Follow-up to the cache/backoff loop fix (aa23cf0): make a device that can't
self-install visible to operators, and fix the signature-verify bug that kept the
whole #139 fix from engaging on the actual Fire OS target.
Dashboard surface (Phase 2):
- devices gains ota_status / ota_target_version / ota_attempts / ota_updated_at
via the idempotent ALTER TABLE ADD COLUMN migration (non-destructive,
default-backfilled, idempotent on re-run).
- The device reports ota_status (OtaThrottle.statusFor -> none | pending |
manual_update_required) in device_info; the server persists it on register
(the reconnect backstop). devices d.* already surfaces it to the dashboard.
- Dashboard shows a non-blocking amber badge when manual_update_required
("Update available (vX) - install failed N times, manual update required");
i18n key in en.js (non-en inherits via the en fallback). Server suite +1 test.
Event-driven status (Option B):
- New device:ota-status WS message, emitted on STATE TRANSITIONS only
(enter-backoff -> manual_update_required, clear -> none), so the badge updates
promptly without waiting for a reconnect and without per-poll/heartbeat chatter.
Server handler persists the same fields; an unknown/forged device_id is a safe
no-op. The register-path persist stays as the reconnect backstop.
Signature-verify fix (the critical piece):
verifyApkSignature read the downloaded APK's signer via
getPackageArchiveInfo(GET_SIGNING_CERTIFICATES).signingInfo, but that field is
null for ARCHIVE files on API 28/29 (populated only from API 30). On Fire OS 8
(Android 9 / API 28) - the actual deployment target - this returned 0 certs from
a correctly-signed APK, so every OTA was refused as "tampered," the cache was
deleted, and the full APK re-downloaded every check cycle. This was the real
cause of the #139 re-download loop, NOT a silent-install failure: the cache and
backoff added in this branch sit behind this verify gate and never engaged on
the target.
Fix: below API 30, read the archive's signer via the legacy GET_SIGNATURES +
.signatures (its v1/JAR cert, which IS populated on 28/29). Keep
GET_SIGNING_CERTIFICATES + signingInfo for API >= 30 and for the installed-app
read (which works on 28+). The archive's signer is still extracted and compared
to the installed app's signer; a mismatch or zero-cert APK is still rejected.
This reads the cert correctly on old APIs - it does not weaken verification.
Verified on emulators:
- API 28: verify now passes for a legit APK (was: 0 certs, refused). Full backoff
then engages - 8.5MB pulled once, cache-hit on retries, backoff after 3,
manual_update_required emitted once; clears on successful update.
- API 28 negative: a re-signed (different-key) APK is still refused on cert
MISMATCH - no hole opened.
- API 30: unchanged path still passes (no regression).
- server suite 173/173, OtaThrottleTest 7/7.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Two independent multi-zone bugs, plus operator-facing warnings, i18n, and
regression tests guarding the data contracts.
Bug 1 — per-item mute was a no-op end to end:
- GET /api/devices/:id dropped the `muted` column from its assignments SELECT,
so the dashboard toggle never reflected state (the muted=false case in
particular). Column restored to the device payload.
- Android player now honours the per-item mute flag for YouTube (initial state
+ live via the IFrame JS API).
Bug 2 — items whose zone_id belongs to a different layout were silently dropped:
- Player fallback (web + Android): an orphaned zone_id is recovered into the
largest zone instead of vanishing, with telemetry.
- server/lib/zone-validate.js is the single source of truth for the orphan rule
(zone not in the device's active layout); used by the device payload
(per-item `orphan` flag + `active_layout_zones`) and the device list
(`orphan_count`).
- Assign-time hardening: a stale zone_id (not in the device's active layout) is
cleared to null on POST/PUT rather than persisted as a new orphan.
- scripts/find-orphan-zone-items.js: read-only sweep for existing orphans.
Dashboard warnings (operator-facing, never on the live player):
- Per-item badge + reassign affordance, device-list glance, preview banner.
- Graceful degradation: the zone selector falls back to /api/layouts/:id so it
can't vanish on a stale payload.
i18n: orphan-zone strings added to en/es/fr/de/pt/it (hi falls back by design;
count strings interpolate through tn()).
Tests: server/test/device-zone-contract.test.js adds 5 regression tests for the
data contracts above (muted true/false round-trip, active_layout_zones, orphan
flag + count, orphan-clears-on-reassign, assign-time clearing). 172/172 pass.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The PiP endpoints and the per-item mute field shipped without OpenAPI coverage.
- openapi.yaml: add POST /pip (show), DELETE /pip + POST /pip/clear (clear), all
x-required-scope: full; add the `muted` boolean to PUT /assignments/{id}; add a `pip` tag.
- openapi-contract.test.js: the scope heuristic only treated `command` paths as full-scope,
so a full-scope non-command route (/pip) would fail it — extend it to recognize /pip.
Docs-only as far as the running build goes (no route/behavior change). Lands on main; not
in the frozen v1.9.1-beta4 tag — ships in the next tag.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(server): proxy remote YouTube thumbnails instead of ENOENT on a local path
YouTube content stores thumbnail_path as a REMOTE URL
(https://img.youtube.com/vi/<id>/hqdefault.jpg), but the thumbnail-serving route
path.resolve'd it into contentDir -> a local file that never existed -> ENOENT logged
a few times a minute (the tester-log spam). Recreating content didn't help (new rows
store the same remote URL).
- GET /api/content/:id/thumbnail now proxies a remote http(s) thumbnail_path
server-side (same-origin, so dashboard CSP img-src is unaffected) via a non-throwing
helper: upstream 404 -> 404, other failure/timeout -> 502, image/* only (modest SSRF
hardening; the URL is server-set at ingest). Local thumbnails keep the sendFile path;
the playlist/widget/workspace access gating is unchanged for both branches.
- routes/widgets.js inlineUserContent skips the disk read for a remote thumbnail and
leaves the /api/content/:id/thumbnail reference in place (the proxy serves it).
- routes/content.js ingest unchanged; a comment notes the future download-at-ingest +
backfill option for CDN independence.
- New test/thumbnail-proxy.test.js: local sendFile still works; a remote thumbnail is
proxied (mock upstream, no local read, no ENOENT); upstream 404 -> clean 404. Full
server suite 164/164.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(server): boot banner shows the real version, not a hardcoded v1.2.0
The startup ASCII banner printed "ScreenTinker Server v1.2.0". Use the already-imported
VERSION (require('./version'), the single source of truth that reads the root VERSION
file) in a fixed-width field (VERSION.padEnd(22).slice(0, 22) — the same padEnd
discipline the port line uses) so the fixed-width box border stays aligned for any
version length. No other behavior changes.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(server): persist + ship + real-time per-item mute (#129)
The dashboard mute toggle was a no-op end to end. The active model is playlist_items
(the device payload is its published_snapshot); the legacy `assignments` table the bug
report cited is unused for devices. Three breaks:
- PUT /api/assignments/:id silently dropped `muted` (only read sort_order/duration_sec/
zone_id). It now accepts muted (coerced 0/1) and ITEM_SELECT returns it, so the toggle
persists and its on/off state sticks.
- playlist_items had no `muted` column — added (schema + idempotent migration).
- buildSnapshotItems didn't select muted, so it never reached the published_snapshot /
device payload — now included.
Real-time: on a mute change, emit device:mute-changed { content_id, widget_id, muted } to
every device on that playlist so the player toggles the matching item's volume live,
decoupled from publish (the value is also in the next snapshot, so it persists). Adds a
[mute] log line (the report noted zero mute log entries).
Test: test/mute.test.js — PUT persists + returns muted, it reaches the published
snapshot, and a non-mute update doesn't reset it. Server suite 164/164.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(player): apply per-item mute live on Android + web (#129)
Honor the new per-item mute from the server, both in real time and on reload.
Android:
- WebSocketService: onMuteChanged callback + main-thread device:mute-changed handler.
- MediaPlayerManager.setVideoMuted(): flips the live ExoPlayer volume on the current
video (YouTube autoplays muted; images/widgets are silent).
- MainActivity: on device:mute-changed, apply immediately if the toggled item is the
one playing now.
- PlaylistController.sig(): include muted so a published mute change re-renders/persists
instead of being de-duped.
Web player (server/player/index.html):
- device:mute-changed handler toggles the current <video>; the video mount now also
honors item.muted so a published mute sticks across reloads.
Tizen intentionally not included: its player mutes ALL video for autoplay, so per-item
unmute isn't achievable there.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* PiP overlay MVP: push image/web overlays to a device or group (#109)
Implements the #109 MVP from docs proposal: a floating overlay PUSHED to a device or
group in real time, rendered above the playlist without disturbing it. Scope is the
MVP only — video/RTSP, MQTT, offline-queue, and the priority/stacking system are
deferred to follow-up PRs as the proposal specifies.
Protocol (/device socket, player-agnostic):
- device:pip-show { pip_id, type:image|web, uri, position, width, height, duration,
title?, title_color?, background_color?, opacity?, border_radius?, close_button? }
- device:pip-clear { pip_id? }
The player fetches uri itself (same trust model as remote_url content; server never
proxies). type:web is full-trust by design, hence the 'full' token scope.
Server (server/routes/pip.js, new; mounted in config/api-surface.js PUBLIC_ROUTERS):
- POST /api/pip and POST /api/pip/clear + DELETE /api/pip, all requireScope('full').
- Resolves device_id to a device OR a group, expands a group to members, and emits
per-device — reusing the group command route's room-size online check and
{device_id, name, status: sent|offline} result shape. Generates pip_id.
- Validates type/position allowlists, uri http(s), numeric bounds on
width/height/duration/opacity/border_radius, colors via the existing VALID_COLOR
(#RRGGBB; transparency is the separate opacity field).
- Workspace-isolated: every target query is scoped to req.workspaceId, so a token
bound to workspace A can't address workspace B (404). Offline devices are reported,
never queued (PiP is ephemeral).
Player overlay layer (Tizen; tizen/js/pip-overlay.js, new):
- A #pip sibling ABOVE #stage that PlaylistPlayer/ZoneRenderer never touch.
- applyOrientation now applies the SAME transform to #pip as #stage, so corner
positions track the visible CONTENT in all four orientations.
- image -> <img>, web -> <iframe> (muted by default: empty allow= denies autoplay),
sized/positioned/styled per payload, optional title bar.
- Single overlay slot, last-show-wins; duration timer (0 = until cleared); pip-clear
(id-aware) or timer tears down; teardown wrapped so a malformed payload can't wedge
the layer. Reports show/clear over device:log (tag 'pip').
Dashboard: a minimal "Send overlay" / "Clear overlay" tester on the device-detail
controls (device/group via the open device, type, uri, position, duration), calling
POST /api/pip through the api helper.
Tests (server suite green, 161/161):
- api.test.js: PiP tier — authz (read/write 403, full passes), workspace isolation
(wsA token -> wsB device 404), payload validation, device + group targeting, clear;
plus the PUBLIC_ROUTERS snapshot-firewall updated for /api/pip.
- pip-overlay.test.js: loads the real player.js + pip-overlay.js in a vm with a DOM
shim; proves the overlay shows, auto-dismisses on the duration timer, and never
changes the playlist signature / touches #stage; web->iframe, last-show-wins,
id-aware clear, malformed-payload safety.
Not in this PR (intentional):
- Android player overlay — fast-follow. Protocol + server are player-agnostic; the
Android layer (an overlay View above the player, orientation-matched to MainActivity's
rootView rotation) is the same shape and lands next.
- OpenAPI docs for POST /api/pip — the contract test's scope heuristic only treats
'command' paths as full-scope, so documenting a full-scope non-command route there
needs that heuristic extended first; deferred with the docs item (proposal §8.6).
- video/rtsp types, MQTT, offline queue-on-reconnect, priority/stacking, arbitrary
(x,y)/selector positioning (proposal §6).
