Board is now a clean master-task view:
- GetBoardAsync returns only top-level (master) tasks; child-tasks render
nested inside their parent card instead of as separate column cards, so a
big task split into many sub-tasks stays one card (orphans treated as master)
- New DoneChildTaskCount on the DTO for real progress bars
- Child/detail consumers (GetChildren endpoint, TaskBridgeService) query
children directly instead of scraping the flat board
Stall watchdog (replaces destructive auto-reset):
- StaleTaskRecoveryService.FlagStalledInProgressTasksAsync marks In-progress
tasks with no activity past the threshold as stalled (activity event +
Iris notification) WITHOUT resetting the column — no work is discarded.
Idempotent: a task is not re-flagged until real progress happens
- BackgroundService now runs this watchdog (TaskRecovery:StalledMinutes=40,
interval 10m); hard reset kept only on the explicit manual endpoint
Review flow (Bao/Iris only):
- POST tasks/{id}/approve (Review -> Done)
- POST tasks/{id}/request-changes (Review -> target, mandatory comment,
ExpectedFrom=iris, notifies Iris)
Frontend:
- BoardCard component: master card with ball chip (who has it), progress from
children, expand to show children grouped by agent with per-child state +
stalled marker, stalled chip on the master, review action buttons
- Request-changes modal; tasks store approveReview/requestChanges actions
Tests: watchdog flag/idempotency + review threshold; 135 backend tests green.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Shell:
- One shared NexusLayout for ALL routes (dashboard + pages): compact 68px
icon rail with hover-expand overlay, replaces both old sidebars + topbars
- Single flat nav source (railNav) — same menu everywhere incl. Settings
- Removed dead shell components (AppSidebar, AppHeader, Topbar, NavGroup,
NavItem, ModuleView) and dead nav routes; App.vue is now just RouterView
Live updates:
- Fix SSE loop in DashboardController: PeriodicTimer.WaitForNextTickAsync and
ChannelReader.ReadAsync were re-invoked while pending — every published
update threw InvalidOperationException and killed ALL live streams
- liveSync store: treat graceful stream close as disconnect (was stuck
connected=true with polling stopped -> page frozen until manual reload),
fast first retry, heartbeat watchdog (65s), reconnect on online/visibility
- One app-wide SSE connection owned by the layout instead of per-view
connect/disconnect churn; removed duplicate live-sync.ts store
Performance:
- Board endpoint: drop nested childTasks duplication (counts stay) — payload
104KB -> 65KB; SSE snapshots shrink equally
- nginx: gzip for JSON/JS/CSS (board 18KB, bundle 95KB over the wire);
text/event-stream excluded to keep SSE unbuffered
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- Postgres memory: 256M→384M limits, 64M→96M reservations
- Added pg_resetwal -f pre-deploy step to recover from corrupt WAL
('PANIC: could not locate a valid checkpoint record' caused by
force-killed postgres during --force-recreate)
- Added data-checksums initdb arg for future corruption detection
- api→postgres and web→api depends_on: service_healthy→service_started
- Deploy wait loop: fail fast on unhealthy, wait on starting (180s)
- Added researcher/executor to ValidAssignees and frontend dropdowns
- Bao darf jetzt Status ändern (neben Iris), Sub-Agents weiterhin nicht
- CanEditContent für Inhaltsbearbeitung durch alle bekannten Caller
- Bao-Content-Änderungen triggern task_content_changed-Notification an Iris
- Bao-Status-Änderungen triggern task_status_changed-Notification an Iris
- Iris-Status-Änderungen triggern task_status_changed-Notification an Bao
- Neue WorkTask-Felder: IsAgentTask (bool), ExpectedFrom (string)
- Agent-Workflow-API: CreateAgentTask, WaitingTasks, AgentOverview
- Frontend: Agent-Task-Badge, Iris-Overview-Panel, isBao-Getter
- Login-Rate-Limiter mit strukturiertem JSON-Fehlermeldungs-Body
- Volume-Name: nexus-postgres → postgres-data (Standardisierung)
devops
reviewer