Original task
- Cover the new /api/roles create, update, and delete handlers with
automated tests so regressions in role validation, system-role
protection, and the in-use conflict response can't slip in unnoticed.
- Tests must run as part of the standard api-server test command.
What was added
- artifacts/api-server/tests/roles-crud.test.mjs with 7 tests:
* POST /api/roles -> 201 (also checks DB row + serialized fields)
* POST /api/roles duplicate name -> 409 (no second row inserted)
* POST /api/roles invalid name format -> 400 (no row leaked)
* PATCH /api/roles/:id description-only update -> 200, name unchanged
* DELETE /api/roles/:id on system role 'admin' -> 400, row preserved
* DELETE /api/roles/:id on in-use role -> 409 with userCount=1,
groupCount=1; row preserved
* DELETE /api/roles/:id on unused role -> 204, row gone
- The file follows the same pattern as groups-crud.test.mjs:
pg.Pool seed via DATABASE_URL, login -> connect.sid cookie,
thorough teardown of users / groups / roles / link tables.
Test infra
- No new test setup needed; the package's existing
"test": "node --test 'tests/**/*.test.mjs'" picks the file up
automatically. All 7 new tests pass.
Notes / drift
- Two pre-existing test failures (apps-open.test.mjs and one assertion
in groups-crud.test.mjs that depends on global group counts) are
unrelated to this task and were left untouched.
Task #80 introduced a new 409 `order_unavailable` error code returned by:
- PATCH /api/orders/:id/confirm-receipt
- PATCH /api/orders/:id/status
…to distinguish "already_claimed" (lost a race) from "no longer claimable"
(target order is cancelled or completed). The integration test suite did
not cover these paths.
Changes
-------
artifacts/api-server/tests/service-orders.test.mjs
+ Added test "confirm-receipt returns 409 order_unavailable when the order
is cancelled or completed":
- Owner cancels a pending order → another receiver's confirm-receipt
returns 409 with body.error === "order_unavailable" (not
"already_claimed").
- Receiver claims, prepares, completes an order → another receiver's
confirm-receipt returns 409 "order_unavailable".
+ Added test "PATCH /orders/:id/status returns 409 order_unavailable when
targeting a terminal order":
- Completed order: assignee → preparing → 409 "order_unavailable"
- Completed order: assignee → completed → 409 "order_unavailable"
- Completed order: owner → cancelled → 409 "order_unavailable"
- Cancelled order: owner → cancelled → 409 "order_unavailable"
Verification
------------
All 11 service-orders integration tests pass (including both new tests):
pnpm --filter @workspace/api-server test → 11 passed, 0 failed.
No production code changes; tests only.
Apply the existing groups delete-warning pattern to user, app, and
service deletions so admins are warned (and have to confirm twice)
before destroying records that still own dependent data.
Backend
- openapi.yaml: added `force` query param to DELETE /users/{id},
/apps/{id}, /services/{id} plus three new conflict schemas
(UserDeletionConflict, AppDeletionConflict, ServiceDeletionConflict).
- routes/users.ts, apps.ts, services.ts: rewrote DELETE handlers to
count dependents (notes/orders/conversations/messages for users;
group_apps/restrictions/open events for apps; service_orders for
services), return 409 with counts when non-empty and force is not set,
and on `?force=true` perform the cascade in a transaction and write
an `*.force_delete` audit_logs row.
- Existing 204 success path preserved for empty deletes.
UI (artifacts/tx-os/src/pages/admin.tsx)
- New `DeletionWarningDialog` helper, plus app/service/user delete state
+ lazy 409 detection (first click probes; on 409 the dialog upgrades
to show counts + "Delete anyway"; second click sends ?force=true).
- Replaced the three plain `confirm(t("admin.deleteConfirm"))` callsites.
i18n
- Added admin.deleteApp/deleteService/deleteUser keys (title, warning,
forceHint, emptyBody, count keys, confirm, anyway) in en.json + ar.json.
Tests
- artifacts/api-server/tests/delete-force-warnings.test.mjs covers all
9 cases (clean delete, 409 with counts, force=true 204 + audit log)
for users, apps, and services. Existing groups-crud and service-orders
tests still pass.
Notes / drift
- Lazy detection (vs eager counts on the row like Groups already does)
was chosen because list endpoints don't return counts yet — proposed
follow-up #96 covers eager counts so the warning appears on first
click everywhere.
