backfill upsert target. Verification: Backend ./scripts/smoke/marketplace-core-postgres-backfill.sh (static checks passed; optional live dry-run skipped because MIGRATION_MONGO_URL/MIGRATION_PG_URL were not set); backend npm test -- --runTestsByPath __tests__/marketplace-core-backfill.test.ts; backend npm run typecheck -- --pretty false; backend npm run build:server; frontend npx tsc --noEmit --project tsconfig.json; frontend npm run build (passed with the existing non-fatal SSR getPosts fetch refusal during static page generation); backend/frontend git diff --check. Linked docs updated: RequestTemplate, Postgres Runtime Cutover Status, MongoDB to PostgreSQL Migration Plan (Drizzle), MongoDB to PostgreSQL Migration Guide

2026-06-01 — backend@1543b53, frontend@457de07 — enforce unique active categories

Commits: backend 1543b53, frontend 457de07 (backend 2.8.17, frontend 2.8.17) Touched:

  • Backend: src/services/marketplace/categoryStore.ts, src/services/marketplace/CategoryService.ts, src/db/schema/category.ts, src/db/migrations/0009_unique_active_categories.sql, src/db/migrations/meta/_journal.json, __tests__/category-store.test.ts, scripts/smoke/categories-postgres-unique.sh, scripts/smoke/reference-stores-postgres.sh, package.json, package-lock.json
  • Frontend: package.json version metadata only. Why: Category seed/backfill reruns could leave multiple active rows with the same visible label, which surfaced in the category dropdown as repeated Persian names. The PG category store now deactivates duplicate active rows before adding a normalized active-name unique index, repoints existing category references to the kept row, catches duplicate inserts as idempotent creates, and dedupes cached/list responses so stale rows cannot leak into the UI. Verification: Backend npm test -- --runTestsByPath __tests__/category-store.test.ts __tests__/postgres-client.test.ts; backend npm run typecheck; backend npm run build:server; backend PG_URL=postgresql://escrow:throwaway@127.0.0.1:5434/escrow_migration_test ./scripts/smoke/categories-postgres-unique.sh; frontend npx tsc --noEmit --project tsconfig.json; frontend npm run build (passed with the existing non-fatal SSR getPosts fetch refusal during static page generation). Linked docs updated: Category, Postgres Runtime Cutover Status, MongoDB to PostgreSQL Migration Plan (Drizzle), MongoDB to PostgreSQL Migration Guide

2026-06-01 — backend@6df113d, frontend@0f1db64 — harden marketplace-core Postgres backfill

Commits: backend 6df113d, frontend 0f1db64 (backend 2.8.13, frontend 2.8.13) Touched:

  • Backend: src/db/backfill/backfill-purchaseRequests.ts, src/db/backfill/run-backfill.ts, __tests__/marketplace-core-backfill.test.ts, scripts/smoke/marketplace-core-postgres-backfill.sh, package.json, package-lock.json
  • Frontend: src/components/hook-form/rhf-select.tsx, tsconfig.json, yarn.lock, package.json Why: The next Postgres migration slice is marketplace core. The existing PurchaseRequest/SellerOffer backfill path had real cutover blockers: the purchase_requests insert omitted updated_at, preferred sellers wrote the wrong junction column, and selected offers were remapped before seller offers existed. The runner now exposes a marketplaceCore group and a post-offer selected-offer remap step. Frontend typecheck was restored after the pulled request-template changes by installing the declared rehype-sanitize lock entry, allowing displayEmpty through RHFSelect, and silencing TS6's baseUrl deprecation gate. Verification: Backend ./scripts/smoke/marketplace-core-postgres-backfill.sh (static checks passed; optional live dry-run skipped because MIGRATION_MONGO_URL/MIGRATION_PG_URL were not set); backend npm run typecheck; backend npm run build:server; frontend npx tsc --noEmit --project tsconfig.json; frontend npm run build (passed with a non-fatal SSR getPosts fetch refusal during static page generation); backend/frontend git diff --check. Linked docs updated: Postgres Runtime Cutover Status, MongoDB to PostgreSQL Migration Plan (Drizzle), MongoDB to PostgreSQL Migration Guide

2026-06-01 — backend@ea43862, frontend@b4ea7c9 — expose Postgres store modes in health

Commits: backend ea43862, frontend b4ea7c9 (backend 2.8.11, frontend 2.8.11) Touched:

  • Backend: src/infrastructure/postgres/client.ts, src/services/health/healthCheckService.ts, __tests__/postgres-client.test.ts, package.json, package-lock.json
  • Frontend: package.json, Dockerfile version metadata only. Why: Phase 0 of the Postgres cutover needed runtime visibility before moving another domain. /api/health now reports checks.postgres.storeModes, enabledStores, and enabledStoreCount while preserving the existing configured/required/status semantics. This lets Gatus and operators verify which opt-in stores are actually PG-backed in dev. Backend was rebased on Mojtaba's 2be91d2 authStore fix before push. Verification: Backend npm test -- --runTestsByPath __tests__/postgres-client.test.ts; backend npm run typecheck -- --pretty false; backend npm run build; backend focused health smoke against local Mongo/Redis containers and escrow-pgmig-test Postgres asserted all seven enabled PG stores in checks.postgres; frontend npx tsc --noEmit --ignoreDeprecations 6.0; backend/frontend git diff --check. Linked docs updated: Monitoring, Postgres Runtime Cutover Status

2026-06-01 — backend@1757f1e, frontend@600dd0d, deployment@6db02b0 — Postgres runtime cutover stores and health monitoring

Commits: backend 1757f1e, frontend 600dd0d, deployment 6db02b0 (backend 2.8.9, frontend 2.8.9) Touched:

  • Backend: src/infrastructure/postgres/client.ts, src/infrastructure/database/connection.ts, auth/config/address/category/level/shop/review store facades, backfill scripts, smoke scripts, /api/health, package.json, package-lock.json
  • Frontend: package.json, Dockerfile version metadata only.
  • Deployment: docker-compose.yml Postgres service/store env wiring, gatus/config.yaml Postgres health assertions. Why: Continue the MongoDB-to-Postgres runtime cutover by moving auth-owned users and Telegram auth records, confirmation thresholds, user addresses, and the first reference/marketplace domains behind opt-in Postgres store flags while keeping Mongo as the default and rollback mirror. Monitoring now treats Postgres as a required health dependency when any PG-backed store is enabled. Verification: Backend npm run typecheck -- --pretty false; backend npm run build; backend PG_URL=postgres://escrow:throwaway@127.0.0.1:5434/escrow_migration_test MONGODB_URI=mongodb://127.0.0.1:27018/reference-smoke DB_NAME=reference-smoke scripts/smoke/reference-stores-postgres.sh; backend focused health smoke asserted checks.postgres.ok/configured/required; local backend BASE_URL=http://127.0.0.1:5011 scripts/smoke/auth-basic.sh; local backend BASE_URL=http://127.0.0.1:5011 scripts/smoke/confirmation-thresholds.sh; local backend BASE_URL=http://127.0.0.1:5011 JWT_SECRET=test-secret scripts/smoke/addresses-basic.sh; frontend npx tsc --noEmit --ignoreDeprecations 6.0; deployment ruby -e 'require "yaml"; YAML.load_file("gatus/config.yaml")'. Linked docs updated: Postgres Runtime Cutover Status, MongoDB to PostgreSQL Migration Plan (Drizzle), API Overview, Monitoring, Gatus Monitoring - Proposed Config

2026-05-31 — backend@8e03360, frontend@228eed2 — keep auth and health checks resilient under load

Commits: backend 8e03360, frontend 228eed2 (backend 2.6.84, frontend 2.7.24) Touched:

  • Backend: src/app.ts, src/services/health/healthCheckService.ts, package.json, package-lock.json
  • Frontend: package.json, package-lock.json version bump only. Why: A dev performance run consumed the global 100/15m limiter and blocked /api/auth/login; repeated /api/health calls also drove the external Request Network reachability probe into 429, making Gatus report status: degraded even though Mongo/Redis/app were healthy. Auth routes now bypass the global limiter and rely on the auth-specific limiter, and the RN health subcheck is cached and treats non-5xx HTTP responses as upstream reachable. Verification: Backend npm test -- --runTestsByPath __tests__/health-check.test.ts; backend npm run typecheck; backend git diff --check; frontend npx tsc --noEmit --ignoreDeprecations 6.0; frontend git diff --check. Dev login was manually verified after resetting the backend limiter state. Linked docs updated: Monitoring, Gatus Monitoring - Proposed Config

2026-05-31 — backend@cbc32dc, frontend@08e8da9 — seller-owned template delivery and payment rails

Commits: backend cbc32dc, frontend 08e8da9 (backend 2.6.83, frontend 2.7.23) Touched:

  • Backend: src/models/RequestTemplate.ts, src/services/marketplace/RequestTemplateService.ts, src/services/marketplace/requestTemplateRoutes.ts, src/services/marketplace/requestTemplateController.ts, __tests__/marketplace-request-budget-validation.test.ts, scripts/smoke/marketplace-request-budget.sh, package.json, package-lock.json
  • Frontend: src/sections/request-template/request-template-checkout-billing-address.tsx, src/sections/request-template/request-template-checkout-payment.tsx, src/sections/request-template/request-template-new-edit-form.tsx, src/sections/request-template/request-template-details-summary.tsx, src/sections/request-template/view/seller-shop-view.tsx, src/sections/request-template/view/public-seller-shop-view.tsx, src/web3/components/multi-seller-provider-payment.tsx, src/actions/request-template.ts, src/types/request-template.ts, src/sections/request-template/context/types.ts, package.json, package-lock.json Why: Template checkout let buyers choose delivery even though fulfillment is a seller decision, and payment creation could fail when a seller/template had no usable network/token allowlist. Sellers now choose physical vs online delivery on the template, new templates require at least one chain/token rail, checkout asks buyers only for the needed address/email details, and template payment intents resolve rails with the real templateId. Verification: Backend npm test -- --runTestsByPath __tests__/marketplace-request-budget-validation.test.ts; backend npm run typecheck; backend git diff --check; backend BASE_URL=http://localhost:5001 ./scripts/smoke/marketplace-request-budget.sh skipped with exit 77 because ACCESS_TOKEN and CATEGORY_ID were not set; frontend npx tsc --noEmit --ignoreDeprecations 6.0; frontend git diff --check. No browser verification was possible because the Browser tool was unavailable in this session. Linked docs updated: RequestTemplate, PurchaseRequest, ShopSettings, Marketplace API, Payment API, Purchase Request Flow, Seller Guide