Refs #109
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* PiP overlay: add Android + web players (#109)
Extends the #109 PiP MVP to the other two players so the protocol (device:pip-show /
device:pip-clear) is honored fleet-wide, not just on Tizen. No server/protocol changes —
the route and socket messages are player-agnostic; these are the two missing surfaces.
Web player (server/player/index.html):
- New #pipContainer layer above #playerContainer, pointer-transparent, that the playlist
render never touches. The same orientation transform is applied to it as to
#playerContainer (extended to also reset width/height on landscape so a
portrait->landscape switch realigns), so corner positions track the visible content.
- Inline PiP logic mirroring tizen/js/pip-overlay.js: image -> <img>, web -> <iframe>
(muted by default via empty allow=), position/size/bg/opacity/radius/title, single slot
last-show-wins, duration timer (0 = until cleared), id-aware clear, wrapped teardown.
- device:pip-show/clear handlers; reports show/clear over device:log (tag "pip").
Android player:
- activity_main.xml: a pipLayout FrameLayout as the LAST child of rootLayout — it draws
above the content AND inherits rootView's orientation rotation/translation, so corner
positioning is orientation-matched for free.
- PipOverlay.kt (new): builds the overlay box into pipLayout. image -> ImageView (decoded
off-thread via ImageLoader, dropped if torn down mid-decode); web -> WebView with
mediaPlaybackRequiresUserGesture=true (mute-by-default). Gravity-based corner/center
placement with a 4% inset, GradientDrawable bg + corner radius, alpha=opacity, optional
title bar. Single slot last-show-wins; duration timer; id-aware clear; teardown wrapped
and also run on activity destroy (WebView cleanup).
- WebSocketService: onPipShow/onPipClear callbacks + safeOn handlers posted to the main
thread (they build Views) + a sendLog(tag, level, message) emitter for device:log.
- MainActivity: instantiate PipOverlay (log -> wsService.sendLog("pip", ...)), wire the
callbacks, tear down on destroy.
Verified: Android assembleDebug builds clean; web player inline JS parses; server suite
still 161/161 (no server changes this commit). Not yet validated on real hardware —
four-orientation corner positioning mirrors the player container/rootView transform but
should be eyeballed on a panel.
Refs #109
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Agencies can only be designated FULL-SCREEN playlists (no item with zone_id) - a full-screen
agency upload can't safely target a zone, so the ambiguous case is excluded rather than
solved. Checked at THREE points:
- Designation (tokens.js create + PUT /:id/targets) -> 400: reject a zoned target.
- Upload (agency.js item-add) -> 409: block if the playlist BECAME zoned after designation.
MANDATORY because auto-publish has no draft net - a full-screen playlist designated to an
auto-publish token, then zone-assigned, would otherwise auto-publish a full-screen upload
into a zoned playlist. The upload check is the only thing that catches it.
- Picker (settings.js): zoned playlists greyed/disabled with the reason (GET /playlists now
returns a zoned flag); backend reject is the guard if the UI is bypassed. i18n x5.
isZonedPlaylist = EXISTS(playlist_items WHERE zone_id IS NOT NULL). Pure restriction - no
zone structure, no api_token_target_zones.
Bite-test (the exact sequence) GREEN and re-proven to bite: full-screen -> designate to an
auto-publish token -> zone-assign the playlist -> agency upload is BLOCKED (409), not
auto-published; neutralizing the upload check makes it go red. 149 suite green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Settings → API Tokens: each agency token gets an "Edit playlists" control that opens the
playlist picker pre-checked with the token's CURRENT designations (from the list GET's
tok.targets), lets the admin add/remove, and calls the existing PUT /:id/targets to
atomically re-designate. Reuses the creation picker pattern; common.save/cancel reused;
edit_targets + targets_updated i18n across all 5 locales. No security-model change - the
endpoint was already proven.
Test (integration): PUT /:id/targets re-designates (add + remove) and the confinement
follows the NEW set - a re-designated token reaches only its new playlists (router.param
403s the removed one). 148 suite green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Investigation found zone placement is a DEVICE property (device.layout_id), not a playlist
property: a normal playlist has no derivable layout (zone_id is NULL unless set in the
device-assignment flow), so a playlist-scoped zone grant can't reach the normal flow. The
right model: placement belongs to the device (same playlist can be full-screen on one screen,
a zone on another); the agency just gets whole-playlist grants + size-guidance.
Removed the zone-grant machinery (security-adjacent dead surface is a liability, not dormant
convenience): api_token_target_zones (schema + a DROP migration for the dev DB where the
short-lived CREATE ran), resolveGrantedZone, grantableZoneIds, buildZoneGrantRows, the
create/PUT zone validation, GET /api/playlists/:id/zones, getPlaylistZones, the settings
zone-picker + its i18n, and the zone-grant bite-test.
KEPT (model-agnostic, good): the reactive per-playlist size-guidance card - GET
/api/agency/playlists/:playlistId/layout (router.param-confined) now reports the zones the
playlist actually feeds (where/what-size content lands), or full-screen when it has no layout.
Whole-playlist grants = today's working model. 147 suite green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Issuance (on the proven seam):
- tokens.js create + PUT /:id/targets accept per-playlist zone grants (target_zones), inserted
into api_token_target_zones inside the same transaction as the playlist grants (FK requires
the parent, so order matters and is correct).
- Issuance validation (the mirror of runtime confinement): grantableZoneIds() - can grant ONLY
a zone the playlist's layout actually feeds; can't grant one it doesn't have or one from
another playlist's layout. Bite-tested. PUT re-designate stays atomic: delete parent rows ->
zone grants cascade out (no manual child delete).
- settings.js: checking a designated playlist reveals its grantable zones (GET
/api/playlists/:id/zones, JWT); leave unchecked = whole-playlist. i18n across all 5 locales.
Card:
- GET /api/agency/playlists/:playlistId/layout (rides router.param - confined; a non-
designated playlist -> 403, asserted). "Your zone" = the GRANTED zones. Retired the
token-wide /layouts (the per-playlist card replaces the disconnected lump).
