feat: added recursive mapping

ref: N25B-400

.gitignore

@@ -224,7 +224,7 @@ docs/*
 
 # Generated files
 agentspeak.asl
-
+experiment-*.log
 
 
 

.logging_config.yaml

@@ -1,36 +1,57 @@
 version: 1
 
 custom_levels:
-  OBSERVATION: 25
-  ACTION: 26
+  OBSERVATION: 24
+  ACTION: 25
+  CHAT: 26
   LLM: 9
 
 formatters:
   # Console output
   colored:
-    (): "colorlog.ColoredFormatter"
+    class: colorlog.ColoredFormatter
     format: "{log_color}{asctime}.{msecs:03.0f} | {levelname:11} | {name:70} | {message}"
     style: "{"
     datefmt: "%H:%M:%S"
 
   # User-facing UI (structured JSON)
-  json_experiment:
-    (): "pythonjsonlogger.jsonlogger.JsonFormatter"
+  json:
+    class: pythonjsonlogger.jsonlogger.JsonFormatter
     format: "{name} {levelname} {levelno} {message} {created} {relativeCreated}"
     style: "{"
 
+  # Experiment stream for console and file output, with optional `role` field
+  experiment:
+    class: control_backend.logging.OptionalFieldFormatter
+    format: "%(asctime)s %(levelname)s %(role?)s %(message)s"
+    defaults:
+      role: "-"
+
+filters:
+  # Filter out any log records that have the extra field "partial" set to True, indicating that they
+  #  will be replaced later.
+  partial:
+    (): control_backend.logging.PartialFilter
+
 handlers:
   console:
     class: logging.StreamHandler
     level: DEBUG
     formatter: colored
+    filters: [partial]
     stream: ext://sys.stdout
   ui:
     class: zmq.log.handlers.PUBHandler
     level: LLM
-    formatter: json_experiment
+    formatter: json
+  file:
+    class: control_backend.logging.DatedFileHandler
+    formatter: experiment
+    filters: [partial]
+    # Directory must match config.logging_settings.experiment_log_directory
+    file_prefix: experiment_logs/experiment
 
-# Level of external libraries
+# Level for external libraries
 root:
   level: WARN
   handlers: [console]
@@ -39,3 +60,6 @@ loggers:
   control_backend:
     level: LLM
     handlers: [ui]
+  experiment:  # This name must match config.logging_settings.experiment_logger_name
+    level: DEBUG
+    handlers: [ui, file]

pyproject.toml

@@ -7,6 +7,7 @@ requires-python = ">=3.13"
 dependencies = [
   "agentspeak>=0.2.2",
   "colorlog>=6.10.1",
+  "deepface>=0.0.96",
   "fastapi[all]>=0.115.6",
   "mlx-whisper>=0.4.3 ; sys_platform == 'darwin'",
   "numpy>=2.3.3",
@@ -21,6 +22,7 @@ dependencies = [
   "silero-vad>=6.0.0",
   "sphinx>=7.3.7",
   "sphinx-rtd-theme>=3.0.2",
+  "tf-keras>=2.20.1",
   "torch>=2.8.0",
   "uvicorn>=0.37.0",
 ]
@@ -48,6 +50,7 @@ test = [
   "pytest-asyncio>=1.2.0",
   "pytest-cov>=7.0.0",
   "pytest-mock>=3.15.1",
+  "python-slugify>=8.0.4",
   "pyyaml>=6.0.3",
   "pyzmq>=27.1.0",
   "soundfile>=0.13.1",

src/control_backend/agents/__init__.py

@@ -1 +1,5 @@
+"""
+This package contains all agent implementations for the PepperPlus Control Backend.
+"""
+
 from .base import BaseAgent as BaseAgent

src/control_backend/agents/actuation/__init__.py

@@ -1,2 +1,6 @@
+"""
+Agents responsible for controlling the robot's physical actions, such as speech and gestures.
+"""
+
 from .robot_gesture_agent import RobotGestureAgent as RobotGestureAgent
 from .robot_speech_agent import RobotSpeechAgent as RobotSpeechAgent

src/control_backend/agents/actuation/robot_gesture_agent.py

@@ -1,4 +1,5 @@
 import json
+import logging
 
 import zmq
 import zmq.asyncio as azmq
@@ -8,6 +9,8 @@ from control_backend.core.agent_system import InternalMessage
 from control_backend.core.config import settings
 from control_backend.schemas.ri_message import GestureCommand, RIEndpoint
 
+experiment_logger = logging.getLogger(settings.logging_settings.experiment_logger_name)
+
 
 class RobotGestureAgent(BaseAgent):
     """
@@ -111,6 +114,7 @@ class RobotGestureAgent(BaseAgent):
                         gesture_command.data,
                     )
                     return
+            experiment_logger.action("Gesture: %s", gesture_command.data)
             await self.pubsocket.send_json(gesture_command.model_dump())
         except Exception:
             self.logger.exception("Error processing internal message.")

src/control_backend/agents/base.py

@@ -1,9 +1,10 @@
 import logging
+from abc import ABC
 
 from control_backend.core.agent_system import BaseAgent as CoreBaseAgent
 
 
-class BaseAgent(CoreBaseAgent):
+class BaseAgent(CoreBaseAgent, ABC):
     """
     The primary base class for all implementation agents.
 

src/control_backend/agents/bdi/__init__.py

@@ -1,8 +1,10 @@
+"""
+Agents and utilities for the BDI (Belief-Desire-Intention) reasoning system,
+implementing AgentSpeak(L) logic.
+"""
+
 from control_backend.agents.bdi.bdi_core_agent import BDICoreAgent as BDICoreAgent
 
-from .belief_collector_agent import (
-    BDIBeliefCollectorAgent as BDIBeliefCollectorAgent,
-)
 from .text_belief_extractor_agent import (
     TextBeliefExtractorAgent as TextBeliefExtractorAgent,
 )

src/control_backend/agents/bdi/agentspeak_ast.py

@@ -77,10 +77,10 @@ class AstTerm(AstExpression, ABC):
         return AstBinaryOp(self, BinaryOperatorType.NOT_EQUALS, _coalesce_expr(other))
 
 
-@dataclass
+@dataclass(eq=False)
 class AstAtom(AstTerm):
     """
-    Grounded expression in all lowercase.
+    Represents a grounded atom in AgentSpeak (e.g., lowercase constants).
     """
 
     value: str
@@ -89,10 +89,10 @@ class AstAtom(AstTerm):
         return self.value.lower()
 
 
-@dataclass
+@dataclass(eq=False)
 class AstVar(AstTerm):
     """
-    Ungrounded variable expression. First letter capitalized.
+    Represents an ungrounded variable in AgentSpeak (e.g., capitalized names).
     """
 
     name: str
@@ -101,24 +101,36 @@ class AstVar(AstTerm):
         return self.name.capitalize()
 
 
-@dataclass
+@dataclass(eq=False)
 class AstNumber(AstTerm):
+    """
+    Represents a numeric constant in AgentSpeak.
+    """
+
     value: int | float
 
     def _to_agentspeak(self) -> str:
         return str(self.value)
 
 
-@dataclass
+@dataclass(eq=False)
 class AstString(AstTerm):
+    """
+    Represents a string literal in AgentSpeak.
+    """
+
     value: str
 
     def _to_agentspeak(self) -> str:
         return f'"{self.value}"'
 
 
-@dataclass
+@dataclass(eq=False)
 class AstLiteral(AstTerm):
+    """
+    Represents a literal (functor and terms) in AgentSpeak.
+    """
+
     functor: str
     terms: list[AstTerm] = field(default_factory=list)
 
@@ -142,6 +154,10 @@ class BinaryOperatorType(StrEnum):
 
 @dataclass
 class AstBinaryOp(AstExpression):
+    """
+    Represents a binary logical or relational operation in AgentSpeak.
+    """
+
     left: AstExpression
     operator: BinaryOperatorType
     right: AstExpression
@@ -167,6 +183,10 @@ class AstBinaryOp(AstExpression):
 
 @dataclass
 class AstLogicalExpression(AstExpression):
+    """
+    Represents a logical expression, potentially negated, in AgentSpeak.
+    """
+
     expression: AstExpression
     negated: bool = False
 
@@ -208,6 +228,10 @@ class AstStatement(AstNode):
 
 @dataclass
 class AstRule(AstNode):
+    """
+    Represents an inference rule in AgentSpeak. If there is no condition, it always holds.
+    """
+
     result: AstExpression
     condition: AstExpression | None = None
 
@@ -231,6 +255,10 @@ class TriggerType(StrEnum):
 
 @dataclass
 class AstPlan(AstNode):
+    """
+    Represents a plan in AgentSpeak, consisting of a trigger, context, and body.
+    """
+
     type: TriggerType
     trigger_literal: AstExpression
     context: list[AstExpression]
@@ -260,6 +288,10 @@ class AstPlan(AstNode):
 
 @dataclass
 class AstProgram(AstNode):
+    """
+    Represents a full AgentSpeak program, consisting of rules and plans.
+    """
+
     rules: list[AstRule] = field(default_factory=list)
     plans: list[AstPlan] = field(default_factory=list)
 

src/control_backend/agents/bdi/agentspeak_generator.py

@@ -22,6 +22,7 @@ from control_backend.schemas.program import (
     BaseGoal,
     BasicNorm,
     ConditionalNorm,
+    EmotionBelief,
     GestureAction,
     Goal,
     InferredBelief,
@@ -40,9 +41,23 @@ from control_backend.schemas.program import (
 
 
 class AgentSpeakGenerator:
+    """
+    Generator class that translates a high-level :class:`~control_backend.schemas.program.Program`
+    into AgentSpeak(L) source code.
+
+    It handles the conversion of phases, norms, goals, and triggers into AgentSpeak rules and plans,
+    ensuring the robot follows the defined behavioral logic.
+    """
+
     _asp: AstProgram
 
     def generate(self, program: Program) -> str:
+        """
+        Translates a Program object into an AgentSpeak source string.
+
+        :param program: The behavior program to translate.
+        :return: The generated AgentSpeak code as a string.
+        """
         self._asp = AstProgram()
 
         if program.phases:
@@ -445,6 +460,10 @@ class AgentSpeakGenerator:
     @_astify.register
     def _(self, sb: SemanticBelief) -> AstExpression:
         return AstLiteral(self.slugify(sb))
+    
+    @_astify.register
+    def _(self, eb: EmotionBelief) -> AstExpression:
+        return AstLiteral("emotion_detected", [AstAtom(eb.emotion)])
 
     @_astify.register
     def _(self, ib: InferredBelief) -> AstExpression:

src/control_backend/agents/bdi/bdi_core_agent.py

@@ -1,6 +1,7 @@
 import asyncio
 import copy
 import json
+import logging
 import time
 from collections.abc import Iterable
 
@@ -19,6 +20,9 @@ from control_backend.schemas.ri_message import GestureCommand, RIEndpoint, Speec
 DELIMITER = ";\n"  # TODO: temporary until we support lists in AgentSpeak
 
 
+experiment_logger = logging.getLogger(settings.logging_settings.experiment_logger_name)
+
+
 class BDICoreAgent(BaseAgent):
     """
     BDI Core Agent.
@@ -167,7 +171,7 @@ class BDICoreAgent(BaseAgent):
                     case "force_next_phase":
                         self._force_next_phase()
                     case _:
-                        self.logger.warning("Received unknow user interruption: %s", msg)
+                        self.logger.warning("Received unknown user interruption: %s", msg)
 
     def _apply_belief_changes(self, belief_changes: BeliefMessage):
         """
@@ -207,6 +211,9 @@ class BDICoreAgent(BaseAgent):
         else:
             term = agentspeak.Literal(name)
 
+        if name != "user_said":
+            experiment_logger.observation(f"Formed new belief: {name}{f'={args}' if args else ''}")
+
         self.bdi_agent.call(
             agentspeak.Trigger.addition,
             agentspeak.GoalType.belief,
@@ -244,6 +251,9 @@ class BDICoreAgent(BaseAgent):
             new_args = (agentspeak.Literal(arg) for arg in args)
             term = agentspeak.Literal(name, new_args)
 
+        if name != "user_said":
+            experiment_logger.observation(f"Removed belief: {name}{f'={args}' if args else ''}")
+
         result = self.bdi_agent.call(
             agentspeak.Trigger.removal,
             agentspeak.GoalType.belief,
@@ -338,7 +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.
@@ -386,6 +396,8 @@ class BDICoreAgent(BaseAgent):
                 body=str(message_text),
             )
 
+            experiment_logger.chat(str(message_text), extra={"role": "assistant"})
+
             self.add_behavior(self.send(chat_history_message))
 
             yield
@@ -441,6 +453,7 @@ class BDICoreAgent(BaseAgent):
             trigger_name = agentspeak.grounded(term.args[0], intention.scope)
 
             self.logger.debug("Started trigger %s", trigger_name)
+            experiment_logger.observation("Triggered: %s", trigger_name)
 
             msg = InternalMessage(
                 to=settings.agent_settings.user_interrupt_name,
@@ -512,10 +525,6 @@ 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/belief_collector_agent.py

@@ -1,152 +0,0 @@
-import json
-
-from pydantic import ValidationError
-
-from control_backend.agents.base import BaseAgent
-from control_backend.core.agent_system import InternalMessage
-from control_backend.core.config import settings
-from control_backend.schemas.belief_message import Belief, BeliefMessage
-
-
-class BDIBeliefCollectorAgent(BaseAgent):
-    """
-    BDI Belief Collector Agent.
-
-    This agent acts as a central aggregator for beliefs derived from various sources (e.g., text,
-    emotion, vision). It receives raw extracted data from other agents,
-    normalizes them into valid :class:`Belief` objects, and forwards them as a unified packet to the
-    BDI Core Agent.
-
-    It serves as a funnel to ensure the BDI agent receives a consistent stream of beliefs.
-    """
-
-    async def setup(self):
-        """
-        Initialize the agent.
-        """
-        self.logger.info("Setting up %s", self.name)
-
-    async def handle_message(self, msg: InternalMessage):
-        """
-        Handle incoming messages from other extractor agents.
-
-        Routes the message to specific handlers based on the 'type' field in the JSON body.
-        Supported types:
-        -   ``belief_extraction_text``: Handled by :meth:`_handle_belief_text`
-        -   ``emotion_extraction_text``: Handled by :meth:`_handle_emo_text`
-
-        :param msg: The received internal message.
-        """
-        sender_node = msg.sender
-
-        # Parse JSON payload
-        try:
-            payload = json.loads(msg.body)
-        except Exception as e:
-            self.logger.warning(
-                "BeliefCollector: failed to parse JSON from %s. Body=%r Error=%s",
-                sender_node,
-                msg.body,
-                e,
-            )
-            return
-
-        msg_type = payload.get("type")
-
-        # Prefer explicit 'type' field
-        if msg_type == "belief_extraction_text":
-            self.logger.debug("Message routed to _handle_belief_text (sender=%s)", sender_node)
-            await self._handle_belief_text(payload, sender_node)
-        # This is not implemented yet, but we keep the structure for future use
-        elif msg_type == "emotion_extraction_text":
-            self.logger.debug("Message routed to _handle_emo_text (sender=%s)", sender_node)
-            await self._handle_emo_text(payload, sender_node)
-        else:
-            self.logger.warning(
-                "Unrecognized message (sender=%s, type=%r). Ignoring.", sender_node, msg_type
-            )
-
-    async def _handle_belief_text(self, payload: dict, origin: str):
-        """
-        Process text-based belief extraction payloads.
-
-        Expected payload format::
-
-            {
-                "type": "belief_extraction_text",
-                "beliefs": {
-                    "user_said": ["Can you help me?"],
-                    "intention": ["ask_help"]
-                }
-            }
-
-        Validates and converts the dictionary items into :class:`Belief` objects.
-
-        :param payload: The dictionary payload containing belief data.
-        :param origin: The name of the sender agent.
-        """
-        beliefs = payload.get("beliefs", {})
-
-        if not beliefs:
-            self.logger.debug("Received empty beliefs set.")
-            return
-
-        def try_create_belief(name, arguments) -> Belief | None:
-            """
-            Create a belief object from name and arguments, or return None silently if the input is
-            not correct.
-
-            :param name: The name of the belief.
-            :param arguments: The arguments of the belief.
-            :return: A Belief object if the input is valid or None.
-            """
-            try:
-                return Belief(name=name, arguments=arguments)
-            except ValidationError:
-                return None
-
-        beliefs = [
-            belief
-            for name, arguments in beliefs.items()
-            if (belief := try_create_belief(name, arguments)) is not None
-        ]
-
-        self.logger.debug("Forwarding %d beliefs.", len(beliefs))
-        for belief in beliefs:
-            for argument in belief.arguments:
-                self.logger.debug("  - %s %s", belief.name, argument)
-
-        await self._send_beliefs_to_bdi(beliefs, origin=origin)
-
-    async def _handle_emo_text(self, payload: dict, origin: str):
-        """
-        Process emotion extraction payloads.
-
-        **TODO**: Implement this method once emotion recognition is integrated.
-
-        :param payload: The dictionary payload containing emotion data.
-        :param origin: The name of the sender agent.
-        """
-        pass
-
-    async def _send_beliefs_to_bdi(self, beliefs: list[Belief], origin: str | None = None):
-        """
-        Send a list of aggregated beliefs to the BDI Core Agent.
-
-        Wraps the beliefs in a :class:`BeliefMessage` and sends it via the 'beliefs' thread.
-
-        :param beliefs: The list of Belief objects to send.
-        :param origin: (Optional) The original source of the beliefs (unused currently).
-        """
-        if not beliefs:
-            return
-
-        msg = InternalMessage(
-            to=settings.agent_settings.bdi_core_name,
-            sender=self.name,
-            body=BeliefMessage(create=beliefs).model_dump_json(),
-            thread="beliefs",
-        )
-
-        await self.send(msg)
-        self.logger.info("Sent %d belief(s) to BDI core.", len(beliefs))

src/control_backend/agents/bdi/text_belief_extractor_agent.py

@@ -18,6 +18,12 @@ type JSONLike = None | bool | int | float | str | list["JSONLike"] | dict[str, "
 
 
 class BeliefState(BaseModel):
+    """
+    Represents the state of inferred semantic beliefs.
+
+    Maintains sets of beliefs that are currently considered true or false.
+    """
+
     true: set[InternalBelief] = set()
     false: set[InternalBelief] = set()
 
