fix: remove duplicate memory extraction call from _finalize_kickoff

`AgentExecutor._save_to_memory` (added in 62d097a7) already handles
memory persistence with full conversation context after each kickoff.
Leaving the older `_save_kickoff_to_memory` call in `_finalize_kickoff`
caused `extract_memories` to be invoked twice, breaking
`test_agent_kickoff_memory_recall_and_save`.

Update the test assertion to match the conversation-aware format now
produced by `AgentExecutor._save_to_memory` (checks for "Conversation:"
or "Final result:" instead of the stale "Input:/Agent:/Result:" markers).

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

lib/crewai/src/crewai/agent/core.py

@@ -1598,8 +1598,6 @@ class Agent(BaseAgent):
                 response_format=response_format,
             )
 
-        self._save_kickoff_to_memory(messages, output.raw)
-
         crewai_event_bus.emit(
             self,
             event=LiteAgentExecutionCompletedEvent(

lib/crewai/src/crewai/mcp/__init__.py

@@ -27,6 +27,7 @@ from crewai.mcp.filters import (
     create_static_tool_filter,
 )
 
+
 if TYPE_CHECKING:
     from crewai.mcp.client import MCPClient
     from crewai.mcp.tool_resolver import MCPToolResolver

lib/crewai/tests/memory/test_unified_memory.py

@@ -704,11 +704,10 @@ def test_agent_kickoff_memory_recall_and_save(tmp_path: Path, mock_embedder: Mag
     recall_mock.assert_called_once()
 
     # Verify extract_memories and remember_many were called (passive batch save)
+    # Memory extraction is now conversation-aware (via AgentExecutor._save_to_memory)
     extract_mock.assert_called_once()
     raw_content = extract_mock.call_args.args[0]
-    assert "Input:" in raw_content
-    assert "Agent:" in raw_content
-    assert "Result:" in raw_content
+    assert "Conversation:" in raw_content or "Final result:" in raw_content
 
     # remember_many was called with the extracted memories
     remember_many_mock.assert_called_once()