Reduce max_steps to 2 to prevent final_answer from reformatting

- Changed max_steps from 3 to 2 to stop after answer_with_sources
- Updated tool description to clarify it generates the FINAL answer
- Removed reference to calling final_answer tool (no longer needed)
- This prevents the built-in final_answer from stripping citations and sources

src/second_brain_online/application/agents/agents.py

@@ -63,7 +63,7 @@ class AgentWrapper:
         agent = ToolCallingAgent(
             tools=[what_can_i_do, retriever_tool, summarizer_tool],
             model=model,
-            max_steps=3,  # Retrieval → answer_with_sources → final_answer (pass-through)
+            max_steps=2,  # Retrieval → answer_with_sources (which IS the final answer)
             verbosity_level=2,
         )
 

src/second_brain_online/application/agents/tools/summarizer.py

@@ -62,11 +62,11 @@ class HuggingFaceEndpointSummarizerTool(Tool):
 
 class OpenAISummarizerTool(Tool):
     name = "answer_with_sources"
-    description = """Use this tool to generate the complete final answer to the user's question based on search results. 
+    description = """Use this tool to generate the FINAL answer to the user's question based on search results. 
 
 After retrieving documents with mongodb_vector_search_retriever, use this tool to synthesize a comprehensive answer with a Sources section.
 
-CRITICAL: This tool's output is the complete answer - after getting results from this tool, you MUST call the built-in final_answer tool and pass this output EXACTLY as-is without any modifications."""
+CRITICAL: This tool generates the final answer that will be returned to the user. Do NOT modify or reformat its output in any way."""
 
     inputs = {
         "search_results": {