@@ -312,6 +318,9 @@ 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}],
@@ -338,7 +347,7 @@ class TextBeliefExtractorAgent(BaseAgent):
 
 class SemanticBeliefInferrer:
     """
-    Class that handles only prompting an LLM for semantic beliefs.
+    Infers semantic beliefs from conversation history using an LLM.
     """
 
     def __init__(
@@ -464,6 +473,10 @@ Respond with a JSON similar to the following, but with the property names as giv
 
 
 class GoalAchievementInferrer(SemanticBeliefInferrer):
+    """
+    Infers whether specific conversational goals have been achieved using an LLM.
+    """
+
     def __init__(self, llm: TextBeliefExtractorAgent.LLM):
         super().__init__(llm)
         self.goals: set[BaseGoal] = set()

src/control_backend/agents/communication/__init__.py

@@ -1 +1,5 @@
+"""
+Agents responsible for external communication and service discovery.
+"""
+
 from .ri_communication_agent import RICommunicationAgent as RICommunicationAgent

src/control_backend/agents/communication/ri_communication_agent.py

@@ -8,6 +8,9 @@ from zmq.asyncio import Context
 
 from control_backend.agents import BaseAgent
 from control_backend.agents.actuation.robot_gesture_agent import RobotGestureAgent
+from control_backend.agents.perception.visual_emotion_recognition_agent.visual_emotion_recognition_agent import (  # noqa
+    VisualEmotionRecognitionAgent,
+)
 from control_backend.core.config import settings
 from control_backend.schemas.internal_message import InternalMessage
 from control_backend.schemas.ri_message import PauseCommand
@@ -52,6 +55,7 @@ class RICommunicationAgent(BaseAgent):
         self.connected = False
         self.gesture_agent: RobotGestureAgent | None = None
         self.speech_agent: RobotSpeechAgent | None = None
+        self.visual_emotion_recognition_agent: VisualEmotionRecognitionAgent | None = None
 
     async def setup(self):
         """
@@ -209,6 +213,14 @@ class RICommunicationAgent(BaseAgent):
                 case "audio":
                     vad_agent = VADAgent(audio_in_address=addr, audio_in_bind=bind)
                     await vad_agent.start()
+                case "video":
+                    visual_emotion_agent = VisualEmotionRecognitionAgent(
+                        settings.agent_settings.visual_emotion_recognition_name,
+                        socket_address=addr,
+                        bind=bind,
+                    )
+                    self.visual_emotion_recognition_agent = visual_emotion_agent
+                    await visual_emotion_agent.start()
                 case _:
                     self.logger.warning("Unhandled negotiation id: %s", id)
 
@@ -313,6 +325,9 @@ class RICommunicationAgent(BaseAgent):
 
         if self.speech_agent is not None:
             await self.speech_agent.stop()
+        
+        if self.visual_emotion_recognition_agent is not None:
+            await self.visual_emotion_recognition_agent.stop()
 
         if self.pub_socket is not None:
             self.pub_socket.close()
@@ -322,9 +337,10 @@ class RICommunicationAgent(BaseAgent):
             self.connected = True
 
     async def handle_message(self, msg: InternalMessage):
+        return 
         try:
             pause_command = PauseCommand.model_validate_json(msg.body)
-            self._req_socket.send_json(pause_command.model_dump())
-            self.logger.debug(self._req_socket.recv_json())
+            await self._req_socket.send_json(pause_command.model_dump())
+            self.logger.debug(await self._req_socket.recv_json())
         except ValidationError:
             self.logger.warning("Incorrect message format for PauseCommand.")

src/control_backend/agents/llm/__init__.py

@@ -1 +1,5 @@
+"""
+Agents that interface with Large Language Models for natural language processing and generation.
+"""
+
 from .llm_agent import LLMAgent as LLMAgent

src/control_backend/agents/llm/llm_agent.py

@@ -1,4 +1,6 @@
+import asyncio
 import json
+import logging
 import re
 import uuid
 from collections.abc import AsyncGenerator
@@ -13,6 +15,8 @@ from control_backend.core.config import settings
 from ...schemas.llm_prompt_message import LLMPromptMessage
 from .llm_instructions import LLMInstructions
 
+experiment_logger = logging.getLogger(settings.logging_settings.experiment_logger_name)
+
 
 class LLMAgent(BaseAgent):
     """
@@ -32,6 +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)
@@ -50,13 +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.")
@@ -73,12 +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):
         """
@@ -132,7 +173,7 @@ class LLMAgent(BaseAgent):
             *self.history,
         ]
 
-        message_id = str(uuid.uuid4())  # noqa
+        message_id = str(uuid.uuid4())
 
         try:
             full_message = ""
@@ -141,10 +182,9 @@ class LLMAgent(BaseAgent):
                 full_message += token
                 current_chunk += token
 
-                self.logger.llm(
-                    "Received token: %s",
+                experiment_logger.chat(
                     full_message,
-                    extra={"reference": message_id},  # Used in the UI to update old logs
+                    extra={"role": "assistant", "reference": message_id, "partial": True},
                 )
 
                 # Stream the message in chunks separated by punctuation.
@@ -160,11 +200,9 @@ class LLMAgent(BaseAgent):
             if current_chunk:
                 yield current_chunk
 
-            self.history.append(
-                {
-                    "role": "assistant",
-                    "content": full_message,
-                }
+            experiment_logger.chat(
+                full_message,
+                extra={"role": "assistant", "reference": message_id, "partial": False},
             )
         except httpx.HTTPError as err:
             self.logger.error("HTTP error.", exc_info=err)
@@ -181,10 +219,13 @@ class LLMAgent(BaseAgent):
         :yield: Raw text tokens (deltas) from the SSE stream.
         :raises httpx.HTTPError: If the API returns a non-200 status.
         """
-        async with httpx.AsyncClient() as client:
+        async with httpx.AsyncClient(timeout=httpx.Timeout(20.0)) as client:
             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/__init__.py

@@ -1,3 +1,8 @@
+"""
+Agents responsible for processing sensory input, such as audio transcription and voice activity
+detection.
+"""
+
 from .transcription_agent.transcription_agent import (
     TranscriptionAgent as TranscriptionAgent,
 )

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/perception/transcription_agent/transcription_agent.py

@@ -1,4 +1,5 @@
 import asyncio
+import logging
 
 import numpy as np
 import zmq
@@ -10,6 +11,8 @@ from control_backend.core.config import settings
 
 from .speech_recognizer import SpeechRecognizer
 
+experiment_logger = logging.getLogger(settings.logging_settings.experiment_logger_name)
+
 
 class TranscriptionAgent(BaseAgent):
     """
@@ -25,6 +28,8 @@ class TranscriptionAgent(BaseAgent):
     :ivar audio_in_socket: The ZMQ SUB socket instance.
     :ivar speech_recognizer: The speech recognition engine instance.
     :ivar _concurrency: Semaphore to limit concurrent transcriptions.
+    :ivar _current_speech_reference: The reference of the current user utterance, for synchronising
+        experiment logs.
     """
 
     def __init__(self, audio_in_address: str):
@@ -39,6 +44,7 @@ class TranscriptionAgent(BaseAgent):
         self.audio_in_socket: azmq.Socket | None = None
         self.speech_recognizer = None
         self._concurrency = None
+        self._current_speech_reference: str | None = None
 
     async def setup(self):
         """
@@ -63,6 +69,10 @@ class TranscriptionAgent(BaseAgent):
 
         self.logger.info("Finished setting up %s", self.name)
 
+    async def handle_message(self, msg: InternalMessage):
+        if msg.thread == "voice_activity":
+            self._current_speech_reference = msg.body
+
     async def stop(self):
         """
         Stop the agent and close sockets.
@@ -74,7 +84,7 @@ class TranscriptionAgent(BaseAgent):
 
     def _connect_audio_in_socket(self):
         """
-        Helper to connect the ZMQ SUB socket for audio input.
+        Connects the ZMQ SUB socket for receiving audio data.
         """
         self.audio_in_socket = azmq.Context.instance().socket(zmq.SUB)
         self.audio_in_socket.setsockopt_string(zmq.SUBSCRIBE, "")
@@ -96,24 +106,25 @@ class TranscriptionAgent(BaseAgent):
 
     async def _share_transcription(self, transcription: str):
         """
-        Share a transcription to the other agents that depend on it.
+        Share a transcription to the other agents that depend on it, and to experiment logs.
 
         Currently sends to:
         - :attr:`settings.agent_settings.text_belief_extractor_name`
+        - The UI via the experiment logger
 
         :param transcription: The transcribed text.
         """
-        receiver_names = [
-            settings.agent_settings.text_belief_extractor_name,
-        ]
+        experiment_logger.chat(
+            transcription,
+            extra={"role": "user", "reference": self._current_speech_reference, "partial": False},
+        )
 
-        for receiver_name in receiver_names:
-            message = InternalMessage(
-                to=receiver_name,
-                sender=self.name,
-                body=transcription,
-            )
-            await self.send(message)
+        message = InternalMessage(
+            to=settings.agent_settings.text_belief_extractor_name,
+            sender=self.name,
+            body=transcription,
+        )
+        await self.send(message)
 
     async def _transcribing_loop(self) -> None:
         """
@@ -129,10 +140,9 @@ class TranscriptionAgent(BaseAgent):
                 audio = np.frombuffer(audio_data, dtype=np.float32)
                 speech = await self._transcribe(audio)
                 if not speech:
-                    self.logger.info("Nothing transcribed.")
+                    self.logger.debug("Nothing transcribed.")
                     continue
 
-                self.logger.info("Transcribed speech: %s", speech)
                 await self._share_transcription(speech)
             except Exception as e:
                 self.logger.error(f"Error in transcription loop: {e}")

src/control_backend/agents/perception/vad_agent.py

@@ -1,4 +1,6 @@
 import asyncio
+import logging
+import uuid
 
 import numpy as np
 import torch
@@ -12,6 +14,8 @@ from control_backend.schemas.internal_message import InternalMessage
 from ...schemas.program_status import PROGRAM_STATUS, ProgramStatus
 from .transcription_agent.transcription_agent import TranscriptionAgent
 
+experiment_logger = logging.getLogger(settings.logging_settings.experiment_logger_name)
+
 
 class SocketPoller[T]:
     """
@@ -252,6 +256,18 @@ class VADAgent(BaseAgent):
             if prob > prob_threshold:
                 if self.i_since_speech > non_speech_patience + begin_silence_length:
                     self.logger.debug("Speech started.")
+                    reference = str(uuid.uuid4())
+                    experiment_logger.chat(
+                        "...",
+                        extra={"role": "user", "reference": reference, "partial": True},
+                    )
+                    await self.send(
+                        InternalMessage(
+                            to=settings.agent_settings.transcription_name,
+                            body=reference,
+                            thread="voice_activity",
+                        )
+                    )
                 self.audio_buffer = np.append(self.audio_buffer, chunk)
                 self.i_since_speech = 0
                 continue

src/control_backend/agents/perception/visual_emotion_recognition_agent/visual_emotion_recognition_agent.py

@@ -0,0 +1,167 @@
+import json
+import time
+from collections import Counter, defaultdict
+
+import cv2
+import numpy as np
+import zmq
+import zmq.asyncio as azmq
+from pydantic_core import ValidationError
+
+from control_backend.agents import BaseAgent
+from control_backend.agents.perception.visual_emotion_recognition_agent.visual_emotion_recognizer import (  # noqa
+    DeepFaceEmotionRecognizer,
+)
+from control_backend.core.agent_system import InternalMessage
+from control_backend.core.config import settings
+from control_backend.schemas.belief_message import Belief
+
+
+class VisualEmotionRecognitionAgent(BaseAgent):
+    def __init__(
+        self,
+        name: str,
+        socket_address: str,
+        bind: bool = False,
+        timeout_ms: int = 1000,
+        window_duration: int = settings.behaviour_settings.visual_emotion_recognition_window_duration_s,  # noqa
+        min_frames_required: int = settings.behaviour_settings.visual_emotion_recognition_min_frames_per_face,  # noqa
+    ):
+        """
+        Initialize the Visual Emotion Recognition Agent.
+
+        :param name: Name of the agent
+        :param socket_address: Address of the socket to connect or bind to
+        :param bind: Whether to bind to the socket address (True) or connect (False)
+        :param timeout_ms: Timeout for socket receive operations in milliseconds
+        :param window_duration: Duration in seconds over which to aggregate emotions
+        :param min_frames_required: Minimum number of frames per face required to consider a face
+        valid
+        """
+        super().__init__(name)
+        self.socket_address = socket_address
+        self.socket_bind = bind
+        self.timeout_ms = timeout_ms
+        self.window_duration = window_duration
+        self.min_frames_required = min_frames_required
+
+    async def setup(self):
+        """
+        Initialize the agent resources.
+        1. Initializes the :class:`VisualEmotionRecognizer`.
+        2. Connects to the video input ZMQ socket.
+        3. Starts the background emotion recognition loop.
+        """
+        self.logger.info("Setting up %s.", self.name)
+
+        self.emotion_recognizer = DeepFaceEmotionRecognizer()
+
+        self.video_in_socket = azmq.Context.instance().socket(zmq.SUB)
+
+        if self.socket_bind:
+            self.video_in_socket.bind(self.socket_address)
+        else:
+            self.video_in_socket.connect(self.socket_address)
+
+        self.video_in_socket.setsockopt_string(zmq.SUBSCRIBE, "")
+        self.video_in_socket.setsockopt(zmq.RCVTIMEO, self.timeout_ms)
+        self.video_in_socket.setsockopt(zmq.CONFLATE, 1)
+
+        self.add_behavior(self.emotion_update_loop())
+
+    async def emotion_update_loop(self):
+        """
+        Background loop to receive video frames, recognize emotions, and update beliefs.
+        1. Receives video frames from the ZMQ socket.
+        2. Uses the :class:`VisualEmotionRecognizer` to detect emotions.
+        3. Aggregates emotions over a time window.
+        4. Sends updates to the BDI Core Agent about detected emotions.
+        """
+        # Next time to process the window and update emotions
+        next_window_time = time.time() + self.window_duration
+
+        # Tracks counts of detected emotions per face index
+        face_stats = defaultdict(Counter)
+
+        prev_dominant_emotions = set()
+
+        while self._running:
+            try:
+                frame_bytes = await self.video_in_socket.recv()
+
+                # Convert bytes to a numpy buffer
+                nparr = np.frombuffer(frame_bytes, np.uint8)
+
+                # Decode image into the generic Numpy Array DeepFace expects
+                frame_image = cv2.imdecode(nparr, cv2.IMREAD_COLOR)
+
+                if frame_image is None:
+                    # Could not decode image, skip this frame
+                    self.logger.warning("Received invalid video frame, skipping.")
+                    continue
+
+                # Get the dominant emotion from each face
+                current_emotions = self.emotion_recognizer.sorted_dominant_emotions(frame_image)
+                # Update emotion counts for each detected face
+                for i, emotion in enumerate(current_emotions):
+                    face_stats[i][emotion] += 1
+
+                # If window duration has passed, process the collected stats
+                if time.time() >= next_window_time:
+                    window_dominant_emotions = set()
+                    # Determine dominant emotion for each face in the window
+                    for _, counter in face_stats.items():
+                        total_detections = sum(counter.values())
+
+                        if total_detections >= self.min_frames_required:
+                            dominant_emotion = counter.most_common(1)[0][0]
+                            window_dominant_emotions.add(dominant_emotion)
+
+                    await self.update_emotions(prev_dominant_emotions, window_dominant_emotions)
+                    prev_dominant_emotions = window_dominant_emotions
+                    face_stats.clear()
+                    next_window_time = time.time() + self.window_duration
+
+            except zmq.Again:
+                self.logger.warning("No video frame received within timeout.")
+
+    async def update_emotions(self, prev_emotions: set[str], emotions: set[str]):
+        """
+        Compare emotions from previous window and current emotions,
+        send updates to BDI Core Agent.
+        """
+        emotions_to_remove = prev_emotions - emotions
+        emotions_to_add = emotions - prev_emotions
+
+        if not emotions_to_add and not emotions_to_remove:
+            return
+
+        emotion_beliefs_remove = []
+        for emotion in emotions_to_remove:
+            self.logger.info(f"Emotion '{emotion}' has disappeared.")
+            try:
+                emotion_beliefs_remove.append(
+                    Belief(name="emotion_detected", arguments=[emotion], remove=True)
+                )
+            except ValidationError:
+                self.logger.warning("Invalid belief for emotion removal: %s", emotion)
+
+        emotion_beliefs_add = []
+        for emotion in emotions_to_add:
+            self.logger.info(f"New emotion detected: '{emotion}'")
+            try:
+                emotion_beliefs_add.append(Belief(name="emotion_detected", arguments=[emotion]))
+            except ValidationError:
+                self.logger.warning("Invalid belief for new emotion: %s", emotion)
+
+        beliefs_list_add = [b.model_dump() for b in emotion_beliefs_add]
+        beliefs_list_remove = [b.model_dump() for b in emotion_beliefs_remove]
+        payload = {"create": beliefs_list_add, "delete": beliefs_list_remove}
+
+        message = InternalMessage(
+            to=settings.agent_settings.bdi_core_name,
+            sender=self.name,
+            body=json.dumps(payload),
+            thread="beliefs",
+        )
+        await self.send(message)

src/control_backend/agents/perception/visual_emotion_recognition_agent/visual_emotion_recognizer.py