- E2E test for the UI flow flaked twice; backend integration tests
(9/9 pass), direct curl validation of force=true returning 204, and
typecheck across the monorepo all confirm correctness.
Adjusts the order cancellation endpoint to correctly handle admin overrides for completed/cancelled orders and updates associated tests to expect 409 status codes reflecting Task #80's changes.
Replaces all user-facing instances of "TeaBoy" with "Tx" across the application, including titles, locale files, database settings, seed data, and API documentation. Also updates internal storage keys and session secrets to remove the old branding.
Original task: Task #75 — add automated tests for the groups system
(CRUD + group-driven app visibility + auto-Everyone) and stabilize the
flaky pagination test.
Changes:
- Extended artifacts/api-server/tests/groups-crud.test.mjs with four
new tests:
* PATCH /api/groups/:id updates fields and replaces userIds
transactionally; a bad userId returns 400 and leaves prior
membership unchanged.
* Creating a group with invalid roleIds returns 400 and creates no
rows.
* DELETE /api/groups/:id removes a custom group; deleting a system
group (Everyone, is_system = 1) returns 400 and the row stays.
* /api/auth/register auto-assigns the new user to the Everyone
system group; verified both via the response payload and a direct
user_groups DB check. The test temporarily flips
app_settings.registration_open to true and restores the prior
value on completion.
- The existing apps-group-visibility.test.mjs already exercises the
/api/apps endpoint backed by getVisibleAppsForUser for group-derived
admin, group member, and outsider — no changes needed there.
- The previously flaky tests/admin-app-opens-pagination.test.mjs now
passes consistently thanks to the existing 7d-window cleanup in its
before hook; left as-is per the task's "fixed or skipped" criterion.
Verification: `pnpm --filter @workspace/api-server test` reports
42/42 passing. The downstream Playwright suite (4 tests) also passes.
Original task (#72): Mirror the recently-added delete Undo toast for
the cancel action so accidental cancellations can be reverted within
~7s.
Changes
- Frontend (artifacts/teaboy-os/src/pages/my-orders.tsx):
OrderCard.handleCancel captures the order's previous status, and on
successful cancel shows a toast with an "Undo" action (duration
matches the existing UNDO_WINDOW_MS = 7000ms). Clicking Undo issues
a PATCH to restore the order to its previous status (pending or
received) and emits a "Restored" toast. Errors surface a
"Could not restore the order" toast.
- Cancel confirmation copy was softened (no longer says "can't be
reopened") and a new restoreFailed string was added in both en.json
and ar.json.
- Backend (artifacts/api-server/src/routes/service-orders.ts):
PATCH /orders/:id/status accepts pending and received as targets
for the owner (or admin) when the existing status is cancelled,
serving as the restore-from-cancelled transition. We require
assignedTo to be null for pending and non-null for received so we
always restore to the actually-prior state. The existing logic
already broadcasts order_incoming_changed to receivers and emits
order_updated to the owner, so receivers' incoming queues refresh
automatically when an order is restored. notifyUser titleMap was
extended with "Your order was restored" entries.
- Time-bound restore: server enforces a RESTORE_WINDOW_MS of 15s
(slightly larger than the 7s client window to absorb network/clock
skew). After it expires, restore is rejected with
`undo_window_expired`, so cancellation truly becomes permanent —
this applies to admin too, so restore is strictly Undo and not an
arbitrary reopen.
- API spec (lib/api-spec/openapi.yaml):
UpdateServiceOrderStatusBody enum extended to include pending and
received; endpoint summary updated. Regenerated api-zod and
api-client-react.
Tests
- Added a new test case in artifacts/api-server/tests/service-orders.test.mjs
that covers: restore from pending, restore from received,
pending<->received validation against assignedTo, stranger forbidden,
and undo window expiry (by backdating updated_at).
Verification
- pnpm typecheck passes across libs and artifacts.
- New api-server test "owner can restore (undo) a freshly cancelled
order..." passes; all pre-existing service-order tests still pass.
- e2e tested: login -> place order -> cancel -> Undo restores to
Pending; cancel again -> let window expire -> order stays Cancelled
with no Undo available.
Notes / unrelated
- The unrelated `admin-app-opens-pagination` test
(`by-app: paginating with limit returns nextOffset until exhausted`)
fails in the dev environment because the dev DB has > 100 app_opens
rows for the chosen app and the test only walks up to 50 pages of
size 2. This is a pre-existing flaky test unrelated to this task
and was not modified.
