S208 Technical Summary: Vitto Metacognition & App Reorganization

Session Overview

• Branch: session-208
• Sub-sessions: S208.1 (libs cleanup), S208.2 (app reorg), S208.3 (backend narrative), S208.4 (frontend components)
• Lines removed: ~20,000 (dead libraries)
• Files created: ~20 new files
• Files modified: ~15 existing files

Phase 0: libs-ai Cleanup

Removed Directories (10)

Removed tsconfig.base.json Paths (8)

• @libs-ai/ai-agent-experience (was redirected to agent-mind alias)
• @libs-ai/ai-agent-registry
• @libs-ai/ai-generation-hub
• @libs-ai/ai-training-models (was redirected to ai-agents-training-center alias)
• @libs-ai/multi-agent-dashboard
• @libs-ai/panel-intelligence
• @libs-ai/task-router (was redirected to agent-orchestrator alias)
• @libs-ai/cursor-agents path entry (directory removed, path kept since alias points to agent-orchestrator)

Preserved Aliases (active imports exist)

• @libs-ai/module-patterns --> ai-module-intelligence (5+ imports)
• @libs-ai/outcome-tracker --> ai-module-intelligence (3+ imports)
• @libs-ai/cursor-agents --> agent-orchestrator (tsconfig entry kept)

Phase 1: App Reorganization

Top Navigation

• Before: Home | Monitor | Vitto
• After: Home | Operations | Command | Vitto
• File: apps/gograb-ai/src/app/app.component.html

Route Changes (app.routes.ts)

NEW: /operations (OperationsPanelComponent)
  |-- /agents (ObservatoryAgentsComponent)
  |-- /system (ObservatorySystemComponent)
  |-- /sandbox (ObservatorySandboxRouteComponent)
  |-- /failures (FailureChainsComponent)
  |-- /specialization (SpecializationMatrixComponent)

NEW: /command (CommandPanelComponent)
  |-- /mission (CommandCenterComponent from @libs-ai/gograb-ai-lib)
  |-- /approvals (ObservatoryApprovalsComponent)
  |-- /audit (ObservatoryAuditComponent)
  |-- /trust (TrustDashboardComponent)
  |-- /costs (CostDashboardComponent)

REDIRECTS: /observatory/* --> /operations/* or /command/*

Vitto Sidebar (vitto-unified-shell.component.ts)

6 groups replacing the old 4:

• Mind (7 items) -- was Brain (9 items, mixed)
• Identity (4 items) -- was Soul (5 items)
• Capabilities (5 items) -- NEW group
• Connections (3 items) -- NEW group
• Developer (5 items, admin-only) -- restructured
• Settings (5 items) -- slimmed

Key moves:

• Channels, Webhooks: Brain --> Connections
• Diary: Soul --> Mind
• Schedules: Settings --> Connections
• Dossier, Mastery, Sub-Agents: hidden flat routes --> visible in sidebar
• Observatory: sidebar --> removed (now top-level /operations + /command)

Phase 2: Backend Narrative Infrastructure

New SSE Event Types (vitto.types.ts)

ConscienceEventType:
  'conscience.dossier.loaded' |
  'conscience.context.assembled' |
  'conscience.skill.selected' |
  'conscience.memory.recalled' |
  'conscience.trust.evaluated' |
  'conscience.personality.applied'

IntrospectionEventType:
  'introspection.cost.alert' |
  'introspection.skill.promoted' |
  'introspection.queue.changed' |
  'introspection.model.routed' |
  'introspection.delegation.decided' |
  'introspection.context.budget'

VittoNarrativeService

• Location: APIs/backend-vitto-api/src/app/engine/vitto-narrative.service.ts
• Pattern: Centralized Subject --> piped into VittoEngineService.sseSubject
• Methods: emitConscience(type, userId, category, narrative, details?) and emitIntrospection(type, category, narrative, details?, severity?)
• Wired in VittoEngineService.onModuleInit(): this.narrative.events$.subscribe(event => this.sseSubject.next(event))

New SSE Endpoint

• @Sse('events/conscience/:userId') on VittoController
• Filters main SSE stream: only conscience.* events where userId matches param
• Privacy: each user only sees their own events

Emission Points

All emissions are @Optional() injected and wrapped in try/catch (fire-and-forget).

Phase 3: Frontend Components

New Files Created

libs/libs-ai/vitto-hub/src/lib/
|-- models/conscience.types.ts
|-- services/vitto-conscience.service.ts
|-- services/vitto-introspection.service.ts
|-- components/conscience/
|   |-- vitto-conscience.component.ts
|   |-- vitto-conscience.component.html
|   |-- vitto-conscience.component.scss
|-- components/introspection/
|   |-- vitto-introspection.component.ts
|   |-- vitto-introspection.component.html
|   |-- vitto-introspection.component.scss
|-- components/operations-panel/
|   |-- operations-panel.component.ts
|   |-- operations-panel.component.html
|   |-- operations-panel.component.scss
|-- components/command-panel/
|   |-- command-panel.component.ts
|   |-- command-panel.component.html
|   |-- command-panel.component.scss
|-- components/shared/external-redirect.component.ts

Service Patterns

• VittoConscienceService: EventSource to /engine/events/conscience/:userId, signal-based state, max 200 entries cap, category filtering via computed signals
• VittoIntrospectionService: EventSource to /engine/events (global stream), filters for conscience.* + introspection.* client-side, per-user filtering, activeUsers computed signal tracks unique users seen

Route Registration (vitto-hub.routes.ts)

• /vitto/conscience --> VittoConscienceComponent
• /vitto/introspection --> VittoIntrospectionComponent
• /vitto/observatory --> ExternalRedirectComponent (redirects to /operations)

Known Issues for S209

1. Command Center OFFLINE

• Symptom: Mission Control tab at /command/mission shows "OFFLINE" status
• Root cause: command-center-client.service.ts line ~443 constructs WebSocket URL using API_BASE_URL (port 3000) but the CommandCenterGateway lives on backend-ai-api (port 3001)
• Fix: Replace API_BASE_URL injection with AI_API_URL from @libs-foundational/api-config

2. Training Pipeline is Mock

• Symptom: "Train" button appears to work but nothing actually trains
• Root cause: training-pipeline.service.ts lines 122-150 generate fake job IDs and simulate delays
• Fix: Replace with few-shot prompt injection (Option B) using existing SubAgentService.buildSystemPrompt() infrastructure

3. Platform Intelligence Hardcoded Values

• Symptom: All agent cards show status: 'training' and successRate: 0
• Root cause: training-center.component.ts hardcodes these values instead of fetching from API
• Fix: Fetch real data from GET /vitto-api/engine/definitions and use performance.* fields

4. Sub-Agents Cannot Self-Execute

• Symptom: Sub-agents have definitions and enriched system prompts but all execution goes through handoff
• Root cause: SubAgentService.spawnFromDefinition() creates instances but has no execute() method
• Fix: Integrate BaseLLMProvider from APIs/backend-vitto-api/src/app/llm/ for direct LLM calls

5. Conscience/Introspection Not Yet Tested Live

• Symptom: Components built, SSE endpoints created, but never tested with running backends
• Fix: Start backend-vitto-api + gograb-ai, trigger chat messages, verify SSE events flow through