@@ -0,0 +1,55 @@
+import abc
+
+import numpy as np
+from deepface import DeepFace
+
+
+class VisualEmotionRecognizer(abc.ABC):
+    @abc.abstractmethod
+    def load_model(self):
+        """Load the visual emotion recognition model into memory."""
+        pass
+
+    @abc.abstractmethod
+    def sorted_dominant_emotions(self, image) -> list[str]:
+        """
+        Recognize dominant emotions from faces in the given image.
+        Emotions can be one of ['angry', 'disgust', 'fear', 'happy', 'sad', 'surprise', 'neutral'].
+        To minimize false positives, consider filtering faces with low confidence.
+
+        :param image: The input image for emotion recognition.
+        :return: List of dominant emotion detected for each face in the image, 
+        sorted per face.
+        """
+        pass
+
+class DeepFaceEmotionRecognizer(VisualEmotionRecognizer):
+    """
+    DeepFace-based implementation of VisualEmotionRecognizer.
+    DeepFape has proven to be quite a pessimistic model, so expect sad, fear and neutral 
+    emotions to be over-represented.
+    """
+    def __init__(self):
+        self.load_model()
+    
+    def load_model(self):
+        print("Loading Deepface Emotion Model...")
+        dummy_img = np.zeros((224, 224, 3), dtype=np.uint8)
+        # analyze does not take a model as an argument, calling it once on a dummy image to load 
+        # the model
+        DeepFace.analyze(dummy_img, actions=['emotion'], enforce_detection=False)
+        print("Deepface Emotion Model loaded.")
+    
+    def sorted_dominant_emotions(self, image) -> list[str]:
+        analysis = DeepFace.analyze(image, 
+                                    actions=['emotion'], 
+                                    enforce_detection=False
+                                    )
+        
+        # Sort faces by x coordinate to maintain left-to-right order
+        analysis.sort(key=lambda face: face['region']['x'])
+
+        analysis = [face for face in analysis if face['face_confidence'] >= 0.90]
+        
+        dominant_emotions = [face['dominant_emotion'] for face in analysis]
+        return dominant_emotions

src/control_backend/agents/user_interrupt/user_interrupt_agent.py

@@ -1,4 +1,5 @@
 import json
+import logging
 
 import zmq
 from zmq.asyncio import Context
@@ -8,7 +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,
@@ -16,6 +17,8 @@ from control_backend.schemas.ri_message import (
     SpeechCommand,
 )
 
+experiment_logger = logging.getLogger(settings.logging_settings.experiment_logger_name)
+
 
 class UserInterruptAgent(BaseAgent):
     """
@@ -50,10 +53,8 @@ class UserInterruptAgent(BaseAgent):
 
     async def setup(self):
         """
-        Initialize the agent.
-
-        Connects the internal ZMQ SUB socket and subscribes to the 'button_pressed' topic.
-        Starts the background behavior to receive the user interrupts.
+        Initialize the agent by setting up ZMQ sockets for receiving button events and
+        publishing updates.
         """
         context = Context.instance()
 
@@ -68,18 +69,15 @@ class UserInterruptAgent(BaseAgent):
 
     async def _receive_button_event(self):
         """
-        The behaviour of the UserInterruptAgent.
-        Continuous loop that receives button_pressed events from the button_pressed HTTP endpoint.
-        These events contain a type and a context.
+        Main loop to receive and process button press events from the UI.
 
-        These are the different types and contexts:
-        - type: "speech", context: string that the robot has to say.
-        - type: "gesture", context: single gesture name that the robot has to perform.
-        - type: "override", context: id that belongs to the goal/trigger/conditional norm.
-        - type: "override_unachieve", context: id that belongs to the conditional norm to unachieve.
-        - type: "next_phase", context: None, indicates to the BDI Core to
-        - type: "pause", context: boolean indicating whether to pause
-        - type: "reset_phase", context: None, indicates to the BDI Core to
+        Handles different event types:
+        - `speech`: Triggers immediate robot speech.
+        - `gesture`: Triggers an immediate robot gesture.
+        - `override`: Forces a belief, trigger, or goal completion in the BDI core.
+        - `override_unachieve`: Removes a belief from the BDI core.
+        - `pause`: Toggles the system's pause state.
+        - `next_phase` / `reset_phase`: Controls experiment flow.
         """
         while True:
             topic, body = await self.sub_socket.recv_multipart()
@@ -172,7 +170,10 @@ class UserInterruptAgent(BaseAgent):
 
     async def handle_message(self, msg: InternalMessage):
         """
-        Handle commands received from other internal Python agents.
+        Handles internal messages from other agents, such as program updates or trigger
+        notifications.
+
+        :param msg: The incoming :class:`~control_backend.core.agent_system.InternalMessage`.
         """
         match msg.thread:
             case "new_program":
@@ -196,6 +197,7 @@ class UserInterruptAgent(BaseAgent):
             case "transition_phase":
                 new_phase_id = msg.body
                 self.logger.info(f"Phase transition detected: {new_phase_id}")
+                experiment_logger.observation("Transitioned to next phase.")
 
                 payload = {"type": "phase_update", "id": new_phase_id}
 
@@ -204,7 +206,7 @@ class UserInterruptAgent(BaseAgent):
                 goal_name = msg.body
                 ui_id = self._goal_reverse_map.get(goal_name)
                 if ui_id:
-                    payload = {"type": "goal_update", "id": ui_id}
+                    payload = {"type": "goal_update", "id": ui_id, "active": True}
                     await self._send_experiment_update(payload)
                     self.logger.info(f"UI Update: Goal {goal_name} started (ID: {ui_id})")
             case "active_norms_update":
@@ -217,8 +219,9 @@ class UserInterruptAgent(BaseAgent):
 
     async def _broadcast_cond_norms(self, active_slugs: list[str]):
         """
-        Sends the current state of all conditional norms to the UI.
-        :param active_slugs: A list of slugs (strings) currently active in the BDI core.
+        Broadcasts the current activation state of all conditional norms to the UI.
+
+        :param active_slugs: A list of sluggified norm names currently active in the BDI core.
         """
         updates = []
         for asl_slug, ui_id in self._cond_norm_reverse_map.items():
@@ -235,7 +238,9 @@ class UserInterruptAgent(BaseAgent):
 
     def _create_mapping(self, program_json: str):
         """
-        Create mappings between UI IDs and ASL slugs for triggers, goals, and conditional norms
+        Creates a bidirectional mapping between UI identifiers and AgentSpeak slugs.
+
+        :param program_json: The JSON representation of the behavioral program.
         """
         try:
             program = Program.model_validate_json(program_json)
@@ -245,6 +250,18 @@ 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)
+
+                # Recursively check steps for subgoals
+                if goal.plan and goal.plan.steps:
+                    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)
@@ -252,8 +269,7 @@ 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}")
@@ -277,8 +293,10 @@ class UserInterruptAgent(BaseAgent):
 
     async def _send_experiment_update(self, data, should_log: bool = True):
         """
-        Sends an update to the 'experiment' topic.
-        The SSE endpoint will pick this up and push it to the UI.
+        Publishes an experiment state update to the internal ZMQ bus for the UI.
+
+        :param data: The update payload.
+        :param should_log: Whether to log the update.
         """
         if self.pub_socket:
             topic = b"experiment"
@@ -293,6 +311,7 @@ class UserInterruptAgent(BaseAgent):
 
         :param text_to_say: The string that the robot has to say.
         """
+        experiment_logger.chat(text_to_say, extra={"role": "user"})
         cmd = SpeechCommand(data=text_to_say, is_priority=True)
         out_msg = InternalMessage(
             to=settings.agent_settings.robot_speech_name,
@@ -358,6 +377,8 @@ class UserInterruptAgent(BaseAgent):
                 thread = "force_next_phase"
             case "reset_phase":
                 thread = "reset_current_phase"
+            case "reset_experiment":
+                thread = "reset_experiment"
             case _:
                 self.logger.warning(
                     "Received unknown experiment control type '%s' to send to BDI Core.",

src/control_backend/api/v1/endpoints/logs.py

@@ -1,8 +1,9 @@
 import logging
+from pathlib import Path
 
 import zmq
-from fastapi import APIRouter
-from fastapi.responses import StreamingResponse
+from fastapi import APIRouter, HTTPException
+from fastapi.responses import FileResponse, StreamingResponse
 from zmq.asyncio import Context
 
 from control_backend.core.config import settings
@@ -38,3 +39,29 @@ async def log_stream():
             yield f"data: {message}\n\n"
 
     return StreamingResponse(gen(), media_type="text/event-stream")
+
+
+LOGGING_DIR = Path(settings.logging_settings.experiment_log_directory).resolve()
+
+
+@router.get("/logs/files")
+@router.get("/api/logs/files")
+async def log_directory():
+    """
+    Get a list of all log files stored in the experiment log file directory.
+    """
+    return [f.name for f in LOGGING_DIR.glob("*.log")]
+
+
+@router.get("/logs/files/{filename}")
+@router.get("/api/logs/files/{filename}")
+async def log_file(filename: str):
+    # Prevent path-traversal
+    file_path = (LOGGING_DIR / filename).resolve()  # This .resolve() is important
+    if not file_path.is_relative_to(LOGGING_DIR):
+        raise HTTPException(status_code=400, detail="Invalid filename.")
+
+    if not file_path.is_file():
+        raise HTTPException(status_code=404, detail="File not found.")
+
+    return FileResponse(file_path, filename=file_path.name)

src/control_backend/api/v1/endpoints/sse.py

@@ -1,12 +0,0 @@
-from fastapi import APIRouter, Request
-
-router = APIRouter()
-
-
-# TODO: implement
-@router.get("/sse")
-async def sse(request: Request):
-    """
-    Placeholder for future Server-Sent Events endpoint.
-    """
-    pass

src/control_backend/api/v1/router.py

@@ -1,13 +1,11 @@
 from fastapi.routing import APIRouter
 
-from control_backend.api.v1.endpoints import logs, message, program, robot, sse, user_interact
+from control_backend.api.v1.endpoints import logs, message, program, robot, user_interact
 
 api_router = APIRouter()
 
 api_router.include_router(message.router, tags=["Messages"])
 
-api_router.include_router(sse.router, tags=["SSE"])
-
 api_router.include_router(robot.router, prefix="/robot", tags=["Pings", "Commands"])
 
 api_router.include_router(logs.router, tags=["Logs"])

src/control_backend/core/agent_system.py

@@ -22,10 +22,22 @@ class AgentDirectory:
 
     @staticmethod
     def register(name: str, agent: "BaseAgent"):
+        """
+        Registers an agent instance with a unique name.
+
+        :param name: The name of the agent.
+        :param agent: The :class:`BaseAgent` instance.
+        """
         _agent_directory[name] = agent
 
     @staticmethod
     def get(name: str) -> "BaseAgent | None":
+        """
+        Retrieves a registered agent instance by name.
+
+        :param name: The name of the agent to retrieve.
+        :return: The :class:`BaseAgent` instance, or None if not found.
+        """
         return _agent_directory.get(name)
 
 

src/control_backend/core/config.py

@@ -35,7 +35,6 @@ class AgentSettings(BaseModel):
     Names of the various agents in the system. These names are used for routing messages.
 
     :ivar bdi_core_name: Name of the BDI Core Agent.
-    :ivar bdi_belief_collector_name: Name of the Belief Collector Agent.
     :ivar bdi_program_manager_name: Name of the BDI Program Manager Agent.
     :ivar text_belief_extractor_name: Name of the Text Belief Extractor Agent.
     :ivar vad_name: Name of the Voice Activity Detection (VAD) Agent.
@@ -50,8 +49,8 @@ class AgentSettings(BaseModel):
 
     # agent names
     bdi_core_name: str = "bdi_core_agent"
-    bdi_belief_collector_name: str = "belief_collector_agent"
     bdi_program_manager_name: str = "bdi_program_manager_agent"
+    visual_emotion_recognition_name: str = "visual_emotion_recognition_agent"
     text_belief_extractor_name: str = "text_belief_extractor_agent"
     vad_name: str = "vad_agent"
     llm_name: str = "llm_agent"
@@ -79,6 +78,10 @@ class BehaviourSettings(BaseModel):
     :ivar transcription_words_per_token: Estimated words per token for transcription timing.
     :ivar transcription_token_buffer: Buffer for transcription tokens.
     :ivar conversation_history_length_limit: The maximum amount of messages to extract beliefs from.
+    :ivar visual_emotion_recognition_window_duration_s: Duration in seconds over which to aggregate 
+    emotions and update emotion beliefs.
+    :ivar visual_emotion_recognition_min_frames_per_face: Minimum number of frames per face required
+    to consider a face valid.
     """
 
     # ATTENTION: When adding/removing settings, make sure to update the .env.example file
@@ -102,6 +105,9 @@ class BehaviourSettings(BaseModel):
     # Text belief extractor settings
     conversation_history_length_limit: int = 10
 
+    # Visual Emotion Recognition settings
+    visual_emotion_recognition_window_duration_s: int = 5
+    visual_emotion_recognition_min_frames_per_face: int = 3
 
 class LLMSettings(BaseModel):
     """
@@ -119,6 +125,7 @@ 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
@@ -155,6 +162,20 @@ class SpeechModelSettings(BaseModel):
     openai_model_name: str = "small.en"
 
 
+class LoggingSettings(BaseModel):
+    """
+    Configuration for logging.
+
+    :ivar logging_config_file: Path to the logging configuration file.
+    :ivar experiment_log_directory: Location of the experiment logs. Must match the logging config.
+    :ivar experiment_logger_name: Name of the experiment logger. Must match the logging config.
+    """
+
+    logging_config_file: str = ".logging_config.yaml"
+    experiment_log_directory: str = "experiment_logs"
+    experiment_logger_name: str = "experiment"
+
+
 class Settings(BaseSettings):
     """
     Global application settings.
@@ -176,6 +197,8 @@ class Settings(BaseSettings):
 
     ri_host: str = "localhost"
 
+    logging_settings: LoggingSettings = LoggingSettings()
+
     zmq_settings: ZMQSettings = ZMQSettings()
 
     agent_settings: AgentSettings = AgentSettings()

src/control_backend/logging/__init__.py

@@ -1 +1,4 @@
+from .dated_file_handler import DatedFileHandler as DatedFileHandler
+from .optional_field_formatter import OptionalFieldFormatter as OptionalFieldFormatter
+from .partial_filter import PartialFilter as PartialFilter
 from .setup_logging import setup_logging as setup_logging

src/control_backend/logging/dated_file_handler.py

@@ -0,0 +1,29 @@
+from datetime import datetime
+from logging import FileHandler
+from pathlib import Path
+
+
+class DatedFileHandler(FileHandler):
+    def __init__(self, file_prefix: str, **kwargs):
+        if not file_prefix:
+            raise ValueError("`file_prefix` argument cannot be empty.")
+        self._file_prefix = file_prefix
+        kwargs["filename"] = self._make_filename()
+        super().__init__(**kwargs)
+
+    def _make_filename(self) -> str:
+        filepath = Path(f"{self._file_prefix}-{datetime.now():%Y%m%d-%H%M%S}.log")
+        if not filepath.parent.is_dir():
+            filepath.parent.mkdir(parents=True, exist_ok=True)
+        return str(filepath)
+
+    def do_rollover(self):
+        self.acquire()
+        try:
+            if self.stream:
+                self.stream.close()
+
+            self.baseFilename = self._make_filename()
+            self.stream = self._open()
+        finally:
+            self.release()

src/control_backend/logging/optional_field_formatter.py

@@ -0,0 +1,67 @@
+import logging
+import re
+
+
+class OptionalFieldFormatter(logging.Formatter):
+    """
+    Logging formatter that supports optional fields marked by `?`.
+
+    Optional fields are denoted by placing a `?` after the field name inside
+    the parentheses, e.g., `%(role?)s`. If the field is not provided in the
+    log call's `extra` dict, it will use the default value from `defaults`
+    or `None` if no default is specified.
+
+    :param fmt: Format string with optional `%(name?)s` style fields.
+    :type fmt: str or None
+    :param datefmt: Date format string, passed to parent :class:`logging.Formatter`.
+    :type datefmt: str or None
+    :param style: Formatting style, must be '%'. Passed to parent.
+    :type style: str
+    :param defaults: Default values for optional fields when not provided.
+    :type defaults: dict or None
+
+    :example:
+
+        >>> formatter = OptionalFieldFormatter(
+        ...     fmt="%(asctime)s %(levelname)s %(role?)s %(message)s",
+        ...     defaults={"role": ""-""}
+        ... )
+        >>> handler = logging.StreamHandler()
+        >>> handler.setFormatter(formatter)
+        >>> logger = logging.getLogger(__name__)
+        >>> logger.addHandler(handler)
+        >>>
+        >>> logger.chat("Hello there!", extra={"role": "USER"})
+        2025-01-15 10:30:00 CHAT USER Hello there!
+        >>>
+        >>> logger.info("A logging message")
+        2025-01-15 10:30:01 INFO - A logging message
+
+    .. note::
+        Only `%`-style formatting is supported. The `{` and `$` styles are not
+        compatible with this formatter.
+
+    .. seealso::
+        :class:`logging.Formatter` for base formatter documentation.
+    """
+
+    # Match %(name?)s or %(name?)d etc.
+    OPTIONAL_PATTERN = re.compile(r"%\((\w+)\?\)([sdifFeEgGxXocrba%])")
+
+    def __init__(self, fmt=None, datefmt=None, style="%", defaults=None):
+        self.defaults = defaults or {}
+
+        self.optional_fields = set(self.OPTIONAL_PATTERN.findall(fmt or ""))
+
+        # Convert %(name?)s to %(name)s for standard formatting
+        normalized_fmt = self.OPTIONAL_PATTERN.sub(r"%(\1)\2", fmt or "")
+
+        super().__init__(normalized_fmt, datefmt, style)
+
+    def format(self, record):
+        for field, _ in self.optional_fields:
+            if not hasattr(record, field):
+                default = self.defaults.get(field, None)
+                setattr(record, field, default)
+
+        return super().format(record)

src/control_backend/logging/partial_filter.py

@@ -0,0 +1,10 @@
+import logging
+
+
+class PartialFilter(logging.Filter):
+    """
+    Class to filter any log records that have the "partial" attribute set to ``True``.
+    """
+
+    def filter(self, record):
+        return getattr(record, "partial", False) is not True

src/control_backend/logging/setup_logging.py

@@ -37,7 +37,7 @@ def add_logging_level(level_name: str, level_num: int, method_name: str | None =
     setattr(logging, method_name, log_to_root)
 