- Portal card reacts to the playlist selector: pick a playlist -> its layout renders, the
granted zone highlighted with px size, siblings as context.
Full suite + agency bite-suite green (154).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Placement-as-grant, replacing the inferred auto-place idea. api_token_target_zones is an
ADDITIVE second table (does NOT touch the proven api_token_targets), structurally anchored:
a composite FK to api_token_targets(token_id, playlist_id) makes a zone grant orphan-
impossible and cascade away when the playlist grant is revoked - "narrow" is structural, not
conventional. zone_id FK -> layout_zones cascades on zone/layout delete.
Confinement (lib/agency-targets.resolveGrantedZone, called in the item-add): grants exist ->
the item MUST land in a granted zone (a body zone_id picks among grants, never escapes them);
none -> whole-playlist/full-screen as before. The item-add stamps the granted zone_id.
Bite-tested (6, all proven incl. neutralize->red on the confinement): granted YES; non-
granted/cross-playlist/ambiguous blocked; orphan-grant rejected by the FK; cascade on
playlist-grant revoke, on playlist delete, on zone/layout delete; and foreign_keys=ON
asserted (a cascade that no-ops because FKs are off is the trap). 153 suite green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Reuses the existing scheduler + sendEmail infra (no new scheduler). The agency endpoint
enqueues one agency_notifications row per item added; a 15-min flush groups unsent rows per
token+playlist+action and sends ONE digest per group to the workspace owner/admins + the
playlist owner (deduped via UNION). Draft -> "added N items, awaiting approval"; published ->
"updated <playlist>".
Two robustness rules, both tested:
- Queue never balloons when SMTP is off: the endpoint skips enqueue when !isConfigured(),
and the flush drains-and-discards unsent rows as a backstop.
- sent_at is stamped ONLY after a successful send, so a failed send retries next cycle
instead of silently dropping.
Wired into boot via startAgencyDigest(). 147 suite green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
So the agency can size/place content: returns the canvas size + zone positions/sizes for the
layouts its designated playlists feed, marking which zones are theirs. DEVICE-FREE BY
CONSTRUCTION - the query path is playlist_items.zone_id -> layout_zones -> layouts and never
touches devices/groups/schedules, so device names/locations/IPs/topology are structurally
absent, not filtered. Geometry only - no sibling-zone content. layout.name included (admin's
canvas name); thumbnail_data omitted (could render other zones' content).
Confinement query in lib/agency-layouts.js, bite-tested: own layout YES, a non-designated
playlist's layout NO, response has NO device fields (asserted on a db where a location-named
device exists), and neutralizing the t.token_id filter goes red. 142 suite green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
api_tokens.auto_publish (DEFAULT 0 = draft, the fail-safe). Admin sets it at token creation
in the designate UI (checkbox, agency scope only). The agency endpoint reads it from the
TOKEN ROW via req.apiToken (apiTokenAuth attaches it) - NEVER from req.body, so an agency
can't opt itself out of approval. 0 -> markDraft; 1 -> the shared publishPlaylist path.
Tests (integration): draft is the default; a draft token with auto_publish:true IN THE BODY
still lands draft (body ignored); an auto-publish token goes live; manual publish still works
(extraction regression). i18n across all 5 locales. 141 suite green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The portal needs to show an agency which playlists it may post to. New read surface on the
security primitive, built with write-path rigor: the confinement query lives in
lib/agency-targets.js (own token + bound workspace only) and is bite-tested four ways -
own targets yes; another token's, outside the allowlist, and cross-workspace all NO;
neutralizing the t.token_id filter makes it go red. Real-path wiring + the portal's
graceful 401 trigger asserted in the integration suite. No :playlistId, so router.param
doesn't apply - the query is the seam.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The agency capability behind the proven off-ladder/agencyGate primitive:
- agencyGate is now SCOPE-only at the mount; the per-target check is router.param
('playlistId') in routes/agency.js - it fires WITH the param before the handler, so no
:playlistId route can skip it (drift-proof). A mount-level target check was silently
bypassed (Express populates req.params only at route match); the integration bite-suite
caught it - this is the fix.
- routes/agency.js: POST /content (shared ingest) + POST /playlists/:id/items (date-bounded
#74/#75 item; lands as draft so the admin's re-publish is the approval gate).
- tokens.js: issue scope='agency' tokens bound to a non-empty in-workspace playlist
allowlist (atomic); PUT /:id/targets re-designates (JWT-only -> can't self-widen).
- server.js: AGENCY_ROUTERS mounted bearerAuth + resolveTenancy + agencyGate.
Full bite-suite (test/agency.test.js) GREEN and re-proven to bite on the SHIPPING path:
neutralizing the router.param check makes non-designated->403 go red. Four assertions at
three seams: target (router.param), off-ladder (tokenScopeGate), can't-widen (tokens
JWT-only), issuance cross-workspace (create validation). 139 suite green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The capability/target-restricted token model for the agency portal (#73 option B),
proven before any endpoint sits on it:
- 'agency' scope value is OFF the read/write/full ladder, so the existing tokenScopeGate
rejects it on every public router by construction (auto-confinement, no new code).
- api_token_targets join table: which playlists an agency token may act on.
- agencyGate: THE single seam - agency-scope-only + (playlist in this token's allowlist
AND in the bound workspace), one query enforcing target + cross-workspace isolation.
- AGENCY_ROUTERS category in config/api-surface.js (mounted with agencyGate, not
tokenScopeGate) - declared; router/mount land with the endpoints.
Both bite-tested: spine (agency 403s on tokenScopeGate; read/write still pass) and the
gate (non-designated/cross-workspace/non-agency/JWT -> 403; neutralizing the target check
goes red). NARROW - not the general capability-scope system.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Review caught the encrypted TOTP secret riding in the login/verify response body:
issueSession receives a SELECT * user row and only destructured out password_hash, so
totp_secret_enc (and the internal replay counter totp_last_step) leaked. Encrypted, so
not catastrophic, but it regresses the API work's "secrets never in responses" rule.
