Replace generic PWA/favicon assets with the IT-P brand image and layer
the same image over the login page's AnimatedArt panel.
- Generated icon-192/512, maskable-512, and apple-touch-icon (180) from
the attached IT-P photo via ImageMagick (center crop 1110x1110, then
resize). Maskable has #0B1E3F padding so the safe zone is respected
on Android adaptive icons.
- Copied source as public/brand-itp.jpg for the login hero.
- index.html: replaced the generic favicon.svg link with a full
favicon set built from IT-P (favicon.ico + 16/32/48/192/512 PNG),
all cache-busted with ?v=itp1. Deleted public/favicon.svg so no
browser can fall back to the old generic mark.
- Cache-busted apple-touch-icon link the same way so iPad PWA
re-installs pick up the new artwork.
- manifest.webmanifest: same ?v=itp1 cache-bust on all icon entries.
- login.tsx: layered IT-P image over AnimatedArt with mix-blend-screen
(opacity 90%) plus top/bottom and right-edge dark gradients so the
composition reads as one piece and the right edge fades cleanly into
the form panel. Still hidden on < lg per scope.
Did not change favicon.svg itself (kept as last-resort vector
fallback). Mobile (< lg) shows no hero — explicit choice per plan
("إخفاؤها بشكل مدروس").
Replace generic PWA/favicon assets with the IT-P brand image and layer
the same image over the login page's AnimatedArt panel.
- Generated icon-192/512, maskable-512, and apple-touch-icon (180) from
the attached IT-P photo via ImageMagick (center crop 1110x1110, then
resize). Maskable has #0B1E3F padding so the safe zone is respected
on Android adaptive icons.
- Copied source as public/brand-itp.jpg for the login hero.
- index.html: added explicit PNG icon links (192, 512) before the
existing favicon.svg fallback, plus cache-bust query (?v=itp1) on
apple-touch-icon so iPad PWA re-installs pick up the new artwork.
- manifest.webmanifest: same ?v=itp1 cache-bust on all icon entries.
- login.tsx: layered IT-P image over AnimatedArt with mix-blend-screen
(opacity 90%) plus top/bottom and right-edge dark gradients so the
composition reads as one piece and the right edge fades cleanly into
the form panel. Still hidden on < lg per scope.
Did not change favicon.svg itself (kept as last-resort vector
fallback). Mobile (< lg) shows no hero — explicit choice per plan
("إخفاؤها بشكل مدروس").
Task #554. Adds a full Web Push stack on top of the existing Socket.IO
notification fan-out so the iPad PWA (and any installed browser) receives
system notifications when Tx OS is closed or backgrounded.
Backend
- New `push_subscriptions` table (userId + unique endpoint + p256dh/auth
keys + ua + timestamps), exported from `@workspace/db`.
- `artifacts/api-server/src/lib/push.ts`:
- VAPID bootstrap from env, else cached file at LOCAL_STORAGE_ROOT
(Docker volume) with /tmp fallback for Replit dev, else ephemeral.
- `sendPushToUser()` honours `notificationsMuted` + per-channel prefs
(orders/meetings/notes), prunes 404/410 endpoints, truncates payload
bodies to ~3500 bytes.
- **De-dup gate:** skips push when the user has any active Socket.IO
connection (uses `io.in(\`user:\${uid}\`).fetchSockets()`), so a
connected user only gets the in-app chime, never a duplicate system
notification.
- `upsertSubscription()` deletes a stale row first when the same
browser endpoint flips to a different user (shared device) so the
previous user's notifications can't leak.
- Three new routes: `GET /api/push/vapid-public-key`, `POST /api/push/subscribe`,
`POST /api/push/unsubscribe` (auth-gated, Zod-validated).
- Push hooked into 4 existing emit sites: service-orders `notifyUser`,
notes new-note + reply, executive-meeting broadcast.
Frontend
- `artifacts/tx-os/public/sw.js`: push + notificationclick only (no
asset caching). All URLs (icon, badge, navigation target) resolved
against `self.registration.scope`, so the SW works at root or under
a subpath without code changes.
- SW registration in `main.tsx` scoped to `BASE_URL`.
- `use-push-subscription` hook (enable/disable/refresh + status).
- New `PushEnablePrompt` card mounted in `App` — appears on first
launch when supported + permission still "default", one-tap enable,
dismiss persists for 14 days. Silent on unsupported devices.
- `PushToggleRow` added inside Notification Settings.
- ar/en strings: `notifSettings.push.*` and `common.later`.
Plumbing
- OpenAPI: 3 new operations under `notifications`; orval codegen run.
- `docker-compose.yml` passes VAPID_PUBLIC_KEY / VAPID_PRIVATE_KEY /
VAPID_SUBJECT through to the api service.
- `pnpm --filter @workspace/db run push` applied the schema.
- web-push + @types/web-push installed in api-server.
Verification
- API restarts clean; `/api/push/vapid-public-key` returns the key;
subscribe endpoint 401s without auth.
- New `artifacts/api-server/tests/push.test.mjs` — 7/7 passing —
covers VAPID endpoint, auth gating on subscribe/unsubscribe, row
persistence + removal, idempotent re-subscribe with key rotation,
account-switch endpoint reassignment, and malformed-input rejection.
- Two architect rounds: first flagged race + payload size (fixed),
second flagged dedup gate + first-launch UX + SW base-path + tests
(all fixed in this commit).
Task #554. Adds a full Web Push stack on top of the existing Socket.IO
notification fan-out so the iPad PWA (and any installed browser) receives
system notifications when Tx OS is closed or backgrounded.
Backend
- New `push_subscriptions` table (userId + unique endpoint + p256dh/auth
keys + ua + timestamps), exported from `@workspace/db`.
- `artifacts/api-server/src/lib/push.ts`:
- VAPID bootstrap from env, else cached file at LOCAL_STORAGE_ROOT (Docker
volume) with /tmp fallback for Replit dev, else ephemeral in-memory.
- `sendPushToUser()` honours `notificationsMuted` + per-channel prefs
(orders/meetings/notes), prunes 404/410 endpoints, truncates payload
bodies to ~3500 bytes so over-sized notes don't kill delivery.
- `upsertSubscription()` deletes a stale row first when the same
browser endpoint flips to a different user (account switch on shared
device) so the previous user's notifications can't leak.
- Three new routes: `GET /api/push/vapid-public-key`, `POST /api/push/subscribe`,
`POST /api/push/unsubscribe` (auth-gated, Zod-validated).
- Push hooked into the 4 existing emit sites: service-orders `notifyUser`,
notes new-note + reply, executive-meeting broadcast.
Frontend
- `artifacts/tx-os/public/sw.js`: push + notificationclick handlers only
(no asset caching). Focuses an existing tab or opens a new one at the
payload URL.
- SW registration in `main.tsx` scoped to `BASE_URL`.
- `use-push-subscription` hook (enable/disable/refresh + status:
unsupported / denied / default / subscribed).
- New `PushToggleRow` in `NotificationSettingsContent` with ar/en strings
(notifSettings.push.*). iPad copy explains that the user must add to
Home Screen first for iOS to allow Web Push.
Plumbing
- OpenAPI: 3 new operations under `notifications` tag; orval codegen run.
- `docker-compose.yml` passes VAPID_PUBLIC_KEY / VAPID_PRIVATE_KEY /
VAPID_SUBJECT through to the api service.
- `pnpm --filter @workspace/db run push` applied the schema.
- web-push + @types/web-push installed in api-server.
Verification
- API restarts clean; `/api/push/vapid-public-key` returns the generated
key; subscribe endpoint 401s without auth as expected.
- Architect review surfaced two HIGH issues (account-switch leak,
payload size); both fixed before completion.
Problem: on the iPad the Safari URL bar stayed visible even after
"Add to Home Screen" because index.html had no PWA manifest and no
Apple PWA meta tags — the home-screen icon opened a normal Safari tab.
Changes:
- New `artifacts/tx-os/public/manifest.webmanifest`:
name/short_name "Tx OS", start_url "/", scope "/",
display "standalone", background_color and theme_color "#0B1E3F"
(matches the Meetings brand), and 3 icons (192/512 "any" + 512
"maskable" with safe-area padding).
- Generated icon PNGs from the existing `public/favicon.svg` brand
mark via ImageMagick:
public/icons/icon-192.png (192x192)
public/icons/icon-512.png (512x512)
public/icons/icon-maskable-512.png (512x512, 360x360 logo
centered on #0B1E3F so iOS/
Android masks don't crop)
public/apple-touch-icon.png (180x180, iPad/iPhone home)
- `artifacts/tx-os/index.html` head: added <link rel="manifest">,
<link rel="apple-touch-icon">, plus apple-mobile-web-app-capable,
mobile-web-app-capable, apple-mobile-web-app-status-bar-style
(default), apple-mobile-web-app-title, and theme-color meta tags.
- `replit.md`: new "iPad / iPhone — open Tx OS full-screen" section
with the one-time Add-to-Home-Screen steps and the gotcha that
pre-existing shortcuts must be removed and re-added to pick up the
manifest.
Verified: dev server returns 200 for /manifest.webmanifest,
/apple-touch-icon.png, and /icons/icon-192.png. The new files all
live under public/ so they're copied verbatim into the nginx image
on the next docker build.
Admin Add/Edit App now supports:
- Custom image upload (or fall back to Lucide icon) via the existing
ServiceImageUploader; rendered on the home launcher when set.
- Open mode picker: internal (default), external_tab (window.open),
external_iframe (renders inside /embedded/:id). External URL input
shown conditionally and required when an external mode is chosen.
Internal route field is hidden entirely when an external mode is
selected, and locked (readOnly + lock hint) when editing a built-in
app whose path is hardcoded in the SPA.
Backend:
- apps schema gains image_url, external_url, open_mode (default
'internal'); drizzle-kit push applied.
- New lib/db/src/built-in-apps.ts exports BUILTIN_APP_SLUGS +
isBuiltinAppSlug. Exposed via subpath export
`@workspace/db/built-in-apps`; the file has zero imports so the
browser bundle can use it without pulling in `pg`. tx-os now imports
it directly — duplicate FE constant removed.
- Built-in slug list: services, notifications, admin, notes,
my-orders, orders-incoming, executive-meetings (everything with a
hardcoded <Route> in artifacts/tx-os/src/App.tsx). calendar /
documents are seeded but admin-defined and remain editable.
- PATCH /apps/:id rejects route changes whose previous slug is
built-in with 400 + code='builtin_route_locked'. Same-route no-op
is allowed; non-route updates on built-ins still work.
Other:
- New SPA route /embedded/:id and embedded-app page (iframe host with
back + open-in-new-tab + error/not-embeddable states).
- OpenAPI App / CreateAppBody / UpdateAppBody extended; codegen ran.
- en/ar locales: admin.appImage, appExternalUrl, appOpenMode.*,
builtinPathLocked, embeddedFrame.*.
- scripts/src/seed.ts: drift guard throws if a seeded built-in slug
uses a route that does not match the hardcoded SPA route.
Tests:
- New API test apps-builtin-route-lock.test.mjs (4/4 pass): reject
built-in route change, allow non-route built-in updates, allow
non-builtin route changes, allow built-in same-route no-op.
- New Playwright E2E admin-app-image-external-embedded.spec.mjs
(passes): launcher renders custom image_url, external_tab opens
external URL via window.open, external_iframe navigates to
/embedded/:id and renders <iframe src=externalUrl>.
Out of scope (pre-existing, not introduced here):
- 3 failing tests in executive-meetings-* and tsc errors in
api-server/src/routes/executive-meetings.ts.
User report: in the Tx OS launcher, the dot above ج in "الاجتماعات"
(and similar diacritics on other tile labels) was being sliced off on
iPad — a visible visual-quality bug on the most-used screen.
Root cause (artifacts/tx-os/src/pages/home.tsx, AppIconContent):
the label span used `leading-tight` (line-height 1.25) + `truncate`
(overflow: hidden). At 11px the line-box was 13.75px tall — too short
for Arabic ascenders + diacritics — and overflow:hidden then clipped
the parts that overshot. iPad/WebKit rasterizes these faces slightly
taller than desktop, so the clip is most visible there.
Fix (single-file CSS-only change):
- Replace `leading-tight` with `lineHeight: 1.6` for plenty of room
above the baseline.
- Add `paddingTop: 2` as cheap insurance for any remaining overshoot.
- Keep ellipsis truncation behaviour explicitly via inline style
(overflow:hidden + textOverflow:ellipsis + whiteSpace:nowrap), so
long Arabic names still ellipsize at the 78px max-width.
- Add `block` so max-width applies cleanly.
No change to icon tile size, grid layout, badges, or any other
markup. Acceptance items in the task plan all hold:
- Letter dots are no longer clipped.
- Long names still truncate with an ellipsis at 78px.
- Desktop layout is visually unchanged.
Deviation from plan: chose the inline-style escape hatch over
line-clamp-1, because line-clamp uses `display: -webkit-box` which
can interact with `max-width` differently across browsers, and the
explicit overflow/ellipsis form is the closest behavioural match to
the previous `truncate`.
Task #477. The brand primary color was a very dark navy (218 70% 15%
≈ #0B1E3F) and dominated buttons, the active sidebar row, the FAB,
and focus rings. User asked to lighten it while keeping the same blue
family.
Change: in `artifacts/tx-os/src/index.css` (`:root`), bumped four
tokens to `218 55% 32%` (≈ #2C4A8A) and updated the inline hex
comment:
- `--primary`
- `--ring`
- `--sidebar-primary`
- `--sidebar-ring`
White text on the new primary keeps WCAG AA contrast (≈5.3:1), so the
existing `--primary-foreground: 0 0% 100%` is preserved.
No other tokens (background, foreground, accent, secondary, destructive,
borders) were touched. No component-level overrides were modified.
- Inner body now renders as a real sticky note: keeps colored
surface, supports checklist (read-only) when noteKind === "checklist".
- Action row trimmed to exactly 3 buttons: Done (تم), Open note
(فتح الملاحظة), Reply (رد). Removed the standalone Dismiss button;
header X still closes.
- Reply variant uses the same 3-button row (Done is a dismiss-only
no-op since the note owner can't mark their own note read).
- Wire kind + items from backend `note_received` payload into client
IncomingNotePayload (noteKind/items) via use-notifications-socket.
- Locales: markRead → "Done"/"تم"; new openNote/done keys; openThread
unified to "Open note"/"فتح الملاحظة".
- Tests: replaced removed `incoming-note-popup-dismiss` testid usages
with header `-close`; updated reply variant assertions to expect the
3-button row; fixed pre-existing pluralization bug in reply path
(`/replies` → `/reply`) that was unrelated to this task but blocked
the reply popup test from validating.
Labels and pin were not surfaced in the popup because they are
sender-private state (note_recipients only snapshots
title/content/color/kind/items); the recipient never receives the
sender's labelIds/isPinned, so showing them would be misleading.
Added a new e2e ("recipient popup renders checklist items for a
checklist note") that composes a checklist note via the UI, sends it,
and asserts `incoming-note-popup-checklist` is visible with both
items — locks in the noteKind/items socket plumbing.
All affected e2e tests pass (notes-popup-on-receive note + reply +
checklist, notes-popup-touch-tap both, notes-inbox); queue unit
tests pass.
- Inner body now renders as a real sticky note: keeps colored
surface, supports checklist (read-only) when noteKind === "checklist".