 
-def setup_logging(path: str = ".logging_config.yaml") -> None:
+def setup_logging(path: str = settings.logging_settings.logging_config_file) -> None:
     """
     Setup logging configuration of the CB. Tries to load the logging configuration from a file,
     in which we specify custom loggers, formatters, handlers, etc.
@@ -65,7 +65,7 @@ def setup_logging(path: str = ".logging_config.yaml") -> None:
 
             # Patch ZMQ PUBHandler to know about custom levels
             if custom_levels:
-                for logger_name in ("control_backend",):
+                for logger_name in config.get("loggers", {}):
                     logger = logging.getLogger(logger_name)
                     for handler in logger.handlers:
                         if isinstance(handler, PUBHandler):

src/control_backend/main.py

@@ -26,7 +26,6 @@ from zmq.asyncio import Context
 
 # BDI agents
 from control_backend.agents.bdi import (
-    BDIBeliefCollectorAgent,
     BDICoreAgent,
     TextBeliefExtractorAgent,
 )
@@ -122,12 +121,6 @@ async def lifespan(app: FastAPI):
                 "name": settings.agent_settings.bdi_core_name,
             },
         ),
-        "BeliefCollectorAgent": (
-            BDIBeliefCollectorAgent,
-            {
-                "name": settings.agent_settings.bdi_belief_collector_name,
-            },
-        ),
         "TextBeliefExtractorAgent": (
             TextBeliefExtractorAgent,
             {
@@ -172,6 +165,8 @@ async def lifespan(app: FastAPI):
 
     await endpoints_pub_socket.send_multipart([PROGRAM_STATUS, ProgramStatus.STOPPING.value])
     # Additional shutdown logic goes here
+    for agent in agents:
+        await agent.stop()
 
     logger.info("Application shutdown complete.")
 

src/control_backend/schemas/belief_list.py

@@ -16,4 +16,10 @@ class BeliefList(BaseModel):
 
 
 class GoalList(BaseModel):
+    """
+    Represents a list of goals, used for communicating multiple goals between agents.
+
+    :ivar goals: The list of goals.
+    """
+
     goals: list[BaseGoal]

src/control_backend/schemas/chat_history.py

@@ -2,9 +2,22 @@ from pydantic import BaseModel
 
 
 class ChatMessage(BaseModel):
+    """
+    Represents a single message in a conversation.
+
+    :ivar role: The role of the speaker (e.g., 'user', 'assistant').
+    :ivar content: The text content of the message.
+    """
+
     role: str
     content: str
 
 
 class ChatHistory(BaseModel):
+    """
+    Represents a sequence of chat messages, forming a conversation history.
+
+    :ivar messages: An ordered list of :class:`ChatMessage` objects.
+    """
+
     messages: list[ChatMessage]

src/control_backend/schemas/events.py

@@ -2,5 +2,13 @@ from pydantic import BaseModel
 
 
 class ButtonPressedEvent(BaseModel):
+    """
+    Represents a button press event from the UI.
+
+    :ivar type: The type of event (e.g., 'speech', 'gesture', 'override').
+    :ivar context: Additional data associated with the event (e.g., speech text, gesture name,
+    or ID).
+    """
+
     type: str
     context: str

src/control_backend/schemas/internal_message.py

@@ -7,7 +7,7 @@ class InternalMessage(BaseModel):
     """
     Standard message envelope for communication between agents within the Control Backend.
 
-    :ivar to: The name of the destination agent.
+    :ivar to: The name(s) of the destination agent(s).
     :ivar sender: The name of the sending agent.
     :ivar body: The string payload (often a JSON-serialized model).
     :ivar thread: An optional thread identifier/topic to categorize the message (e.g., 'beliefs').

src/control_backend/schemas/program.py

@@ -20,19 +20,23 @@ class ProgramElement(BaseModel):
 
 
 class LogicalOperator(Enum):
+    """
+    Logical operators for combining beliefs.
+    """
+
     AND = "AND"
     OR = "OR"
 
 
-type Belief = KeywordBelief | SemanticBelief | InferredBelief
-type BasicBelief = KeywordBelief | SemanticBelief
+type Belief = KeywordBelief | SemanticBelief | InferredBelief | EmotionBelief
+type BasicBelief = KeywordBelief | SemanticBelief | EmotionBelief
 
 
 class KeywordBelief(ProgramElement):
     """
-    Represents a belief that is set when the user spoken text contains a certain keyword.
+    Represents a belief that is activated when a specific keyword is detected in the user's speech.
 
-    :ivar keyword: The keyword on which this belief gets set.
+    :ivar keyword: The string to look for in the transcription.
     """
 
     name: str = ""
@@ -41,9 +45,11 @@ class KeywordBelief(ProgramElement):
 
 class SemanticBelief(ProgramElement):
     """
-    Represents a belief that is set by semantic LLM validation.
+    Represents a belief whose truth value is determined by an LLM analyzing the conversation
+    context.
 
-    :ivar description: Description of how to form the belief, used by the LLM.
+    :ivar description: A natural language description of what this belief represents,
+    used as a prompt for the LLM.
     """
 
     description: str
@@ -51,13 +57,11 @@ class SemanticBelief(ProgramElement):
 
 class InferredBelief(ProgramElement):
     """
-    Represents a belief that gets formed by combining two beliefs with a logical AND or OR.
+    Represents a belief derived from other beliefs using logical operators.
 
-    These beliefs can also be :class:`InferredBelief`, leading to arbitrarily deep nesting.
-
-    :ivar operator: The logical operator to apply.
-    :ivar left: The left part of the logical expression.
-    :ivar right: The right part of the logical expression.
+    :ivar operator: The :class:`LogicalOperator` (AND/OR) to apply.
+    :ivar left: The left operand (another belief).
+    :ivar right: The right operand (another belief).
     """
 
     name: str = ""
@@ -65,8 +69,24 @@ class InferredBelief(ProgramElement):
     left: Belief
     right: Belief
 
+class EmotionBelief(ProgramElement):
+    """
+    Represents a belief that is set when a certain emotion is detected.
+
+    :ivar emotion: The emotion on which this belief gets set.
+    """
+
+    name: str = ""
+    emotion: str
 
 class Norm(ProgramElement):
+    """
+    Base class for behavioral norms that guide the robot's interactions.
+
+    :ivar norm: The textual description of the norm.
+    :ivar critical: Whether this norm is considered critical and should be strictly enforced.
+    """
+
     name: str = ""
     norm: str
     critical: bool = False
@@ -74,10 +94,7 @@ class Norm(ProgramElement):
 
 class BasicNorm(Norm):
     """
-    Represents a behavioral norm.
-
-    :ivar norm: The actual norm text describing the behavior.
-    :ivar critical: When true, this norm should absolutely not be violated (checked separately).
+    A simple behavioral norm that is always considered for activation when its phase is active.
     """
 
     pass
@@ -85,9 +102,9 @@ class BasicNorm(Norm):
 
 class ConditionalNorm(Norm):
     """
-    Represents a norm that is only active when a condition is met (i.e., a certain belief holds).
+    A behavioral norm that is only active when a specific condition (belief) is met.
 
-    :ivar condition: When to activate this norm.
+    :ivar condition: The :class:`Belief` that must hold for this norm to be active.
     """
 
     condition: Belief
@@ -140,9 +157,9 @@ type Action = SpeechAction | GestureAction | LLMAction
 
 class SpeechAction(ProgramElement):
     """
-    Represents the action of the robot speaking a literal text.
+    An action where the robot speaks a predefined literal text.
 
-    :ivar text: The text to speak.
+    :ivar text: The text content to be spoken.
     """
 
     name: str = ""
@@ -151,11 +168,10 @@ class SpeechAction(ProgramElement):
 
 class Gesture(BaseModel):
     """
-    Represents a gesture to be performed. Can be either a single gesture,
-    or a random gesture from a category (tag).
+    Defines a physical gesture for the robot to perform.
 
-    :ivar type: The type of the gesture, "tag" or "single".
-    :ivar name: The name of the single gesture or tag.
+    :ivar type: Whether to use a specific "single" gesture or a random one from a "tag" category.
+    :ivar name: The identifier for the gesture or tag.
     """
 
     type: Literal["tag", "single"]
@@ -164,9 +180,9 @@ class Gesture(BaseModel):
 
 class GestureAction(ProgramElement):
     """
-    Represents the action of the robot performing a gesture.
+    An action where the robot performs a physical gesture.
 
-    :ivar gesture: The gesture to perform.
+    :ivar gesture: The :class:`Gesture` definition.
     """
 
     name: str = ""
@@ -175,10 +191,9 @@ class GestureAction(ProgramElement):
 
 class LLMAction(ProgramElement):
     """
-    Represents the action of letting an LLM generate a reply based on its chat history
-    and an additional goal added in the prompt.
+    An action that triggers an LLM-generated conversational response.
 
-    :ivar goal: The extra (temporary) goal to add to the LLM.
+    :ivar goal: A temporary conversational goal to guide the LLM's response generation.
     """
 
     name: str = ""
@@ -187,10 +202,10 @@ class LLMAction(ProgramElement):
 
 class Trigger(ProgramElement):
     """
-    Represents a belief-based trigger. When a belief is set, the corresponding plan is executed.
+    Defines a reactive behavior: when the condition (belief) is met, the plan is executed.
 
-    :ivar condition: When to activate the trigger.
-    :ivar plan: The plan to execute.
+    :ivar condition: The :class:`Belief` that triggers this behavior.
+    :ivar plan: The :class:`Plan` to execute upon activation.
     """
 
     condition: Belief
@@ -199,11 +214,11 @@ class Trigger(ProgramElement):
 
 class Phase(ProgramElement):
     """
-    A distinct phase within a program, containing norms, goals, and triggers.
+    A logical stage in the interaction program, grouping norms, goals, and triggers.
 
-    :ivar norms: List of norms active in this phase.
-    :ivar goals: List of goals to pursue in this phase.
-    :ivar triggers: List of triggers that define transitions out of this phase.
+    :ivar norms: List of norms active during this phase.
+    :ivar goals: List of goals the robot pursues in this phase.
+    :ivar triggers: List of reactive behaviors defined for this phase.
     """
 
     name: str = ""
@@ -214,9 +229,15 @@ class Phase(ProgramElement):
 
 class Program(BaseModel):
     """
-    Represents a complete interaction program, consisting of a sequence or set of phases.
+    The top-level container for a complete robot behavior definition.
 
-    :ivar phases: The list of phases that make up the program.
+    :ivar phases: An ordered list of :class:`Phase` objects defining the interaction flow.
     """
 
     phases: list[Phase]
+
+
+if __name__ == "__main__":
+    input = input("Enter program JSON: ")
+    program = Program.model_validate_json(input)
+    print(program)

test/integration/agents/perception/vad_agent/test_vad_agent.py

@@ -40,7 +40,7 @@ async def test_normal_setup(per_transcription_agent):
     per_vad_agent = VADAgent("tcp://localhost:12345", False)
     per_vad_agent._streaming_loop = AsyncMock()
 
-    async def swallow_background_task(coro):
+    def swallow_background_task(coro):
         coro.close()
 
     per_vad_agent.add_behavior = swallow_background_task
@@ -106,7 +106,7 @@ async def test_out_socket_creation_failure(zmq_context):
     per_vad_agent._streaming_loop = AsyncMock()
     per_vad_agent._connect_audio_out_socket = MagicMock(return_value=None)
 
-    async def swallow_background_task(coro):
+    def swallow_background_task(coro):
         coro.close()
 
     per_vad_agent.add_behavior = swallow_background_task
@@ -126,7 +126,7 @@ async def test_stop(zmq_context, per_transcription_agent):
     per_vad_agent._reset_stream = AsyncMock()
     per_vad_agent._streaming_loop = AsyncMock()
 
-    async def swallow_background_task(coro):
+    def swallow_background_task(coro):
         coro.close()
 
     per_vad_agent.add_behavior = swallow_background_task
@@ -150,6 +150,7 @@ async def test_application_startup_complete(zmq_context):
     vad_agent._running = True
     vad_agent._reset_stream = AsyncMock()
     vad_agent.program_sub_socket = AsyncMock()
+    vad_agent.program_sub_socket.close = MagicMock()
     vad_agent.program_sub_socket.recv_multipart.side_effect = [
         (PROGRAM_STATUS, ProgramStatus.RUNNING.value),
     ]

test/unit/agents/actuation/test_robot_gesture_agent.py

@@ -28,7 +28,11 @@ async def test_setup_bind(zmq_context, mocker):
     settings = mocker.patch("control_backend.agents.actuation.robot_gesture_agent.settings")
     settings.zmq_settings.internal_sub_address = "tcp://internal:1234"
 
-    agent.add_behavior = MagicMock()
+    def close_coro(coro):
+        coro.close()
+        return MagicMock()
+
+    agent.add_behavior = MagicMock(side_effect=close_coro)
 
     await agent.setup()
 
@@ -55,7 +59,11 @@ async def test_setup_connect(zmq_context, mocker):
     settings = mocker.patch("control_backend.agents.actuation.robot_gesture_agent.settings")
     settings.zmq_settings.internal_sub_address = "tcp://internal:1234"
 
-    agent.add_behavior = MagicMock()
+    def close_coro(coro):
+        coro.close()
+        return MagicMock()
+
+    agent.add_behavior = MagicMock(side_effect=close_coro)
 
     await agent.setup()
 
@@ -119,6 +127,65 @@ async def test_handle_message_rejects_invalid_gesture_tag():
     pubsocket.send_json.assert_not_awaited()
 
 
+@pytest.mark.asyncio
+async def test_handle_message_sends_valid_single_gesture_command():
+    """Internal message with valid single gesture is forwarded."""
+    pubsocket = AsyncMock()
+    agent = RobotGestureAgent("robot_gesture", single_gesture_data=["wave", "point"], address="")
+    agent.pubsocket = pubsocket
+
+    payload = {
+        "endpoint": RIEndpoint.GESTURE_SINGLE,
+        "data": "wave",
+    }
+    msg = InternalMessage(to="robot", sender="tester", body=json.dumps(payload))
+
+    await agent.handle_message(msg)
+
+    pubsocket.send_json.assert_awaited_once()
+
+
+@pytest.mark.asyncio
+async def test_handle_message_rejects_invalid_single_gesture():
+    """Internal message with invalid single gesture is not forwarded."""
+    pubsocket = AsyncMock()
+    agent = RobotGestureAgent("robot_gesture", single_gesture_data=["wave", "point"], address="")
+    agent.pubsocket = pubsocket
+
+    payload = {
+        "endpoint": RIEndpoint.GESTURE_SINGLE,
+        "data": "dance",
+    }
+    msg = InternalMessage(to="robot", sender="tester", body=json.dumps(payload))
+
+    await agent.handle_message(msg)
+
+    pubsocket.send_json.assert_not_awaited()
+
+
+@pytest.mark.asyncio
+async def test_zmq_command_loop_valid_single_gesture_payload():
+    """UI command with valid single gesture is read from SUB and published."""
+    command = {"endpoint": RIEndpoint.GESTURE_SINGLE, "data": "wave"}
+    fake_socket = AsyncMock()
+
+    async def recv_once():
+        agent._running = False
+        return b"command", json.dumps(command).encode("utf-8")
+
+    fake_socket.recv_multipart = recv_once
+    fake_socket.send_json = AsyncMock()
+
+    agent = RobotGestureAgent("robot_gesture", single_gesture_data=["wave", "point"], address="")
+    agent.subsocket = fake_socket
+    agent.pubsocket = fake_socket
+    agent._running = True
+
+    await agent._zmq_command_loop()
+
+    fake_socket.send_json.assert_awaited_once()
+
+
 @pytest.mark.asyncio
 async def test_handle_message_invalid_payload():
     """Invalid payload is caught and does not send."""
@@ -411,8 +478,7 @@ async def test_stop_closes_sockets():
 
     pubsocket.close.assert_called_once()
     subsocket.close.assert_called_once()
-    # Note: repsocket is not closed in stop() method, but you might want to add it
-    # repsocket.close.assert_called_once()
+    repsocket.close.assert_called_once()
 
 
 @pytest.mark.asyncio

test/unit/agents/actuation/test_robot_speech_agent.py

@@ -30,7 +30,11 @@ async def test_setup_bind(zmq_context, mocker):
     settings = mocker.patch("control_backend.agents.actuation.robot_speech_agent.settings")
     settings.zmq_settings.internal_sub_address = "tcp://internal:1234"
 
-    agent.add_behavior = MagicMock()
+    def close_coro(coro):
+        coro.close()
+        return MagicMock()
+
+    agent.add_behavior = MagicMock(side_effect=close_coro)
 
     await agent.setup()
 