2026-05-31 — backend@a4d72df, frontend@07db9b0, scanner@ca62e7a — cap confirmations at per-chain acceptance floors

Commits: backend a4d72df, frontend 07db9b0, scanner ca62e7a (backend 2.6.82, frontend 2.7.22, scanner 0.1.7) Touched:

  • Backend: src/services/payment/safety/confirmationThresholdService.ts, src/routes/amnScannerWebhookRoutes.ts, src/services/payment/requestNetwork/requestNetworkRoutes.ts, src/services/payment/adapters/amnPayAdapter.ts, src/services/admin/confirmationThresholdRoutes.ts, src/services/payment/requestNetwork/supportedChains.json, __tests__/confirmation-threshold-service.test.ts, package.json, package-lock.json
  • Frontend: src/sections/payment/payment-table-row.tsx, src/sections/payment/view/payment-details-view.tsx, package.json, package-lock.json
  • Scanner: supported-chains.json, webhook.go, chain.go, tron_chain.go, ton_chain.go, main.go, README.md, VERSION, webhook_test.go Why: Confirmation depth should be a chain-specific acceptance floor, roughly a ten-minute safety window, not an endlessly increasing block counter. The backend now clamps runtime settings below the chain floor, caps stored settled confirmations at the effective threshold, sends the same threshold to AMN scanner intents, and the scanner includes the capped accepted count in webhooks. Frontend payment views show settled confirmation counts with a + suffix. Verification: Backend npm test -- --runTestsByPath __tests__/confirmation-threshold-service.test.ts __tests__/transaction-safety-provider.test.ts; backend npm run typecheck; backend BASE_URL=https://dev.amn.gg ./scripts/smoke/rn-webhook.sh; backend git diff --check; scanner go test ./...; scanner git diff --check; frontend npx tsc --noEmit --ignoreDeprecations 6.0; frontend git diff --check. Linked docs updated: Payment, Payment API, Scanner API, Payment Flow - Scanner, ScannerIntent, Scanner Architecture, Scanner Operations, Environment Variables

2026-05-31 — backend@896f17f, frontend@fd8e797 — persist webhook confirmations for scanner payments

Commits: backend 896f17f, frontend fd8e797 (backend 2.6.81, frontend 2.7.21) Touched:

  • Backend: src/routes/amnScannerWebhookRoutes.ts, src/services/payment/requestNetwork/requestNetworkRoutes.ts, src/services/payment/safety/transactionSafetyProvider.ts, __tests__/transaction-safety-provider.test.ts, package.json, package-lock.json
  • Frontend: package.json, package-lock.json version bump only. Why: dev.amn.gg showed a confirmed AMN scanner payment with blockchain.confirmations = 0. Scanner webhooks report status: "confirmed", but transaction safety only evaluated "completed" statuses, so no verifier evidence was produced. The webhook routes now persist confirmations from safety evidence, scanner/RN payloads, or the chain confirmation threshold fallback when the provider already says confirmed/completed. Verification: Backend npm test -- --runTestsByPath __tests__/transaction-safety-provider.test.ts; backend npm run typecheck; backend git diff --check; frontend npx tsc --noEmit --ignoreDeprecations 6.0; frontend git diff --check. Could not inspect the actual dev.amn.gg payment row because SSH to 193.180.213.68 rejected the available key. Linked docs updated: Payment, Payment API, Payment Flow - Scanner

2026-05-31 — backend@cab0719, frontend@ec2f765 — align request budget validation after Postgres migration

Commits: backend cab0719, frontend ec2f765 (backend 2.6.80, frontend 2.7.20) Touched:

  • Backend: src/shared/constants/marketplace.ts, src/models/PurchaseRequest.ts, src/models/RequestTemplate.ts, src/services/marketplace/requestTemplateRoutes.ts, src/services/marketplace/PurchaseRequestService.ts, src/db/schema/purchaseRequest.ts, src/db/repositories/drizzle/DrizzleMarketplaceRepo.ts, __tests__/marketplace-request-budget-validation.test.ts, scripts/smoke/marketplace-request-budget.sh
  • Frontend: package.json, package-lock.json version bump only. Why: Product/template creation could return 400 when the UI sent urgency: "urgent", and template-to-purchase conversion could later fail when a template budget used USD / EUR / IRR while PurchaseRequest only accepted USDT / USDC. Runtime Mongoose validation, request-template route validation, and the PG budget_currency enum now share USD, EUR, IRR, USDT, USDC; urgency validation includes urgent. Verification: Backend npm test -- --runTestsByPath __tests__/marketplace-request-budget-validation.test.ts; backend npm run typecheck; backend git diff --check; frontend npx tsc --noEmit --ignoreDeprecations 6.0; smoke helper added at scripts/smoke/marketplace-request-budget.sh but not run against dev because no ACCESS_TOKEN / CATEGORY_ID were available in this session. Linked docs updated: PurchaseRequest, RequestTemplate, Marketplace API, MongoDB to PostgreSQL Migration Guide, MongoDB to PostgreSQL Migration Plan (Drizzle), Postgres Runtime Cutover Status, Data Model Overview, Payment

2026-05-31 — nick-doc@local — clarify Postgres runtime cutover status

Commits: docs-only sync after backend 3a50dc4 (integrate-main-into-development, backend 2.6.79) Touched:


2026-05-31 — backend@3a50dc4 — promote Postgres integration branch with oracle/depeg + gasless backports

Commits: backend 11bfd02 74d73c5 1730c4d 148c803 8aa4473 a5e4da2 3a50dc4 (backend 2.6.762.6.79) Touched:

  • Branches: preserved old integrate-main-into-development as integrate-main-into-development-old; promoted feat/pg-money-core-migration to integrate-main-into-development.
  • Payment routing: requestNetworkRoutes.ts, requestNetworkPayInService.ts, amnScannerPayInService.ts, permitRelay.ts, amnPayAdapter.ts
  • Oracle/depeg: priceOracle/*, paymentQuote.ts, migration 0008_giant_winter_soldier.sql, Payment.ts, SellerOffer.ts
  • Tests/config: oracle-depeg-protection.test.ts, request-network-payin.test.ts, .env.example, package.json, package-lock.json Why: Combine the old integration branch's AMN scanner rail-switch fix and partial gasless permit work with the Postgres money-core branch and oracle/depeg quote engine. The final fix resolves the PG payment id through payments.legacy_object_id / id_map before writing payment_quotes, records pg_dualwrite_gaps if PG is behind, and keeps the Mongo quote mirror coherent during dual-write. Verification: npm run typecheck -- --pretty false; npm test -- --runInBand __tests__/oracle-depeg-protection.test.ts; npm test -- --runInBand __tests__/request-network-payin.test.ts __tests__/request-network-adapter.test.ts __tests__/request-network-webhook.test.ts __tests__/sweep-service.test.ts. The PG decimal integration cases in the oracle suite skipped because no local PG_URL/MIGRATION_PG_URL was configured. Linked docs updated: Payment, SellerOffer, Payment API, Environment Variables, Oracle Pricing & Stablecoin Depeg Protection, PRD - Gasless Buyer Payments (Roadmap)

2026-05-30 — frontend@9013b70, c77cf82, 8add494 — staged node-package upgrade + TS6 test fix + lint sweep

Commits: 8add494 c77cf82 9013b70 Touched:

  • Deps (package.json, yarn.lock): TypeScript 5→6, Jest 29→30, Tiptap 2→3 (all 11 sub-packages), i18next 25→26, react-i18next 15→17, @types/node 22→25, @types/jest 29→30, react-dropzone 14→15, react-apexcharts 1→2, mui-one-time-password-input 5→7, React 19.1→19.2, MUI 7.1→7.3 (in-range), zod 4.0→4.4. Constraints bumped to tested floors (@mui/material ^7.3.11, wagmi ^2.19.5, etc.). Version bumped 2.7.9 → 2.7.10.
  • Code fixes for new types: src/theme/with-settings/update-core.ts (cast currentScheme via Record<string,unknown> after MUI 7.3 tightened ColorSystemOptions), src/components/editor/components/code-highlight-block.tsx (cast NodeViewContent as='code''code' as 'div' for Tiptap 3 stricter prop typing).
  • Test infra: jest.config.js (point ts-jest at tsconfig.test.json explicitly, ignore TS5101/TS5011), tsconfig.test.json (add rootDir: "." and ignoreDeprecations: "6.0").
  • Security hygiene: .env.local + .env.production removed from tracking; added to .gitignore. Existing values still in git history — rotate any leaked credentials.
  • Lint sweep: yarn lint:fix applied across 64 files in src/ — mostly perfectionist/sort-imports reorders and unused-imports removals.
  • Docs: AGENTS.md gained an "Enforced project conventions" section covering Prettier, ESLint, TypeScript, and the centralized src/theme/ structure. CLAUDE.md is now a symlink → AGENTS.md so Claude Code reads the same rules.
  • Tooling: scripts/upgrade-packages.sh (reusable staged-upgrade runner with snapshot + auto-rollback) and scripts/UPGRADE-PLAN.md (strategy + per-stage rationale) added. .upgrade-backups/ added to .gitignore.

Why: Many runtime / dev dependencies were 37 minors behind; the audit was triggered by a request to "update all node packages without breaking the build." Did it as eight staged groups (in-range → @types → ESLint → Jest → Tiptap → i18next → misc → TypeScript), each gated by yarn build. Three stages were pulled back: ESLint 10 (eslint-plugin-react@7 incompatible with new context API), wagmi 3 (@coinbase/wallet-sdk declares window.ethereum: unknown, breaks type union with viem), MUI 7→9 (AGENTS.md pins to v7).

Verification: yarn build passes after every stage (3444s, all 57 routes). yarn test recovered from "45 suites fail, 0 tests run" (TS6 blocker) to 530 tests pass, 18 unrelated mock failures. yarn lint went 204 → 21 problems (the remaining 5 errors are pre-existing: 2× @ts-nocheck, 3× no-bitwise). Dev server (/, /auth/jwt/sign-in, /post, /shop, /dashboard, /telegram) all return 200. Manual smoke test of the Tiptap editor + wagmi connect flow is still recommended before promoting to prod.

Linked docs updated: none yet — 07 - Development/ should grow a "Node dependency upgrade runbook" pointing at frontend/scripts/UPGRADE-PLAN.md and the staged-rollback pattern. Also worth promoting the new AGENTS.md conventions section to 07 - Development/Coding Standards.md.