Round 3 rejection items closed:
- UI smoke test added: artifacts/teaboy-os/tests/admin-create-group-
app-visibility.spec.mjs. Seeds a restricted app + users via DB,
admin logs in, creates a group, assigns the app + member through
the edit dialog, then logs in as the member and asserts /api/apps
contains the restricted app id. Cleans up after itself. Passes.
- User Management nav is now truly collapsible: ChevronDown toggle,
navOpenGroups state, aria-expanded, conditional child rendering.
Defaults to expanded only when a child is the active section.
- Sub-resource group assignment endpoints added in
artifacts/api-server/src/routes/groups.ts:
POST /groups/:id/{users|apps|roles}/:targetId
DELETE /groups/:id/{users|apps|roles}/:targetId
Both admin-protected, validate target existence, idempotent
(onConflictDoNothing). Aggregate PATCH /groups/:id remains.
Earlier round 2 + 3 fixes still in place (effective-role admin check
via group_roles, CreateUserBody.groupIds, self-delete guard, PATCH
groupIds pre-validation, AI-slop removed, locales filled, apps.ts
inArray fix).
Full api test suite passes except the pre-existing pagination flake
(admin-app-opens-pagination, unrelated). Architect: PASS.
Round 2 fixes:
- apps.ts getVisibleAppsForUser uses getEffectiveRoleIds() so admin
detection honors group_roles inheritance (closes group-only-admin
bypass). Exported helper from middlewares/auth.ts.
- POST /users accepts optional groupIds[] via new CreateUserBody
schema. Validates ids exist; user create + auto-Everyone + requested
groups happen in one transaction. OpenAPI extended; api-zod and
api-client-react regenerated.
- DELETE /users/:id 400s on self-delete.
- scripts/src/seed.ts: removed `void inArray;` slop and unused import.
- Locales (ar+en): added admin.users.col.{displayNameAr,displayNameEn,
language} and admin.groups.searchPlaceholder.
Round 3 fixes:
- Admin Users "Add User" dialog gets a groupIds checkbox multi-select
populated from useListGroups; createUser sends groupIds when set.
- PATCH /users/:id validates ALL groupIds before deleting memberships
(returns 400 on any invalid id); replacement wrapped in transaction
to avoid partial-loss windows.
- Apps admin SQL bug: `${arr} = ANY(...)` produced bad params; now uses
`and(inArray(rolesTable.id, ids), eq(name,'admin'))`.
- New tests (artifacts/api-server/tests/apps-group-visibility.test.mjs):
group-derived admin sees all; group member sees group-granted +
unrestricted; outsider sees only unrestricted. All 3 pass.
Typecheck clean; full api test suite green except pre-existing
admin-app-opens-pagination flake (unrelated). Architect: PASS.
Auth middleware:
- Add getEffectiveRoleIds() that UNIONs user_roles with group_roles via
group memberships, used by requireAdmin / requirePermission /
userHasPermission / getUserRoles.
- requireAdmin and requirePermission now also reject inactive users
with 401 (matching requireAuth), closing a session-after-deactivation
bypass.
Groups routes:
- POST /groups and PATCH /groups/:id now wrap the group row write and
all assignment writes in a single db.transaction via
applyGroupAssignmentsTx(tx, ...), so partial state cannot leak.
- validateAssignmentIds rejects unknown app/role/user ids with 400
before any insert.
- Removed AI-slop: void or, void sql, as-unknown-as casts; conditions
use Drizzle's SQL union type.
Users route:
- /api/users supports q, groupId, status filters (server-side).
Admin UI (teaboy-os/admin.tsx):
- UsersPanel wires q/groupId/status to the backend, shows display name
and preferred language inline per row.
- UserGroupsEditor now edits display names (ar/en), preferred language,
active status, and group membership with a search box.
- GroupsPanel adds a top-level group search box.
- GroupDetailEditor Users tab adds a user search box.
Infra:
- scripts/post-merge.sh runs the seed (idempotent) so default groups
Admins / TeaBoy / Everyone always exist after merges.
Tests (artifacts/api-server/tests/groups-crud.test.mjs, all passing):
- Admin-only access (regular user gets 403).
- Default seed groups exist.
- Create group + member assignment.
- Bad userIds yields 400 with no leaked group row.