@@ -48,7 +52,11 @@ async def test_setup_connect(zmq_context, mocker):
     settings = mocker.patch("control_backend.agents.actuation.robot_speech_agent.settings")
     settings.zmq_settings.internal_sub_address = "tcp://internal:1234"
 
-    agent.add_behavior = MagicMock()
+    def close_coro(coro):
+        coro.close()
+        return MagicMock()
+
+    agent.add_behavior = MagicMock(side_effect=close_coro)
 
     await agent.setup()
 

test/unit/agents/bdi/test_agentspeak_ast.py

@@ -0,0 +1,186 @@
+import pytest
+
+from control_backend.agents.bdi.agentspeak_ast import (
+    AstAtom,
+    AstBinaryOp,
+    AstLiteral,
+    AstLogicalExpression,
+    AstNumber,
+    AstPlan,
+    AstProgram,
+    AstRule,
+    AstStatement,
+    AstString,
+    AstVar,
+    BinaryOperatorType,
+    StatementType,
+    TriggerType,
+    _coalesce_expr,
+)
+
+
+def test_ast_atom():
+    atom = AstAtom("test")
+    assert str(atom) == "test"
+    assert atom._to_agentspeak() == "test"
+
+
+def test_ast_var():
+    var = AstVar("Variable")
+    assert str(var) == "Variable"
+    assert var._to_agentspeak() == "Variable"
+
+
+def test_ast_number():
+    num = AstNumber(42)
+    assert str(num) == "42"
+    num_float = AstNumber(3.14)
+    assert str(num_float) == "3.14"
+
+
+def test_ast_string():
+    s = AstString("hello")
+    assert str(s) == '"hello"'
+
+
+def test_ast_literal():
+    lit = AstLiteral("functor", [AstAtom("atom"), AstNumber(1)])
+    assert str(lit) == "functor(atom, 1)"
+    lit_empty = AstLiteral("functor")
+    assert str(lit_empty) == "functor"
+
+
+def test_ast_binary_op():
+    left = AstNumber(1)
+    right = AstNumber(2)
+    op = AstBinaryOp(left, BinaryOperatorType.GREATER_THAN, right)
+    assert str(op) == "1 > 2"
+
+    # Test logical wrapper
+    assert isinstance(op.left, AstLogicalExpression)
+    assert isinstance(op.right, AstLogicalExpression)
+
+
+def test_ast_binary_op_parens():
+    # 1 > 2
+    inner = AstBinaryOp(AstNumber(1), BinaryOperatorType.GREATER_THAN, AstNumber(2))
+    # (1 > 2) & 3
+    outer = AstBinaryOp(inner, BinaryOperatorType.AND, AstNumber(3))
+    assert str(outer) == "(1 > 2) & 3"
+
+    # 3 & (1 > 2)
+    outer_right = AstBinaryOp(AstNumber(3), BinaryOperatorType.AND, inner)
+    assert str(outer_right) == "3 & (1 > 2)"
+
+
+def test_ast_binary_op_parens_negated():
+    inner = AstLogicalExpression(AstAtom("foo"), negated=True)
+    outer = AstBinaryOp(inner, BinaryOperatorType.AND, AstAtom("bar"))
+    # The current implementation checks `if self.left.negated: l_str = f"({l_str})"`
+    # str(inner) is "not foo"
+    # so we expect "(not foo) & bar"
+    assert str(outer) == "(not foo) & bar"
+
+    outer_right = AstBinaryOp(AstAtom("bar"), BinaryOperatorType.AND, inner)
+    assert str(outer_right) == "bar & (not foo)"
+
+
+def test_ast_logical_expression_negation():
+    expr = AstLogicalExpression(AstAtom("true"), negated=True)
+    assert str(expr) == "not true"
+
+    expr_neg_neg = ~expr
+    assert str(expr_neg_neg) == "true"
+    assert not expr_neg_neg.negated
+
+    # Invert a non-logical expression (wraps it)
+    term = AstAtom("true")
+    inverted = ~term
+    assert isinstance(inverted, AstLogicalExpression)
+    assert inverted.negated
+    assert str(inverted) == "not true"
+
+
+def test_ast_logical_expression_no_negation():
+    # _as_logical on already logical expression
+    expr = AstLogicalExpression(AstAtom("x"))
+    # Doing binary op will call _as_logical
+    op = AstBinaryOp(expr, BinaryOperatorType.AND, AstAtom("y"))
+    assert isinstance(op.left, AstLogicalExpression)
+    assert op.left is expr  # Should reuse instance
+
+
+def test_ast_operators():
+    t1 = AstAtom("a")
+    t2 = AstAtom("b")
+
+    assert str(t1 & t2) == "a & b"
+    assert str(t1 | t2) == "a | b"
+    assert str(t1 >= t2) == "a >= b"
+    assert str(t1 > t2) == "a > b"
+    assert str(t1 <= t2) == "a <= b"
+    assert str(t1 < t2) == "a < b"
+    assert str(t1 == t2) == "a == b"
+    assert str(t1 != t2) == r"a \== b"
+
+
+def test_coalesce_expr():
+    t = AstAtom("a")
+    assert str(t & "b") == 'a & "b"'
+    assert str(t & 1) == "a & 1"
+    assert str(t & 1.5) == "a & 1.5"
+
+    with pytest.raises(TypeError):
+        _coalesce_expr(None)
+
+
+def test_ast_statement():
+    stmt = AstStatement(StatementType.DO_ACTION, AstLiteral("action"))
+    assert str(stmt) == ".action"
+
+
+def test_ast_rule():
+    # Rule with condition
+    rule = AstRule(AstLiteral("head"), AstLiteral("body"))
+    assert str(rule) == "head :- body."
+
+    # Rule without condition
+    rule_simple = AstRule(AstLiteral("fact"))
+    assert str(rule_simple) == "fact."
+
+
+def test_ast_plan():
+    plan = AstPlan(
+        TriggerType.ADDED_GOAL,
+        AstLiteral("goal"),
+        [AstLiteral("context")],
+        [AstStatement(StatementType.DO_ACTION, AstLiteral("action"))],
+    )
+    output = str(plan)
+    # verify parts exist
+    assert "+!goal" in output
+    assert ":   context" in output
+    assert "<-  .action." in output
+
+
+def test_ast_plan_no_context():
+    plan = AstPlan(
+        TriggerType.ADDED_GOAL,
+        AstLiteral("goal"),
+        [],
+        [AstStatement(StatementType.DO_ACTION, AstLiteral("action"))],
+    )
+    output = str(plan)
+    assert "+!goal" in output
+    assert ":   " not in output
+    assert "<-  .action." in output
+
+
+def test_ast_program():
+    prog = AstProgram(
+        rules=[AstRule(AstLiteral("fact"))],
+        plans=[AstPlan(TriggerType.ADDED_BELIEF, AstLiteral("b"), [], [])],
+    )
+    output = str(prog)
+    assert "fact." in output
+    assert "+b" in output

test/unit/agents/bdi/test_agentspeak_generator.py

@@ -0,0 +1,187 @@
+import uuid
+
+import pytest
+
+from control_backend.agents.bdi.agentspeak_ast import AstProgram
+from control_backend.agents.bdi.agentspeak_generator import AgentSpeakGenerator
+from control_backend.schemas.program import (
+    BasicNorm,
+    ConditionalNorm,
+    Gesture,
+    GestureAction,
+    Goal,
+    InferredBelief,
+    KeywordBelief,
+    LLMAction,
+    LogicalOperator,
+    Phase,
+    Plan,
+    Program,
+    SemanticBelief,
+    SpeechAction,
+    Trigger,
+)
+
+
+@pytest.fixture
+def generator():
+    return AgentSpeakGenerator()
+
+
+def test_generate_empty_program(generator):
+    prog = Program(phases=[])
+    code = generator.generate(prog)
+    assert 'phase("end").' in code
+    assert "!notify_cycle" in code
+
+
+def test_generate_basic_norm(generator):
+    norm = BasicNorm(id=uuid.uuid4(), name="n1", norm="be nice")
+    phase = Phase(id=uuid.uuid4(), norms=[norm], goals=[], triggers=[])
+    prog = Program(phases=[phase])
+
+    code = generator.generate(prog)
+    assert f'norm("be nice") :- phase("{phase.id}").' in code
+
+
+def test_generate_critical_norm(generator):
+    norm = BasicNorm(id=uuid.uuid4(), name="n1", norm="safety", critical=True)
+    phase = Phase(id=uuid.uuid4(), norms=[norm], goals=[], triggers=[])
+    prog = Program(phases=[phase])
+
+    code = generator.generate(prog)
+    assert f'critical_norm("safety") :- phase("{phase.id}").' in code
+
+
+def test_generate_conditional_norm(generator):
+    cond = KeywordBelief(id=uuid.uuid4(), name="k1", keyword="please")
+    norm = ConditionalNorm(id=uuid.uuid4(), name="n1", norm="help", condition=cond)
+    phase = Phase(id=uuid.uuid4(), norms=[norm], goals=[], triggers=[])
+    prog = Program(phases=[phase])
+
+    code = generator.generate(prog)
+    assert 'norm("help")' in code
+    assert 'keyword_said("please")' in code
+    assert f"force_norm_{generator._slugify_str(norm.norm)}" in code
+
+
+def test_generate_goal_and_plan(generator):
+    action = SpeechAction(id=uuid.uuid4(), name="s1", text="hello")
+    plan = Plan(id=uuid.uuid4(), name="p1", steps=[action])
+    # IMPORTANT: can_fail must be False for +achieved_ belief to be added
+    goal = Goal(id=uuid.uuid4(), name="g1", description="desc", plan=plan, can_fail=False)
+    phase = Phase(id=uuid.uuid4(), norms=[], goals=[goal], triggers=[])
+    prog = Program(phases=[phase])
+
+    code = generator.generate(prog)
+    # Check trigger for goal
+    goal_slug = generator._slugify_str(goal.name)
+    assert f"+!{goal_slug}" in code
+    assert f'phase("{phase.id}")' in code
+    assert '!say("hello")' in code
+
+    # Check success belief addition
+    assert f"+achieved_{goal_slug}" in code
+
+
+def test_generate_subgoal(generator):
+    subplan = Plan(id=uuid.uuid4(), name="p2", steps=[])
+    subgoal = Goal(id=uuid.uuid4(), name="sub1", description="sub", plan=subplan)
+
+    plan = Plan(id=uuid.uuid4(), name="p1", steps=[subgoal])
+    goal = Goal(id=uuid.uuid4(), name="g1", description="main", plan=plan)
+    phase = Phase(id=uuid.uuid4(), norms=[], goals=[goal], triggers=[])
+    prog = Program(phases=[phase])
+
+    code = generator.generate(prog)
+    subgoal_slug = generator._slugify_str(subgoal.name)
+    # Main goal calls subgoal
+    assert f"!{subgoal_slug}" in code
+    # Subgoal plan exists
+    assert f"+!{subgoal_slug}" in code
+
+
+def test_generate_trigger(generator):
+    cond = SemanticBelief(id=uuid.uuid4(), name="s1", description="desc")
+    plan = Plan(id=uuid.uuid4(), name="p1", steps=[])
+    trigger = Trigger(id=uuid.uuid4(), name="t1", condition=cond, plan=plan)
+    phase = Phase(id=uuid.uuid4(), norms=[], goals=[], triggers=[trigger])
+    prog = Program(phases=[phase])
+
+    code = generator.generate(prog)
+    # Trigger logic is added to check_triggers
+    assert f"{generator.slugify(cond)}" in code
+    assert f'notify_trigger_start("{generator.slugify(trigger)}")' in code
+    assert f'notify_trigger_end("{generator.slugify(trigger)}")' in code
+
+
+def test_phase_transition(generator):
+    phase1 = Phase(id=uuid.uuid4(), name="p1", norms=[], goals=[], triggers=[])
+    phase2 = Phase(id=uuid.uuid4(), name="p2", norms=[], goals=[], triggers=[])
+    prog = Program(phases=[phase1, phase2])
+
+    code = generator.generate(prog)
+    assert "transition_phase" in code
+    assert f'phase("{phase1.id}")' in code
+    assert f'phase("{phase2.id}")' in code
+    assert "force_transition_phase" in code
+
+
+def test_astify_gesture(generator):
+    gesture = Gesture(type="single", name="wave")
+    action = GestureAction(id=uuid.uuid4(), name="g1", gesture=gesture)
+    ast = generator._astify(action)
+    assert str(ast) == 'gesture("single", "wave")'
+
+
+def test_astify_llm_action(generator):
+    action = LLMAction(id=uuid.uuid4(), name="l1", goal="be funny")
+    ast = generator._astify(action)
+    assert str(ast) == 'reply_with_goal("be funny")'
+
+
+def test_astify_inferred_belief_and(generator):
+    left = KeywordBelief(id=uuid.uuid4(), name="k1", keyword="a")
+    right = KeywordBelief(id=uuid.uuid4(), name="k2", keyword="b")
+    inf = InferredBelief(
+        id=uuid.uuid4(), name="i1", operator=LogicalOperator.AND, left=left, right=right
+    )
+
+    ast = generator._astify(inf)
+    assert 'keyword_said("a") & keyword_said("b")' == str(ast)
+
+
+def test_astify_inferred_belief_or(generator):
+    left = KeywordBelief(id=uuid.uuid4(), name="k1", keyword="a")
+    right = KeywordBelief(id=uuid.uuid4(), name="k2", keyword="b")
+    inf = InferredBelief(
+        id=uuid.uuid4(), name="i1", operator=LogicalOperator.OR, left=left, right=right
+    )
+
+    ast = generator._astify(inf)
+    assert 'keyword_said("a") | keyword_said("b")' == str(ast)
+
+
+def test_astify_semantic_belief(generator):
+    sb = SemanticBelief(id=uuid.uuid4(), name="s1", description="desc")
+    ast = generator._astify(sb)
+    assert str(ast) == f"semantic_{generator._slugify_str(sb.name)}"
+
+
+def test_slugify_not_implemented(generator):
+    with pytest.raises(NotImplementedError):
+        generator.slugify("not a program element")
+
+
+def test_astify_not_implemented(generator):
+    with pytest.raises(NotImplementedError):
+        generator._astify("not a program element")
+
+
+def test_process_phase_transition_from_none(generator):
+    # Initialize AstProgram manually as we are bypassing generate()
+    generator._asp = AstProgram()
+    # Should safely return doing nothing
+    generator._add_phase_transition(None, None)
+
+    assert len(generator._asp.plans) == 0

test/unit/agents/bdi/test_bdi_core_agent.py

@@ -45,23 +45,34 @@ async def test_setup_no_asl(mock_agentspeak_env, agent):
 
 
 @pytest.mark.asyncio
-async def test_handle_belief_collector_message(agent, mock_settings):
+async def test_handle_belief_message(agent, mock_settings):
     """Test that incoming beliefs are added to the BDI agent"""
     beliefs = [Belief(name="user_said", arguments=["Hello"])]
     msg = InternalMessage(
         to="bdi_agent",
-        sender=mock_settings.agent_settings.bdi_belief_collector_name,
+        sender=mock_settings.agent_settings.text_belief_extractor_name,
         body=BeliefMessage(create=beliefs).model_dump_json(),
         thread="beliefs",
     )
 
     await agent.handle_message(msg)
 
-    # Expect bdi_agent.call to be triggered to add belief
-    args = agent.bdi_agent.call.call_args.args
-    assert args[0] == agentspeak.Trigger.addition
-    assert args[1] == agentspeak.GoalType.belief
-    assert args[2] == agentspeak.Literal("user_said", (agentspeak.Literal("Hello"),))
+    # Check for the specific call we expect among all calls
+    # bdi_agent.call is called multiple times (for transition_phase, check_triggers)
+    # We want to confirm the belief addition call exists
+    found_call = False
+    for call in agent.bdi_agent.call.call_args_list:
+        args = call.args
+        if (
+            args[0] == agentspeak.Trigger.addition
+            and args[1] == agentspeak.GoalType.belief
+            and args[2].functor == "user_said"
+            and args[2].args[0].functor == "Hello"
+        ):
+            found_call = True
+            break
+
+    assert found_call, "Expected belief addition call not found in bdi_agent.call history"
 
 
 @pytest.mark.asyncio
@@ -71,25 +82,33 @@ async def test_handle_delete_belief_message(agent, mock_settings):
 
     msg = InternalMessage(
         to="bdi_agent",
-        sender=mock_settings.agent_settings.bdi_belief_collector_name,
+        sender=mock_settings.agent_settings.text_belief_extractor_name,
         body=BeliefMessage(delete=beliefs).model_dump_json(),
         thread="beliefs",
     )
     await agent.handle_message(msg)
 
-    # Expect bdi_agent.call to be triggered to remove belief
-    args = agent.bdi_agent.call.call_args.args
-    assert args[0] == agentspeak.Trigger.removal
-    assert args[1] == agentspeak.GoalType.belief
-    assert args[2] == agentspeak.Literal("user_said", (agentspeak.Literal("Hello"),))
+    found_call = False
+    for call in agent.bdi_agent.call.call_args_list:
+        args = call.args
+        if (
+            args[0] == agentspeak.Trigger.removal
+            and args[1] == agentspeak.GoalType.belief
+            and args[2].functor == "user_said"
+            and args[2].args[0].functor == "Hello"
+        ):
+            found_call = True
+            break
+
+    assert found_call
 