Strip both in issueSession (covers /login and /totp/verify); add an assertion that a
verify response carries no totp_secret_enc.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Unit: the mfa_pending BITE (db-injected so removing the rejection goes red), lockout, replay,
recovery-hash, decrypt-null graceful. Integration: enrollment, login->mfa_required, route-level
bite, recovery single-use, API-token bypass, verify lockout. Key-rotation: enroll under key A,
reboot under key B -> recovery still works, TOTP fails cleanly (no 500).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The 6-digit pairing code is generated client-side, so the server can't raise its entropy
without a player change. Instead, harden server-side (no client change):
- lib/pair-lockout.js: lock an IP out of POST /api/provision/pair after 5 failed claims
(15-min lockout), and expire stale provisioning codes after 15 min so a code is not
claimable indefinitely. A successful claim resets the IP.
- /pair enforces both. Only an UNKNOWN code (404) counts toward the lockout (a real guess);
an EXPIRED code (410) is a legitimate-but-stale code and does NOT count, so a slow bulk
rollout from one shared-NAT IP can't lock itself out. getClientIp is Cloudflare-aware
(CF-Connecting-IP validated against a trusted edge peer), so the lockout keys on the real
per-client IP, never a shared edge.
Unit-tested deterministically with injected time, incl. the bulk-rollout-never-locks case.
Closes#87
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
POST /api/provision was a second pairing endpoint that paired a device by code but,
unlike POST /api/provision/pair, did NOT assign a workspace, enforce checkDeviceLimit, or
emit device:paired / dashboard:device-added - a silently-diverging duplicate that no
client ever called. It now returns 410 Gone and points callers at /pair, so
/api/provision/pair is the single, fully-protected pairing endpoint. The mount stays in
the JWT-only partition, so a Bearer st_ token still gets 401 (requireAuth) before the 410.
Closes#90
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The non-security gaps named in the public-API self-review:
- gap-fix: zone_id (playlist items) + layout_id (device PUT) accepted and returned on read,
INCLUDING the cross-tenant rejection (the is_template OR workspace_id guard - the
security-relevant one).
- docs serving: /openapi.yaml serves the spec, /docs returns the Redoc page.
- i18n drift-guard: apitoken.* keys have full parity across en/es/fr/de/pt (a key missing
in one locale fails CI).
- token lifecycle branches: token-create workspace-membership validation and last_used_at
stamping (integration), plus the must_change_password gate (unit test via the in-memory
DB injection - cross-process WAL visibility is unreliable for that branch in-process).
119 tests total (was 108), all in the existing node --test job.
Closes#92
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Self-review follow-ups, kept as a separate commit so the review trail is honest.
- Spec drift: POST /widgets/preview was documented scope 'read' but the method-based
tokenScopeGate enforces 'write' for any POST, so a read-token integrator following the
published docs would hit a surprise 403. The code is right; fix the SPEC to match it.
- Guard it forever: test/openapi-contract.test.js cross-checks every spec operation's
x-required-scope against the enforcement rule, and that every documented path is a
public (token-reachable) router - both derived from the same config/api-surface.js.
Adds js-yaml (devDep) to parse the spec. Spec/enforcement drift now fails CI.
- Vendored Redoc: add frontend/vendor/README.md (library, version 2.3.9, source, update
steps) and drop the dangling //# sourceMappingURL line so /docs doesn't 404 in devtools.
Remaining (non-security) test-coverage gaps tracked in #92.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A dedicated public-API suite (boots the real server as a subprocess) so CI green proves
the token layer, not just the pre-existing tests:
- Partition firewall, derived from the SAME config/api-surface.js server.js mounts from:
every JWT-only router 401s a token; a public-surface snapshot fails if any router is
added to the token door; known-privileged routers asserted JWT-only.
- Threat model: role-strip gates, workspace-binding both directions (token ignores
X-Workspace-Id, JWT honors it), the scope ladder, the render bypass, token lifecycle,
and JWT no-regression.
- Device WS round-trip via socket.io-client (added as a devDep): valid device_token
registers + receives its playlist; wrong token rejected.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Each playlist item can carry schedule blocks (active days, start/end
time-of-day, optional start/end dates). An item plays when the screen's
local "now" matches at least one block; an item with no blocks always
plays. #74 covers time-of-day/day-of-week windows including overnight
wrap; #75 covers inclusive date ranges (auto-expiry). Evaluation is
on-device, so dayparting and expiry work offline.
- Shared evaluator contract: shared/schedule-vectors.json (39 vectors —
DST US+AU, overnight-wrap anchoring, timezone correctness, date
boundaries). Canonical JS evaluator in server/lib/schedule-eval.js;
Kotlin and Tizen ports kept in lockstep by drift guards (Tizen byte-diff
test, Kotlin JUnit reads the shared JSON, new android-test CI job).
- All three players (web, Android, Tizen) filter by schedule against their
own clock, idle with a "Nothing scheduled" message + 30s re-check when
everything is filtered, and fail open on any evaluator error.
- Editor: per-item schedule modal + row badge in the playlist editor;
client validation mirrors the server; editing marks the playlist draft.
- Part B (behaviour change): device/group schedule overrides now evaluate
in each device's effective timezone instead of server-local time.
- Device detail shows the reported timezone + a clock-skew warning.
- i18n for en/es/fr/de/pt across all new strings (namespaced itemsched.*
to avoid colliding with the device-schedule calendar's schedule.*).
- CHANGELOG documents the feature, the Part B change, the fail-open
guarantee, and the scheduled-single-video re-render tradeoff.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- server/version.js: shared version helper that reads the root VERSION file once
(fallback 0.0.0). Replaces the stale hardcoded 1.2.0 / 1.5.1 / 1.0.0 fallbacks
in /api/version, /api/update/check, and /api/status.
- config.js: DATA_DIR / DB_PATH / UPLOADS_DIR / CERTS_DIR env overrides for the
db, uploads, and certs/jwt-secret locations. Unset resolves to exactly the
legacy in-repo paths, so existing installs (including production) are
byte-for-byte unchanged. Guarded by test/config-paths.test.js.
