Fix tool propagation bug in hierarchical crews (#3679)

- Remove check_tools validator from task.py that was extending task.tools
  with agent.tools at creation time
- This caused manager agents in hierarchical crews to incorrectly inherit
  tools from task agents
- The crew.py execution logic already handles tool resolution correctly
  at execution time via fallback: task.tools or agent_to_use.tools or []
- Add reproduction test test_hierarchical_crew_does_not_propagate_agent_tools_to_manager
- Update test_task_tool_reflect_agent_tools to verify execution-time behavior

Fixes #3679

Co-Authored-By: João <joao@crewai.com>

src/crewai/task.py

@@ -336,12 +336,6 @@ class Task(BaseModel):
                 setattr(self, key, value)
         return self
 
-    @model_validator(mode="after")
-    def check_tools(self):
-        """Check if the tools are set."""
-        if not self.tools and self.agent and self.agent.tools:
-            self.tools.extend(self.agent.tools)
-        return self
 
     @model_validator(mode="after")
     def check_output(self):

tests/test_crew.py

@@ -3814,6 +3814,46 @@ def test_fetch_inputs():
     )
 
 
+def test_hierarchical_crew_does_not_propagate_agent_tools_to_manager():
+    """
+    Test that in hierarchical crews, manager agent doesn't inherit task agents' tools.
+    This verifies that the check_tools validator doesn't pollute task.tools at creation time.
+    Fixes issue #3679: https://github.com/crewAIInc/crewAI/issues/3679
+    """
+    from crewai.tools import tool
+
+    @tool
+    def agent_specific_tool() -> str:
+        """A tool that should only be available to the specific agent."""
+        return "agent specific result"
+
+    agent_with_tools = Agent(
+        role="Specialist",
+        goal="Do specialized work with custom tools",
+        backstory="You are a specialist with specific tools",
+        tools=[agent_specific_tool],
+        allow_delegation=False,
+    )
+
+    # Create a task with an agent that has tools, but don't assign tools to the task
+    task = Task(
+        description="Perform a specialized task",
+        expected_output="Task result",
+        agent=agent_with_tools,
+    )
+
+    crew = Crew(
+        agents=[agent_with_tools],
+        tasks=[task],
+        process=Process.hierarchical,
+        manager_llm="gpt-4o",
+    )
+
+    # Verify that task.tools is empty (not populated with agent's tools)
+    assert task.tools == []
+    assert len(task.tools) == 0
+
+
 def test_task_tools_preserve_code_execution_tools():
     """
     Test that task tools don't override code execution tools when allow_code_execution=True

tests/test_task.py

@@ -20,11 +20,13 @@ from crewai.utilities.string_utils import interpolate_only
 
 
 def test_task_tool_reflect_agent_tools():
+    """Test that agent tools are available during task execution via crew fallback logic."""
     from crewai.tools import tool
 
     @tool
-    def fake_tool() -> None:
+    def fake_tool() -> str:
         "Fake tool"
+        return "result"
 
     researcher = Agent(
         role="Researcher",
@@ -40,7 +42,9 @@ def test_task_tool_reflect_agent_tools():
         agent=researcher,
     )
 
-    assert task.tools == [fake_tool]
+    assert task.tools == []
+    
+    assert researcher.tools == [fake_tool]
 
 
 def test_task_tool_takes_precedence_over_agent_tools():