 
 @pytest.mark.asyncio
-async def test_incorrect_belief_collector_message(agent, mock_settings):
+async def test_incorrect_belief_message(agent, mock_settings):
     """Test that incorrect message format triggers an exception."""
     msg = InternalMessage(
         to="bdi_agent",
-        sender=mock_settings.agent_settings.bdi_belief_collector_name,
+        sender=mock_settings.agent_settings.text_belief_extractor_name,
         body=json.dumps({"bad_format": "bad_format"}),
         thread="beliefs",
     )
@@ -171,7 +190,11 @@ def test_remove_belief_success_wakes_loop(agent):
     agent._remove_belief("remove_me", ["x"])
 
     assert agent.bdi_agent.call.called
-    trigger, goaltype, literal, *_ = agent.bdi_agent.call.call_args.args
+
+    call_args = agent.bdi_agent.call.call_args.args
+    trigger = call_args[0]
+    goaltype = call_args[1]
+    literal = call_args[2]
 
     assert trigger == agentspeak.Trigger.removal
     assert goaltype == agentspeak.GoalType.belief
@@ -288,3 +311,216 @@ async def test_deadline_sleep_branch(agent):
 
     duration = time.time() - start_time
     assert duration >= 0.004  # loop slept until deadline
+
+
+@pytest.mark.asyncio
+async def test_handle_new_program(agent):
+    agent._load_asl = AsyncMock()
+    agent.add_behavior = MagicMock()
+    # Mock existing loop task so it can be cancelled
+    mock_task = MagicMock()
+    mock_task.cancel = MagicMock()
+    agent._bdi_loop_task = mock_task
+
+    def close_coro(coro):
+        coro.close()
+        return MagicMock()
+
+    agent.add_behavior = MagicMock(side_effect=close_coro)
+
+    msg = InternalMessage(to="bdi_agent", thread="new_program", body="path/to/asl.asl")
+
+    await agent.handle_message(msg)
+
+    mock_task.cancel.assert_called_once()
+    agent._load_asl.assert_awaited_once_with("path/to/asl.asl")
+    agent.add_behavior.assert_called()
+
+
+@pytest.mark.asyncio
+async def test_handle_user_interrupts(agent, mock_settings):
+    mock_settings.agent_settings.user_interrupt_name = "user_interrupt_agent"
+    # force_phase_transition
+    agent._set_goal = MagicMock()
+    msg = InternalMessage(
+        to="bdi_agent",
+        sender=mock_settings.agent_settings.user_interrupt_name,
+        thread="force_phase_transition",
+        body="",
+    )
+    await agent.handle_message(msg)
+    agent._set_goal.assert_called_with("transition_phase")
+
+    # force_trigger
+    agent._force_trigger = MagicMock()
+    msg.thread = "force_trigger"
+    msg.body = "trigger_x"
+    await agent.handle_message(msg)
+    agent._force_trigger.assert_called_with("trigger_x")
+
+    # force_norm
+    agent._force_norm = MagicMock()
+    msg.thread = "force_norm"
+    msg.body = "norm_y"
+    await agent.handle_message(msg)
+    agent._force_norm.assert_called_with("norm_y")
+
+    # force_next_phase
+    agent._force_next_phase = MagicMock()
+    msg.thread = "force_next_phase"
+    msg.body = ""
+    await agent.handle_message(msg)
+    agent._force_next_phase.assert_called_once()
+
+    # unknown interrupt
+    agent.logger = MagicMock()
+    msg.thread = "unknown_thing"
+    await agent.handle_message(msg)
+    agent.logger.warning.assert_called()
+
+
+@pytest.mark.asyncio
+async def test_custom_action_reply_with_goal(agent):
+    agent._send_to_llm = MagicMock(side_effect=agent.send)
+    agent._add_custom_actions()
+    action_fn = agent.actions.actions[(".reply_with_goal", 3)]
+
+    mock_term = MagicMock(args=["msg", "norms", "goal"])
+    gen = action_fn(agent, mock_term, MagicMock())
+    next(gen)
+    agent._send_to_llm.assert_called_with("msg", "norms", "goal")
+
+
+@pytest.mark.asyncio
+async def test_custom_action_notify_norms(agent):
+    agent._add_custom_actions()
+    action_fn = agent.actions.actions[(".notify_norms", 1)]
+
+    mock_term = MagicMock(args=["norms_list"])
+    gen = action_fn(agent, mock_term, MagicMock())
+    next(gen)
+
+    agent.send.assert_called()
+    msg = agent.send.call_args[0][0]
+    assert msg.thread == "active_norms_update"
+    assert msg.body == "norms_list"
+
+
+@pytest.mark.asyncio
+async def test_custom_action_say(agent):
+    agent._add_custom_actions()
+    action_fn = agent.actions.actions[(".say", 1)]
+
+    mock_term = MagicMock(args=["hello"])
+    gen = action_fn(agent, mock_term, MagicMock())
+    next(gen)
+
+    assert agent.send.call_count == 2
+    msgs = [c[0][0] for c in agent.send.call_args_list]
+    assert any(m.to == settings.agent_settings.robot_speech_name for m in msgs)
+    assert any(
+        m.to == settings.agent_settings.llm_name and m.thread == "assistant_message" for m in msgs
+    )
+
+
+@pytest.mark.asyncio
+async def test_custom_action_gesture(agent):
+    agent._add_custom_actions()
+    # Test single
+    action_fn = agent.actions.actions[(".gesture", 2)]
+    mock_term = MagicMock(args=["single", "wave"])
+    gen = action_fn(agent, mock_term, MagicMock())
+    next(gen)
+    msg = agent.send.call_args[0][0]
+    assert "actuate/gesture/single" in msg.body
+
+    # Test tag
+    mock_term.args = ["tag", "happy"]
+    gen = action_fn(agent, mock_term, MagicMock())
+    next(gen)
+    msg = agent.send.call_args[0][0]
+    assert "actuate/gesture/tag" in msg.body
+
+
+@pytest.mark.asyncio
+async def test_custom_action_notify_user_said(agent):
+    agent._add_custom_actions()
+    action_fn = agent.actions.actions[(".notify_user_said", 1)]
+    mock_term = MagicMock(args=["hello"])
+    gen = action_fn(agent, mock_term, MagicMock())
+    next(gen)
+    msg = agent.send.call_args[0][0]
+    assert msg.to == settings.agent_settings.llm_name
+    assert msg.thread == "user_message"
+
+
+@pytest.mark.asyncio
+async def test_custom_action_notify_trigger_start_end(agent):
+    agent._add_custom_actions()
+    # Start
+    action_fn = agent.actions.actions[(".notify_trigger_start", 1)]
+    gen = action_fn(agent, MagicMock(args=["t1"]), MagicMock())
+    next(gen)
+    assert agent.send.call_args[0][0].thread == "trigger_start"
+
+    # End
+    action_fn = agent.actions.actions[(".notify_trigger_end", 1)]
+    gen = action_fn(agent, MagicMock(args=["t1"]), MagicMock())
+    next(gen)
+    assert agent.send.call_args[0][0].thread == "trigger_end"
+
+
+@pytest.mark.asyncio
+async def test_custom_action_notify_goal_start(agent):
+    agent._add_custom_actions()
+    action_fn = agent.actions.actions[(".notify_goal_start", 1)]
+    gen = action_fn(agent, MagicMock(args=["g1"]), MagicMock())
+    next(gen)
+    assert agent.send.call_args[0][0].thread == "goal_start"
+
+
+@pytest.mark.asyncio
+async def test_custom_action_notify_transition_phase(agent):
+    agent._add_custom_actions()
+    action_fn = agent.actions.actions[(".notify_transition_phase", 2)]
+    gen = action_fn(agent, MagicMock(args=["old", "new"]), MagicMock())
+    next(gen)
+    msg = agent.send.call_args[0][0]
+    assert msg.thread == "transition_phase"
+    assert "old" in msg.body and "new" in msg.body
+
+
+def test_remove_belief_no_args(agent):
+    agent._wake_bdi_loop = MagicMock()
+    agent.bdi_agent.call.return_value = True
+    agent._remove_belief("fact", None)
+    assert agent.bdi_agent.call.called
+
+
+def test_set_goal_with_args(agent):
+    agent._wake_bdi_loop = MagicMock()
+    agent._set_goal("goal", ["arg1", "arg2"])
+    assert agent.bdi_agent.call.called
+
+
+def test_format_belief_string():
+    assert BDICoreAgent.format_belief_string("b") == "b"
+    assert BDICoreAgent.format_belief_string("b", ["a1", "a2"]) == "b(a1,a2)"
+
+
+def test_force_norm(agent):
+    agent._add_belief = MagicMock()
+    agent._force_norm("be_polite")
+    agent._add_belief.assert_called_with("force_be_polite")
+
+
+def test_force_trigger(agent):
+    agent._set_goal = MagicMock()
+    agent._force_trigger("trig")
+    agent._set_goal.assert_called_with("trig")
+
+
+def test_force_next_phase(agent):
+    agent._set_goal = MagicMock()
+    agent._force_next_phase()
+    agent._set_goal.assert_called_with("force_transition_phase")

test/unit/agents/bdi/test_bdi_program_manager.py

@@ -1,13 +1,13 @@
 import asyncio
+import json
 import sys
 import uuid
-from unittest.mock import AsyncMock
+from unittest.mock import AsyncMock, MagicMock, mock_open, patch
 
 import pytest
 
 from control_backend.agents.bdi.bdi_program_manager import BDIProgramManager
 from control_backend.core.agent_system import InternalMessage
-from control_backend.schemas.belief_message import BeliefMessage
 from control_backend.schemas.program import BasicNorm, Goal, Phase, Plan, Program
 
 # Fix Windows Proactor loop for zmq
@@ -48,24 +48,26 @@ def make_valid_program_json(norm="N1", goal="G1") -> str:
     ).model_dump_json()
 
 
-@pytest.mark.skip(reason="Functionality being rebuilt.")
 @pytest.mark.asyncio
-async def test_send_to_bdi():
+async def test_create_agentspeak_and_send_to_bdi(mock_settings):
     manager = BDIProgramManager(name="program_manager_test")
     manager.send = AsyncMock()
 
     program = Program.model_validate_json(make_valid_program_json())
-    await manager._create_agentspeak_and_send_to_bdi(program)
+
+    with patch("builtins.open", mock_open()) as mock_file:
+        await manager._create_agentspeak_and_send_to_bdi(program)
+
+        # Check file writing
+        mock_file.assert_called_with("src/control_backend/agents/bdi/agentspeak.asl", "w")
+        handle = mock_file()
+        handle.write.assert_called()
 
     assert manager.send.await_count == 1
     msg: InternalMessage = manager.send.await_args[0][0]
-    assert msg.thread == "beliefs"
-
-    beliefs = BeliefMessage.model_validate_json(msg.body)
-    names = {b.name: b.arguments for b in beliefs.beliefs}
-
-    assert "norms" in names and names["norms"] == ["N1"]
-    assert "goals" in names and names["goals"] == ["G1"]
+    assert msg.thread == "new_program"
+    assert msg.to == mock_settings.agent_settings.bdi_core_name
+    assert msg.body == "src/control_backend/agents/bdi/agentspeak.asl"
 
 
 @pytest.mark.asyncio
@@ -81,6 +83,9 @@ async def test_receive_programs_valid_and_invalid():
     manager.sub_socket = sub
     manager._create_agentspeak_and_send_to_bdi = AsyncMock()
     manager._send_clear_llm_history = AsyncMock()
+    manager._send_program_to_user_interrupt = AsyncMock()
+    manager._send_beliefs_to_semantic_belief_extractor = AsyncMock()
+    manager._send_goals_to_semantic_belief_extractor = AsyncMock()
 
     try:
         # Will give StopAsyncIteration when the predefined `sub.recv_multipart` side-effects run out
@@ -94,10 +99,9 @@ async def test_receive_programs_valid_and_invalid():
     assert forwarded.phases[0].norms[0].name == "N1"
     assert forwarded.phases[0].goals[0].name == "G1"
 
-    # Verify history clear was triggered
-    assert (
-        manager._send_clear_llm_history.await_count == 2
-    )  # first sends program to UserInterrupt, then clears LLM
+    # Verify history clear was triggered exactly once (for the valid program)
+    # The invalid program loop `continue`s before calling _send_clear_llm_history
+    assert manager._send_clear_llm_history.await_count == 1
 
 
 @pytest.mark.asyncio
@@ -115,4 +119,179 @@ async def test_send_clear_llm_history(mock_settings):
 
     # Verify the content and recipient
     assert msg.body == "clear_history"
-    assert msg.to == "llm_agent"
+
+
+@pytest.mark.asyncio
+async def test_handle_message_transition_phase(mock_settings):
+    mock_settings.agent_settings.user_interrupt_name = "user_interrupt_agent"
+    manager = BDIProgramManager(name="program_manager_test")
+    manager.send = AsyncMock()
+
+    # Setup state
+    prog = Program.model_validate_json(make_valid_program_json(norm="N1", goal="G1"))
+    manager._initialize_internal_state(prog)
+
+    # Test valid transition (to same phase for simplicity, or we need 2 phases)
+    # Let's create a program with 2 phases
+    phase2_id = uuid.uuid4()
+    phase2 = Phase(id=phase2_id, name="Phase 2", norms=[], goals=[], triggers=[])
+    prog.phases.append(phase2)
+    manager._initialize_internal_state(prog)
+
+    current_phase_id = str(prog.phases[0].id)
+    next_phase_id = str(phase2_id)
+
+    payload = json.dumps({"old": current_phase_id, "new": next_phase_id})
+    msg = InternalMessage(to="me", sender="bdi", body=payload, thread="transition_phase")
+
+    await manager.handle_message(msg)
+
+    assert str(manager._phase.id) == next_phase_id
+
+    # Allow background tasks to run (add_behavior)
+    await asyncio.sleep(0)
+
+    # Check notifications sent
+    # 1. beliefs to extractor
+    # 2. goals to extractor
+    # 3. notification to user interrupt
+
+    assert manager.send.await_count >= 3
+
+    # Verify user interrupt notification
+    calls = manager.send.await_args_list
+    ui_msgs = [
+        c[0][0] for c in calls if c[0][0].to == mock_settings.agent_settings.user_interrupt_name
+    ]
+    assert len(ui_msgs) > 0
+    assert ui_msgs[-1].body == next_phase_id
+
+
+@pytest.mark.asyncio
+async def test_handle_message_transition_phase_desync():
+    manager = BDIProgramManager(name="program_manager_test")
+    manager.logger = MagicMock()
+
+    prog = Program.model_validate_json(make_valid_program_json())
+    manager._initialize_internal_state(prog)
+
+    current_phase_id = str(prog.phases[0].id)
+
+    # Request transition from WRONG old phase
+    payload = json.dumps({"old": "wrong_id", "new": "some_new_id"})
+    msg = InternalMessage(to="me", sender="bdi", body=payload, thread="transition_phase")
+
+    await manager.handle_message(msg)
+
+    # Should warn and do nothing
+    manager.logger.warning.assert_called_once()
+    assert "Phase transition desync detected" in manager.logger.warning.call_args[0][0]
+    assert str(manager._phase.id) == current_phase_id
+
+
+@pytest.mark.asyncio
+async def test_handle_message_transition_phase_end(mock_settings):
+    mock_settings.agent_settings.user_interrupt_name = "user_interrupt_agent"
+    manager = BDIProgramManager(name="program_manager_test")
+    manager.send = AsyncMock()
+
+    prog = Program.model_validate_json(make_valid_program_json())
+    manager._initialize_internal_state(prog)
+    current_phase_id = str(prog.phases[0].id)
+
+    payload = json.dumps({"old": current_phase_id, "new": "end"})
+    msg = InternalMessage(to="me", sender="bdi", body=payload, thread="transition_phase")
+
+    await manager.handle_message(msg)
+
+    assert manager._phase is None
+
+    # Allow background tasks to run (add_behavior)
+    await asyncio.sleep(0)
+
+    # Verify notification to user interrupt
+    assert manager.send.await_count == 1
+    msg_sent = manager.send.await_args[0][0]
+    assert msg_sent.to == mock_settings.agent_settings.user_interrupt_name
+    assert msg_sent.body == "end"
+
+
+@pytest.mark.asyncio
+async def test_handle_message_achieve_goal(mock_settings):
+    mock_settings.agent_settings.text_belief_extractor_name = "text_belief_extractor_agent"
+    manager = BDIProgramManager(name="program_manager_test")
+    manager.send = AsyncMock()
+
+    prog = Program.model_validate_json(make_valid_program_json(goal="TargetGoal"))
+    manager._initialize_internal_state(prog)
+
+    goal_id = str(prog.phases[0].goals[0].id)
+
+    msg = InternalMessage(to="me", sender="ui", body=goal_id, thread="achieve_goal")
+
+    await manager.handle_message(msg)
+
+    # Should send achieved goals to text extractor
+    assert manager.send.await_count == 1
+    msg_sent = manager.send.await_args[0][0]
+    assert msg_sent.to == mock_settings.agent_settings.text_belief_extractor_name
+    assert msg_sent.thread == "achieved_goals"
+
+    # Verify body
+    from control_backend.schemas.belief_list import GoalList
+
+    gl = GoalList.model_validate_json(msg_sent.body)
+    assert len(gl.goals) == 1
+    assert gl.goals[0].name == "TargetGoal"
+
+
+@pytest.mark.asyncio
+async def test_handle_message_achieve_goal_not_found():
+    manager = BDIProgramManager(name="program_manager_test")
+    manager.send = AsyncMock()
+    manager.logger = MagicMock()
+
+    prog = Program.model_validate_json(make_valid_program_json())
+    manager._initialize_internal_state(prog)
+
+    msg = InternalMessage(to="me", sender="ui", body="non_existent_id", thread="achieve_goal")
+
+    await manager.handle_message(msg)
+
+    manager.send.assert_not_called()
+    manager.logger.debug.assert_called()
+
+
+@pytest.mark.asyncio
+async def test_setup(mock_settings):
+    manager = BDIProgramManager(name="program_manager_test")
+    manager.send = AsyncMock()
+
+    def close_coro(coro):
+        coro.close()
+        return MagicMock()
+
+    manager.add_behavior = MagicMock(side_effect=close_coro)
+
+    mock_context = MagicMock()
+    mock_sub = MagicMock()
+    mock_context.socket.return_value = mock_sub
+
+    with patch(
+        "control_backend.agents.bdi.bdi_program_manager.Context.instance", return_value=mock_context
+    ):
+        # We also need to mock file writing in _create_agentspeak_and_send_to_bdi
+        with patch("builtins.open", new_callable=MagicMock):
+            await manager.setup()
+
+    # Check logic
+    # 1. Sends default empty program to BDI
+    assert manager.send.await_count == 1
+    assert manager.send.await_args[0][0].to == mock_settings.agent_settings.bdi_core_name
+
+    # 2. Connects SUB socket
+    mock_sub.connect.assert_called_with(mock_settings.zmq_settings.internal_sub_address)
+    mock_sub.subscribe.assert_called_with("program")
+
+    # 3. Adds behavior
+    manager.add_behavior.assert_called()

