feat: Node Self-Healing, DAGI Audit, Agent Prompts, Infra Invariants

### Backend (city-service)
- Node Registry + Self-Healing API (migration 039)
- Improved get_all_nodes() with robust fallback for node_registry/node_cache
- Agent Prompts Runtime API for DAGI Router integration
- DAGI Router Audit endpoints (phantom/stale detection)
- Node Agents API (Guardian/Steward)
- Node metrics extended (CPU/GPU/RAM/Disk)

### Frontend (apps/web)
- Node Directory with improved error handling
- Node Cabinet with metrics cards
- DAGI Router Card component
- Node Metrics Card component
- useDAGIAudit hook

### Scripts
- check-invariants.py - deploy verification
- node-bootstrap.sh - node self-registration
- node-guardian-loop.py - continuous self-healing
- dagi_agent_audit.py - DAGI audit utility

### Migrations
- 034: Agent prompts seed
- 035: Agent DAGI audit
- 036: Node metrics extended
- 037: Node agents complete
- 038: Agent prompts full coverage
- 039: Node registry self-healing

### Tests
- test_infra_smoke.py
- test_agent_prompts_runtime.py
- test_dagi_router_api.py

### Documentation
- DEPLOY_CHECKLIST_2024_11_30.md
- Multiple TASK_PHASE docs

services/router/main.py

@@ -19,6 +19,7 @@ SWAPPER_URL = os.getenv("SWAPPER_URL", "http://192.168.1.33:8890")
 STT_URL = os.getenv("STT_URL", "http://192.168.1.33:8895")
 VISION_URL = os.getenv("VISION_URL", "http://192.168.1.33:11434")
 OCR_URL = os.getenv("OCR_URL", "http://192.168.1.33:8896")
+CITY_SERVICE_URL = os.getenv("CITY_SERVICE_URL", "http://daarion-city-service:7001")
 
 # HTTP client for backend services
 http_client: Optional[httpx.AsyncClient] = None
@@ -56,7 +57,27 @@ def load_config():
         }
     }
 
+def load_router_config():
+    """Load main router-config.yml with agents and LLM profiles"""
+    # Try multiple locations
+    paths = [
+        "router-config.yml",
+        "/app/router-config.yml",
+        "../router-config.yml",
+        "../../router-config.yml"
+    ]
+    
+    for path in paths:
+        if os.path.exists(path):
+            with open(path, 'r') as f:
+                logger.info(f"✅ Loaded router config from {path}")
+                return yaml.safe_load(f)
+    
+    logger.warning("⚠️ router-config.yml not found, using empty config")
+    return {"agents": {}}
+
 config = load_config()
+router_config = load_router_config()
 
 @app.on_event("startup")
 async def startup_event():
@@ -363,10 +384,30 @@ async def agent_infer(agent_id: str, request: InferRequest):
     - Agent configuration (model, capabilities)
     - Request type (text, vision, audio)
     - Backend availability
+    
+    System prompt is fetched from database via city-service API.
     """
     logger.info(f"🔀 Inference request for agent: {agent_id}")
     logger.info(f"📝 Prompt: {request.prompt[:100]}...")
     
+    # Get system prompt from database or config
+    system_prompt = request.system_prompt
+    
+    if not system_prompt:
+        try:
+            from prompt_builder import get_agent_system_prompt
+            system_prompt = await get_agent_system_prompt(
+                agent_id,
+                city_service_url=CITY_SERVICE_URL,
+                router_config=router_config
+            )
+            logger.info(f"✅ Loaded system prompt from database for {agent_id}")
+        except Exception as e:
+            logger.warning(f"⚠️ Could not load prompt from database: {e}")
+            # Fallback to config
+            agent_config = router_config.get("agents", {}).get(agent_id, {})
+            system_prompt = agent_config.get("system_prompt")
+    
     # Determine which backend to use
     model = request.model or "gpt-oss:latest"
     
@@ -389,7 +430,7 @@ async def agent_infer(agent_id: str, request: InferRequest):
                     json={
                         "model": model,
                         "prompt": request.prompt,
-                        "system": request.system_prompt,
+                        "system": system_prompt,
                         "stream": False,
                         "options": {
                             "num_predict": request.max_tokens,