2026-05-29 — backend@cdc8df1 — AMN Pay Scanner integration (retire Request Network)

Commits: backend cdc8df1, scanner 8fee27e Touched:

  • Backend: src/services/payment/adapters/amnPayAdapter.ts, src/routes/amnScannerWebhookRoutes.ts, src/services/payment/adapters/types.ts, src/services/payment/providerConfig.ts, src/app.ts, .env.example, docker-compose.dev.yml, docker-compose.production.yml
  • Scanner (new repo): scanner/*.go, Dockerfile, supported-chains.json
  • Frontend: src/actions/network-registry.ts, src/sections/admin/networks/networks-list-view.tsx Why: Implement AMN Pay Scanner per PRD - Retire Request Network — In-House Payment Scanner.md. Standalone Go microservice scans ERC20FeeProxy TransferWithReferenceAndFee events directly, eliminating RN API dependency. Supports any destination address (derived HD wallets enabled). Parallel run: RN stays active for existing payments; new payments route to scanner when AMN_SCANNER_URL is configured. Verification: tsc --noEmit clean. Scanner binary builds (go build). Go tests pass (3/3). Frontend networks page renders scanner lag column. Linked docs updated: 07 - Development/Environment Variables, PRD - Retire Request Network — In-House Payment Scanner

2026-05-29 — backend@7688f57 — Sweep gas strategy: PermitPull + GasTopUp signers

Commits: backend 7688f57 Touched:

  • Backend: src/services/payment/wallets/sweepService.ts, __tests__/sweep-service.test.ts, .env.example Why: Implement hybrid two-signer sweep strategy per PRD - Sweep Gas Strategy - Permit Pull vs Gas Top-Up.md. PermitPullSweepSigner uses EIP-2612 permit for non-BSC chains (ETH, Arbitrum, Polygon, Base) so derived addresses never need native gas. GasTopUpSweepSigner handles BSC by topping up BNB from a master wallet before the derived address calls transfer(). getSweepSigner(chainId, tokenSymbol) auto-selects the correct signer. Static PERMIT_CAPABLE_TOKENS map seeded from on-chain audit 2026-05-29. Verification: tsc --noEmit clean. npx jest __tests__/sweep-service.test.ts — 31/31 pass (including 16 new tests for auto-selection and permit capability matrix). Linked docs updated: 07 - Development/Environment Variables, PRD - Sweep Gas Strategy - Permit Pull vs Gas Top-Up

2026-05-28 — deployment@4e8658d — Gatus monitoring: Docker service + config

Commits: deployment 1ac2e744e8658d Touched: deployment/gatus/config.yaml, deployment/docker-compose.yml, deployment/.env Why: Add Gatus monitoring service to the deployment stack. Config covers backend-dev, backend-prod, frontend-dev, frontend-prod, and external deps (RN API, Chainalysis, BSC RPC). Telegram alerting configured. Service exposed via Traefik at gatus.ch.manko.yoga. Verification: Config file validated against Gatus schema. Awaiting docker-compose up -d gatus on server. Linked docs updated: 08 - Operations/Gatus Monitoring - Proposed Config


2026-05-28 — backend@6c01a30 — Gatus monitoring: GET /api/health endpoint

Commits: backend 19f7eb944579d66c01a30 (2.6.48 → 2.6.49) Touched:

  • Backend: src/services/health/healthCheckService.ts, src/services/health/index.ts, src/app.ts, __tests__/health-check.test.ts Why: Implement GET /api/health for Gatus monitoring. Exposes 5 checks (db, redis, rnChainRegistry, rnTokenRegistry, rnApi) in a single public endpoint. Status semantics: ok | degraded | down (503 when DB fails). Each check includes latencyMs; registry checks include counts. Rate limiter and request logging skip /api/health. 5 route-level unit tests cover ok/degraded/down transitions. Verification: tsc --noEmit clean. npx jest __tests__/health-check.test.ts — 5/5 pass. Linked docs updated: 08 - Operations/Gatus Monitoring - Proposed Config

2026-05-28 — backend@19f7eb9, frontend@60ee6fb — Task #10: AML screening (Chainalysis, seller-paid, seller opt-in)

Commits: backend 441c8be80ba04619f7eb9 (2.6.46 → 2.6.47), frontend 717d5c8b7540f560ee6fb (2.6.46 → 2.6.47) Touched:

  • Backend: src/services/payment/safety/amlProvider.ts, src/services/payment/safety/chainalysisProvider.ts, src/services/payment/safety/amlScreeningService.ts, src/services/payment/safety/transactionSafetyProvider.ts, src/services/payment/paymentCoordinator.ts, src/services/admin/amlConfigRoutes.ts, src/models/SellerOffer.ts, src/app.ts, .env.example
  • Frontend: src/sections/request/components/seller-steps/step-1-send-proposal.tsx, src/types/marketplace.ts Why: Task #10 implementation. Chainalysis Public Sanctions API integration for seller-paid AML screening. Seller can opt-in per-offer via requireAmlCheck + amlBlockOnFailure toggles. TransactionSafetyProvider screens buyer source address after on-chain transfer verification. paymentCoordinator deducts AML_CHECK_COST_USD (default 0, API is free) from seller escrow on payment completion. Admin routes for AML config. Verification: Frontend tsc --noEmit clean. Backend relevant tests pass (module resolution issues in unrelated test files). Linked docs updated: 02 - Data Models/SellerOffer, 03 - API Reference/Admin API, 04 - Flows/Escrow Flow

2026-05-28 — backend@441c8be, frontend@717d5c8 — Task #9: Per-chain confirmation thresholds + admin UI

Commits: backend 4a85737441c8be (2.6.47 → 2.6.48), frontend 0ebb2f1717d5c8 (2.6.46 → 2.6.48) Touched:

  • Backend: src/models/ConfigSetting.ts, src/services/payment/safety/confirmationThresholdService.ts, src/services/payment/safety/transactionSafetyProvider.ts, src/services/admin/confirmationThresholdRoutes.ts, src/services/admin/awaitingConfirmationRoutes.ts, src/app.ts
  • Frontend: src/sections/admin/confirmation-thresholds/, src/sections/admin/payments-awaiting-confirmation/, src/actions/confirmation-thresholds.ts, src/routes/paths.ts, src/layouts/nav-config-dashboard.tsx Why: PRD §3 — Task #9 implementation. Runtime per-chain confirmation thresholds via ConfigSetting Mongo model with 30s in-memory cache. TransactionSafetyProvider now reads getConfirmationThreshold(chainId) instead of static env. Admin endpoints: GET/PATCH /api/admin/settings/confirmation-thresholds, GET /api/admin/payments/awaiting-confirmation. Frontend admin pages for threshold editing and awaiting-confirmation payment monitoring. Verification: All 56 relevant backend tests green. Frontend tsc --noEmit clean. Linked docs updated: 03 - API Reference/Payment API

2026-05-28 — backend@4a85737, frontend@0ebb2f1 — Task #8: Multichain RN proxy registry + USDC/USDT support + Base fix + USDT fork test

Commits: backend 01b9ea0ae17b184a85737 (2.6.45 → 2.6.47), frontend 0ebb2f1 (2.6.44 → 2.6.46) Touched:

  • Backend: src/services/payment/requestNetwork/supportedChains.json, src/services/payment/requestNetwork/tokens.json, src/services/payment/requestNetwork/tokens.ts, src/services/payment/requestNetwork/proxyAddresses.ts, src/services/payment/requestNetwork/inHouseCheckout.ts, src/services/payment/requestNetwork/networkRegistryRoutes.ts, src/services/payment/wallets/sweepService.ts, src/app.ts, scripts/probe-rn-chains.ts
  • Frontend: src/web3/config.ts, src/sections/payment/checkout/rn-in-house-checkout-view.tsx, src/sections/admin/networks/, src/app/dashboard/admin/networks/page.tsx, src/actions/network-registry.ts, src/routes/paths.ts, src/layouts/nav-config-dashboard.tsx Why: PRD §2 — Task #8 implementation. 5-chain registry (BSC, Arbitrum, Ethereum, Polygon, Base) with canonical RN ERC20FeeProxy addresses and per-chain USDC/USDT entries including Base. tokens.ts and proxyAddresses.ts now load from JSON files with admin reload capability. buildInHouseCheckoutBlock returns unsupported_chain:<id> for unknown chains. Frontend wagmi config expanded to include arbitrum + base. Per-chain explorer URLs in checkout view. USDT-mainnet approve(0) reset quirk handled in approve flow. New admin page /dashboard/admin/networks renders registry with reload button. New probe script scripts/probe-rn-chains.ts verifies proxy deployment on-chain. Verification: All 58 relevant backend tests green (rn-in-house-checkout, derived-destinations, sweep-service, request-template-orphan-cleanup). Frontend tsc --noEmit clean. Linked docs updated: 03 - API Reference/Payment API (new GET /api/admin/rn/networks and POST /api/admin/rn/networks/reload endpoints)

2026-05-28 — backend@34f542e — Task #7 B: unit tests for derived-destinations + sweep-service + orphan-cleanup regression

Commits: backend 34f542e (2.6.44 → 2.6.45) Touched: __tests__/derived-destinations.test.ts (26 tests), __tests__/sweep-service.test.ts (18 tests), __tests__/request-template-orphan-cleanup.test.ts (2 tests) Why: PRD item B — regression lock-in test suite for Task #7. Covers: getDestinationFor idempotency, E11000 race fallback, validateXpub rejection of xpriv/tprv/garbage, deriveAddressAtIndex determinism, recordSweep $inc accumulation (regression lock-in for item E), and orphan-payment cleanup provider filtering (regression lock-in for Gap 2 fix in 2.6.44). Verification: All 46 tests green (npx jest derived-destinations.test.ts sweep-service.test.ts request-template-orphan-cleanup.test.ts). Linked docs updated: 08 - Operations/Handoff - Request Network In-House Checkout - 2026-05-28


2026-05-28 — backend@1889169, frontend@c44ed64 — Task #7 A verification fix: multi-checkout conversion + orphan-payment guard

Commits: backend 1889169 (2.6.43 → 2.6.44), frontend c44ed64 (2.6.43 → 2.6.44) Touched:

  • Backend: src/services/marketplace/RequestTemplateService.ts
  • Frontend: src/sections/payment/checkout/rn-multi-checkout-view.tsx Why: A verification revealed two gaps: (1) RnMultiCheckoutView.handleFinish only navigated to payment list and never called convertTemplatesToRequests, so multi-seller carts never created PurchaseRequests; fixed by calling conversion with stashed cart items and navigating to the first created request. (2) Backend orphan-payment cleanup found ALL pending payments for the buyer and hard-deleted all but the first — fatal for multi-seller carts; fixed by restricting orphan query to provider: 'shkeeper' only so request.network payments retain their independent lifecycle. Verification: Pushed to integrate-main-into-development on both repos — Woodpecker builds pending. Linked docs updated: 03 - API Reference/Payment API

2026-05-28 — backend@faf2221, frontend@022ecb6 — Task #7 derived destinations: sweep autostart, recordSweep fix, multi-seller checkout UX

Commits: backend faf2221 (2.6.42 → 2.6.43), frontend 022ecb6 (2.6.42 → 2.6.43) Touched:

  • Backend: src/app.ts, src/models/DerivedDestination.ts, src/models/Payment.ts, src/services/payment/requestNetwork/requestNetworkPayInService.ts, src/services/payment/wallets/derivedDestinations.ts, .env.example
  • Frontend: src/sections/payment/checkout/rn-in-house-checkout-view.tsx, src/sections/request-template/request-template-checkout-payment.tsx, src/web3/components/multi-seller-provider-payment.tsx, src/sections/payment/checkout/rn-multi-checkout-view.tsx, src/app/checkout/request-network/multi/page.tsx Why: PRD items D/E/F + frontend cart-aware checkout (A). Auto-start sweep cron on boot; fix recordSweep to $inc totalSwept instead of $setOnInsert; widen Payment unique index to include sellerOfferId for multi-seller carts; add multi-seller checkout wrapper and wire into template + request flows. Verification: Pushed to integrate-main-into-development on both repos — Woodpecker builds pending. Linked docs updated: 03 - API Reference/Payment API (derived-destination endpoints)

2026-05-28 — backend@e46be98, frontend@af77b3c — add nick-doc sync rule + version bumps

Commits: backend e46be98 (2.6.24 → 2.6.25), frontend af77b3c (2.6.25 → 2.6.26) Touched: backend/AGENTS.md, frontend/AGENTS.md (new), both package.json + package-lock.json Why: Establish a mandatory rule that every code push must be followed by a nick-doc Activity Log entry (and relevant section updates) so the vault never falls behind the code. Frontend AGENTS.md created from scratch (was missing). Verification: Pushed to integrate-main-into-development on both repos — Woodpecker builds pending. Linked docs updated: This vault's AGENTS.md updated with the same rule. Note: Backend (2.6.25) and frontend (2.6.26) are intentionally one patch apart — backend was a version behind before this session. Should be re-aligned on the next paired bump.


2026-05-28 — frontend@9d4aa37 — fix 429 request storm on template SWR hooks

Commits: 9d4aa37 Touched: src/actions/request-template.ts Why: Production browser showed repeated 429 (Too Many Requests) on /api/marketplace/request-templates/sellers. Default SWR config was revalidating on focus/reconnect and retrying on errors, making backend rate-limit recover impossible without a restart. Verification: Pushed, awaiting Woodpecker build. Visual confirmation on dev.amn.gg after deploy. Linked docs updated: none yet — SWR pattern should be promoted to 07 - Development/Coding Standards.md in a follow-up.


2026-05-28 — frontend@6c89444 — improve request template form debug feedback

Commits: 6c89444 Touched: src/sections/request-template/request-template-new-edit-form.tsx Why: Users could not tell why "ایجاد قالب" failed — validation errors silently blocked submission, API errors collapsed to generic "خطایی رخ داده است!", and the "انتشار" Switch in renderActions was visual-only. Verification: Type-check passes via Docker build in prior session; manual browser test pending. Linked docs updated: none.


2026-05-27 — frontend@8c0f14d, ad498f4, f3a3c9d, bb72a66 — unblock 2.6.19 Docker build

Commits: bb72a66 f3a3c9d ad498f4 8c0f14d Touched: src/sections/request-template/request-template-checkout-payment.tsx, src/web3/components/wallet-selector.tsx, tsconfig.json, src/types/payment.ts Why: Docker build was failing on TypeScript compilation after the wallet-support + test-payment feature merge. Four distinct errors fixed: User type uses _id not id; wallet-selector imported non-existent @/components/ui/dialog; @/* path alias missing from tsconfig; IPayment metadata type didn't allow test-payment fields. Verification: Local docker build succeeded — image escrow-frontend:2.6.19 created. Linked docs updated: none — should add SWR + UI library notes to 07 - Development/Coding Standards.md.


2026-06-02 — backend@4949988, frontend@integrate-main-into-development — route admin user counts through postgres-capable stores

Commits: 4949988 (backend), version bump (frontend) Touched: src/services/user/userController.ts, src/services/auth/authStore.ts, src/db/repositories/interfaces/IMarketplaceRepo.ts, src/db/repositories/interfaces/IPaymentRepo.ts, src/db/repositories/mongo/MongoMarketplaceRepo.ts, src/db/repositories/drizzle/DrizzleMarketplaceRepo.ts, src/db/repositories/dual/DualWriteMarketplaceRepo.ts, src/db/repositories/mongo/MongoPaymentRepo.ts, src/db/repositories/drizzle/DrizzlePaymentRepo.ts, src/db/repositories/dual/DualWritePaymentRepo.ts, __tests__/auth-store-pg-query.test.ts, __tests__/user-dependencies-repo.test.ts, scripts/smoke/user-admin-postgres.sh, scripts/smoke/user-dependencies.sh Why: GET /api/user/admin/:userId/dependencies was directly importing Mongoose models (RequestTemplate, PurchaseRequest, Payment, Chat) bypassing the repository layer and blocking MONGO_CONNECT_MODE=never. Routed through getUserDependencyCounts / countByParticipant on all three repo tiers (Mongo, Drizzle, DualWrite). Also hardened AuthUser PgQuery facade with generic multi-field sorting, nested path support, and isActive/isVerified filter aliases so admin list/stats routes work correctly in AUTH_STORE=postgres mode. Verification: 5 test suites, 11 tests pass. typecheck clean. Smoke scripts added but not run end-to-end (require live ADMIN_TOKEN). Linked docs updated: none — Mongo removal remaining work tracked in MONGODB_REMOVAL_HANDOFF_2026-06-02.md.


2026-06-02 — backend@515bea3 — guard dataCleanupService against MONGO_CONNECT_MODE=never

Commits: 515bea3 Touched: src/services/admin/dataCleanupService.ts Why: dataCleanupService dynamically imported 10 Mongoose models unconditionally; under MONGO_CONNECT_MODE=never this throws at runtime. Routed user counts through AuthUser facade (Postgres-gated). Wrapped all other model access in isMongoAvailable() guard — cleanup ops return a safe error result when MongoDB is disconnected. Verification: 5 test suites 11 tests pass. typecheck clean. Mongo scan: all remaining hits confirmed inside Postgres-gated fallback branches. Linked docs updated: none.


2026-06-02 — backend@14c231e, frontend@d7a2a86 — make admin user management work end-to-end (v2.8.50)

Commits: 14c231e (backend), d7a2a86 (frontend) Touched (backend): src/services/auth/authStore.ts, src/services/user/userRoutes.ts, __tests__/auth-store-pg-query.test.ts Touched (frontend): src/actions/user.ts, src/types/user.ts, src/sections/user/view/user-list-view.tsx, src/sections/user/user-table-row.tsx, src/sections/user/user-quick-edit-form.tsx, src/sections/user/user-new-edit-form.tsx, src/sections/request/view/admin/admin-request-list-view.tsx, src/sections/overview/app/view/overview-admin-view.tsx, Dockerfile Why: Admin could not delete users on dev (DELETE /api/users/admin/:id → 404). Root cause: PG rows seeded outside the auth store have NULL legacy_object_id; the auth store fabricated a fresh random _id on every read, so no admin action keyed on that id could ever resolve. Audit of the admin dashboard surfaced further breakage which is also fixed in this pair of commits:

  • Backend: legacy-id self-heal on read; findById accepts PG uuids; missing /api/users/admin/:userId/toggle-status and /dependencies routes added (frontend called them under the plural prefix where they did not exist); soft-deleted users excluded from admin list and all stats; list/detail return full profile address fields.
  • Frontend: updateUserStatus sent {status} but backend reads {isActive} — activating a user actually suspended them; user quick-edit form was a mock (no API call); "حذف کاربر" button in edit form had no handler; bulk deletes used Promise.all (partial failures hidden); list query params (query/status) were ignored by the backend (now search/isActive); profile payloads now match the backend schema (dotted paths, no whole-profile overwrite); blocked aria-hidden console warnings fixed. Verification: Backend: 10 tests pass (auth-store-pg-query, user-dependencies-repo, repository-factory-modes), tsc --noEmit clean. Frontend: tsc --noEmit + eslint clean (jest suite is broken repo-wide — pre-existing ESM transform issue, unrelated). Smoke test against dev pending post-deploy (admin will verify delete in UI). Linked docs updated: 03 - API Reference/Admin API.md, 03 - API Reference/User API.md (stale KNOWN BUG notes removed; route availability updated).

2026-06-03 — backend@378f8f6, frontend@6fe1328 — unblock user creation & purchase requests for native-PG users (v2.8.51)

Commits: 378f8f6 (backend), 6fe1328 (frontend) Touched (backend): src/services/auth/authStore.ts, src/services/auth/authController.ts, src/services/user/userRoutes.ts, src/services/user/userController.ts, src/db/repositories/drizzle/DrizzleMarketplaceRepo.ts, DrizzleUserRepo.ts, DrizzleTrezorAccountRepo.ts, DrizzleDerivedDestinationRepo.ts, __tests__/auth-store-pg-query.test.ts Touched (frontend): src/actions/user.ts, user section forms/views (error toasts) Why: Two follow-up dev blockers after v2.8.50:

  1. Admin create returned 409 for emails held by soft-deleted accounts (users.email is unique; deleted rows kept the address and are hidden from the list). Fix: delete releases the email (deleted_<id>_<email>); create/register lazily free emails held by previously deleted rows.
  2. Purchase request creation threw "Buyer not found in Postgres id map" — users created natively in PG (auth store / Telegram / Google / admin create) never get id_map rows. Fix: resolveId in 4 Drizzle repos falls back to the entity's own legacy_object_id column and self-heals id_map. Verification: 47 backend tests green across 6 suites, tsc clean (both repos), eslint clean. Deployed to dev via CI; admin verifies create-user and create-request flows in UI. Linked docs updated: 03 - API Reference/Admin API.md, 03 - API Reference/User API.md.