test/unit/agents/bdi/test_belief_collector.py

@@ -1,135 +0,0 @@
-import json
-from unittest.mock import AsyncMock
-
-import pytest
-
-from control_backend.agents.bdi import (
-    BDIBeliefCollectorAgent,
-)
-from control_backend.core.agent_system import InternalMessage
-from control_backend.core.config import settings
-from control_backend.schemas.belief_message import Belief
-
-
-@pytest.fixture
-def agent():
-    agent = BDIBeliefCollectorAgent("belief_collector_agent")
-    return agent
-
-
-def make_msg(body: dict, sender: str = "sender"):
-    return InternalMessage(to="collector", sender=sender, body=json.dumps(body))
-
-
-@pytest.mark.asyncio
-async def test_handle_message_routes_belief_text(agent, mocker):
-    """
-    Test that when a message is received, _handle_belief_text is called with that message.
-    """
-    payload = {"type": "belief_extraction_text", "beliefs": {"user_said": [["hi"]]}}
-    spy = mocker.patch.object(agent, "_handle_belief_text", new_callable=AsyncMock)
-
-    await agent.handle_message(make_msg(payload))
-
-    spy.assert_awaited_once_with(payload, "sender")
-
-
-@pytest.mark.asyncio
-async def test_handle_message_routes_emotion(agent, mocker):
-    payload = {"type": "emotion_extraction_text"}
-    spy = mocker.patch.object(agent, "_handle_emo_text", new_callable=AsyncMock)
-
-    await agent.handle_message(make_msg(payload))
-
-    spy.assert_awaited_once_with(payload, "sender")
-
-
-@pytest.mark.asyncio
-async def test_handle_message_bad_json(agent, mocker):
-    agent._handle_belief_text = AsyncMock()
-    bad_msg = InternalMessage(to="collector", sender="sender", body="not json")
-
-    await agent.handle_message(bad_msg)
-
-    agent._handle_belief_text.assert_not_awaited()
-
-
-@pytest.mark.asyncio
-async def test_handle_belief_text_sends_when_beliefs_exist(agent, mocker):
-    payload = {"type": "belief_extraction_text", "beliefs": {"user_said": ["hello"]}}
-    spy = mocker.patch.object(agent, "_send_beliefs_to_bdi", new_callable=AsyncMock)
-    expected = [Belief(name="user_said", arguments=["hello"])]
-
-    await agent._handle_belief_text(payload, "origin")
-
-    spy.assert_awaited_once_with(expected, origin="origin")
-
-
-@pytest.mark.asyncio
-async def test_handle_belief_text_no_send_when_empty(agent, mocker):
-    payload = {"type": "belief_extraction_text", "beliefs": {}}
-    spy = mocker.patch.object(agent, "_send_beliefs_to_bdi", new_callable=AsyncMock)
-
-    await agent._handle_belief_text(payload, "origin")
-
-    spy.assert_not_awaited()
-
-
-@pytest.mark.asyncio
-async def test_send_beliefs_to_bdi(agent):
-    agent.send = AsyncMock()
-    beliefs = [Belief(name="user_said", arguments=["hello", "world"])]
-
-    await agent._send_beliefs_to_bdi(beliefs, origin="origin")
-
-    agent.send.assert_awaited_once()
-    sent: InternalMessage = agent.send.call_args.args[0]
-    assert sent.to == settings.agent_settings.bdi_core_name
-    assert sent.thread == "beliefs"
-    assert json.loads(sent.body)["create"] == [belief.model_dump() for belief in beliefs]
-
-
-@pytest.mark.asyncio
-async def test_setup_executes(agent):
-    """Covers setup and asserts the agent has a name."""
-    await agent.setup()
-    assert agent.name == "belief_collector_agent"  # simple property assertion
-
-
-@pytest.mark.asyncio
-async def test_handle_message_unrecognized_type_executes(agent):
-    """Covers the else branch for unrecognized message type."""
-    payload = {"type": "unknown_type"}
-    msg = make_msg(payload, sender="tester")
-    # Wrap send to ensure nothing is sent
-    agent.send = AsyncMock()
-    await agent.handle_message(msg)
-    # Assert no messages were sent
-    agent.send.assert_not_awaited()
-
-
-@pytest.mark.asyncio
-async def test_handle_emo_text_executes(agent):
-    """Covers the _handle_emo_text method."""
-    # The method does nothing, but we can assert it returns None
-    result = await agent._handle_emo_text({}, "origin")
-    assert result is None
-
-
-@pytest.mark.asyncio
-async def test_send_beliefs_to_bdi_empty_executes(agent):
-    """Covers early return when beliefs are empty."""
-    agent.send = AsyncMock()
-    await agent._send_beliefs_to_bdi({})
-    # Assert that nothing was sent
-    agent.send.assert_not_awaited()
-
-
-@pytest.mark.asyncio
-async def test_handle_belief_text_invalid_returns_none(agent, mocker):
-    payload = {"type": "belief_extraction_text", "beliefs": {"user_said": "invalid-argument"}}
-
-    result = await agent._handle_belief_text(payload, "origin")
-
-    # The method itself returns None
-    assert result is None

test/unit/agents/bdi/test_text_belief_extractor.py

@@ -14,6 +14,7 @@ from control_backend.schemas.belief_message import Belief as InternalBelief
 from control_backend.schemas.belief_message import BeliefMessage
 from control_backend.schemas.chat_history import ChatHistory, ChatMessage
 from control_backend.schemas.program import (
+    BaseGoal,  # Changed from Goal
     ConditionalNorm,
     KeywordBelief,
     LLMAction,
@@ -28,7 +29,8 @@ from control_backend.schemas.program import (
 @pytest.fixture
 def llm():
     llm = TextBeliefExtractorAgent.LLM(MagicMock(), 4)
-    llm._query_llm = AsyncMock()
+    # We must ensure _query_llm returns a dictionary so iterating it doesn't fail
+    llm._query_llm = AsyncMock(return_value={})
     return llm
 
 
@@ -357,6 +359,30 @@ async def test_simulated_real_turn_remove_belief(agent, llm, sample_program):
     assert any(b.name == "no_more_booze" for b in agent._current_beliefs.false)
 
 
+@pytest.mark.asyncio
+async def test_infer_goal_completions_sends_beliefs(agent, llm):
+    """Test that inferred goal completions are sent to the BDI core."""
+    goal = BaseGoal(
+        id=uuid.uuid4(), name="Say Hello", description="The user said hello", can_fail=True
+    )
+    agent.goal_inferrer.goals = {goal}
+
+    # Mock goal inference: goal is achieved
+    llm.query = AsyncMock(return_value=True)
+
+    await agent._infer_goal_completions()
+
+    # Should send belief change to BDI core
+    agent.send.assert_awaited_once()
+    sent: InternalMessage = agent.send.call_args.args[0]
+    assert sent.to == settings.agent_settings.bdi_core_name
+    assert sent.thread == "beliefs"
+
+    parsed = BeliefMessage.model_validate_json(sent.body)
+    assert len(parsed.create) == 1
+    assert parsed.create[0].name == "achieved_say_hello"
+
+
 @pytest.mark.asyncio
 async def test_llm_failure_handling(agent, llm, sample_program):
     """
@@ -374,3 +400,155 @@ async def test_llm_failure_handling(agent, llm, sample_program):
 
     assert len(belief_changes.true) == 0
     assert len(belief_changes.false) == 0
+
+
+def test_belief_state_bool():
+    # Empty
+    bs = BeliefState()
+    assert not bs
+
+    # True set
+    bs_true = BeliefState(true={InternalBelief(name="a", arguments=None)})
+    assert bs_true
+
+    # False set
+    bs_false = BeliefState(false={InternalBelief(name="a", arguments=None)})
+    assert bs_false
+
+
+@pytest.mark.asyncio
+async def test_handle_beliefs_message_validation_error(agent, mock_settings):
+    # Invalid JSON
+    mock_settings.agent_settings.bdi_program_manager_name = "bdi_program_manager_agent"
+    msg = InternalMessage(
+        to="me",
+        sender=mock_settings.agent_settings.bdi_program_manager_name,
+        thread="beliefs",
+        body="invalid json",
+    )
+    # Should log warning and return
+    agent.logger = MagicMock()
+    await agent.handle_message(msg)
+    agent.logger.warning.assert_called()
+
+    # Invalid Model
+    msg.body = json.dumps({"beliefs": [{"invalid": "obj"}]})
+    await agent.handle_message(msg)
+    agent.logger.warning.assert_called()
+
+
+@pytest.mark.asyncio
+async def test_handle_goals_message_validation_error(agent, mock_settings):
+    mock_settings.agent_settings.bdi_program_manager_name = "bdi_program_manager_agent"
+    msg = InternalMessage(
+        to="me",
+        sender=mock_settings.agent_settings.bdi_program_manager_name,
+        thread="goals",
+        body="invalid json",
+    )
+    agent.logger = MagicMock()
+    await agent.handle_message(msg)
+    agent.logger.warning.assert_called()
+
+
+@pytest.mark.asyncio
+async def test_handle_goal_achieved_message_validation_error(agent, mock_settings):
+    mock_settings.agent_settings.bdi_program_manager_name = "bdi_program_manager_agent"
+    msg = InternalMessage(
+        to="me",
+        sender=mock_settings.agent_settings.bdi_program_manager_name,
+        thread="achieved_goals",
+        body="invalid json",
+    )
+    agent.logger = MagicMock()
+    await agent.handle_message(msg)
+    agent.logger.warning.assert_called()
+
+
+@pytest.mark.asyncio
+async def test_goal_inferrer_infer_from_conversation(agent, llm):
+    # Setup goals
+    # Use BaseGoal object as typically received by the extractor
+    g1 = BaseGoal(id=uuid.uuid4(), name="g1", description="desc", can_fail=True)
+
+    # Use real GoalAchievementInferrer
+    from control_backend.agents.bdi.text_belief_extractor_agent import GoalAchievementInferrer
+
+    inferrer = GoalAchievementInferrer(llm)
+    inferrer.goals = {g1}
+
+    # Mock LLM response
+    llm._query_llm.return_value = True
+
+    completions = await inferrer.infer_from_conversation(ChatHistory(messages=[]))
+    assert completions
+    # slugify uses slugify library, hard to predict exact string without it,
+    # but we can check values
+    assert list(completions.values())[0] is True
+
+
+def test_apply_conversation_message_limit(agent):
+    with patch("control_backend.agents.bdi.text_belief_extractor_agent.settings") as mock_s:
+        mock_s.behaviour_settings.conversation_history_length_limit = 2
+        agent.conversation.messages = []
+
+        agent._apply_conversation_message(ChatMessage(role="user", content="1"))
+        agent._apply_conversation_message(ChatMessage(role="assistant", content="2"))
+        agent._apply_conversation_message(ChatMessage(role="user", content="3"))
+
+        assert len(agent.conversation.messages) == 2
+        assert agent.conversation.messages[0].content == "2"
+        assert agent.conversation.messages[1].content == "3"
+
+
+@pytest.mark.asyncio
+async def test_handle_program_manager_reset(agent):
+    with patch("control_backend.agents.bdi.text_belief_extractor_agent.settings") as mock_s:
+        mock_s.agent_settings.bdi_program_manager_name = "pm"
+        agent.conversation.messages = [ChatMessage(role="user", content="hi")]
+        agent.belief_inferrer.available_beliefs = [
+            SemanticBelief(id=uuid.uuid4(), name="b", description="d")
+        ]
+
+        msg = InternalMessage(to="me", sender="pm", thread="conversation_history", body="reset")
+        await agent.handle_message(msg)
+
+        assert len(agent.conversation.messages) == 0
+        assert len(agent.belief_inferrer.available_beliefs) == 0
+
+
+def test_split_into_chunks():
+    from control_backend.agents.bdi.text_belief_extractor_agent import SemanticBeliefInferrer
+
+    items = [1, 2, 3, 4, 5]
+    chunks = SemanticBeliefInferrer._split_into_chunks(items, 2)
+    assert len(chunks) == 2
+    assert len(chunks[0]) + len(chunks[1]) == 5
+
+
+@pytest.mark.asyncio
+async def test_infer_beliefs_call(agent, llm):
+    from control_backend.agents.bdi.text_belief_extractor_agent import SemanticBeliefInferrer
+
+    inferrer = SemanticBeliefInferrer(llm)
+    sb = SemanticBelief(id=uuid.uuid4(), name="is_happy", description="User is happy")
+
+    llm.query = AsyncMock(return_value={"is_happy": True})
+
+    res = await inferrer._infer_beliefs(ChatHistory(messages=[]), [sb])
+    assert res == {"is_happy": True}
+    llm.query.assert_called_once()
+
+
+@pytest.mark.asyncio
+async def test_infer_goal_call(agent, llm):
+    from control_backend.agents.bdi.text_belief_extractor_agent import GoalAchievementInferrer
+
+    inferrer = GoalAchievementInferrer(llm)
+    goal = BaseGoal(id=uuid.uuid4(), name="g1", description="d")
+
+    llm.query = AsyncMock(return_value=True)
+
+    res = await inferrer._infer_goal(ChatHistory(messages=[]), goal)
+    assert res is True
+    llm.query.assert_called_once()

test/unit/agents/communication/test_ri_communication_agent.py

@@ -4,6 +4,8 @@ from unittest.mock import ANY, AsyncMock, MagicMock, patch
 import pytest
 
 from control_backend.agents.communication.ri_communication_agent import RICommunicationAgent
+from control_backend.core.agent_system import InternalMessage
+from control_backend.schemas.ri_message import PauseCommand, RIEndpoint
 
 
 def speech_agent_path():
@@ -53,7 +55,11 @@ async def test_setup_success_connects_and_starts_robot(zmq_context):
         MockGesture.return_value.start = AsyncMock()
         agent = RICommunicationAgent("ri_comm", address="tcp://localhost:5555", bind=False)
 
-        agent.add_behavior = MagicMock()
+        def close_coro(coro):
+            coro.close()
+            return MagicMock()
+
+        agent.add_behavior = MagicMock(side_effect=close_coro)
 
         await agent.setup()
 
@@ -83,7 +89,11 @@ async def test_setup_binds_when_requested(zmq_context):
 
     agent = RICommunicationAgent("ri_comm", address="tcp://localhost:5555", bind=True)
 
-    agent.add_behavior = MagicMock()
+    def close_coro(coro):
+        coro.close()
+        return MagicMock()
+
+    agent.add_behavior = MagicMock(side_effect=close_coro)
 
     with (
         patch(speech_agent_path(), autospec=True) as MockSpeech,
@@ -151,6 +161,7 @@ async def test_handle_negotiation_response_updates_req_socket(zmq_context):
 @pytest.mark.asyncio
 async def test_handle_disconnection_publishes_and_reconnects():
     pub_socket = AsyncMock()
+    pub_socket.close = MagicMock()
     agent = RICommunicationAgent("ri_comm")
     agent.pub_socket = pub_socket
     agent.connected = True
@@ -233,6 +244,25 @@ async def test_handle_negotiation_response_unhandled_id():
     )
 
 
+@pytest.mark.asyncio
+async def test_handle_negotiation_response_audio(zmq_context):
+    agent = RICommunicationAgent("ri_comm")
+
+    with patch(
+        "control_backend.agents.communication.ri_communication_agent.VADAgent", autospec=True
+    ) as MockVAD:
+        MockVAD.return_value.start = AsyncMock()
+
+        await agent._handle_negotiation_response(
+            {"data": [{"id": "audio", "port": 7000, "bind": False}]}
+        )
+
+        MockVAD.assert_called_once_with(
+            audio_in_address="tcp://localhost:7000", audio_in_bind=False
+        )
+        MockVAD.return_value.start.assert_awaited_once()
+
+
 @pytest.mark.asyncio
 async def test_stop_closes_sockets():
     req = MagicMock()
@@ -323,6 +353,7 @@ async def test_listen_loop_generic_exception():
 @pytest.mark.asyncio
 async def test_handle_disconnection_timeout(monkeypatch):
     pub = AsyncMock()
+    pub.close = MagicMock()
     pub.send_multipart = AsyncMock(side_effect=TimeoutError)
 
     agent = RICommunicationAgent("ri_comm")
@@ -365,3 +396,38 @@ async def test_negotiate_req_socket_none_causes_retry(zmq_context):
     result = await agent._negotiate_connection(max_retries=1)
 
     assert result is False
+
+
+@pytest.mark.asyncio
+async def test_handle_message_pause_command(zmq_context):
+    """Test handle_message with a valid PauseCommand."""
+    agent = RICommunicationAgent("ri_comm")
+    agent._req_socket = AsyncMock()
+    agent.logger = MagicMock()
+
+    agent._req_socket.recv_json.return_value = {"status": "ok"}
+
+    pause_cmd = PauseCommand(data=True)
+    msg = InternalMessage(to="ri_comm", sender="user_int", body=pause_cmd.model_dump_json())
+
+    await agent.handle_message(msg)
+
+    agent._req_socket.send_json.assert_awaited_once()
+    args = agent._req_socket.send_json.await_args[0][0]
+    assert args["endpoint"] == RIEndpoint.PAUSE.value
+    assert args["data"] is True
+
+
+@pytest.mark.asyncio
+async def test_handle_message_invalid_pause_command(zmq_context):
+    """Test handle_message with invalid JSON."""
+    agent = RICommunicationAgent("ri_comm")
+    agent._req_socket = AsyncMock()
+    agent.logger = MagicMock()
+
+    msg = InternalMessage(to="ri_comm", sender="user_int", body="invalid json")
+
+    await agent.handle_message(msg)
+
+    agent.logger.warning.assert_called_with("Incorrect message format for PauseCommand.")
+    agent._req_socket.send_json.assert_not_called()

test/unit/agents/llm/test_llm_agent.py

@@ -58,17 +58,64 @@ async def test_llm_processing_success(mock_httpx_client, mock_settings):
         to="llm_agent",
         sender=mock_settings.agent_settings.bdi_core_name,
         body=prompt.model_dump_json(),
+        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
+    # The agent should call send once with the full sentence, PLUS once more for full reply
     assert agent.send.called
-    args = agent.send.call_args_list[0][0][0]
-    assert args.to == mock_settings.agent_settings.bdi_core_name
-    assert "Hello world." in args.body
+
+    # Check args. We expect at least one call sending "Hello world."
+    calls = agent.send.call_args_list
+    bodies = [c[0][0].body for c in calls]
+    assert any("Hello world." in b for b in bodies)
 