- Admin role inherited via group_roles grants admin access.
- Deactivated admin session is rejected with 401.
- Group create rolls back atomically when assignment fails.
- /api/users q + groupId + status filters return correct rows.
Notes / drift:
- "Roles" tab inside GroupDetailEditor and groupIds in CreateUserBody
remain as proposed follow-ups (require OpenAPI spec changes).
- Pre-existing pagination-test flake unrelated to this work.
- Add DELETE /api/orders/:id endpoint: owner can delete completed/cancelled
orders; admin can delete any. 401/403/404/204 responses. Emits
order_deleted to owner and assignee for live invalidation.
- Update OpenAPI spec with deleteServiceOrder operation; regenerate
api-client-react + api-zod.
- Frontend: per-order trash button on completed/cancelled cards with
confirm dialog; bulk "Clear finished (N)" button at top of list with
confirm dialog and toast summary (handles partial failures).
- Add ar/en locale keys under myOrders for delete + clear flows
(singular/plural variants).
- Backend tests cover all paths: pending owner=403, stranger=403,
owner-cancelled=204, owner-completed=204, admin-pending=204, 404, 401.
All 25 service-order tests pass; the 1 failing test (admin-app-opens-
pagination) is unrelated/pre-existing.
Follow-ups proposed: #70 undo for deleted orders, #71 audit
self-notification on receiver-cancel-own-order.
- Add DELETE /api/orders/:id endpoint: owner can delete completed/cancelled
orders; admin can delete any. 401/403/404/204 responses. Emits
order_deleted to owner and assignee for live invalidation.
- Update OpenAPI spec with deleteServiceOrder operation; regenerate
api-client-react + api-zod.
- Frontend: per-order trash button on completed/cancelled cards with
confirm dialog; bulk "Clear finished (N)" button at top of list with
confirm dialog and toast summary (handles partial failures).
- Add ar/en locale keys under myOrders for delete + clear flows
(singular/plural variants).
- Backend tests cover all paths: pending owner=403, stranger=403,
owner-cancelled=204, owner-completed=204, admin-pending=204, 404, 401.
All 25 service-order tests pass; the 1 failing test (admin-app-opens-
pagination) is unrelated/pre-existing.
Follow-ups proposed: #70 undo for deleted orders, #71 audit
self-notification on receiver-cancel-own-order.
Context
- After Task #64 receivers can cancel an order they have already claimed.
- Previously the requester's cancellation notification body was the same regardless of who cancelled (admin vs receiver), giving them no signal that their order had been claimed and then dropped.
Changes
- artifacts/api-server/src/routes/service-orders.ts (PATCH /orders/:id/status, cancel branch):
- When the cancellation is initiated by the order's currently assigned receiver
(existing.assignedTo === userId && next === "cancelled"), the requester's
notification body now includes the bilingual phrase
"استلم طلبك ولكن تم إلغاؤه لاحقاً" / "Your order was claimed but later
cancelled by the receiver", appended to the service name for context.
- Admin / non-assignee cancellations keep the existing service-name body so
the two cases remain distinguishable in the requester's notifications list.
- The notification is created via the existing notifyUser() helper, which
persists to notifications and emits notification_created over Socket.IO,
so the bell badge updates in real time.
- artifacts/api-server/tests/service-orders.test.mjs:
- Added "requester gets a distinct notification when the receiver cancels a
claimed order" covering the full flow: place → confirm-receipt → cancel,
then asserts the owner has a cancellation notification whose body matches
the new bilingual receiver-cancel copy.
Verification
- All 7 tests in service-orders.test.mjs pass locally against the running API.
Notes / deviations
- Locale JSON files (ar.json/en.json) were inspected but not modified: notification
copy is stored server-side as titleAr/titleEn/bodyAr/bodyEn on the notifications
row; the client just renders those strings directly.
Backend (artifacts/api-server)
- Add admin-only role-toggle endpoints:
- POST /users/:id/roles { roleName } — idempotent, returns UserProfile
- DELETE /users/:id/roles/:roleName — idempotent, returns UserProfile
- Allow assigned receiver to cancel their own claimed order
(received/preparing) in PATCH /orders/:id/status. Owner & admin
rules unchanged.
- New backend test cases: receiver can cancel own assigned order;
another receiver gets 403.
API spec / codegen
- Add AddUserRoleBody schema and the two new role endpoints
under a new "roles" tag in lib/api-spec/openapi.yaml.