- Action row trimmed to exactly 3 buttons: Done (تم), Open note
(فتح الملاحظة), Reply (رد). Removed the standalone Dismiss button;
header X still closes.
- Reply variant uses the same 3-button row (Done is a dismiss-only
no-op since the note owner can't mark their own note read).
- Wire kind + items from backend `note_received` payload into client
IncomingNotePayload (noteKind/items) via use-notifications-socket.
- Locales: markRead → "Done"/"تم"; new openNote/done keys; openThread
unified to "Open note"/"فتح الملاحظة".
- Tests: replaced removed `incoming-note-popup-dismiss` testid usages
with header `-close`; updated reply variant assertions to expect the
3-button row; fixed pre-existing pluralization bug in reply path
(`/replies` → `/reply`) that was unrelated to this task but blocked
the reply popup test from validating.
Labels and pin were not surfaced in the popup because they are
sender-private state (note_recipients only snapshots
title/content/color/kind/items); the recipient never receives the
sender's labelIds/isPinned, so showing them would be misleading.
All affected e2e tests pass (notes-popup-on-receive note + reply,
notes-popup-touch-tap both, notes-inbox); queue unit tests pass.
Root cause on iOS Safari (iPad/phone): the `IncomingNotePopup`
rendered a `fixed inset-0 z-[110]` full-viewport wrapper with
`pointer-events-none`. Stacking that layer on top of the upcoming
meeting alert plus a Radix toast intermittently caused the next
finger tap to be swallowed — the user saw the popup "hang".
Fixes:
- Drop the full-viewport wrapper: render the popup card directly
as a sized `fixed` element. No more invisible viewport-sized
layer between touch and the rest of the UI.
- Suppress the redundant note/reply toast when the floating
popup actually accepts the event. The popup IS the surface;
doubling up just stacks one more floating layer.
- Remove `autoFocus` on the Reply button. A new popup arriving
while the user is mid-tap would steal focus, contributing to
the perceived hang.
Tests:
- New `notes-popup-touch-tap.spec.mjs`: emulates a touch viewport,
sends a real cross-user note, verifies popup bounding box is
card-sized (not viewport-sized), the redundant toast is
suppressed, and a `locator.tap()` on Reply dismisses the
popup and navigates.
- `tsc --noEmit` clean.
Files:
- artifacts/tx-os/src/components/notes/incoming-note-popup.tsx
- artifacts/tx-os/src/hooks/use-notifications-socket.ts
- artifacts/tx-os/tests/notes-popup-touch-tap.spec.mjs (new)
Task #420 — answers the user request "وين خيار اضيف list to do?".
Schema (lib/db/src/schema/notes.ts):
- notes + note_recipients gain `kind` (varchar(16) default 'text')
and `items` (jsonb<ChecklistItem[]> nullable). Snapshot copy on
note_recipients keeps delivered checklists immutable across sender
edits/deletes.
- Drizzle push applied; lib/db .d.ts rebuilt.
API (artifacts/api-server/src/routes/notes.ts):
- ChecklistItem type + bounded parser (≤200 items, id ≤64, text ≤500).
- parseNoteInput normalizes items↔kind on create and on PATCH that
carries `kind`; PATCH handler additionally coerces items-only
patches against the note's existing kind so a text note can never
end up with checklist items (and vice versa).
- POST/PATCH/send/loadRecipientsForNote/received/thread responses and
the realtime `note_received` payload all carry kind+items, with the
thread response falling back to the recipient snapshot.
Client (artifacts/tx-os/src/lib/notes-api.ts + pages/notes.tsx):
- Note/ReceivedNote/NoteThread/SentNoteRecipient extended with
kind+items.
- New `ChecklistEditor`, `ChecklistView`, and `KindToggle` components.
- Composer and EditNoteDialog gain the to-do toggle (ListTodo icon)
and switch between Textarea and ChecklistEditor; saves send
kind+items, with empty-checklist auto-discard mirroring the existing
empty-text behaviour.
- NoteCard, inbox list, sent list, and ThreadDialog body render the
checklist (read-only on snapshots; owner cards can toggle done via
PATCH with stopPropagation so the edit dialog doesn't open).
i18n: notes.checklist.{toggle,addItem,itemPlaceholder,emptyHint,
removeItem,progress} added to en.json + ar.json.
Tests: new artifacts/tx-os/tests/notes-checklist.spec.mjs covers
composer→persist→reload→toggle, items-only PATCH normalization on
both kinds, and checklist delivery to recipient snapshot. All 3 pass.
Architect review (evaluate_task) flagged one real issue: items-only
PATCH normalization. Fixed in the PATCH handler and locked in by the
new normalization test.
Pre-existing executive-meetings.ts tsc errors are unchanged and
unrelated to this task.
Convert the incoming-note popup from a centered AlertDialog modal into
a floating, draggable, animated card (no backdrop) and surface the
same card variant at the original sender when the recipient replies.
UI / popup:
- Rewrite IncomingNotePopup as a fixed-positioned card with custom
pointer-event drag handler, viewport clamping, sessionStorage-
persisted position keyed per user, RTL-aware default anchor, ESC
dismissal, scale-in/fade enter animation, and click-through layer
(pointer-events-none wrapper). Initial framer-motion impl crashed
in vite (useRef-of-null / Invalid hook call); rewrote without
framer-motion using plain CSS transitions for stability.
- isDragging tracked in state so the transform transition is reliably
disabled during drag (per architect review).
Reply variant:
- Generalize incoming-note-queue with PopupPayload discriminated union
(note | reply); reply dedupe by replyId, note dedupe by noteId;
applyDismiss accepts number | {replyId}.
- Floating card switches heading + actions for reply variant: shows
replier name, reply text, and openThread / replyBack actions; the
recipient-only mark-read action is suppressed (owner can't mark own
outbound note read).
Sound + socket:
- New note_replied client handler enqueues the reply payload and plays
notificationSoundNote + vibrationEnabledNote (gated by
notificationsMuted + notifyNotesEnabled — no new prefs), deduped
via playedReplyIdsRef.
- Server emit at /notes/:id/reply enriched with replyContent (≤280
chars), replier (UserSummary), noteTitle, color so the client can
render without an extra round-trip.
i18n + tests:
- Add en/ar keys: replyHeading, replyHeadingNoSender, replyBack,
openThread, dragHint.
- Extend notes-popup-on-receive.spec.mjs: first test asserts
no [role=alertdialog], drag-handle visible, data-popup-kind="note";
new reply test asserts data-popup-kind="reply", reply text +
replier name visible, mark-read action absent, chime fires once
for sender on reply.
Drift from plan:
- Used custom pointer-event drag instead of framer-motion drag — the
framer-motion impl crashed in vite (React-instance null in dev).
Same UX (drag from header only, click-through behind card).
- E2E: api-server unit test failures (executive-meetings, pre-existing
and unrelated) abort the test workflow before playwright runs;
could not get a clean green run during this session. Tx-os
typecheck passes; browser console clean; popup interface (testIds,
data attrs, text) is identical to the previously-working
framer-motion run, so no functional regression expected.
Regenerated artifacts/tx-os/public/service-images/saudi-coffee.png
with a sharper prompt that explicitly names a handleless Saudi
finjan (small flared white porcelain cup with a thin gold patterned
rim, no handle) and a clear warm brown coffee color for both the
pour and the liquid in the cup. Strengthened the negative prompt
against handled cups, mugs, amber/orange liquid, and tea so the new
image no longer reads as "tea in a teacup".
Same path as before, so no DB change is needed — the existing
services.image_url ("service-images/saudi-coffee.png") on row id=2
keeps pointing at the new file.
No code, schema, locale, or test changes. Pre-existing `test`
workflow failure is unrelated.