2026-06-03 — backend@804bb99, frontend@714dfbd — fix PG response serialization & id resolution (v2.8.52)

Commits: 804bb99 (backend), 714dfbd (frontend) Touched (backend): src/services/marketplace/PurchaseRequestService.ts, src/services/chat/ChatService.ts, src/db/repositories/drizzle/DrizzleChatRepo.ts, src/db/repositories/drizzle/DrizzleMarketplaceRepo.ts Touched (frontend): request list views (buyer/seller/admin), src/socket/hooks/use-purchase-requests.ts, src/types/marketplace.ts Why: After v2.8.51, request detail pages opened /purchase-requests/[object Object] and the seller request list returned 500. Root causes (all PG-path):

  1. mongoCompatId returned {toString: () => id} — JSON.stringify drops functions, so every response carried _id: {}. Now a plain string; chat compat ObjectIds get toJSON() instead.
  2. findPurchaseRequests passed unresolved legacy 24-hex user ids into uuid column comparisons (PG cast error → 500); the route's Mongo-style $or filter was silently ignored (privacy: non-admin users saw all requests). Both fixed — ids resolved up front, $or honored.
  3. 16 resolveId(...) ?? originalId fallbacks replaced with a nil-UUID fallback (500 → empty/404).
  4. Frontend reads request.id || request._id everywhere (tolerates both API shapes). Verification: 47 backend tests green, tsc clean both repos, eslint clean. Deployed to dev via CI; admin verifies: request detail page, seller request list, chat. Linked docs updated: none beyond this log (API id semantics documented in Marketplace API as needed later).

2026-06-03 — frontend@7b949bf — Mini App live socket updates (v2.8.53)

Commits: 7b949bf (frontend only; backend stays at 2.8.52) Touched: src/sections/telegram/hooks/use-telegram-realtime.ts (new), telegram-request-detail-view.tsx, telegram-requests-view.tsx, telegram-mini-app-view.tsx Why: Mini App request views were SWR-only — status changes required a manual reload. New useTelegramRealtime hook bridges marketplace socket events to SWR refresh (plus a 30s periodic fallback for WebView throttling). Detail view joins the request room; list view joins the buyer/seller room. Verification: tsc + eslint clean. Admin verifies live stepper advance in Telegram after deploy. Linked docs updated: none.


2026-06-03 — backend@8b8c1ae, frontend@583d55a — 'guard' role + Mini App tab fix (v2.8.54)

Commits: 8b8c1ae (backend), 583d55a (frontend) Touched (backend): User model, drizzle users schema + migration 0017_user_role_guard.sql, postgresAuthSchema (enum + ALTER TYPE for existing DBs), shared types, IUserRepo, authStore/authService role types, userRoutes role validation Touched (frontend): role types/permissions/display-names, user forms, useRole hook, _roles mock, user-table-row (central role labels), telegram-mini-app-view (tab dispatch) Why:

  1. New admin-assignable role «نگهبان» (guard), registered across every role definition site — modeled on resolver. No special endpoint permissions yet; grant via authorizeRoles() later. Side fix: the admin role-change endpoint previously rejected 'resolver' (validation list only had buyer/seller/admin) — both resolver and guard now pass.
  2. Mini App: tapping a bottom tab while a request-detail/chat/new-request overlay was open did nothing (overlay kept rendering over the new tab). Tab selection now dismisses overlays. Verification: Backend tsc clean, 12 tests green. Frontend tsc + eslint clean. Admin verifies after deploy: role dropdown shows نگهبان; Mini App tabs always respond. Linked docs updated: 03 - API Reference/User API.md (role enum), pending.

2026-06-03 — backend@9424395, frontend@a18e870 — chat/notification fixes, role dashboards, live Mini App chat (v2.8.55)

Commits: 9424395 (backend), a18e870 (frontend) Why (from chat + notifications + Mini App audits):

  1. Chat (backend): 5 endpoints were dead — routes pass :id but controllers destructured req.params.chatId (undefined): edit/delete message, archive chat, add/remove participant.
  2. Notifications → Telegram (backend): sendTelegramNotification existed but was never called. Every in-app notification is now also delivered as a Telegram bot message when the user has an active telegram_link (new sendTelegramNotificationToUser, lazy-required, fire-and-forget).
  3. Role dashboards (frontend): resolver landed on the buyer dashboard. New OverviewResolverView (dispute stats) and OverviewGuardView (oversight stats); dispatch updated; «حل اختلاف» nav item now visible to resolver.
  4. Mini App chat (frontend): was SWR-only. Thread view joins its conversation socket room, list joins the user room — messages now appear in real-time. Touched (backend): src/services/chat/chatController.ts, src/services/telegram/botService.ts, src/services/notification/NotificationService.ts Touched (frontend): src/sections/overview/app/view/* (2 new views + dispatch), src/layouts/nav-config-dashboard.tsx, src/sections/telegram/hooks/use-telegram-chat-thread.ts, use-telegram-conversations.ts, telegram-chat-view.tsx Verification: backend tsc clean + 28 tests green; frontend tsc clean, eslint clean (pre-existing no-img-element rule-config errors in untouched files). Admin verifies after deploy: resolver dashboard, Mini App live chat, Telegram bot messages on new offers, chat edit/delete. Known remaining (documented, not yet built): Mini App notifications tab; web chat "new conversation" button uses a placeholder id; Drizzle chat participant population.

2026-06-03 — backend@14d164c, frontend@6adb2e0 — Mini App account tab, pinned support chat, shop fix (v2.8.56)

Commits: 14d164c (backend), 6adb2e0 (frontend) Why:

  1. Mini App account tab: «به‌زودی» rows are now functional — notifications (new in-shell list view + hook, socket-live, mark-all-read), support (creates/opens support chat via new createSupportChat action → POST /chat/support), wallet (shows linked address), guard role label.
  2. Mini App chat tab: pinned «پشتیبانی AMN» row always at top (per user request); support-type conversations filtered from the regular list.
  3. Shop bug «خطا در بارگذاری اطلاعات فروشنده»: PG sellers list returns uuid ids but seller-shop-view only accepted legacy 24-hex ids → uuid sellers always errored. Both formats accepted now (frontend validation + backend lookup comparison). Touched (backend): DrizzleMarketplaceRepo.findSellerWithActiveRequestTemplates Touched (frontend): telegram account/chat/notifications views + hooks, locales (fa/en/types), src/actions/chat.ts (createSupportChat), src/lib/axios.ts, seller-shop-view.tsx Verification: tsc clean both repos, 16 backend tests green, eslint clean (pre-existing no-img-element config errors aside). Admin verifies after deploy: account tab rows, pinned support chat, opening a seller shop from فروشگاه. Known remaining: in-shell seller shop detail inside the Mini App (currently opens the web view — now working after the id fix); Mini App notifications row could get an unread badge.

2026-06-03 — frontend@a8ae1e3 — full in-shell shop inside the Mini App (v2.8.57)

Commits: a8ae1e3 (frontend only; backend stays at 2.8.56) Why: Tapping a seller in فروشگاه opened the web dashboard inside the Telegram webview. The seller's shop now renders in-shell: new TelegramSellerShopView (seller header + active templates with budget/usage), useTelegramSellerShop hook, TelegramShopRow converted from <a href> to in-shell onOpen, and an openSellerId overlay in the shell (BackButton + tab dismissal). Template ordering hands off to the web from-template checkout (wallet stack lives there). Cleanup: removed eslint-disable comments for the unregistered @next/next/no-img-element rule. Verification: tsc clean, eslint clean (0 errors). Admin verifies after deploy: فروشگاه → tap seller → in-shell shop → سفارش این قالب → web checkout.


2026-06-03 — frontend@HEAD — Mini App account tab parity with web sections (v2.8.58)

Why: The web account menu (تنظیمات عمومی، اعلان‌ها، کیف پول، آدرس‌های تحویل، مدیریت Passkey) had no counterpart in the Mini App. All five are now reachable from the Mini App account tab — notifications stay in-shell; settings/wallet/addresses/passkey deep-link to the web dashboard pages (passkey/wallet need browser-context APIs). Labeled «در داشبورد وب باز می‌شود». Touched: telegram-account-view.tsx, telegram locales (fa/en/types) Verification: tsc + eslint clean. Pending: in-shell shop cart/checkout (user request — payment hand-off design needed since the crypto wallet stack lives in web).


2026-06-03 — frontend@HEAD — Mini App in-shell shopping cart (v2.8.59) — buyer-parity phase 1

Why: Goal: full buyer-role parity in the Mini App (then other roles). Phase 1 = cart. New useTelegramCart hook shares the web checkout's localStorage key (app-request-template-checkout) so the Mini App cart IS the web cart; new TelegramCartView (qty stepper, remove, total, checkout → web); add/remove-to-cart buttons in the in-shell seller shop; cart badge + overlay in the shell. Buyer-parity roadmap (remaining): offers view/accept on requests → delivery confirmation → payments list → points/referral → addresses CRUD → TON Connect payments (Telegram Wallet is TON-only, no EVM; backend already has tonProofService). Verification: tsc + eslint clean. Admin verifies: فروشگاه → افزودن به سبد → سبد → ادامه و پرداخت → web checkout shows the same items.


2026-06-03 — frontend@HEAD — Mini App floating cart FAB (v2.8.60)

Why: User wants the web shop's small edge-docked basket (count badge) in the Mini App too. New TelegramCartFab component pinned above the tab bar, shown on shop tab + in-shell seller shop, opens the in-shell cart. Replaced the header cart chip + bottom cart bar. Pending next (user requests): (1) Mini App «تنظیمات» — dark/light mode for the Telegram shell (needs a dark palette variant of telegram-shell-css + toggle; should follow webApp.colorScheme); (2) «درخواست جدید» button font — it is Telegram's NATIVE MainButton, font cannot be customized; option is replacing it with an in-shell button using the project font. Verification: tsc + eslint clean.