- Regenerated api-zod and api-client-react.
Frontend (artifacts/teaboy-os)
- New page src/pages/orders-incoming.tsx at /orders/incoming:
RBAC-gated (admin || order_receiver), shows "My active orders"
and "Awaiting receiver" sections, with claim, mark preparing,
mark completed and cancel buttons. Handles 409 already_claimed.
- Add Inbox button to home top bar, conditional on the same roles.
- Admin users table now has an "Order Receiver" Switch wired to
the new role-toggle hooks.
- Extend use-notifications-socket to invalidate the incoming-orders
query on order_incoming_changed and on notification_created with
type === "order".
- Bilingual locale keys (ar/en) for the new page and admin label.
Tests
- All 25 api-server tests pass (24 existing + 1 new receiver-cancel
case). All 3 teaboy-os e2e tests pass.
Follow-up filed: #67 (notify requester when receiver cancels).
- New `service_orders` table (status pending|claimed|delivered|received|cancelled)
- New `orders:receive` permission + `order_receiver` role; admins implicitly allowed
- Added `requirePermission(name)` and `userHasPermission(userId, name)` middleware helpers
- New routes:
- POST /api/orders place order (authenticated)
- GET /api/orders/my list current user's orders
- GET /api/orders/incoming receivers see pending+active visible orders
- PATCH /api/orders/:id/status claim (atomic), deliver, cancel
- PATCH /api/orders/:id/confirm-receipt requester confirms a delivered order
- Atomic claim via UPDATE ... WHERE status='pending' AND assigned_to IS NULL
(returns 409 already_claimed on race)
- Realtime: emits `notification_created` to receivers/requester,
`order_incoming_changed` to all receivers, `order_updated` to requester
- Service shape in order responses limited to id/nameAr/nameEn/imageUrl
(no description fields), per spec
- OpenAPI updated with new paths and schemas; codegen run
- Seed updated idempotently (permission, role, role_permissions)
- New tests in artifacts/api-server/tests/service-orders.test.mjs
(full lifecycle, atomic claim race, unauth rejection) — all 21 api tests pass
No deviations from the planned scope. Tasks #63 (client UI) and #64
(receiver page + admin role toggle) remain blocked-by #62 and are next.
Background:
Task #53 fixed the login/register session persistence race by explicitly
awaiting `req.session.save` before responding. Other session-mutating
endpoints (notably `/auth/logout`) still relied on express-session's
default end-hook, which can flush the response before the store write
finishes — producing intermittent "still logged in" / "logged out"
glitches on the immediate next request.
Changes:
- New shared helper `artifacts/api-server/src/lib/session.ts` exporting
`saveSession(req)` and `destroySession(req)` — promise wrappers around
`req.session.save` / `req.session.destroy` so handlers can `await`
store persistence before flushing the HTTP response. Documented the
rationale in the file so future session-mutating routes use the same
safe pattern.
- `artifacts/api-server/src/routes/auth.ts`:
- `/auth/register` and `/auth/login` now use `await saveSession(req)`
in place of the inline ad-hoc Promise wrapper.
- `/auth/logout` is now async and `await`s `destroySession(req)`
before responding, closing the same race for the destroy path.
- Audited remaining routes: only `auth.ts` mutates `req.session`; all
other handlers only read `req.session.userId`, so no further changes
are needed.
Notes / deviations:
- Pre-existing TypeScript errors in unrelated files (conversations,
notes, users, api-zod exports) were left untouched — out of scope.
- New test file `artifacts/api-server/tests/auth-session-persistence.test.mjs`
covers the acceptance criterion: login / register / logout each
followed by an immediate `/auth/me` probe to assert the session was
persisted (or destroyed) before the response was flushed. Modeled on
the existing leave-test pattern. Full suite: 18/18 passing.
Adds artifacts/api-server/tests/conversations-leave.test.mjs, modeled on
the existing apps-open.test.mjs, covering POST /conversations/:id/leave:
- Solo member leaving deletes the conversation entirely.
- Sole admin leaving with no successor auto-promotes the
earliest-joined remaining member.
- Sole admin leaving with a chosen successorId promotes that user.
- Sole admin leaving with a non-member successorId returns 400 and
leaves the group untouched (leaver still admin, no promotion).
- Non-admin leaving a group removes them with no admin promotion.