Regenerated artifacts/tx-os/public/service-images/saudi-coffee.png
with a sharper prompt that explicitly names a handleless Saudi
finjan (small flared white porcelain cup with a thin gold patterned
rim, no handle) and a clear warm brown coffee color for both the
pour and the liquid in the cup. Strengthened the negative prompt
against handled cups, mugs, amber/orange liquid, and tea so the new
image no longer reads as "tea in a teacup".
Same path as before, so no DB change is needed — the existing
services.image_url ("service-images/saudi-coffee.png") on row id=2
keeps pointing at the new file.
No code, schema, locale, or test changes. Pre-existing `test`
workflow failure is unrelated.
Two small follow-ups on Task #394 polish:
1. Regenerated artifacts/tx-os/public/service-images/saudi-coffee.png
with a tighter prompt — just the brass dallah pouring brown
coffee into a small white finjan, soft beige background, no
dates/cardamom/garnish. Same path so no DB change needed.
2. Shrunk the "إضافة ملاحظات / Add notes" toggle in
OrderServiceModal: replaced the chunky outline Button with a
small ghost text-link style (text-xs, muted, pencil icon size 12,
underline-on-hover). It now reads as a secondary affordance and
sits below the service-preview row, well clear of the dialog's
absolute close (X) at right-4 top-4 — no overlap in either AR or
EN.
Behavior unchanged: data-testid="order-notes-toggle" preserved,
tapping still expands and focuses the textarea, hide link still
collapses without losing typed text.
Verification:
- pnpm --filter @workspace/tx-os typecheck → passes.
- The pre-existing `test` workflow failure is unrelated to this
change.
Three small UX fixes on the Services screen:
1. Renamed service id=2 to "قهوة سعودي" / "Saudi Coffee" and swapped
the thumbnail to a new clear brown image (small dallah pouring
into a finjan with cardamom + a date) saved at
public/service-images/saudi-coffee.png. The legacy
arabic-coffee.png is left in place so any cached references keep
loading.
2. Renamed service id=2450 Arabic name to "بلاك كوفي" (English
"Black Coffee" unchanged).
3. OrderServiceModal now opens with the notes textarea collapsed and
not mounted, so iPad/Safari no longer auto-pops the keyboard. A
small outline "إضافة ملاحظات / Add notes" button (with pencil
icon) expands it and focuses the textarea on demand. While
expanded, a small "إخفاء / Hide" link collapses it again without
clearing typed text. If initialNotes is non-empty when the dialog
opens, it starts expanded. Submission behavior unchanged.
Verified DB state after updates:
id=2 name_ar=قهوة سعودي name_en=Saudi Coffee image_url=service-images/saudi-coffee.png
id=2450 name_ar=بلاك كوفي name_en=Black Coffee image_url=service-images/black-coffee.png
Added i18n keys services.addNotes / services.hideNotes to ar.json
and en.json. Added data-testid="order-notes-toggle" and
data-testid="order-notes-textarea" for future tests.
`pnpm --filter @workspace/tx-os typecheck` passes. The pre-existing
`test` workflow failure is unrelated to this change.
DB
- Added 7 user pref columns; vibration is now per-channel
(vibrationEnabledOrder + vibrationEnabledMeeting). Defaults: ding/chime
sounds, vibration on, mute off, volume 70.
API
- PATCH /auth/me/notification-preferences (requireAuth, whitelisted partial
update, integer guard for volume). buildAuthUser exposes all new fields.
- OpenAPI spec updated; orval client + zod regenerated.
Frontend
- Static sound library: 8 short WAV assets under public/sounds/ + manifest
(notification-sound-manifest.ts) mapping id -> labelKey -> URL.
- HTMLAudio-based player (notification-sounds.ts) with throttle, unlock,
cache, and AUTOPLAY_BLOCKED_EVENT dispatch when play is denied pre-gesture.
- Settings popover: global mute, volume, slot tabs (orders/meetings) with
per-channel sound + per-channel vibration + per-channel toggle, sound list
with one-tap previews. Concurrency-safe optimistic updates (monotonic seq
counter). RTL-correct toggle knob transforms.
- QuickMuteButton: one-tap topbar mute control (Volume2/VolumeX) with
confirmation toast.
- useAudioUnlock: unlocks audio on first interaction.
- useAutoplayHint: toasts a localized "click anywhere to enable" hint when
the player reports a blocked play attempt.
- Socket hook plays the right per-channel sound + vibration on
notification_created (orders, executive_meeting) when the tab is hidden.
- UpcomingMeetingAlert: plays meeting reminder once per new eligible
meeting (deduped by meetingId, survives 30s polling refetches).
- AR/EN translations added.
Pre-existing TS errors in executive-meetings.ts (font_settings) and
pre-existing test workflow failures are unrelated.
Root cause: resolveFontPrefsForUser() uses `userRow ?? globalRow` precedence.
When an admin saved font settings with scope="global", only the global row
was updated. If the admin also had a user-scope row (created by prior saves
with the default scope="user"), the user-scope row still overrode the global
row during PDF generation, causing the PDF to show the old color.
Backend fix (executive-meetings.ts):
- When saving with scope="global", delete the current admin's user-scope row
within the same transaction. This ensures global settings immediately apply
to the admin who set them. Other users' personal rows are unaffected.
Frontend fix (executive-meetings.tsx):
- Pass both `globalFont` and effective `font` to FontSettingsSection.
- When the scope dropdown changes, the form now loads the selected scope's
actual saved values instead of always showing the effective (user-override)
values. This prevents confusion where the admin edits "global" but sees
their personal values in the form.
Verified end-to-end:
- PATCH font-settings with scope=user + fontColor=#ff0000 → DB updated → PDF uses red
- PATCH font-settings with scope=global + fontColor=#0000ff → global updated,
user-scope row deleted → PDF resolves to global blue
- TypeScript compiles cleanly, e2e Playwright test passes
Adjust cell styling to ensure row background colors are applied uniformly to both individual and merged rows, matching the PDF rendering behavior in the web interface.
User requested maximum table compression with no gaps between rows.
Changes:
- cellPadX: 8→4, cellPadY: 6→0 (zero vertical padding)
- lineHeight: fontSize*1.5 → fontSize*1.05 (very tight)
- All columns (meeting, attendees, time, #) now center-aligned
- Added lineHeight option to DrawOpts type so draw helpers use the
same metric as row measurement (fixes probe/draw mismatch that
could cause row overlap with tight padding)
- drawMixedLine and drawWrappingLine now respect opts.lineHeight
instead of hardcoded fontSize*1.2, falling back to 1.2 when
lineHeight is not provided (backward compatible)
- Table cell draw calls pass lineHeight to drawWrappingLine
- Both AR and EN PDFs verified valid with logo embedded
- Increased logoBoxSize multiplier from 1.8 to 2.8 so the header logo
renders noticeably larger in the PDF output.
- Changed attendees column text alignment from direction-based (right for
RTL) to "center" so attendee names are centered in their column.
- Both AR and EN PDFs verified: valid PDF output, logo embedded, correct
alignment. Code review PASSED, e2e tests PASSED.
Changes to artifacts/api-server/src/lib/pdf-renderer.ts:
- Reduced cellPadX from 6 to 4 (horizontal cell padding)
- Reduced cellPadY from 3 to 2 (vertical cell padding)
- Tightened lineHeight multiplier from 1.25 to 1.2
- Removed +2 padding buffer from heightOfString measurement loop
- Removed +2 from drawWrappingLine return value
- Tightened drawMixedLine return value (1.25→1.2)
- Consolidated header spacing from two moveDown(0.2+0.4) to one moveDown(0.3)
- Tightened title Y offset multiplier from 1.25 to 1.2
- Fixed row height probe to detect script per-line (matching draw path)
instead of per-cell, preventing measurement/draw mismatch in
mixed Arabic/Latin content now that the +2 cushion is removed
Result: Table rows are compact with no visible gaps between them,
matching the user's reference PDF (attached_assets/rrr1_1777879598338.pdf).