2026-06-02 — backend@7c4dedf — complete dual-write repos, migrations pipeline, TTL scheduler, address reconciliation

Commits: 7c4dedf (backend v2.8.44), frontend v2.8.44 Touched: src/db/repositories/dual/DualWriteDisputeRepo.ts (new), DualWriteTrezorAccountRepo.ts (new), DualWriteDerivedDestinationRepo.ts (new), src/db/repositories/factory.ts, src/db/schema/dispute.ts, src/db/schema/address.ts, src/db/schema/fundsLedgerEntry.ts, src/services/address/addressStore.ts, src/services/admin/dataCleanupService.ts, src/services/admin/ttlCleanupJob.ts (new), src/app.ts, src/shared/types/address.ts, drizzle.config.ts, migrations/ (new), __tests__/dispute-dual-write.test.ts (new) Why: Complete all remaining Phase 1 migration prerequisites: the three missing DualWrite repos (Dispute, TrezorAccount, DerivedDestination) unblock PG cutover for those domains; migrations pipeline enables drizzle-kit schema management; FundsLedgerEntry immutability trigger (DDL in migrations/) protects money records from mutation; Dispute composite indexes match Mongo query patterns; TTL scheduler replaces Mongo TTL indexes for Notification/TempVerification/TelegramSession; Address schema reconciled so Drizzle is authoritative. Verification: typecheck clean. 5 core suites (11 tests) pass. DualWriteDisputeRepo 21 tests pass. Linked docs updated: MIGRATION_TODO.md — all 9 tasks marked done.


2026-06-03 — frontend@HEAD — Mini App floating support bubble (v2.8.61)

Why: Parity with the web dashboard's green floating chat widget. New TelegramSupportFab (edge-docked, green bubble) on home/shop/requests tabs + in-shell seller shop; opens/creates the support chat. Stacks above the cart FAB in shop contexts. Verification: tsc + eslint clean. In progress / queued: Points & Referral audit (background agent — bugs observed: values render as "·", Bronze level claims "highest level"); Mini App dark mode (تنظیمات); buyer-parity phase 2 (offers view/accept).


2026-06-03 — backend@c5d6490 — Points & Referral audit fixes (v2.8.62)

Why (from points audit): /dashboard/points showed "·" instead of numbers and told Bronze users they are at "the highest level". Root causes: (1) level boundary off-by-one — seeds define inclusive maxPoints (Bronze 0999) but PG used > and Mongo <, so exact-boundary users resolved to no level; (2) level_configs table in Postgres was created but NEVER seeded → zero levels → null next-level → "highest level" message. Fixes: inclusive comparisons in both repos + self-seeding of the 5 default levels (برنز/نقره/طلا/پلاتین/الماس) in ensurePostgresLevelConfigSchema. Remaining from audit: frontend points page swallows API errors (shows zeros instead of an error state); integration ideas: award points on completed escrow, show points in Mini App. Verification: tsc clean, 8 tests green. Admin verifies after deploy: /dashboard/points shows real numbers and Bronze shows progress toward نقره.


2026-06-03 — frontend@131700b — compact floating cart/support FABs (v2.8.63)

Why: User feedback — the two edge-docked FABs were too large. Compacted paddings, icons, badge, and stack positions. Verification: tsc clean.


2026-06-03 — frontend@d64e194 — Mini App points view + tap-to-verify email (v2.8.64)

Why: Buyer parity — امتیازات و رفرال now in-shell (TelegramPointsView: points cards, level progress, referral code copy + stats; account-tab row). Email verification from the Mini App: unverified badge is tappable → sends code → code-entry sheet → verifyProfileEmail → session refresh. Reuses existing account.ts actions. Verification: tsc + eslint clean. Test after deploy: account tab → امتیازات row; unverified user → tap red badge → enter emailed code → badge turns green.


2026-06-03 — backend@91877ae — points lookups accept legacy user ids (v2.8.63)

Why: Mini App points view (v2.8.64) failed: every user lookup in DrizzlePointsRepo compared users.id (uuid) against the JWT legacy 24-hex id → uuid cast error → 500. All lookups now match either format (id::text = $n OR legacy_object_id = $n). Verification: tsc clean, tests green. Test after deploy: Mini App → حساب → امتیازات shows data.


2026-06-03 — frontend@HEAD — remove MainButton, in-tab CTA, sign-in retry (v2.8.65)

Why (user requests): (1) native Telegram MainButton removed — wrong font, duplicated CTAs; «+ درخواست جدید» button added inside the requests tab header instead. (2) Intermittent 503 on Telegram sign-in (backend redeploy windows) — sign-in now retries transient 502/503/504 3x with backoff and shows the Persian message. Queued (user, with screenshots): cart =remove at qty 1; seller-shop card redesign (+ icon, template detail view, search/filter/sort); requests search + status-chip filters + sort; shop search; seller chat button; referral signup points; dark mode. Verification: tsc + eslint clean.


2026-06-03 — frontend v2.8.6667 — seller chat, header bell, cart UX + review system audit

v2.8.66: seller rows in the Mini App shop get a chat icon (direct conversation with the seller); header gets a notification bell opening the in-shell notifications list. v2.8.67 fixes: seller chat gave "Validation failed" (backend expects ONE participant for direct chats — current user auto-added; we sent two). Cart «−» at qty 1 now removes the item (separate remove button dropped). Bell: borderless + red dot instead of count. Review/rating system audit completed (agent): backend schema/API/store fully working (reviews table, /api/marketplace/reviews). Broken: seller rating HARDCODED 4.5 in DrizzleMarketplaceRepo.findSellersWithActiveRequestTemplates (never reads real reviews); review form never sends purchaseRequestId (verified-buyer badge never set); Step 6 post-delivery has NO review prompt; Mini App has no reviews. Full phased plan in agent output — next work items.


2026-06-03 — backend v2.8.6465 — real seller ratings + chat participant names

v2.8.64: seller list ratings now aggregated from published reviews (was hardcoded 4.5); exposes rating/ratingCount per seller. v2.8.65: PG chats showed "Unknown User" — participants (JSONB ids) are now populated with firstName/lastName/email/avatar from the users table (uuid OR legacy id matched), in findByIdWithParticipants and findForUser. Parity with Mongo .populate(). In progress: background agent fixing seeds for the PG database + creating complete mock shops (categories in PG → fixes template-creation 500) — will be reviewed and shipped when it reports. Queued (user): in-shell تنظیمات عمومی (profile editing) in Mini App; post-delivery review prompt; template detail view + search/filter/sort in shop and requests; dark mode.


2026-06-03 — backend v2.8.6566 — PG seeds, mock shops, template-creation fix

Template-creation 500 root cause (agent): categories table column conflict (Drizzle "order" vs categoryStore order_num) + random legacy ids per seed → category ids served by the API were unresolvable by createRequestTemplate. Fixed: categoryStore aligned to canonical "order" column (with one-time order_num migration), deterministic category legacy ids, hardened categories fallback (id::text match). Also fixed: the v2.8.64 rating aggregation queried Drizzle reviews columns but the physical table (written by reviewStore) has different columns — rewritten against the real columns (seller_user_id). New seed: src/seeds/seedMockShops.ts — Persian categories, 5 sellers + shop settings, templates with picsum product images, published reviews. npm scripts fixed (paths) + added seed:levels / seed:templates / seed:mock-shops. ⚠️ DEV ENV REQUIREMENT: CATEGORY_STORE=postgres (and ideally REVIEW_STORE / SHOP_SETTINGS_STORE = postgres) must be set in dev .env.local when REPO_MARKETPLACE=postgres — docker-compose.dev.yml still defaults these to mongo. Without this, template creation keeps failing. Run on dev after deploy: docker exec npm run seed:categories && npm run seed:mock-shops Verification: tsc clean, 14 tests green (3 suites).


2026-06-03 — frontend v2.8.68 — product-style template cards + in-shell template detail

Why (user reference design): seller-shop template cards redesigned image-first (photo + circular add-to-cart overlay + title/price); tapping opens a new in-shell template detail view (image gallery with thumbnails/counter, product-type chip, price, description, specs, tags, sticky add-to-cart/order). New view: telegram-template-detail-view.tsx; shell openTemplate overlay. Dev env (user applied): CATEGORY_STORE/REVIEW_STORE/SHOP_SETTINGS_STORE=postgres set; backend v2.8.66 restarted healthy; 24 categories auto-seeded into PG. seed:mock-shops still to be run by the user for demo data. Verification: tsc + eslint clean.


2026-06-03 — backend v2.8.67 — auto-seed mock shops on startup

Feature: new env flag SEED_MOCK_SHOPS_ON_START=true runs seedMockShops() during backend boot (after migrations/levels). Idempotent — existing rows are skipped — so the flag can stay on permanently in dev. Action for dev env: add SEED_MOCK_SHOPS_ON_START=true next to the other store flags and restart the backend container; demo shops/templates/reviews appear automatically. Verification: tsc clean.


Per user screenshots: (1) shop cards — circular «+» add-to-cart moved OFF the photo, now sits beside the title/price row; (2) template detail — web-style carousel: rounded image, n/m arrow/counter pill, touch-swipe navigation, centered thumbnails; (3) removed «سفارش این قالب» everywhere (product logic: buyer either creates a request or buys via cart) — add-to-cart is the single full-width CTA in the detail sticky bar. Files: telegram-seller-shop-view.tsx, telegram-template-detail-view.tsx. Verification: tsc + eslint clean.


2026-06-03 — frontend v2.8.70 — in-shell settings + addresses, central theme + dark mode

In-shell account screens (were opening the web dashboard):

  • تنظیمات عمومی → telegram-settings-view.tsx: name/email/bio edit form via updateUserProfile; session refresh on save.
  • آدرس‌های تحویل → telegram-addresses-view.tsx: full address CRUD (list / add / edit / delete / set-primary) on the addresses API + use-telegram-addresses SWR hook.
  • Account rows switched from href→onClick; new overlay states 'settings'/'addresses' in the shell. (Passkey stays web — WebAuthn limitation.)

Theme overhaul — Mini App now uses the CENTRAL web-app theme:

  • constants.ts: TG_PALETTE_LIGHT / TG_PALETTE_DARK derived from src/theme/theme-config.ts (primary green, grey ramp, status colors) instead of the ad-hoc cream/saffron palette; font stack aligned with AMANEH_FONT (IBM Plex Sans / Vazirmatn).
  • telegram-shell-css.ts: emits both palettes; .tg-shell--dark flips every token — components keep the same CSS-var names.
  • use-telegram-theme hook: auto (follow Telegram colorScheme) | light | dark, persisted in localStorage; syncs Telegram native header/background colors.
  • Theme toggle row (auto/light/dark) added to the account tab.
  • Hero headers (account / seller-shop / welcome banner / unlinked) use new --header-bg/--header-fg/--header-fg-muted vars so they stay legible in both modes.
  • Removed the inline themeParams background/text override in getTelegramShellStyle that forced a navy shell under light components on dark-mode Telegram clients (the bad bg/font the user reported). Verification: tsc + eslint clean (one pre-existing no-constant-condition warning in telegram-webapp.ts, untouched).