- package.json: rename remote-display-server -> screentinker (+ lockfile name).
- scripts/bump-version.sh: one-shot bump across VERSION, package.json (+lock),
android (versionName and versionCode + 1), and the tizen widget version; makes
one commit plus an annotated tag; prints the push command, never pushes.
- .gitignore: global *.db / *.db-wal / *.db-shm / *.db.* so no database file
(including .db.devbak backups, at any path) can be committed.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Models sometimes stacked text lines at the same y (unreadable) and emitted accent
shapes after text, so a band could hide the words.
- deoverlapTexts: push a line down only when it also overlaps horizontally
(leaves side-by-side text alone), with conservative line-height clearance so
real rendering doesn't re-overlap; shift the stack up if it ran past the bottom.
- Order shapes before text in the output so accent bands always render behind the
words.
Verified: 0 text-on-text overlaps across multiple prompts (Playwright DOM check);
unit test asserts overlapping lines get separated + shapes precede text. 63/63.
Text could run off the edge (long/large headlines, nowrap) and shapes placed at
the far edge (e.g. a bottom band at y=100) spilled over.
- Server-side fit pass on every generated element: shrink text fontSize so it
fits the canvas width (chars*fontSize*0.075, tuned for bold/uppercase
headlines) and height (incl. line-height), then nudge x/y within 4% margins;
clamp shapes so x+width<=100 and y+height<=100. Deterministic - doesn't rely on
the model getting layout right.
- Designer preview: vw -> cqw (+ container-type on the canvas) so text scales to
the canvas, not the browser window. The preview was overstating size vs what
actually publishes; now it matches. Published widget keeps vw (scales on the
player).
Verified: Playwright DOM check shows zero elements overflowing the canvas after
generation; unit test asserts long text is shrunk + repositioned in-bounds. 62/62.
Competitor pressure (Mandoe 'AI Magic Create'): prompt -> signage. We answer it
in a way that's actually BETTER for signage and costs the operator nothing.
Key idea: don't generate raw images (AI garbles text - fatal for menus/promos).
The LLM returns a STRUCTURED design spec (headline, supporting text, accent
shapes, palette) that the existing Designer renders with real fonts - crisp and
fully editable. Reuses the whole Designer.
BYOK, fully under the customer's control: each workspace configures its own
OpenAI-COMPATIBLE endpoint + key - OpenAI cloud OR self-hosted (Ollama / LM Studio
/ llama.cpp). Operator bears zero AI cost/liability.
- server/lib/secretbox.js: AES-256-GCM for the key at rest (never returned).
- routes/ai.js: GET/PUT /api/ai/settings (admin; key write-only) + POST
/generate-design (editor+). Output is strictly validated/normalized (cap count,
clamp ranges, px->%, strip HTML, validate colors) - never trust the model.
SSRF guard: hosted instances block private/internal targets; self-hosted (the
whole point of local AI) may point at localhost/LAN.
- Designer: an 'AI generate' panel (prompt + Generate) + a settings modal.
Verified end-to-end against local Ollama (llama3.1:8b): prompt -> editable design
on the canvas. Unit tests cover normalization + the SSRF guard. Suite 61/61.
Phase 2 (next): AI background images (OpenAI images / AUTOMATIC1111).
Self-hosters rebuilding could end up schema-behind-code, failing only at runtime
(a missing users.must_change_password locked out all logins). Two root causes:
1. The migration loop swallowed EVERY error (catch {}), so a real ALTER failure
was indistinguishable from the benign 'duplicate column' on an already-migrated
DB. Now only 'duplicate column'/'already exists' is treated as a no-op; any
other error is logged loudly, and a one-line summary reports how many new
column migrations actually applied this boot.
2. Nothing verified the schema after migrations. Added lib/schema-check.js:
verifyAndRepairSchema() checks the tables + columns the request path REQUIRES,
idempotently repairs missing repairable columns (logging each), and if anything
required is STILL missing, prints a loud FATAL block and exits - failing fast at
boot instead of at the first authed request.
Note: the reported 'audit_log missing' was a misdiagnosis - the code uses
activity_log (0 refs to audit_log), created by schema.sql on every boot.
Tests: healthy (no-op), auto-repair of must_change_password, missing-table report.
Platform admins can now cleanly remove a customer org (account ends) or a stray
workspace from the UI, instead of raw SQL that risks orphaning resources.
The tenant cascade isn't pure DB CASCADE - workspace-scoped tables (devices,
content, playlists, ...) are NO ACTION and must be purged before the workspace.
Extracted that logic out of deleteUserCascade into shared deleteWorkspaceCascade /
deleteOrgCascade helpers (one tested implementation; deleteUserCascade now reuses
the purgeWorkspaces extraction).
Backend (platform-admin only): GET /api/admin/orgs (list + owner + counts +
workspaces), DELETE /api/admin/orgs/:id, DELETE /api/admin/workspaces/:id.
UI: an Organizations section in Admin listing every org/workspace with a
type-the-name confirmation before the irreversible delete.
Tests: org/workspace cascade (real FKs) + endpoint gating/404. Suite 53/53.
MSPs onboarding customers as separate orgs had no way to create one with
AUTO_CREATE_ORG_ON_SIGNUP=false (the only path was signup auto-org). Add a
platform-admin 'Create organization' action.
POST /api/admin/orgs (requirePlatformAdmin) creates the org + its first 'Default'
workspace. organizations.owner_user_id is NOT NULL, so an org can't be ownerless;
the creating admin becomes org_owner + workspace_admin (mirrors the signup
bootstrap in routes/auth.js) - which also surfaces the org in their switcher.
Customer users are then added via the existing Add User / manage-memberships flow.
UI: 'Create organization' button + single-field modal in the Admin area (gated).
Tests: create (201 + memberships + audit), empty-name 400, non-admin/operator 403.