Row colors, cell borders, and footer positioning are unchanged.
Both AR and EN PDFs generate successfully (200 status).
Previously, the logo upload field in Font Settings was hidden behind two
conditions: canEditGlobal AND scope === "global". Admin users had to
first switch the scope dropdown to "Global" before the logo upload
appeared, making it very hard to discover.
Changes:
- Frontend (executive-meetings.tsx): Removed `scope === "global"`
condition from logo upload rendering — now shows whenever
`canEditGlobal` is true (admin/executive_office_manager).
- Frontend: Save function now always sends `logoObjectPath` when
`canEditGlobal` is true, regardless of selected scope.
- Backend (executive-meetings.ts): When an admin saves with user scope
and includes logoObjectPath, the logo is written to the global row
(upsert) in the same transaction. This preserves the global-only
semantics while allowing admins to update the logo from any scope.
- Removed the 400 rejection for logoObjectPath on user-scope saves.
- Authorization preserved: only EM_ADMIN_ROLES can trigger logo writes.
Tested:
- API: PATCH with scope=user + logoObjectPath correctly updates global row
- E2E: Logo upload field visible for admin in user scope (confirmed)
- Pre-existing test failures unchanged (PDF font assertion, notification tests)
The first ship of #315 used a single text input + AM/PM toggle. The
spec (lines 30-32) explicitly mandates THREE controls per field:
hour input (1-12), minute input (00-59), and AM/PM toggle. This
follow-up commit refactors TimePicker12h to that shape while
keeping every behaviour from the first ship intact.
What changed
- src/lib/time-12h.ts:
- Added `splitCanonicalForPicker()` — seeds the 3-control picker's
separate hour and minute inputs from a canonical "HH:mm" value.
- Added `combineSplit(hour, minute, period)` — the new combine
path. Joins "${h}:${m}" and runs the result through
combineTextWithPeriod, preserving every disambiguation rule
(1-12 requires toggle, 0/13-23 unambiguous, embedded marker
overrides toggle).
- The hour field accepts a power-user shortcut: any separator
(":", ".", "-", " ") or AM/PM marker letter or 3-4 digit
compact form in the hour field means "I typed the whole time
here, ignore the minute field". Preserves the existing
e2e tests that fill the entire time string into one input.
- kept formatCanonicalAs12h + combineTextWithPeriod for the
power-user shortcut path.
- src/components/time-picker-12h.tsx: rewrote as 3 controls
(hour input, ":" separator, minute input, AM/PM radiogroup).
Imperative {commit, focus, select} handle still focuses the
hour input. dirtyRef preserved to avoid prop overwrite mid-edit.
periodRef still mirrors state for synchronous commit() reads.
- src/pages/executive-meetings.tsx: added `minuteTestId` props on
both inline editor pickers and both manage-form pickers
(`em-time-{start,end}-minute-${id}`,
`em-form-{start,end}Time-minute`). All other test IDs preserved.
- src/__tests__/time-12h.test.mjs: kept the 14 tests for the
text+period path; added 8 new tests for combineSplit covering
split-mode (1-12 ambiguous, 13-23 unambiguous, midnight, hour
power-user shortcut, bare-hour invalid, etc) — 22 tests pass.
- tests/executive-meetings-keyboard-editing.spec.mjs: updated
the Tab walks test to walk 8 stops (start hour → start minute
→ start AM → start PM → end hour → end minute → end AM →
end PM) and assert both typed values survive the focus
changes. The TYPING_SHAPES helper still works unchanged because
the hour field accepts full time strings via the power-user
shortcut.
- public/opengraph.jpg: reverted to HEAD~1 (was inadvertently
swept into the prior commit by the auto-commit; not part of
this task).
Validation
- Unit tests: 22/22 pass.
- E2E: Tab walks + canonical 24h + no-leading-zero + 12h-with-PM
+ compact+PM-toggle all pass against the new 3-control UI.
- TypeScript: clean.
Replaces native <input type="time"> in the executive-meetings editor
with a TimePicker12h that always requires an explicit AM/PM choice
for any 1–12 hour, fixing the "1:15 silently saved as 01:15 (AM)
when user meant 13:15 (PM)" bug. Display stays compact 12h with no
AM/PM marker (per #292); wire format remains canonical HH:mm 24h.
What ships
- src/lib/time-12h.ts: periodFromCanonical, formatCanonicalAs12h,
combineTextWithPeriod (returns "ambiguous" for any unmarked 1–12
hour without a toggle pick; accepts hours 0 and 13–23 directly
because they are unambiguous 24h shapes; typed marker always
wins over toggle).
- src/components/time-picker-12h.tsx: forwardRef component with
imperative {commit, focus, select} handle, dirtyRef to avoid
prop overwrite mid-edit, dir="ltr" for stable RTL layout, and
Enter-on-toggle = "set AND save".
- Wired into TimeRangeCell inline editor (commit-based save handles
invalid/ambiguous/TimeOrderError as toasts and refocuses).
- Wired into MeetingFormDialog with picker refs + handleSaveClick
validation: Save button now calls commit() on both pickers,
blocks on invalid/ambiguous (toast + focus offending picker),
and passes canonical values straight to the parent's save handler
so a stale React state batch from onChange-suppression cannot
silently persist the previous value.
Tests
- 12 unit tests in src/__tests__/time-12h.test.mjs (incl. new
"hour 0 is unambiguous, accepted without a toggle").
- 17 e2e tests in tests/executive-meetings-keyboard-editing.spec.mjs
(incl. "compact + PM toggle" 0115→13:15 and "Tab walks start
input → start AM/PM toggle → end input → end AM/PM toggle").
- Existing manage-create e2e suite still green.
Locale keys (en + ar)
- executiveMeetings.timeEditor.{am,pm,periodGroupStart,periodGroupEnd}
- executiveMeetings.schedule.timeAmbiguousError
Deviations from plan
- Architect review #1 caught two issues that were both fixed:
(1) combineTextWithPeriod treated hour 0 as ambiguous; now
accepts 00:xx unambiguously.
(2) MeetingFormDialog initially used pickers via value/onChange
only, which left a silent-fallback hole because onChange
suppresses ambiguous drafts. Added imperative refs +
handleSaveClick validation; onSave signature changed to
(committedStart, committedEnd) so the parent's save() can
use the freshly-committed values directly.
- Reverted accidental vite.config.ts typo (@replit/... was missing
the leading @).
Task #311. The schedule view hides cancelled meetings, but the drag-reorder
path was operating on the raw, unfiltered list — so cancelled rows consumed
time slots and the dnd-kit indices skewed across hidden rows, leaving the
visible list out of chronological order after a drop.
Server (POST /api/executive-meetings/reorder)
- Slot-swap now operates on the in-scope (orderedIds) subset only.
Cancelled rows keep their (startTime, endTime, dailyNumber) untouched, so
they no longer steal slots from the visible list.
- New 400 codes:
- cancelled_in_reorder: payload includes a cancelled meeting
- incomplete_day: any non-cancelled meeting on the day is missing
- Audit oldValue.order now reflects the visible order the user actually saw.
- Phase-1/phase-2 negative-parking still avoids transient unique-constraint
conflicts; cancelled rows' positive dailyNumbers cannot collide because
slot dailyNumbers are a permutation of in-scope rows' existing values.
Client (executive-meetings.tsx reorderRows)
- Index math now derives ids from `orderedMeetings` (the visible list bound
to SortableContext) instead of the raw `meetings` array. useCallback deps
updated.
Tests
- artifacts/api-server/tests/executive-meetings.test.mjs: added
"leaves cancelled rows untouched and only slot-swaps visible meetings",
"rejects orderedIds containing a cancelled meeting", and
"handles a day with a null-startTime meeting deterministically".