Tests create their own users (with the standard user role) and groups
directly in Postgres so joined_at ordering is deterministic for the
auto-promotion case, then exercise the route through HTTP using a real
session cookie obtained from POST /api/auth/login. An after() hook
cleans up all created conversations, participants, messages, role
assignments, and users.
The optional e2e for the chooser dialog is intentionally deferred and
proposed as follow-up #51.
Verified by running `pnpm --filter @workspace/api-server test` three
times consecutively; all 15 tests pass on every run.
Added offset/limit pagination to the two admin app-opens drill-in
endpoints so admins can investigate spikes that span more than the
default 100 most-recent opens.
Backend (artifacts/api-server/src/routes/stats.ts):
- New parsePaging() helper validates `limit` (1..200, default 100) and
`offset` (>=0, default 0); invalid values return 400.
- Both `/stats/admin/app-opens/by-app/:appId` and
`/stats/admin/app-opens/by-user/:userId` accept the new params, apply
`.limit(limit).offset(offset)`, and return `limit`, `offset`, and a
`nextOffset` (number | null) computed from `totalCount`.
- Added a stable secondary sort (`id desc`) so paged results don't
shuffle when timestamps tie.
Spec & client (lib/api-spec/openapi.yaml + regenerated clients):
- Added `limit`/`offset` query params and `limit`/`offset`/`nextOffset`
response fields to AdminAppOpensByApp/AdminAppOpensByUser.
- Re-ran `pnpm --filter @workspace/api-spec run codegen`.
Frontend (artifacts/teaboy-os/src/pages/admin.tsx + locales):
- AppOpensDrillIn / UserOpensDrillIn now accumulate extra pages in
local state and expose a "Load more" button via a shared
LoadMoreSection footer that also shows "Showing X of Y".
- Extra pages are fetched via the generated `getAdminAppOpensByApp` /
`getAdminAppOpensByUser` functions; accumulated state resets when
the appId/userId or stats query params change.
- Added en/ar translations for `loadMore`, `loadMoreError`, `shownOf`.
Tests:
- New artifacts/api-server/tests/admin-app-opens-pagination.test.mjs
covers happy-path paging for both endpoints, the default page size,
and 400 responses for invalid limit/offset.
- All 10 api-server tests pass; full workspace typecheck passes.
Original task: verify POST /api/apps/:id/open behaves correctly under slow
networks (the keepalive POST must survive the user navigating away),
unauthenticated callers (401 with no row inserted), and unknown app ids
(404 with no row inserted).
Changes
- New committed test file artifacts/api-server/tests/apps-open.test.mjs
using Node's built-in node:test runner (no new test framework added):
* happy path: authenticated POST returns 204 and inserts an app_opens row
* unauthenticated POST returns 401 and inserts no row
* authenticated POST to a non-existent app id (max(id)+100000) returns 404
and inserts no row
* slow-network simulation: opens a raw http.request to /api/apps/:id/open,
aborts the socket ~50 ms after sending so the client never reads the
response (mimicking a navigation-aborted keepalive POST), then asserts
the server still inserted the row. This proves the route's
`await db.insert(...)` runs to completion independently of whether the
client is still around to read the 204.
- The tests create a dedicated test user with a precomputed bcrypt hash for
"TestPass123!", assign the standard "user" role, log in via
POST /api/auth/login to obtain a connect.sid cookie, run the four cases,
and clean up (app_opens / user_roles / users) in an `after` hook.
- Added `pg` as a devDependency on @workspace/api-server (used by the
tests for direct DB assertions) and a `test` script:
`node --test 'tests/**/*.test.mjs'`.
- Also ran in-browser end-to-end coverage via the testing skill that
exercised the keepalive + wouter navigation flow against a live home
page with a 3 s route delay; that run also passed.
Schema drift fixed during the run
- The dev DB was missing the `app_opens` table and the `users.clock_style`
column referenced by the running schema. `pnpm --filter @workspace/db
push` blocked on an interactive rename/create prompt that could not be
answered non-interactively, so I brought the dev DB in line with the
Drizzle schema using idempotent SQL (CREATE TABLE IF NOT EXISTS for
app_opens with its two indexes; ALTER TABLE users ADD COLUMN IF NOT
EXISTS clock_style varchar(30)). No schema files were modified.
No production code changes were required — the existing route already
returns 401/404/204 correctly and the tests now lock that behavior in.
Riyadh