Lorenze/metrics for human feedback flows (#4188)

* measuring human feedback feat

* add some tests

lib/crewai/tests/utilities/test_events.py

@@ -25,6 +25,8 @@ from crewai.events.types.flow_events import (
     FlowCreatedEvent,
     FlowFinishedEvent,
     FlowStartedEvent,
+    HumanFeedbackReceivedEvent,
+    HumanFeedbackRequestedEvent,
     MethodExecutionFailedEvent,
     MethodExecutionFinishedEvent,
     MethodExecutionStartedEvent,
@@ -45,6 +47,7 @@ from crewai.events.types.tool_usage_events import (
     ToolUsageFinishedEvent,
 )
 from crewai.flow.flow import Flow, listen, start
+from crewai.flow.human_feedback import human_feedback
 from crewai.llm import LLM
 from crewai.task import Task
 from crewai.tools.base_tool import BaseTool
@@ -1273,3 +1276,135 @@ def test_llm_emits_event_with_lite_agent():
 
     assert set(all_agent_roles) == {agent.role}
     assert set(all_agent_id) == {str(agent.id)}
+
+
+# ----------- HUMAN FEEDBACK EVENTS -----------
+
+
+@patch("builtins.input", return_value="looks good")
+@patch("builtins.print")
+def test_human_feedback_emits_requested_and_received_events(mock_print, mock_input):
+    """Test that @human_feedback decorator emits HumanFeedbackRequested and Received events."""
+    requested_events = []
+    received_events = []
+    events_received = threading.Event()
+
+    @crewai_event_bus.on(HumanFeedbackRequestedEvent)
+    def handle_requested(source, event):
+        requested_events.append(event)
+
+    @crewai_event_bus.on(HumanFeedbackReceivedEvent)
+    def handle_received(source, event):
+        received_events.append(event)
+        events_received.set()
+
+    class TestFlow(Flow):
+        @start()
+        @human_feedback(
+            message="Review:",
+            emit=["approved", "rejected"],
+            llm="gpt-4o-mini",
+        )
+        def review(self):
+            return "test content"
+
+    flow = TestFlow()
+
+    with patch.object(flow, "_collapse_to_outcome", return_value="approved"):
+        flow.kickoff()
+
+    assert events_received.wait(timeout=5), (
+        "Timeout waiting for human feedback events"
+    )
+
+    assert len(requested_events) == 1
+    assert requested_events[0].type == "human_feedback_requested"
+    assert requested_events[0].emit == ["approved", "rejected"]
+    assert requested_events[0].message == "Review:"
+    assert requested_events[0].output == "test content"
+
+    assert len(received_events) == 1
+    assert received_events[0].type == "human_feedback_received"
+    assert received_events[0].feedback == "looks good"
+    assert received_events[0].outcome is None
+
+    assert flow.last_human_feedback is not None
+    assert flow.last_human_feedback.outcome == "approved"
+
+
+@patch("builtins.input", return_value="feedback text")
+@patch("builtins.print")
+def test_human_feedback_without_routing_emits_events(mock_print, mock_input):
+    """Test that @human_feedback without emit still emits events."""
+    requested_events = []
+    received_events = []
+    events_received = threading.Event()
+
+    @crewai_event_bus.on(HumanFeedbackRequestedEvent)
+    def handle_requested(source, event):
+        requested_events.append(event)
+
+    @crewai_event_bus.on(HumanFeedbackReceivedEvent)
+    def handle_received(source, event):
+        received_events.append(event)
+        events_received.set()
+
+    class SimpleFlow(Flow):
+        @start()
+        @human_feedback(message="Please review:")
+        def review(self):
+            return "content to review"
+
+    flow = SimpleFlow()
+    flow.kickoff()
+
+    assert events_received.wait(timeout=5), (
+        "Timeout waiting for human feedback events"
+    )
+
+    assert len(requested_events) == 1
+    assert requested_events[0].emit is None
+
+    assert len(received_events) == 1
+    assert received_events[0].feedback == "feedback text"
+    assert received_events[0].outcome is None
+
+
+@patch("builtins.input", return_value="")
+@patch("builtins.print")
+def test_human_feedback_empty_feedback_emits_events(mock_print, mock_input):
+    """Test that empty feedback (skipped) still emits events correctly."""
+    received_events = []
+    events_received = threading.Event()
+
+    @crewai_event_bus.on(HumanFeedbackReceivedEvent)
+    def handle_received(source, event):
+        received_events.append(event)
+        events_received.set()
+
+    class SkipFlow(Flow):
+        @start()
+        @human_feedback(
+            message="Review:",
+            emit=["approved", "rejected"],
+            llm="gpt-4o-mini",
+            default_outcome="rejected",
+        )
+        def review(self):
+            return "content"
+
+    flow = SkipFlow()
+    flow.kickoff()
+
+    assert events_received.wait(timeout=5), (
+        "Timeout waiting for human feedback events"
+    )
+
+
+    assert len(received_events) == 1
+    assert received_events[0].feedback == ""
+    assert received_events[0].outcome is None
+
+
+    assert flow.last_human_feedback is not None
+    assert flow.last_human_feedback.outcome == "rejected"