The public, CSP-exempt widget render (GET /api/widgets/:id/render) inlined
config values straight into <style>/CSS and (for the text widget) raw into the
same-origin document. A workspace editor could store `}</style><script>...` in a
color/background/size field (bypassing the UI pickers via the API) → stored XSS
executing in the app origin for anyone who opens the render URL (JWT theft).
- safeCss(): allow colors/gradients but reject CSS breakout / url() / @import /
expression / javascript:. Applied to background/color across clock, weather,
rss, social renders.
- safeNumber(): coerce font_size / scroll_speed / max_items to a finite number
so they can't smuggle markup.
- Text widget keeps its intentional raw HTML/CSS feature, but it now renders
inside an <iframe sandbox="allow-scripts"> (NO allow-same-origin) - scripts run
in a null origin that can't reach the dashboard's localStorage/JWT.
Tests: test/widget-render-xss.test.js (breakout rejected, numbers coerced, text
isolated, legit colors/gradients preserved). Full suite green.
Five low-risk, high-value fixes surfaced by the security review:
#3 Branding lockdown — `custom_domain`/`custom_css` (which feed the PUBLIC,
pre-auth branding resolver and the login-page <style>) are now settable only
by platform admins; a workspace_admin can no longer hijack the platform login
page by claiming its domain. The public /api/branding (+ /domain) now return
only presentational fields via publicBranding() (no id/user_id/workspace_id/
custom_domain/timestamps leak).
#6 Strip device_token — the device WS auth secret (validated with
timingSafeEqual) was returned in device list/get/update + pairing responses
(SELECT d.* / *). New lib/device-sanitize.js strips it everywhere; prevents
device impersonation by any workspace user.
#7 must_change_password enforced server-side — was a frontend-only redirect, so
a provisioned temp password worked indefinitely via the API. requireAuth now
403s every route except GET/PUT /api/auth/me (the password change, which
clears the flag) and logout while the flag is set.
#8 XSS — escape user data interpolated into innerHTML in teams.js, kiosk.js,
layout-editor.js (team/page/layout/zone names, member name/email, kiosk
config fields). scriptSrcAttr 'unsafe-inline' made these exploitable via
injected event handlers, not just markup.
#9 Thumbnail IDOR — /api/content/:id/thumbnail had no auth/scope gate (any UUID
served any tenant's thumbnail). Now mirrors the /file route's playlist/widget
workspace-scoped reference check.
Tests: new test/security-fixes.test.js (device strip, publicBranding field
allowlist, must_change_password gate). Full suite 41/41. Verified live against a
prod-data copy: device_token absent from /api/devices, /api/branding trimmed.
Not addressed here (tracked for follow-up): Android OTA signature verification
(Critical), public widget-render XSS, token revocation/logout, pairing-code
strength, validateRemoteUrl hardening, import quota.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
White-label is stored per-workspace (white_labels.workspace_id); unbranded and
new workspaces - and the login page - fell back to hardcoded ScreenTinker. Add a
single platform default that everything inherits beneath the per-workspace layer.
Resolution (lib/branding.js): workspace row -> custom-domain match -> platform
default -> hardcoded ScreenTinker. Row-level override: a workspace with its own
row keeps it (current behavior); only row-less workspaces inherit the default,
so editing the default propagates instantly (no row-copying at creation).
The platform default is a white_labels row with a FIXED id ('platform-default'),
not a "workspace_id IS NULL" sentinel - legacy pre-multitenancy rows can also
have a null workspace_id, which would be ambiguous.
- routes/admin.js: GET/PUT /api/admin/branding (requirePlatformAdmin) to read/
upsert the single platform-default row; audit-logged.
- server.js: public GET /api/branding (domain match -> platform default ->
hardcoded) for pre-login/pre-workspace contexts.
- routes/white-label.js: authed GET now falls back to the platform default
(was hardcoded) for row-less workspaces.
- Frontend: login page resolves + applies branding (logo, name, colors, favicon,
custom CSS) pre-auth; Admin page gets a "Default branding" form.
Tests: resolver order incl. legacy null-ws safety; admin GET/PUT (single row,
upsert, platform-admin-only 403). Full suite 37/37. Verified end-to-end:
public + authed + login-page all inherit the platform default; per-workspace
override preserved.
Closes#15.
The Workspace column on the platform Users page could only move a 0/1-workspace
user and showed a dead "N workspaces" label for multi-membership users. Replace
it with a "Manage workspaces" modal that handles the full picture.
Backend (routes/admin.js, requirePlatformAdmin):
- GET /api/admin/users/:id/workspaces list memberships (+org/ws names, role)
- POST /api/admin/users/:id/workspaces add to a workspace (upsert role)
- PUT /api/admin/users/:id/workspaces/:wsId change role in a workspace
- DELETE /api/admin/users/:id/workspaces/:wsId remove (last one allowed -> unassigned)
Roles validated against WORKSPACE_ROLES; each mutation writes an audit row.
Frontend:
- Workspace cell is now a summary (Unassigned / <name> / N workspaces /
"Platform (all)" for staff) + a Manage button.
- New admin-user-workspaces-modal: lists every membership with an inline role
dropdown + Remove, plus a type-to-filter "Add to workspace" picker (org-grouped,
excludes current memberships) with a role select. Staff get a note that they
already have platform-wide access. Refreshes the table on close if changed.
- Removed the old single-select inline move control (superseded by the modal).
Tests: 6 added (add to multiple workspaces, per-workspace role change, upsert,
remove incl. last->unassigned, validation 400/404, non-platform-admin 403).
Full suite 33/33. Verified headless: Manage opens, lists memberships, filtered
picker, add/role-change/remove round-trips persist (throwaway user, cleaned up).
The #18 user-delete bug was the first symptom of a broader gap: 13 tables
reference workspaces(id) (and activity_log also organizations(id)) with NO
ACTION, so deleting a workspace or organization fails the same FK wall once it
holds any content. SQLite can't ALTER an FK action, so this migration rebuilds
each table (the create-copy-rename pattern the assignments/schedules migrations
already use), changing only the tenant FK clause:
workspace_id -> ON DELETE CASCADE (resources belong to the workspace)
activity_log.workspace_id / organization_id -> ON DELETE SET NULL (keep audit)
user_id FKs are intentionally left as-is - user deletion stays handled app-side
by lib/user-deletion.js (the #18 fix).
