feat(rag): Complete RAG prompt refactoring and bug fixes

- Removed all hardcoded prompts from RAG agents
- Created 3 new prompts for RAG Swarm Coordinator:
  * document_grader_v1: Document relevance grading
  * query_rewriter_v1: Query optimization for retrieval
  * answer_generator_v1: Answer synthesis from context
- Refactored rag_swarm_coordinator.py to use AgentPromptLoader
- Refactored writer_agent.py to use AgentPromptLoader
- All 9 RAG prompts now in LangSmith Hub with proper tags
- Fixed critical bug: Changed ToolMessage detection from msg.type to isinstance()
- Fixed bug: Collect ALL tool results, not just last message
- Changed model from gemini-2.0-flash-exp to gemini-2.5-flash (consistency)
- Added system message to force tool usage in agent node
- Enhanced LangSmith tracing setup in RAG Management App
- Added visual tracing indicator in UI

Related: US-RAG-004, US-SWARM-002, Sprint 6

Author: gitwalter

.gitignore

@@ -150,6 +150,7 @@ chroma.sqlite3
 .env.temp
 secrets.json
 secrets.toml
+.google_cloud/
 
 # Build artifacts
 dist/

agents/rag/__init__.py

@@ -10,12 +10,14 @@
 from agents.rag.re_ranker_agent import ReRankerAgent
 from agents.rag.quality_assurance_agent import QualityAssuranceAgent
 from agents.rag.writer_agent import WriterAgent
+from agents.rag.rag_swarm_coordinator import RAGSwarmCoordinator
 
 __all__ = [
     'QueryAnalystAgent',
     'RetrievalSpecialistAgent',
     'ReRankerAgent',
     'QualityAssuranceAgent',
     'WriterAgent',
+    'RAGSwarmCoordinator',
 ]
 

agents/rag/query_analyst_agent.py

@@ -179,25 +179,10 @@ async def _analyze_query_with_llm(self, query: str) -> Dict[str, Any]:
             convert_system_message_to_human=True
         )
 
-        # Create analysis prompt
-        system_prompt = """You are a Query Analysis Expert for a RAG (Retrieval Augmented Generation) system.
-
-Your task is to analyze user queries and provide:
-1. Intent classification (factual, conceptual, procedural, multi-hop, exploratory)
-2. 3-5 rewritten query variants optimized for semantic search
-3. Key concepts/terms that should be searched
-4. Recommended search strategy (focused, broad, multi-stage)
-5. Complexity score (0.0-1.0)
-
-Respond in this exact JSON format:
-{
-    "intent": "factual|conceptual|procedural|multi-hop|exploratory",
-    "rewritten_queries": ["variant1", "variant2", "variant3"],
-    "key_concepts": ["concept1", "concept2", "concept3"],
-    "search_strategy": "focused|broad|multi-stage",
-    "complexity": 0.5,
-    "reasoning": "Brief explanation"
-}"""
+        # Load prompt from LangSmith (following langgraph_workflow pattern)
+        from prompts.agent_prompt_loader import get_agent_prompt_loader
+        prompt_loader = get_agent_prompt_loader("query_analyst")
+        system_prompt = prompt_loader.get_system_prompt()
 
         user_prompt = f"""Analyze this query: