Merge remote-tracking branch 'origin/dev' into feat/use-experiment-logs

# Conflicts:
#	src/control_backend/agents/llm/llm_agent.py

src/control_backend/agents/bdi/bdi_core_agent.py

@@ -348,7 +348,7 @@ class BDICoreAgent(BaseAgent):
             yield
 
         @self.actions.add(".reply_with_goal", 3)
-        def _reply_with_goal(agent: "BDICoreAgent", term, intention):
+        def _reply_with_goal(agent, term, intention):
             """
             Let the LLM generate a response to a user's utterance with the current norms and a
             specific goal.
@@ -524,10 +524,6 @@ class BDICoreAgent(BaseAgent):
 
             yield
 
-        @self.actions.add(".notify_ui", 0)
-        def _notify_ui(agent, term, intention):
-            pass
-
     async def _send_to_llm(self, text: str, norms: str, goals: str):
         """
         Sends a text query to the LLM agent asynchronously.

src/control_backend/agents/bdi/text_belief_extractor_agent.py

@@ -345,6 +345,9 @@ class TextBeliefExtractorAgent(BaseAgent):
                 async with httpx.AsyncClient() as client:
                     response = await client.post(
                         settings.llm_settings.local_llm_url,
+                        headers={"Authorization": f"Bearer {settings.llm_settings.api_key}"}
+                        if settings.llm_settings.api_key
+                        else {},
                         json={
                             "model": settings.llm_settings.local_llm_model,
                             "messages": [{"role": "user", "content": prompt}],

src/control_backend/agents/llm/llm_agent.py

@@ -1,3 +1,4 @@
+import asyncio
 import json
 import logging
 import re
@@ -35,6 +36,10 @@ class LLMAgent(BaseAgent):
     def __init__(self, name: str):
         super().__init__(name)
         self.history = []
+        self._querying = False
+        self._interrupted = False
+        self._interrupted_message = ""
+        self._go_ahead = asyncio.Event()
 
     async def setup(self):
         self.logger.info("Setting up %s.", self.name)
@@ -53,13 +58,13 @@ class LLMAgent(BaseAgent):
                 case "prompt_message":
                     try:
                         prompt_message = LLMPromptMessage.model_validate_json(msg.body)
-                        await self._process_bdi_message(prompt_message)
+                        self.add_behavior(self._process_bdi_message(prompt_message))  # no block
                     except ValidationError:
                         self.logger.debug("Prompt message from BDI core is invalid.")
                 case "assistant_message":
-                    self.history.append({"role": "assistant", "content": msg.body})
+                    self._apply_conversation_message({"role": "assistant", "content": msg.body})
                 case "user_message":
-                    self.history.append({"role": "user", "content": msg.body})
+                    self._apply_conversation_message({"role": "user", "content": msg.body})
         elif msg.sender == settings.agent_settings.bdi_program_manager_name:
             if msg.body == "clear_history":
                 self.logger.debug("Clearing conversation history.")
@@ -76,12 +81,45 @@ class LLMAgent(BaseAgent):
 
         :param message: The parsed prompt message containing text, norms, and goals.
         """
+        if self._querying:
+            self.logger.debug("Received another BDI prompt while processing previous message.")
+            self._interrupted = True  # interrupt the previous processing
+            await self._go_ahead.wait()  # wait until we get the go-ahead
+
+        message.text = f"{self._interrupted_message} {message.text}"
+
+        self._go_ahead.clear()
+        self._querying = True
         full_message = ""
         async for chunk in self._query_llm(message.text, message.norms, message.goals):
+            if self._interrupted:
+                self._interrupted_message = message.text
+                self.logger.debug("Interrupted processing of previous message.")
+                break
             await self._send_reply(chunk)
             full_message += chunk
-        self.logger.debug("Finished processing BDI message. Response sent in chunks to BDI core.")
-        await self._send_full_reply(full_message)
+        else:
+            self._querying = False
+
+            self._apply_conversation_message(
+                {
+                    "role": "assistant",
+                    "content": full_message,
+                }
+            )
+            self.logger.debug(
+                "Finished processing BDI message. Response sent in chunks to BDI core."
+            )
+            await self._send_full_reply(full_message)
+
+        self._go_ahead.set()
+        self._interrupted = False
+
+    def _apply_conversation_message(self, message: dict[str, str]):
+        if len(self.history) > 0 and message["role"] == self.history[-1]["role"]:
+            self.history[-1]["content"] += " " + message["content"]
+            return
+        self.history.append(message)
 
     async def _send_reply(self, msg: str):
         """
@@ -166,13 +204,6 @@ class LLMAgent(BaseAgent):
                 full_message,
                 extra={"role": "assistant", "reference": message_id, "partial": False},
             )
-
-            self.history.append(
-                {
-                    "role": "assistant",
-                    "content": full_message,
-                }
-            )
         except httpx.HTTPError as err:
             self.logger.error("HTTP error.", exc_info=err)
             yield "LLM service unavailable."
@@ -192,6 +223,9 @@ class LLMAgent(BaseAgent):
             async with client.stream(
                 "POST",
                 settings.llm_settings.local_llm_url,
+                headers={"Authorization": f"Bearer {settings.llm_settings.api_key}"}
+                if settings.llm_settings.api_key
+                else {},
                 json={
                     "model": settings.llm_settings.local_llm_model,
                     "messages": messages,

src/control_backend/agents/perception/transcription_agent/speech_recognizer.py

@@ -145,4 +145,6 @@ class OpenAIWhisperSpeechRecognizer(SpeechRecognizer):
 
     def recognize_speech(self, audio: np.ndarray) -> str:
         self.load_model()
-        return whisper.transcribe(self.model, audio, **self._get_decode_options(audio))["text"]
+        return whisper.transcribe(self.model, audio, **self._get_decode_options(audio))[
+            "text"
+        ].strip()

src/control_backend/agents/user_interrupt/user_interrupt_agent.py

@@ -9,7 +9,7 @@ from control_backend.agents.bdi.agentspeak_generator import AgentSpeakGenerator
 from control_backend.core.agent_system import InternalMessage
 from control_backend.core.config import settings
 from control_backend.schemas.belief_message import Belief, BeliefMessage
-from control_backend.schemas.program import ConditionalNorm, Program
+from control_backend.schemas.program import ConditionalNorm, Goal, Program
 from control_backend.schemas.ri_message import (
     GestureCommand,
     PauseCommand,
@@ -249,6 +249,16 @@ class UserInterruptAgent(BaseAgent):
             self._cond_norm_map = {}
             self._cond_norm_reverse_map = {}
 
+            def _register_goal(goal: Goal):
+                """Recursively register goals and their subgoals."""
+                slug = AgentSpeakGenerator.slugify(goal)
+                self._goal_map[str(goal.id)] = slug
+                self._goal_reverse_map[slug] = str(goal.id)
+
+                for step in goal.plan.steps:
+                    if isinstance(step, Goal):
+                        _register_goal(step)
+
             for phase in program.phases:
                 for trigger in phase.triggers:
                     slug = AgentSpeakGenerator.slugify(trigger)
@@ -256,8 +266,7 @@ class UserInterruptAgent(BaseAgent):
                     self._trigger_reverse_map[slug] = str(trigger.id)
 
                 for goal in phase.goals:
-                    self._goal_map[str(goal.id)] = AgentSpeakGenerator.slugify(goal)
-                    self._goal_reverse_map[AgentSpeakGenerator.slugify(goal)] = str(goal.id)
+                    _register_goal(goal)
 
                 for goal, id in self._goal_reverse_map.items():
                     self.logger.debug(f"Goal mapping: UI ID {goal} -> {id}")

src/control_backend/core/config.py

@@ -123,6 +123,7 @@ class LLMSettings(BaseModel):
 
     local_llm_url: str = "http://localhost:1234/v1/chat/completions"
     local_llm_model: str = "gpt-oss"
+    api_key: str = ""
     chat_temperature: float = 1.0
     code_temperature: float = 0.3
     n_parallel: int = 4

test/unit/agents/llm/test_llm_agent.py

@@ -67,8 +67,52 @@ async def test_llm_processing_success(mock_httpx_client, mock_settings):
         thread="prompt_message",  # REQUIRED: thread must match handle_message logic
     )
 
+    agent._process_bdi_message = AsyncMock()
+
     await agent.handle_message(msg)
 
+    agent._process_bdi_message.assert_called()
+
+
+@pytest.mark.asyncio
+async def test_process_bdi_message_success(mock_httpx_client, mock_settings):
+    # Setup the mock response for the stream
+    mock_response = MagicMock()
+    mock_response.raise_for_status = MagicMock()
+
+    # Simulate stream lines
+    lines = [
+        b'data: {"choices": [{"delta": {"content": "Hello"}}]}',
+        b'data: {"choices": [{"delta": {"content": " world"}}]}',
+        b'data: {"choices": [{"delta": {"content": "."}}]}',
+        b"data: [DONE]",
+    ]
+
+    async def aiter_lines_gen():
+        for line in lines:
+            yield line.decode()
+
+    mock_response.aiter_lines.side_effect = aiter_lines_gen
+
+    mock_stream_context = MagicMock()
+    mock_stream_context.__aenter__ = AsyncMock(return_value=mock_response)
+    mock_stream_context.__aexit__ = AsyncMock(return_value=None)
+
+    # Configure the client
+    mock_httpx_client.stream = MagicMock(return_value=mock_stream_context)
+
+    # Setup Agent
+    agent = LLMAgent("llm_agent")
+    agent.send = AsyncMock()  # Mock the send method to verify replies
+
+    mock_logger = MagicMock()
+    agent.logger = mock_logger
+
+    # Simulate receiving a message from BDI
+    prompt = LLMPromptMessage(text="Hi", norms=[], goals=[])
+
+    await agent._process_bdi_message(prompt)
+
     # Verification
     # "Hello world." constitutes one sentence/chunk based on punctuation split
     # The agent should call send once with the full sentence, PLUS once more for full reply
@@ -85,28 +129,16 @@ async def test_llm_processing_errors(mock_httpx_client, mock_settings):
     agent = LLMAgent("llm_agent")
     agent.send = AsyncMock()
     prompt = LLMPromptMessage(text="Hi", norms=[], goals=[])
-    msg = InternalMessage(
-        to="llm",
-        sender=mock_settings.agent_settings.bdi_core_name,
-        body=prompt.model_dump_json(),
-        thread="prompt_message",
-    )
 
     # HTTP Error: stream method RAISES exception immediately
     mock_httpx_client.stream = MagicMock(side_effect=httpx.HTTPError("Fail"))
 
-    await agent.handle_message(msg)
+    await agent._process_bdi_message(prompt)
 
     # Check that error message was sent
     assert agent.send.called
     assert "LLM service unavailable." in agent.send.call_args_list[0][0][0].body
 
-    # General Exception
-    agent.send.reset_mock()
-    mock_httpx_client.stream = MagicMock(side_effect=Exception("Boom"))
-    await agent.handle_message(msg)
-    assert "Error processing the request." in agent.send.call_args_list[0][0][0].body
-
 
 @pytest.mark.asyncio
 async def test_llm_json_error(mock_httpx_client, mock_settings):
@@ -131,13 +163,7 @@ async def test_llm_json_error(mock_httpx_client, mock_settings):
     agent.logger = MagicMock()
 
     prompt = LLMPromptMessage(text="Hi", norms=[], goals=[])
-    msg = InternalMessage(
-        to="llm",
-        sender=mock_settings.agent_settings.bdi_core_name,
-        body=prompt.model_dump_json(),
-        thread="prompt_message",
-    )
-    await agent.handle_message(msg)
+    await agent._process_bdi_message(prompt)
 
     agent.logger.error.assert_called()  # Should log JSONDecodeError
 

test/unit/agents/user_interrupt/test_user_interrupt.py

@@ -535,3 +535,158 @@ async def test_send_experiment_control_unknown(agent):
     agent.send.assert_awaited()
     msg = agent.send.call_args[0][0]
     assert msg.thread == ""
+
+
+@pytest.mark.asyncio
+async def test_create_mapping_recursive_goals(agent):
+    """Verify that nested subgoals are correctly registered in the mapping."""
+    import uuid
+
+    # 1. Setup IDs
+    parent_goal_id = uuid.uuid4()
+    child_goal_id = uuid.uuid4()
+
+    # 2. Create the child goal
+    child_goal = Goal(
+        id=child_goal_id,
+        name="child_goal",
+        description="I am a subgoal",
+        plan=Plan(id=uuid.uuid4(), name="p_child", steps=[]),
+    )
+
+    # 3. Create the parent goal and put the child goal inside its plan steps
+    parent_goal = Goal(
+        id=parent_goal_id,
+        name="parent_goal",
+        description="I am a parent",
+        plan=Plan(id=uuid.uuid4(), name="p_parent", steps=[child_goal]),  # Nested here
+    )
+
+    # 4. Build the program
+    phase = Phase(
+        id=uuid.uuid4(),
+        name="phase1",
+        norms=[],
+        goals=[parent_goal],  # Only the parent is top-level
+        triggers=[],
+    )
+    prog = Program(phases=[phase])
+
+    # 5. Execute mapping
+    msg = InternalMessage(to="me", thread="new_program", body=prog.model_dump_json())
+    await agent.handle_message(msg)
+
+    # 6. Assertions
+    # Check parent
+    assert str(parent_goal_id) in agent._goal_map
+    assert agent._goal_map[str(parent_goal_id)] == "parent_goal"
+
+    # Check child (This confirms the recursion worked)
+    assert str(child_goal_id) in agent._goal_map
+    assert agent._goal_map[str(child_goal_id)] == "child_goal"
+    assert agent._goal_reverse_map["child_goal"] == str(child_goal_id)
+
+
+@pytest.mark.asyncio
+async def test_receive_loop_advanced_scenarios(agent):
+    """
+    Covers:
+    - JSONDecodeError (lines 86-88)
+    - Override: Trigger found (lines 108-109)
+    - Override: Norm found (lines 114-115)
+    - Override: Nothing found (line 134)
+    - Override Unachieve: Success & Fail (lines 136-145)
+    - Pause: Context true/false logs (lines 150-157)
+    - Next Phase (line 160)
+    """
+    # 1. Setup Data Maps
+    agent._trigger_map["101"] = "trigger_slug"
+    agent._cond_norm_map["202"] = "norm_slug"
+
+    # 2. Define Payloads
+    # A. Invalid JSON
+    bad_json = b"INVALID{JSON"
+
+    # B. Override -> Trigger
+    override_trigger = json.dumps({"type": "override", "context": "101"}).encode()
+
+    # C. Override -> Norm
+    override_norm = json.dumps({"type": "override", "context": "202"}).encode()
+
+    # D. Override -> Unknown
+    override_fail = json.dumps({"type": "override", "context": "999"}).encode()
+
+    # E. Unachieve -> Success
+    unachieve_success = json.dumps({"type": "override_unachieve", "context": "202"}).encode()
+
+    # F. Unachieve -> Fail
+    unachieve_fail = json.dumps({"type": "override_unachieve", "context": "999"}).encode()
+
+    # G. Pause (True)
+    pause_true = json.dumps({"type": "pause", "context": "true"}).encode()
+
+    # H. Pause (False/Resume)
+    pause_false = json.dumps({"type": "pause", "context": ""}).encode()
+
+    # I. Next Phase
+    next_phase = json.dumps({"type": "next_phase", "context": ""}).encode()
+
+    # 3. Setup Socket
+    agent.sub_socket.recv_multipart.side_effect = [
+        (b"topic", bad_json),
+        (b"topic", override_trigger),
+        (b"topic", override_norm),
+        (b"topic", override_fail),
+        (b"topic", unachieve_success),
+        (b"topic", unachieve_fail),
+        (b"topic", pause_true),
+        (b"topic", pause_false),
+        (b"topic", next_phase),
+        asyncio.CancelledError,  # End loop
+    ]
+
+    # Mock internal helpers to verify calls
+    agent._send_to_bdi = AsyncMock()
+    agent._send_to_bdi_belief = AsyncMock()
+    agent._send_pause_command = AsyncMock()
+    agent._send_experiment_control_to_bdi_core = AsyncMock()
+
+    # 4. Run Loop
+    try:
+        await agent._receive_button_event()
+    except asyncio.CancelledError:
+        pass
+
+    # 5. Assertions
+
+    # JSON Error
+    agent.logger.error.assert_called_with("Received invalid JSON payload on topic %s", b"topic")
+
+    # Override Trigger
+    agent._send_to_bdi.assert_awaited_with("force_trigger", "trigger_slug")
+
+    # Override Norm
+    # We expect _send_to_bdi_belief to be called for the norm
+    # Note: The loop calls _send_to_bdi_belief(asl_cond_norm, "cond_norm")
+    agent._send_to_bdi_belief.assert_any_call("norm_slug", "cond_norm")
+
+    # Override Fail (Warning log)
+    agent.logger.warning.assert_any_call("Could not determine which element to override.")
+
+    # Unachieve Success
+    # Loop calls _send_to_bdi_belief(asl_cond_norm, "cond_norm", True)
+    agent._send_to_bdi_belief.assert_any_call("norm_slug", "cond_norm", True)
+
+    # Unachieve Fail
+    agent.logger.warning.assert_any_call("Could not determine which conditional norm to unachieve.")
+
+    # Pause Logic
+    agent._send_pause_command.assert_any_call("true")
+    agent.logger.info.assert_any_call("Sent pause command.")
+
+    # Resume Logic
+    agent._send_pause_command.assert_any_call("")
+    agent.logger.info.assert_any_call("Sent resume command.")
+
+    # Next Phase
+    agent._send_experiment_control_to_bdi_core.assert_awaited_with("next_phase")