- lib/tenant-cascade-migration.js: pure, idempotent core (table-existence
guarded; transforms the stored CREATE text, copies rows verbatim, recreates
indexes; fixes activity_log's AUTOINCREMENT sequence; baseline-vs-after
foreign_key_check so pre-existing orphan rows don't abort it but a botched
rebuild does).
- db/database.js: boot wrapper owns the pre-migration snapshot + process.exit
on failure, matching the other heavy migrations.
Tests (node:test): reproduces the workspace-delete FK failure, applies the
migration, verifies FK actions (CASCADE / SET NULL), index recreation, data
preserved, and that workspace/org delete now cascades (activity_log preserved).
Full suite 27/27. Verified on a copy of a real DB: 13 tables rebuilt,
integrity_check ok, workspace delete cascades, no new FK violations.
DELETE /api/auth/users/:id ran a bare `DELETE FROM users`, but 23 columns
reference users(id) and only 4 cascade, so with foreign_keys=ON the delete
fails the moment the user is referenced anywhere - and a real user always is
(owns an org, created a workspace, has login activity). Reproduces on a fresh
DB, exactly as reported.
The schema also lacks cascades from workspaces -> tenant resources, so the DB
can't clean up on its own. New lib/user-deletion.js resolves every reference in
one transaction (defer_foreign_keys=ON for forgiving order; table-existence
guard for resilience):
- Refuse (409) if the user OWNS an organization that has other members -
don't nuke a shared tenant; transfer ownership first.
- Hard-delete the organizations they SOLELY own (workspaces + all contents).
- In orgs they don't own, PRESERVE resources: SET NULL the nullable
creator/inviter columns, and reassign the NOT NULL legacy creator user_id to
the resource's org owner (fallback: the acting admin).
- Memberships (organization_members/workspace_members/team_members/
content_folders) cascade on the user delete; pending invites they sent and
legacy teams they own are removed.
The handler now 404s an unknown id and 409s the shared-org case.
Tests (node:test): reproduces the FK failure, then verifies provisioned-member
delete (resources preserved + unlinked/reassigned), solo-org-owner cascade,
shared-org refusal (409), self-delete 400, non-superadmin 403, unknown 404.
Full suite 22/22. Verified end-to-end on a copy of a real DB: deleted a user
owning 2 solo orgs, foreign_key_check clean.
Closes#18.
Adds a "Workspace" column (after Plan) to the platform Users admin table so a
platform_admin can see and reassign a user's workspace inline, alongside the
Role/Plan dropdowns. Single-workspace move/assign model.
Backend:
- GET /api/auth/users (platform branch): one aggregate query adds
workspace_count and, for exactly-one membership, the workspace id/name + org
name (no N+1).
- PUT /api/admin/users/:id/workspace (requirePlatformAdmin - operator excluded):
move (1 membership) or assign (0) into the chosen workspace, default role
workspace_viewer, in a transaction; no-op if already there; REFUSES (400) a
user with >1 membership (manage in the members view). logActivity
admin_set_user_workspace.
Frontend (admin.js):
- Editable <select> only for a 'user' with 0/1 membership; multi-membership ->
read-only "N workspaces", platform staff -> read-only "Platform (all)".
- Options grouped by org via <optgroup>, built ONCE from /me's
accessible_workspaces (same source as the Add User picker) and reused per row.
- Picking "Unassigned" or the same workspace is a no-op so a stray pick can't
strip a membership. Success -> toast + refresh. EN i18n only.
Tests: 4 added (single-membership move 200 + changed, zero-membership assign
200, multi-membership 400 refused, non-platform-admin/operator 403). npm test
16/16. Verified headless: column renders, selected value correct, "Platform
(all)" for staff, and a dropdown move persisted (throwaway user, cleaned up).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The bug: #13 added 'platform_operator' to the frontend role dropdown
(PLATFORM_ROLE_OPTIONS) but #14's PUT /api/auth/users/:id/role whitelist
(ASSIGNABLE_PLATFORM_ROLES) only listed ['user','platform_admin'], so
selecting "Platform operator" returned 400 "Invalid role" - the role was
unassignable via the UI.
Fix: add 'platform_operator' to ASSIGNABLE_PLATFORM_ROLES. One line; the
self-demote guard is intentionally left untouched (a platform_admin still
cannot self-assign the non-owner operator role and lock themselves out).
Tests (node:test, isolated in-memory DB injection - no DB_PATH change):
- admin-users.test.js: platform_admin can PUT role=platform_operator on a
target user -> 200 and the row persists as platform_operator (regression
guard for the whitelist gap).
- operator-permissions.test.js (new): verify-then-test of the highest-blast
-radius deny. Operator CAN update/delete a workspace-scoped content row
(cross-org write works) but is denied (403) updating or deleting a shared
(workspace_id IS NULL) row - proving the separate PLATFORM_ROLES gate in
content.js's checkContentWrite still holds after canWrite was broadened to
isPlatformStaff.
Verified read-only (no leak): the other shared-asset write sites keep their
PLATFORM_ROLES gate that excludes operator - kiosk.js:57, widgets.js:110,
folders.js:31, layouts.js:59/117/133.
cd server && npm test -> 12 pass / 0 fail.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds server/test/admin-users.test.js and a `npm test` (node --test) script.
No DB_PATH override: the suite mounts the real routers against an isolated
in-memory better-sqlite3 instance injected into the require cache, seeded by
the test itself. Node v20 built-ins only (node:test, node:assert, fetch).
Covers: Add User success (response omits password/hash, hash stored not
plaintext, membership written, hosted lifecycle sentinels stamped, audit row
without the password), duplicate-email 409 (no overwrite), non-admin 403,
platform_operator denied (403), org_admin scoped to their own org only,
input validation, and the must_change_password lifecycle (set on create,
surfaced on login, cleared on PUT /api/auth/me).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