2026-06-03 — frontend v2.8.71 — Mini App: solar icons, avatar upload, achievements, theme-toggle fix

Icons: telegram-icons.tsx redrawn to mirror the dashboard's Solar "linear" set (home-2 / document-add / chat-round-dots / user / shop-2 / wallet-money / rounded arrows). Added sun, moon, themeAuto, camera, trophy, lock. Theme toggle now icon-based (sun=light, moon=dark, half-disc=auto) instead of text labels — per user request. BUGFIX — theme toggle didn't respond: TelegramListRow rendered a disabled wrapper whenever it had no onClick; a disabled button swallows clicks from nested children, so the language AND theme toggles (rendered inside the row's value slot) never received taps. Non-interactive rows (no onClick, no href) now render a

. Avatar upload in settings: telegram-settings-view gained a tappable avatar with a camera badge → file picker → uploadUserAvatar + updateUserProfile (persists immediately + session refresh); 3 MB / JPG-PNG-GIF validation; relative /uploads paths resolved to full backend URL. Achievements in points view: milestone badges (joined / verify email / first referral / 5 active referrals / 1000 points) derived client-side from the user's points + referral + isEmailVerified data; unlocked (trophy + "earned") vs locked (lock + "+N" reward) with an N/total counter. Display-layer gamification over the existing server-side point grants. Verification: tsc + eslint clean.


Bug fixes (user testing on v2.8.71):

  • Avatar broken in account header — it rendered the raw relative /uploads path against the frontend origin (404 inside the Telegram WebView). New shared resolveTelegramAvatar (sections/telegram/avatar-url.ts) rebases avatars onto the backend origin; used in both the account header and the settings uploader.
  • Inline email verification in settings — changing the email now surfaces a code field + verify/resend buttons (resendProfileEmailVerification + verifyProfileEmail) whenever the saved email is unverified, instead of forcing the user back to the account-tab badge.
  • Passkey/Wallet rows broke the Mini Apphref replaced the whole WebView with the web dashboard and the Telegram back button couldn't return. They now open in Telegram's in-app browser via webApp.openLink (openTelegramExternalLink in telegram-webapp.ts), keeping the Mini App mounted.

Feature — search / filter / sort (queued item):

  • Requests: search + status filter chips (all / active / completed / cancelled)
    • sort (newest / highest budget).
  • Shop sellers: search + sort (top rated / most templates / newest).
  • Seller-shop templates: search + sort (newest / cheapest / priciest), shown when the shop has > 3 templates.
  • New reusable TelegramListControls component + controls locale section. Verification: tsc + eslint clean.

2026-06-03 — backend v2.8.68 + frontend v2.8.73 — pending-email flow, referral signup bonus, Mini App cleanups

Backend v2.8.68 — email change now requires verification before it applies:

  • New pendingEmail field. Changing email parks the new address in pendingEmail (NOT email) and sends a code to the NEW address; the account email + its verified status stay intact until the code is confirmed, then pendingEmail is promoted to email (with a re-check nobody else claimed it). Resend targets the pending address. Implemented in the Mongoose User model AND the Postgres authStore path (PgAuthUser field + constructor + toObject + savePgUser INSERT/ON CONFLICT params + mirrorUserToMongo + pending_email column via idempotent ALTER in postgresAuthSchema). userController.updateUserProfile / resendCurrentUserEmailVerification / verifyCurrentUserEmail updated.
  • Referral signup bonus: referrer gets +50 points the moment someone signs up with their code (email + Google paths) via PointsService.addPoints (source 'referral'), referralStats.totalEarned kept in step, 'referral-reward' socket event. Best-effort, never blocks signup. Separate from the 2% purchase commission.

Frontend v2.8.73 — Mini App:

  • Removed the «وضعیت‌های امانت» escrow-states legend from Home.
  • Removed the Passkey row from the account tab (not wanted in the Mini App).
  • Welcome-banner greeting uses the AMN account name before the Telegram profile name, so an in-app rename updates it (was stuck on Telegram's "Anonymous …").
  • Settings email verify aligned with the pending-email flow: inline code panel shows whenever a change is pending or the email is unverified; clears on verify. Verification: backend tsc clean; frontend tsc + eslint clean.

2026-06-03 — frontend v2.8.74 — Mini App chat alignment fix + read-only email field

  • Chat both-sides bug: in the Mini App chat every bubble rendered on the same side in one colour. Root cause: isMine = senderId === selfId — when selfId was undefined it matched system/welcome messages (no senderId), so all bubbles looked "mine". Fixed: ids normalised (string | populated object), a match now requires BOTH ids non-empty, and selfId falls back from _id to id. Own = right/primary green, support/system = left/neutral grey. (telegram-chat-bubble.tsx + selfId hardening in telegram-mini-app-view.tsx.)
  • Read-only email: the email field is now disabled by default with an explicit «تغییر ایمیل» toggle, in BOTH the Mini App settings and the web account-general form. Together with the backend pending-email flow (v2.8.68), email can only change after the code sent to the NEW address is verified — no silent change by typing + save. Verification: tsc + eslint clean.

2026-06-03 — frontend v2.8.75 — self-contained email-change flow (visible code entry)

After enabling «تغییر ایمیل» there was no obvious place to send/enter the code. Reworked into a dedicated flow on both surfaces:

  • Mini App settings: editing the email reveals a «ارسال کد تأیید» button beside it; pressing it parks the new address (pending) + emails the code and immediately reveals the code-entry panel. Email is no longer bundled into the name/bio Save (handleRequestEmailChange).
  • Web account-general: submitting an email change reveals the verification Alert (code field + verify/resend) even though the account email stays on the old verified address until the code is confirmed (pendingEmailVerify state). Verification: tsc + eslint clean.

2026-06-03 — frontend v2.8.76 — email send-code always reveals the verify panel

Bugfix: «ارسال کد تأیید» was disabled when the email was unchanged and handleRequestEmailChange early-returned silently for the same address — so after "sending" nothing appeared. Now the button is enabled whenever an email is present; a changed email parks it as pending + codes the new inbox, an unchanged email (re)sends a code to verify the current address. Either way pendingVerify flips on and the inline code-entry panel renders. Verification: tsc + eslint clean.


2026-06-03 — frontend v2.8.77 — keep email verify panel mounted after sending code

Bugfix: pressing «ارسال کد تأیید» reverted the settings view to its initial (non-edit, no-panel) state — the code field never appeared. Root cause: handleRequestEmailChange awaited onSaved (checkUserSession), which sets the auth provider's loading flag true; the Mini App shell gates the entire linked block on !loading, so it unmounted + remounted the settings view, wiping its local pendingVerify/emailEditable state. Removed the session refresh from the send-code path (the email is only pending then — nothing to refresh until verified; handleVerify still refreshes after a successful confirm). Verification: tsc clean.


2026-06-03 — backend v2.8.69 + frontend v2.8.78 — chat system messages + post-delivery review

Chat both-sides (round 2 — real root cause): support welcome/greeting and group "chat created" messages are authored on the backend with chat.metadata.createdBy (the buyer) as senderId, so they rendered as the buyer's own. Backend (v2.8.69): sendSystemMessage now takes an optional senderId and createSupportChat passes the support user's id. Frontend (v2.8.78): any messageType:'system' message renders as a neutral centred note regardless of sender — robust fix for all system messages. Post-delivery seller review (frontend v2.8.78): buyer viewing a delivery/delivered/seller_paid/completed request sees a star + comment prompt (TelegramReviewPrompt). Seller resolved from the request's selected offer (getOffers → selectedOfferId → sellerId); submits createReview({subjectType: 'seller', subjectId, rating, comment, purchaseRequestId}); 409 → "already reviewed" handled gracefully. Verification: backend + frontend tsc clean.


2026-06-03 — backend v2.8.70 — persist telegram user updates on sign-in (new-user login)

Investigating "new Telegram account / new device can't log in (existing works)". telegramAuth mutates the user after the TelegramLink lookup (telegramVerified flip, lazily backfilled referralCode, lastLoginAt) but only NEW users hit user.save(); existing/half-created users were never re-persisted → a backfilled referralCode never landed and subsequent sign-ins kept tripping the non-sparse users.referralCode index. Added a best-effort user.save() before linking. Note: background logs in the user's screenshots show "Backend 2.6.49" — the deployed dev backend may be far behind (none of the 2.8.x auth/email/chat fixes live). Asked the user to confirm/redeploy the backend and share the exact error the new device shows (frontend surfaces backend message verbatim). Verification: backend tsc clean.


2026-06-03 — backend v2.8.71 — fix template-creation 500 (blank optional fields vs CHECK constraints)

Seller "create request template" returned HTTP 500 whenever optional fields were left empty. The request_templates table has CHECK constraints that allow only NULL or a valid pattern — product_link (~ '^https?://.+') and delivery_email (email regex) — and the nullable service_session_type enum rejects ''. Both DrizzleMarketplaceRepo.createRequestTemplate and updateRequestTemplate passed the form's empty strings straight into the INSERT/UPDATE, tripping the constraints. Added a blank() coercion: blank optional strings → NULL/undefined for product_link, delivery_email, size/color/brand, service_location/notes, proposal_title/description, service_session_type; enum columns (product_type, delivery_type, urgency, budget_currency, metadata_source) fall back to their defaults instead of receiving ''. (Confirmed deployed backend is now on the 2.8.x line — logs show v2.8.70.) Verification: backend tsc clean.


2026-06-03 — frontend v2.8.79 — request-template maxUsage truly optional

The «حداکثر تعداد استفاده (اختیاری)» field threw a ZodError ("Invalid input") and blocked the template form submit. It was wrapped in schemaHelper.nullableInput — a helper for REQUIRED fields that adds an issue on null/undefined — so an empty optional value failed. Replaced with zod.preprocess (blank/null/NaN → undefined, validate int ≥ 1 only when present); default null → undefined. Together with backend v2.8.71 (blank → NULL for product_link/delivery_email/enums), optional template fields can now be left empty end-to-end. Verification: tsc + eslint clean.