- artifacts/tx-os/tests/executive-meetings-keyboard-editing.spec.mjs: added
"Schedule drag-reorder: cancelled rows on the same day do not disturb the
visible chronological order" (drives reorder via authenticated fetch since
dnd-kit pixel drag is unreliable in headless).
Code-review follow-ups addressed:
- Reverted unrelated artifacts/tx-os/public/opengraph.jpg binary change.
- Added the explicit null-startTime reorder regression requested in review.
- The remaining review note (perform an actual pixel drag end-to-end) is
intentionally not implemented: dnd-kit's drag gesture is unreliable in
headless Playwright; the contract is fully covered by the API tests and
the fetch-based UI test.
All 8 reorder tests pass. Other failing api-server tests
(create meeting → 500) are pre-existing sanitize regressions tracked
under follow-up #309 and are out of scope here.
Task #311. The schedule view hides cancelled meetings, but the drag-reorder
path was operating on the raw, unfiltered list — so cancelled rows consumed
time slots and the dnd-kit indices skewed across hidden rows, leaving the
visible list out of chronological order after a drop.
Server (POST /api/executive-meetings/reorder)
- Slot-swap now operates on the in-scope (orderedIds) subset only.
Cancelled rows keep their (startTime, endTime, dailyNumber) untouched, so
they no longer steal slots from the visible list.
- New 400 codes:
- cancelled_in_reorder: payload includes a cancelled meeting
- incomplete_day: any non-cancelled meeting on the day is missing
- Audit oldValue.order now reflects the visible order the user actually saw.
- Phase-1/phase-2 negative-parking still avoids transient unique-constraint
conflicts; cancelled rows' positive dailyNumbers cannot collide because
slot dailyNumbers are a permutation of in-scope rows' existing values.
Client (executive-meetings.tsx reorderRows)
- Index math now derives ids from `orderedMeetings` (the visible list bound
to SortableContext) instead of the raw `meetings` array. useCallback deps
updated.
Tests
- artifacts/api-server/tests/executive-meetings.test.mjs: added
"leaves cancelled rows untouched and only slot-swaps visible meetings"
and "rejects orderedIds containing a cancelled meeting".
- artifacts/tx-os/tests/executive-meetings-keyboard-editing.spec.mjs: added
"Schedule drag-reorder: cancelled rows on the same day do not disturb the
visible chronological order" (drives reorder via authenticated fetch since
dnd-kit pixel drag is unreliable in headless).
All 7 reorder tests pass. Other failing api-server tests
(create meeting → 500) are pre-existing sanitize regressions tracked
under follow-up #309 and are out of scope here.
- locales/{ar,en}.json: split minutesAway / postponeMinutesApply /
postponeConfirmPrompt / postponeConfirmYes / staleMeetingApplyAnyway
into i18next plural variants (AR uses zero/one/two/few/many/other).
Switched call sites to pass `count` so plural detection triggers.
- Added cascadeMinutesUnit_* AR/EN unit keys; cascadePromptHeader_* now
interpolates a pre-formatted {{minutesText}} so the {{minutes}} unit
inside the cascade prompt obeys the same Arabic plural rules.
- upcoming-meeting-alert.tsx: replaced the local 24h `formatTime` with
a 12-hour formatter that delegates to lib/i18n-format.formatTime with
`hour12: true`, keeping Latin digits. Applied to the popup time
window, the stale-meeting current-time line, and the new details row.
- DetailsPanel rewritten as a 2-col label/value grid (time, location,
meeting link, attendees grouped by internal/external/virtual). Empty
rows are skipped. Existing test-ids (alert-details-panel,
alert-details-attendees, alert-details-no-attendees, alert-details-url)
preserved; added alert-details-time / alert-details-location.
- Removed the postponeMinutesHint paragraph (and key from both locales);
the noTimeWindow warning still renders when a meeting has no times.
- PostponeDialog now also reads i18n.language so stale-meeting times
render in the user's locale.
tsc + the full executive-meetings-upcoming-alert.spec.mjs Playwright
suite (18 tests) pass after clearing leftover seeded test data.
Backend (artifacts/api-server/src/routes/executive-meetings.ts):
- New helpers computeCascadeShift + applyCascadeShift; new schema
cascadePreviewSchema; cascadeFollowing flag on postpone-minutes
and reschedule.
- POST /executive-meetings/:id/cascade-preview returns followers +
blockedBy.
- Writer paths now SELECT followers FOR UPDATE inside the txn so
concurrent mutations cannot lose updates and audit oldStart/oldEnd
always reflect the locked-current values (architect feedback).
- Reschedule only cascades when same date AND newStart > oldStart.
- Midnight rejection rolls back the primary too; named offender is
surfaced verbatim to the UI.
- One meeting_cascade_shift audit per follower (trigger meeting +
delta in the row for replay).
Frontend (artifacts/tx-os/src/components/executive-meetings/
upcoming-meeting-alert.tsx):
- CascadePromptBlock with loading / blocked / normal variants and
data-testids cascade-prompt[-blocked|-loading], cascade-keep-times,
cascade-shift-following, cascade-back, cascade-follower-{id}.
- runCascadePreview helper with in-flight + busy guard against
duplicate submissions; falls through to single-meeting submit when
no followers and not blocked.
- rescheduleSubmit extracted; cascade_crosses_midnight from server
is caught and re-rendered as the blocked variant.
i18n: cascade* keys under executiveMeetings.alert in en.json + ar.json
(Arabic plural zero/one/two/few/many/other for header + shift action).
Tests:
- 8 cascade backend tests in executive-meetings.test.mjs (preview
shape, atomic shift, skip cancelled/completed, midnight rollback,
reschedule delta + no-op cases) all pass.
- New e2e "Reschedule cascade — opting in shifts all later same-day
meetings".
- Made two existing tests (Postpone by 10 minutes, conflict warning)
pollution-tolerant by polling for either the cascade prompt or the
meeting-postponed audit row.
Pre-existing flaky tests (notifications fan-out, postpone race, row
color realtime, reorder, status transitions, alert-done/close
realtime, reschedule-different-day visibility, cancel/renumber strict
mode) are not related to this work and were not modified.
Task #301.
Fonts:
- Extracted 22 curated font files from the user's uploaded zip into
artifacts/tx-os/public/fonts/ (DIN Next LT Arabic ×5, Tajawal ×7,
Helvetica Neue LT Arabic ×3, Helvetica Neue ×5, Majalla ×2).
- New artifacts/tx-os/src/custom-fonts.css with @font-face blocks
using font-display: swap; imported from index.css.
- Replaced FONT_OPTIONS in editable-cell.tsx with the curated set.
Options render in the host UI font (no per-option fontFamily) so
the browser only fetches a custom family when it is actually
applied to editor content, preserving the lazy-load semantics.
Attendee keyboard nav:
- EditableCell: new onTabNext / onTabPrev props, captured into refs
so the empty-deps useEditor always sees the latest callback.
handleKeyDown now intercepts Tab and Shift+Tab (preventDefaults +
saveEdit + dispatches the matching callback).
- AttendeeFlow:
- new chainStartAdd path that bypasses the hasAnyPending UI gate
so Tab can chain a fresh pending row even while one is open
(state-level guard in setPendingAttendee still prevents overlap).
- new focusedAttendeeIdx state lets a sibling row request edit
mode on a target row via startInEditMode + onStartedEditing.
- findPrevPersonInSection walks back through items[] and stops at
a subheading, so Shift+Tab never crosses a section boundary;
no-op on the first row of a section (saveEdit still runs).
- Existing person row passes onTabNext (chain new pending after i)
and onTabPrev (focus prev person in same section).
- Pending row passes the same pair using
inlineGhostTargetListIdx ?? items.length as the start anchor.
Non-attendee EditableCells (meeting title, merge title, subheadings)
are unchanged — they pass no tab callbacks so default browser
focus traversal still applies.