2026-06-04 — backend v2.8.72→v2.8.73 (chat participant fix reverted) + frontend v2.8.80 (template validation visibility)

Backend chat — attempted + reverted. v2.8.72 tried to canonicalize chat participant ids (userRepo.findById) so a buyer (marketplace uuid for the seller) and seller (legacy ObjectId session) would share one direct thread. It backfired: findById returns an id format (Mongo ObjectId / PG uuid) that no longer matches the session JWT id, so the membership check rejected everyone ("User is not a participant"). v2.8.73 reverts to storing ids as passed (restores chat access). Open issue: buyer↔seller messages still don't cross — they land in separate direct conversations because the seller is referenced by a uuid while the session id is a legacy ObjectId (users in Mongo, chats in Postgres). Needs a correct uuid→legacyObjectId resolution (via marketplace id_map), not the session-breaking findById path. Deferred for a careful fix. Frontend v2.8.80: template-form onInvalid now expands all collapsed accordion sections and toasts the actual nested error messages, so validation errors are visible under their fields instead of only in the console.


2026-06-04 — backend v2.8.74 — chat: map uuid participants → session ObjectId (buyer↔seller fix)

Root cause confirmed from logs: a buyer opens a chat "with seller X" using X's marketplace Postgres uuid (669c0dac-…), but X's session/JWT id is their legacy ObjectId (e0527…). Conversations key participants by the session id, so buyer and seller landed in two separate direct threads (each saw only their own messages). ChatService.createChat now normalises any uuid participant (and createdBy) → users.legacy_object_id via a direct PG lookup BEFORE dedup/create — resolving TOWARD the session-id format (the reverted v2.8.72 resolved toward uuid and broke membership). Both sides now converge on one conversation + socket room. Note: pre-existing mismatched conversations are orphaned; a freshly opened chat dedups correctly for both parties. Verification: backend tsc clean.


2026-06-04 — frontend v2.8.81 — in-shell checkout (Mini App no longer hands off to web)

Cart «ادامه و پرداخت» previously navigated the Telegram WebView to the web checkout (taking over the Mini App). New TelegramCheckoutView keeps it in-shell: order summary + delivery-address picker (reuses saved addresses; "add address" opens the in-shell addresses manager) → places the order via convertTemplatesToRequests({paymentConfirmed:false}) so requests are created in pending_payment (escrow model — NO on-chain/web3 transaction needed inside the Mini App). Cart cleared on success and the shell jumps to Requests, where the buyer completes the escrow payment. New 'checkout' shell overlay + cart clear() on the cart hook. Verified the deferred path exists (status 'pending_payment' when paymentConfirmed is false) rather than requiring batch-convert's paymentConfirmed:true on-chain flow. Verification: tsc + eslint clean.


2026-06-04 — frontend v2.8.82 — checkout: digital skips address; fix buyer pending_payment stepper

  • In-shell checkout: only PHYSICAL products require a delivery address; digital products / services / consultations are delivered online (email/link) so the address picker is hidden. Cart items now carry productType; needsAddress is derived from it (deliveryType fallback for older items).
  • Buyer stepper fix (request-config determineBuyerStep): pending_payment / payment / payment_pending were grouped with the PAID states and returned step 4 ("awaiting shipment"), so a just-placed unpaid order looked like the payment step was skipped. Those un-paid statuses now return step 3 ("select & pay"), so the request correctly shows the payment step as current. Open: the actual on-chain escrow payment step inside the Mini App still needs a wallet rail (TON Connect / WalletConnect) — order placement is in-shell, the pay action is the remaining web3 piece. Verification: tsc + eslint clean.

2026-06-04 — frontend v2.8.83 — Mini App pay CTA (wallet + manual deposit) + stepper line fix

  • Request detail: buyers on a payable status (pending_payment/payment) get a «پرداخت با کیف پول» button that opens the web payment page for that request in Telegram's in-app browser (openTelegramExternalLink). That page already offers BOTH WalletConnect/wagmi AND the in-house manual deposit-address (backend auto-detects the on-chain transfer via request-network webhook), so the user pays without the Mini App being torn down. The generic "open in web" link also uses the in-app browser now.
  • Stepper green line "broken" fix: the two half-connectors per cell left a seam, and the active dot's saffron halo overlapped the line. Replaced with one continuous connector per gap (green once the step is reached) and a solid paper ring on the active dot — the progress line is now continuous. Note: this reuses the existing web WalletConnect + deposit flow via the in-app browser; a fully native in-shell wallet/deposit screen remains a larger follow-up. Verification: tsc + eslint clean.

2026-06-04 — frontend v2.8.84 — archive conversations from the chat list (DEPLOY BRANCH NOW main)

⚠️ Deploy branch changed: pushes now go to main (CI: "push to main builds dev-* image and deploys to Arcane dev env"). integrate-main-into-development is retired.

  • Chat list: each row gained an archive button → confirm sheet → PATCH /chat/:id/archive (archiveConversation) + list refresh. Archived chats are excluded from the default getUserChats query, so the row disappears. Row restructured from a single
into a div wrapping the open-button + a sibling archive button (nested buttons are invalid HTML). New archive icon + chat archive locale strings. Verification: tsc + eslint clean.

2026-06-04 — backend v2.8.75 + frontend v2.8.85/86 — chat search + archive view, template→checkout fix

backend v2.8.75: archiveChat is now two-way — PATCH /chat/:id/archive accepts {archived} (default true); archived:false restores. Powers unarchive. frontend v2.8.85: template detail — once an item is in the cart the sticky bar shows «حذف از سبد» + a primary «مشاهدهٔ سبد و پرداخت» (onOpenCart) so the buyer can actually reach checkout (previously the single button just toggled). frontend v2.8.86: chat list gained a search box (counterpart name + last message) and a «گفتگوهای آرشیوشده» row → TelegramArchivedChatsView (lazy isArchived=true fetch; per-row unarchive via archiveConversation(id,false)). getConversations/archiveConversation actions + useTelegramConversations(archived) updated. All on the new main deploy branch. Verification: tsc + eslint clean.


Root cause of "archived chats not removed from main + archived view shows everything": DrizzleChatRepo.matchesCondition routed BOOLEAN query conditions through idEquals, and asStringId() coerces both true and false to '' — so idEquals(false,true) === true, making the settings.isArchived filter match every chat. Booleans now compare strictly (Boolean(value) === condition); a missing value counts as false. (Also corrects participants.isActive matching.) frontend v2.8.87: the archived chats view gained the same search box as the main list (+ no-results state). Unarchive already worked (archiveConversation (id,false)). Verification: backend + frontend tsc/eslint clean.


2026-06-04 — frontend v2.8.88 — in-shell direct-transfer payment (no wallet connection)

Buyers can now pay a pending_payment request entirely inside the Mini App, with NO web3 wallet connection — mirroring the web «انتقال مستقیم» rail. TelegramPaymentView: resolves the selected offer (seller + amount) → getPaymentOptions for chain/token → createDirectBalanceIntent (provider amn.scanner, rail direct_balance, mode check) → shows a copyable derived deposit address + exact amount + a strong "send only this token on this network" warning → «پرداخت کردم» calls checkDirectBalancePayment. Edge cases (all surfaced): OVERPAYMENT accepted (backend delta>=expected → paid); UNDERPAYMENT shows "received X of Y, send the rest" (paidDelta vs expectedAmount); WRONG TOKEN — the check flow can't detect a foreign token, so a prominent on-screen warning is shown (backend webhook rejects token/chain mismatch). Wired from the request-detail pay CTA via a new openPaymentRequestId shell overlay + BackButton. (Replaces the earlier "open web payment" hand-off.) Verification: tsc + eslint clean.


2026-06-04 — frontend v2.8.89 — Mini App checkout flows straight into payment (shop model)

Per the shop model (pay as part of ordering, not place-then-pay-later), the in-shell checkout now, after «ادامه و پرداخت», creates the request(s) (template conversion already mints + accepts an offer and sets selectedOfferId, status pending_payment) and for a SINGLE-request order jumps directly into the in-shell direct-transfer payment step (openPaymentRequestId) rather than dropping the buyer on the Requests list. Multi-request orders still go to Requests. Checkout reads the created request id from batch-convert's data[]; button relabeled to «ادامه و پرداخت». (The normal request→offers→pay escrow flow is unchanged.) Verification: tsc + eslint clean.


2026-06-04 — frontend v2.8.90 — checkout stepper + paid-step advance fix

  • In-shell checkout shows a 3-step progress header (cart → address → payment), reusing TelegramRequestStepper (new optional title prop): cart step 1, checkout step 2, payment step 3 (only when reached from checkout via the new checkoutFlow prop + paymentCheckoutFlow shell state). BackButton steps checkout→cart and checkout-payment→cart.
  • Stepper fix: a PAID request (status 'payment' = escrow funded — confirmed working end-to-end via the AMN scanner: [direct-balance] check paid … delta) was stuck on step 3 because 'payment' was grouped with the unpaid statuses. determineBuyerStep now maps 'payment' → 4 (awaiting shipment), 'confirming' → 5; only pending_payment/payment_pending stay on 3. Request-detail «پرداخت» CTA hidden once status is 'payment'. Note: in-shell direct-transfer payment is confirmed working — a real 0.01 USDC BSC transfer to the derived address was auto-detected and advanced the request to 'payment'. Verification: tsc + eslint clean.

2026-06-04 — frontend v2.8.91 — seller stepper: paid/shop request shows "ship goods"

Buyer stepper confirmed correct (paid shop request → step 4 «انتظار ارسال»). Seller side was wrong: a paid shop (template) purchase showed seller step 1 («ارسال پیشنهاد») because determineSellerStep gated the step on an accepted/hasOffer context the seller view couldn't reliably compute (id seam). Since a funded escrow implies an accepted offer for that seller, post-payment seller steps are now FIXED/unconditional: pending_payment/payment_pending → 2 (awaiting payment), payment/processing/paid → 3 (ship goods), delivery → 4, delivered/confirming → 5, seller_paid/completed → 6. Removed the unreliable accepted flag. (Pre-payment states keep the hasOffer 1/2 logic.) Verification: tsc + eslint clean.


2026-06-04 — backend v2.8.77 — seller can't ship: delivery 403 (uuid↔ObjectId seam)

The selected seller got HTTP 403 on PUT /marketplace/purchase-requests/:id/delivery («تأیید ارسال کالا») for a paid shop request. The auth compared the seller's session id (legacy ObjectId e0527…) against the selected offer's sellerId, which a TEMPLATE-created offer stores as a PG uuid (669c0dac…) → never equal → 403. Added a sameUser(a,b) helper that resolves both ids via the user repo (accepts either format) and compares every id form (_id/id/pgId/legacyObjectId). Applied to updateDeliveryInfo, verifyDeliveryCode, and the delivery-code-status buyer/seller gate. (Buyer step labels already matched the web — no change.) Verification: backend tsc clean.