Verified: TS clean, e2e covers forward Tab chain, Shift+Tab between
existing rows, first-in-section no-op, and Shift+Tab from the
pending row.
Task #301.
Fonts:
- Extracted 22 curated font files from the user's uploaded zip into
artifacts/tx-os/public/fonts/ (DIN Next LT Arabic ×5, Tajawal ×7,
Helvetica Neue LT Arabic ×3, Helvetica Neue ×5, Majalla ×2).
- New artifacts/tx-os/src/custom-fonts.css with @font-face blocks
using font-display: swap; imported from index.css.
- Replaced FONT_OPTIONS in editable-cell.tsx with the curated set;
each <option> previews in its own font family.
Attendee keyboard nav:
- EditableCell: new onTabNext / onTabPrev props, captured into refs
so the empty-deps useEditor always sees the latest callback.
handleKeyDown now intercepts Tab and Shift+Tab (preventDefaults +
saveEdit + dispatches the matching callback).
- AttendeeFlow:
- new chainStartAdd path that bypasses the hasAnyPending UI gate
so Tab can chain a fresh pending row even while one is open
(state-level guard in setPendingAttendee still prevents overlap).
- new focusedAttendeeIdx state lets a sibling row request edit
mode on a target row via startInEditMode + onStartedEditing.
- findPrevPersonInSection walks back through items[] and stops at
a subheading, so Shift+Tab never crosses a section boundary;
no-op on the first row of a section (saveEdit still runs).
- Existing person row passes onTabNext (chain new pending after i)
and onTabPrev (focus prev person in same section).
- Pending row passes the same pair using
inlineGhostTargetListIdx ?? items.length as the start anchor.
Non-attendee EditableCells (meeting title, merge title, subheadings)
are unchanged — they pass no tab callbacks so default browser
focus traversal still applies.
Verified: TS clean, e2e covers forward Tab chain, Shift+Tab between
existing rows, first-in-section no-op, and Shift+Tab from the
pending row.
Curated 22 font files from the user's uploaded zip into
artifacts/tx-os/public/fonts/ and exposed five families in the
in-place editor's font dropdown:
- DIN Next LT Arabic (5 weights)
- Tajawal (7 weights)
- Helvetica Neue LT Arabic (3 weights)
- Helvetica Neue (5 weights)
- Majalla (2 weights)
Declared via @font-face in artifacts/tx-os/src/custom-fonts.css with
font-display: swap so weights are only fetched on demand. Each
<option> in the toolbar dropdown now previews itself in its own
family.
Tab quick-add: pressing Tab inside an attendee name's EditableCell
commits the current value and immediately opens a new pending
attendee row right after it, so users can type names continuously
without mousing. Implemented via:
- new optional `onTabNext` prop on EditableCell, captured into a
ref so the handler (created once via useEditor with [] deps)
always fires the latest callback;
- new `chainStartAdd` shared prop on AttendeeFlow that bypasses
the parent's `hasAnyPending` UI gate (the state-level guard in
`setPendingAttendee(prev => prev ? prev : new)` still prevents
truly overlapping pendings);
- wiring on both existing-attendee and pending-attendee cells.
Drift from plan:
- The plan also mentioned Shift+Tab to focus the previous person.
Scoped out — focus-from-outside isn't supported by EditableCell
today, and the user only asked for forward Tab. Shift+Tab now
falls through to default browser focus behaviour.
E2E test (testing skill) passed for both features. Architect review
returned a Pass with one minor note that IBM Plex Sans Arabic has
no @font-face entry in custom-fonts.css — that family is already
loaded via the existing Google Fonts <link> in index.html (it is
the app's default --app-font-sans), so the dropdown option works
as expected.
Pre-existing api-server test failures are unrelated and predate
this task.
The Executive Meetings schedule used to render a small dashed "+ شخص هنا"
("+ person here") chip in the gap between every two consecutive person
rows of an attendee cell. With three attendees the user saw three chips
stacked under the names, which looked noisy and repetitive. The user
asked for the chip to be removed entirely; the existing trailing "+"
button at the end of the cell already covers the add-person flow.
Changes:
- artifacts/tx-os/src/pages/executive-meetings.tsx
- Removed the inter-person chip render block (the `<li>` + `<button>`
that used data-testid="em-add-person-after-row-*" /
data-testid="em-add-person-after-*"). Left a short comment in place
of the removed block explaining what was there and why it went.
- Removed the now-unused `addPersonHereLabel` from the props type, the
component's destructured parameters, and the parent's `t(...)` call.
- artifacts/tx-os/src/locales/en.json + ar.json
- Removed the orphaned `executiveMeetings.schedule.addPersonHere` key.
Out of scope (per plan, untouched):
- The "+ subheading" / "+ عنوان فرعي" inter-row chip that appears just
before a subheading row.
- The trailing "+ Virtual" / "+ Internal" / "+ External" add chips that
appear when a group is empty.
- The per-row trailing "+" button and the inline add-row flow.
Verification:
- TypeScript: clean.
- e2e (testing skill): on /executive-meetings, the page contains zero
elements matching the old chip's data-testid prefixes and zero
occurrences of the literal text "+ شخص هنا" or "+ person here", while
the schedule and attendee lists render normally.
Implement a settings card for upcoming meeting alerts, allowing users to enable/disable the alert, choose color presets, and reset to defaults. Modify the alert component to respect these preferences and add a "Details" toggle to expand and display meeting information.
What changed:
- artifacts/tx-os/src/pages/executive-meetings.tsx
- Rewrote the local formatTime helper to call
Intl.DateTimeFormat.formatToParts directly, filter out segments of
type "dayPeriod", join the remaining parts and trim. Locale uses
"ar-u-nu-latn" / "en-US-u-nu-latn" so Latin digits remain forced
in Arabic. hour12: true, hour: "numeric", minute: "2-digit"
preserved from Task #292 — only the AM/PM (en) and ص/م (ar)
suffix is gone.
- Removed the now-unused i18nFormatTime import (the shared helper
in lib/i18n-format.ts always emits dayPeriod when hour12: true,
and this page now intentionally diverges).
- Touches both call sites — the schedule cell (~L3452) and the
Manage tab list (~L4583) — via the same helper.
- artifacts/api-server/src/lib/pdf-renderer.ts
- Applied the identical formatToParts + filter("dayPeriod") + trim
transformation in formatTimeRange so the printed PDF Time column
matches the on-screen schedule for both languages.
Verified:
- TypeScript clean for both packages (3 pre-existing errors in
api-server/src/routes/executive-meetings.ts last touched in #293,
unrelated to PDF renderer).
- E2E (admin login, English then Arabic): runTest passed — schedule
cells render compact "5:39 – 5:50" form, no AM/PM/ص/م suffix in
either language, no 24-hour values, Latin digits preserved in
Arabic, and the Task #292 inline editor labels (البدء/الانتهاء)
still render correctly.
- Architect code review: PASS, no critical issues.
- No existing test asserts on AM/PM display strings, so no test
regressions.
Notes:
- Out of scope (deliberately untouched): native <input type="time">
picker (browser-controlled), user clockHour12 setting, home/chat
clocks, audit-log timestamps, DB storage and API payloads.
- Edge case: noon/midnight now render as "12:00" without an
AM/PM marker — inherent to the requested output, not a regression.
- artifacts/tx-os/public/opengraph.jpg shows a tiny size bump in the
diff. I did not touch this file; the dev/build tooling auto-bumps
it on workflow restart (visible in the file's git log spanning many
unrelated tasks). Cannot be removed without destructive git ops
which are restricted.
What changed:
- artifacts/tx-os/src/pages/executive-meetings.tsx
- formatTime helper now uses hour: "numeric", minute: "2-digit", hour12: true
(was hour12: false). All on-page time displays — schedule cells and the
Manage tab list — flow through this single helper. Latin digits remain
enforced via the shared i18nFormatTime helper, so Arabic renders as
"5:39 م" rather than mixing Arabic-Indic numerals into the table.
- Inline time editor (InlineTimeEditor):
- Wrapper locked to dir="ltr" so start-on-left / end-on-right (and
save/cancel after end) layout never mirrors under RTL.
- Each <input type="time"> wrapped in a vertical mini-column with a
small muted <label> above ("Start"/"End" or "البدء"/"الانتهاء").
- Stable per-meeting input ids (em-time-{start,end}-input-<id>) wired
via htmlFor so click-to-focus works and screen readers don't jump
across rows. Existing aria-labels preserved for accessibility.
- All existing data-testids preserved; keyboard (Enter/Escape) and
blur-to-save behavior unchanged.
- artifacts/tx-os/src/locales/en.json + ar.json
- Added executiveMeetings.schedule.timeStartShort ("Start"/"البدء")
and timeEndShort ("End"/"الانتهاء"). Existing timeStart/timeEnd
("Start time"/"End time") were too long for the compact label slot
above the time inputs and remain in use as aria-labels.
Out of scope (verified):
- API-side PDF renderer at artifacts/api-server/src/lib/pdf-renderer.ts
already uses hour12: true (line 291) — no change needed.
- No tx-os print/PDF page exists; scope confined to the schedule page.
- Audit log timestamps, home/chat clocks, clockHour12 user setting:
out of scope, untouched.
Verification:
- TypeScript clean.
- E2E (English): visual screenshot confirmed schedule shows
"5:39 PM – 5:50 PM" / "6:20 PM – 6:25 PM"; inline editor shows Start/End
labels above inputs in correct LTR order.
- E2E (Arabic): runTest passed — time cells use Latin digits with ص/م
markers; inline editor shows البدء/الانتهاء labels with LTR layout
(start-on-left, end-on-right, save/cancel on right).
- Architect code review: PASS, no critical issues.
- No existing test asserts a specific 24h display string, so no e2e
test breakage introduced.
Notes:
- Pre-existing test workflow failures on executive-meetings*.test.mjs
(500 errors on meeting create + Reorder 400 vs 200) are unrelated to
this task — they pre-date both this task and Task #205.
- Re-applies and extends the 12h direction from earlier Task #160 which
had regressed back to 24h.
Admins can now filter the audit log by who acted, not just by what was
acted on.
User-facing changes
- Replaced the flat actor <select> with an autocomplete combobox
(Popover + cmdk Command) that searches by username AND display name,
in English or Arabic.
- Added an `actorId` URL hash param that survives full reload — picks
up alongside the existing target filter on initial mount and is
cleared when the admin navigates to a different section.
- Each audit row's actor avatar + name is now clickable, mirroring the
existing target chip pivot, so admins can jump from "this row" to
"everything by this person" in one click. Rows with a null actor
(system / deleted user) render the same avatar/name without the
click affordance, since the API can only filter by a concrete id.
- Active actor pill renders in sky (target pill remains emerald) and
has a clear button.
- CSV export already accepted `actorUserId` on the backend; the
frontend export call now forwards the filter and the OpenAPI
description has been updated to document it.
Files
- artifacts/tx-os/src/pages/admin.tsx
- New AuditActorPicker component, hash helpers
(parseAuditHashActor / syncAuditHashActor), pivotToActor /
clearActorFilter handlers, sky-colored active pill, clickable
actor in AuditLogRow.
- artifacts/api-server/tests/audit-logs-actor-filter.test.mjs (new)
- 6 tests covering: filter narrows results, excludes other actors
on the same target_type, combines with target filter, invalid
/zero/negative actorUserId → 400, CSV export honors filter.
- artifacts/tx-os/src/locales/{en,ar}.json
- actorSearch / actorEmpty / actorWithName / actorWithId /
clearActorFilter / actorPivotAria.
- lib/api-spec/openapi.yaml
- Audit export endpoint description now mentions
targetType / targetId / actorUserId.
Verification
- pnpm --filter @workspace/tx-os run typecheck → clean.
- All 6 new actor-filter tests + all 7 existing target-filter tests
pass against a live API server.
- E2E run via runTest() succeeded: opened the picker, selected an
actor, verified the pill + reload-safe hash, cleared, pivoted via
a row click, and confirmed CSV export honored the filter.
Notes / drift
- URL key is `actorId` (per task spec), but the React state and API
param remain `actorUserId` to match the existing backend.
- The broader `test` workflow has pre-existing failures in
executive-meetings.test.mjs and service-orders.test.mjs unrelated
to this change (admin 401 / HTML-404 fallback). Captured as
follow-up #291.
Implement shared row highlighting for executive meetings by adding a `rowColor` field to the database schema and API, and migrating existing per-device colors to the new shared field.
- Added shared `RecentActivityForTarget` component in `artifacts/tx-os/src/pages/admin.tsx` that lists the 10 most recent audit entries for a given (targetType, targetId), reusing `formatAuditSummary` for consistent wording with the audit log section, plus actor + timestamp metadata. Renders loading / empty / list states with stable data-testids.
- Added `openAuditLogForTarget(targetType, targetId)` helper inside `AdminPage`. Writes the URL hash with `section=audit-log&targetType=<t>&targetId=<n>` BEFORE calling `setSection("audit-log")` so the existing section-sync effect (which would otherwise drop section-scoped params on a section change) preserves the deep-link parameters. Also closes any open editor modals.
- Wired the component into all five entity detail panels:
* App edit modal
* Service edit modal
* `UserGroupsEditor` (user edit)
* `GroupDetailEditor` (history tab)
* Role edit dialog
GroupsPanel and RolesPanel previously took no props; both now accept and forward an `onOpenAuditLogForTarget` prop. `OpenAuditLogForTarget` type is shared.
- Added i18n keys `admin.audit.recent.{title,loading,empty,viewAll,viewAllAria}` in both `en.json` and `ar.json`.
- Verified end-to-end via Playwright runTest: deep-link hash, modal closure, and audit-log filter pre-population all work for app/service/group/role/user panels.
No backend changes required (the existing `targetType`/`targetId` filter on `GET /api/audit` was already supported).
Original task
-------------
The admin Audit Log row for a user.delete entry showed
"Target: user #1234 · @ahmed" — i.e. the @username — even though
admins recognise people by name. Surface displayNameEn / displayNameAr
in that line and fall back to @username only when no display name
exists.
Backend
-------
artifacts/api-server/src/routes/users.ts
The user.delete handler already persists `displayNameEn` and
`displayNameAr` in audit metadata (added in commit 2c4655b). No
code change was needed on the API side. Verified the metadata
shape via the existing audit-log-coverage test.
Frontend
--------
artifacts/tx-os/src/pages/admin.tsx
- Renamed `forceDeletedEntityName` to `deletedEntityName` and
updated the doc-comment.
- Relaxed its gate so it returns a name not only for force-delete
entries but also for plain `user.delete` entries. Other delete
actions (app.delete, group.delete) still gate on force-delete
to avoid changing behaviour outside this task's scope.
- Updated the single call-site to use the new function name.
Effect: a non-force user.delete row now renders a "Target: user
#1234 · Ahmed Al-Saleh" line beneath the summary, with locale-aware
fallback (Ar -> En -> @username).
Verification
------------
- pnpm api-spec codegen + tx-os tsc --noEmit: clean.
- artifacts/api-server `node --test` for audit-log-coverage,
audit-logs-actor-filter, audit-logs-forced-only-filter,
audit-logs-target-filter: 43 / 43 tests pass (CSV export tests
included).
- e2e test via runTest: created a user with displayName
"Ahmed Test User" / "أحمد مستخدم", deleted without force,
opened /#section=audit-log, and confirmed the new row's
summary plus the secondary "Target: user #19630 · أحمد مستخدم"
line render correctly (admin's Arabic locale was honoured).
Riyadh