 
 @pytest.mark.asyncio
@@ -76,22 +123,15 @@ 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(),
-    )
 
-    # HTTP Error
+    # HTTP Error: stream method RAISES exception immediately
     mock_httpx_client.stream = MagicMock(side_effect=httpx.HTTPError("Fail"))
-    await agent.handle_message(msg)
-    assert "LLM service unavailable." in agent.send.call_args[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[0][0].body
+    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
 
 
 @pytest.mark.asyncio
@@ -113,16 +153,13 @@ async def test_llm_json_error(mock_httpx_client, mock_settings):
 
     agent = LLMAgent("llm_agent")
     agent.send = AsyncMock()
+    # Ensure logger is mocked
+    agent.logger = MagicMock()
 
-    with patch.object(agent.logger, "error") as log:
-        prompt = LLMPromptMessage(text="Hi", norms=[], goals=[])
-        msg = InternalMessage(
-            to="llm",
-            sender=mock_settings.agent_settings.bdi_core_name,
-            body=prompt.model_dump_json(),
-        )
-        await agent.handle_message(msg)
-        log.assert_called()  # Should log JSONDecodeError
+    prompt = LLMPromptMessage(text="Hi", norms=[], goals=[])
+    await agent._process_bdi_message(prompt)
+
+    agent.logger.error.assert_called()  # Should log JSONDecodeError
 
 
 def test_llm_instructions():
@@ -157,6 +194,7 @@ async def test_handle_message_validation_error_branch_no_send(mock_httpx_client,
         to="llm_agent",
         sender=mock_settings.agent_settings.bdi_core_name,
         body=invalid_json,
+        thread="prompt_message",
     )
 
     await agent.handle_message(msg)
@@ -285,3 +323,28 @@ async def test_clear_history_command(mock_settings):
     )
     await agent.handle_message(msg)
     assert len(agent.history) == 0
+
+
+@pytest.mark.asyncio
+async def test_handle_assistant_and_user_messages(mock_settings):
+    agent = LLMAgent("llm_agent")
+
+    # Assistant message
+    msg_ast = InternalMessage(
+        to="llm_agent",
+        sender=mock_settings.agent_settings.bdi_core_name,
+        thread="assistant_message",
+        body="I said this",
+    )
+    await agent.handle_message(msg_ast)
+    assert agent.history[-1] == {"role": "assistant", "content": "I said this"}
+
+    # User message
+    msg_usr = InternalMessage(
+        to="llm_agent",
+        sender=mock_settings.agent_settings.bdi_core_name,
+        thread="user_message",
+        body="User said this",
+    )
+    await agent.handle_message(msg_usr)
+    assert agent.history[-1] == {"role": "user", "content": "User said this"}

test/unit/agents/perception/transcription_agent/test_speech_recognizer.py

@@ -55,4 +55,6 @@ def test_get_decode_options():
     assert isinstance(options["sample_len"], int)
 
     # When disabled, it should not limit output length based on input size
-    assert "sample_rate" not in options
+    recognizer = OpenAIWhisperSpeechRecognizer(limit_output_length=False)
+    options = recognizer._get_decode_options(audio)
+    assert "sample_len" not in options

test/unit/agents/perception/transcription_agent/test_transcription_agent.py

@@ -36,7 +36,12 @@ async def test_transcription_agent_flow(mock_zmq_context):
         agent.send = AsyncMock()
 
         agent._running = True
-        agent.add_behavior = AsyncMock()
+
+        def close_coro(coro):
+            coro.close()
+            return MagicMock()
+
+        agent.add_behavior = MagicMock(side_effect=close_coro)
 
         await agent.setup()
 
@@ -143,7 +148,12 @@ async def test_transcription_loop_continues_after_error(mock_zmq_context):
         agent = TranscriptionAgent("tcp://in")
         agent._running = True  # ← REQUIRED to enter the loop
         agent.send = AsyncMock()  # should never be called
-        agent.add_behavior = AsyncMock()  # match other tests
+
+        def close_coro(coro):
+            coro.close()
+            return MagicMock()
+
+        agent.add_behavior = MagicMock(side_effect=close_coro)  # match other tests
 
         await agent.setup()
 
@@ -180,7 +190,12 @@ async def test_transcription_continue_branch_when_empty(mock_zmq_context):
         # Make loop runnable
         agent._running = True
         agent.send = AsyncMock()
-        agent.add_behavior = AsyncMock()
+
+        def close_coro(coro):
+            coro.close()
+            return MagicMock()
+
+        agent.add_behavior = MagicMock(side_effect=close_coro)
 
         await agent.setup()
 

test/unit/agents/perception/vad_agent/test_vad_agent_unit.py

@@ -0,0 +1,152 @@
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import pytest
+
+from control_backend.agents.perception.vad_agent import VADAgent
+from control_backend.core.agent_system import InternalMessage
+from control_backend.schemas.program_status import PROGRAM_STATUS, ProgramStatus
+
+
+@pytest.fixture(autouse=True)
+def mock_zmq():
+    with patch("zmq.asyncio.Context") as mock:
+        mock.instance.return_value = MagicMock()
+        yield mock
+
+
+@pytest.fixture
+def agent():
+    return VADAgent("tcp://localhost:5555", False)
+
+
+@pytest.mark.asyncio
+async def test_handle_message_pause(agent):
+    agent._paused = MagicMock()
+    # It starts set (not paused)
+
+    msg = InternalMessage(to="vad", sender="user_interrupt_agent", body="PAUSE")
+
+    # We need to mock settings to match sender name
+    with patch("control_backend.agents.perception.vad_agent.settings") as mock_settings:
+        mock_settings.agent_settings.user_interrupt_name = "user_interrupt_agent"
+
+        await agent.handle_message(msg)
+
+        agent._paused.clear.assert_called_once()
+        assert agent._reset_needed is True
+
+
+@pytest.mark.asyncio
+async def test_handle_message_resume(agent):
+    agent._paused = MagicMock()
+    msg = InternalMessage(to="vad", sender="user_interrupt_agent", body="RESUME")
+
+    with patch("control_backend.agents.perception.vad_agent.settings") as mock_settings:
+        mock_settings.agent_settings.user_interrupt_name = "user_interrupt_agent"
+
+        await agent.handle_message(msg)
+
+        agent._paused.set.assert_called_once()
+
+
+@pytest.mark.asyncio
+async def test_handle_message_unknown_command(agent):
+    agent._paused = MagicMock()
+    msg = InternalMessage(to="vad", sender="user_interrupt_agent", body="UNKNOWN")
+
+    with patch("control_backend.agents.perception.vad_agent.settings") as mock_settings:
+        mock_settings.agent_settings.user_interrupt_name = "user_interrupt_agent"
+        agent.logger = MagicMock()
+
+        await agent.handle_message(msg)
+
+        agent._paused.clear.assert_not_called()
+        agent._paused.set.assert_not_called()
+
+
+@pytest.mark.asyncio
+async def test_handle_message_unknown_sender(agent):
+    agent._paused = MagicMock()
+    msg = InternalMessage(to="vad", sender="other_agent", body="PAUSE")
+
+    with patch("control_backend.agents.perception.vad_agent.settings") as mock_settings:
+        mock_settings.agent_settings.user_interrupt_name = "user_interrupt_agent"
+
+        await agent.handle_message(msg)
+
+        agent._paused.clear.assert_not_called()
+
+
+@pytest.mark.asyncio
+async def test_status_loop_waits_for_running(agent):
+    agent._running = True
+    agent.program_sub_socket = AsyncMock()
+    agent.program_sub_socket.close = MagicMock()
+    agent._reset_stream = AsyncMock()
+
+    # Sequence of messages:
+    # 1. Wrong topic
+    # 2. Right topic, wrong status (STARTING)
+    # 3. Right topic, RUNNING -> Should break loop
+
+    agent.program_sub_socket.recv_multipart.side_effect = [
+        (b"wrong_topic", b"whatever"),
+        (PROGRAM_STATUS, ProgramStatus.STARTING.value),
+        (PROGRAM_STATUS, ProgramStatus.RUNNING.value),
+    ]
+
+    await agent._status_loop()
+
+    assert agent._reset_stream.await_count == 1
+    agent.program_sub_socket.close.assert_called_once()
+
+
+@pytest.mark.asyncio
+async def test_setup_success(agent, mock_zmq):
+    def close_coro(coro):
+        coro.close()
+        return MagicMock()
+
+    agent.add_behavior = MagicMock(side_effect=close_coro)
+
+    mock_context = mock_zmq.instance.return_value
+    mock_sub = MagicMock()
+    mock_pub = MagicMock()
+
+    # We expect multiple socket calls:
+    # 1. audio_in (SUB)
+    # 2. audio_out (PUB)
+    # 3. program_sub (SUB)
+    mock_context.socket.side_effect = [mock_sub, mock_pub, mock_sub]
+
+    with patch("control_backend.agents.perception.vad_agent.torch.hub.load") as mock_load:
+        mock_load.return_value = (MagicMock(), None)
+
+        with patch("control_backend.agents.perception.vad_agent.TranscriptionAgent") as MockTrans:
+            mock_trans_instance = MockTrans.return_value
+            mock_trans_instance.start = AsyncMock()
+
+            await agent.setup()
+
+            mock_trans_instance.start.assert_awaited_once()
+
+    assert agent.add_behavior.call_count == 2  # streaming_loop + status_loop
+    assert agent.audio_in_socket is not None
+    assert agent.audio_out_socket is not None
+    assert agent.program_sub_socket is not None
+
+
+@pytest.mark.asyncio
+async def test_reset_stream(agent):
+    mock_poller = MagicMock()
+    agent.audio_in_poller = mock_poller
+
+    # poll(1) returns not None twice, then None
+    mock_poller.poll = AsyncMock(side_effect=[b"data", b"data", None])
+
+    agent._ready = MagicMock()
+
+    await agent._reset_stream()
+
+    assert mock_poller.poll.await_count == 3
+    agent._ready.set.assert_called_once()

test/unit/agents/perception/vad_agent/test_vad_streaming.py

@@ -5,6 +5,7 @@ import pytest
 import zmq
 
 from control_backend.agents.perception.vad_agent import VADAgent
+from control_backend.core.config import settings
 
 
 # We don't want to use real ZMQ in unit tests, for example because it can give errors when sockets
@@ -135,6 +136,54 @@ async def test_no_data(audio_out_socket, vad_agent):
     assert len(vad_agent.audio_buffer) == 0
 
 
+@pytest.mark.asyncio
+async def test_streaming_loop_reset_needed(audio_out_socket, vad_agent):
+    """Test that _reset_needed branch works as expected."""
+    vad_agent._reset_needed = True
+    vad_agent._ready.set()
+    vad_agent._paused.set()
+    vad_agent._running = True
+    vad_agent.audio_buffer = np.array([1.0], dtype=np.float32)
+    vad_agent.i_since_speech = 0
+
+    # Mock _reset_stream to stop the loop by setting _running=False
+    async def mock_reset():
+        vad_agent._running = False
+
+    vad_agent._reset_stream = mock_reset
+
+    # Needs a poller to avoid AssertionError
+    vad_agent.audio_in_poller = AsyncMock()
+    vad_agent.audio_in_poller.poll.return_value = None
+
+    await vad_agent._streaming_loop()
+
+    assert vad_agent._reset_needed is False
+    assert len(vad_agent.audio_buffer) == 0
+    assert vad_agent.i_since_speech == settings.behaviour_settings.vad_initial_since_speech
+
+
+@pytest.mark.asyncio
+async def test_streaming_loop_no_data_clears_buffer(audio_out_socket, vad_agent):
+    """Test that if poll returns None, buffer is cleared if not empty."""
+    vad_agent.audio_buffer = np.array([1.0], dtype=np.float32)
+    vad_agent._ready.set()
+    vad_agent._paused.set()
+    vad_agent._running = True
+
+    class MockPoller:
+        async def poll(self, timeout_ms=None):
+            vad_agent._running = False  # stop after one poll
+            return None
+
+    vad_agent.audio_in_poller = MockPoller()
+
+    await vad_agent._streaming_loop()
+
+    assert len(vad_agent.audio_buffer) == 0
+    assert vad_agent.i_since_speech == settings.behaviour_settings.vad_initial_since_speech
+
+
 @pytest.mark.asyncio
 async def test_vad_model_load_failure_stops_agent(vad_agent):
     """

test/unit/agents/test_base.py

@@ -0,0 +1,24 @@
+import logging
+
+from control_backend.agents.base import BaseAgent
+
+
+class MyAgent(BaseAgent):
+    async def setup(self):
+        pass
+
+    async def handle_message(self, msg):
+        pass
+
+
+def test_base_agent_logger_init():
+    # When defining a subclass, __init_subclass__ runs
+    # The BaseAgent in agents/base.py sets the logger
+    assert hasattr(MyAgent, "logger")
+    assert isinstance(MyAgent.logger, logging.Logger)
+    # The logger name depends on the package.
+    # Since this test file is running as a module, __package__ might be None or the test package.
+    # In 'src/control_backend/agents/base.py', it uses __package__ of base.py which is
+    #  'control_backend.agents'.
+    # So logger name should be control_backend.agents.MyAgent
+    assert MyAgent.logger.name == "control_backend.agents.MyAgent"

test/unit/agents/user_interrupt/test_user_interrupt.py

@@ -7,6 +7,15 @@ import pytest
 from control_backend.agents.user_interrupt.user_interrupt_agent import UserInterruptAgent
 from control_backend.core.agent_system import InternalMessage
 from control_backend.core.config import settings
+from control_backend.schemas.program import (
+    ConditionalNorm,
+    Goal,
+    KeywordBelief,
+    Phase,
+    Plan,
+    Program,
+    Trigger,
+)
 from control_backend.schemas.ri_message import RIEndpoint
 
 
@@ -16,6 +25,7 @@ def agent():
     agent.send = AsyncMock()
     agent.logger = MagicMock()
     agent.sub_socket = AsyncMock()
+    agent.pub_socket = AsyncMock()
     return agent
 
 
@@ -49,21 +59,18 @@ async def test_send_to_gesture_agent(agent):
 
 
 @pytest.mark.asyncio
-async def test_send_to_program_manager(agent):
+async def test_send_to_bdi_belief(agent):
     """Verify belief update format."""
-    context_str = "2"
+    context_str = "some_goal"
 
-    await agent._send_to_program_manager(context_str)
+    await agent._send_to_bdi_belief(context_str, "goal")
 
-    agent.send.assert_awaited_once()
-    sent_msg: InternalMessage = agent.send.call_args.args[0]
+    assert agent.send.await_count == 1
+    sent_msg = agent.send.call_args.args[0]
 
-    assert sent_msg.to == settings.agent_settings.bdi_program_manager_name
-    assert sent_msg.thread == "belief_override_id"
-
-    body = json.loads(sent_msg.body)
-
-    assert body["belief"] == context_str
+    assert sent_msg.to == settings.agent_settings.bdi_core_name
+    assert sent_msg.thread == "beliefs"
+    assert "achieved_some_goal" in sent_msg.body
 
 
 @pytest.mark.asyncio
@@ -77,6 +84,10 @@ async def test_receive_loop_routing_success(agent):
     # Prepare JSON payloads as bytes
     payload_speech = json.dumps({"type": "speech", "context": "Hello Speech"}).encode()
     payload_gesture = json.dumps({"type": "gesture", "context": "Hello Gesture"}).encode()
+    # override calls _send_to_bdi (for trigger/norm) OR _send_to_bdi_belief (for goal).
+
+    # To test routing, we need to populate the maps
+    agent._goal_map["Hello Override"] = "some_goal_slug"
     payload_override = json.dumps({"type": "override", "context": "Hello Override"}).encode()
 
     agent.sub_socket.recv_multipart.side_effect = [
@@ -88,7 +99,7 @@ async def test_receive_loop_routing_success(agent):
 
     agent._send_to_speech_agent = AsyncMock()
     agent._send_to_gesture_agent = AsyncMock()
-    agent._send_to_program_manager = AsyncMock()
+    agent._send_to_bdi_belief = AsyncMock()
 
     try:
         await agent._receive_button_event()
@@ -103,12 +114,12 @@ async def test_receive_loop_routing_success(agent):
     # Gesture
     agent._send_to_gesture_agent.assert_awaited_once_with("Hello Gesture")
 
-    # Override
-    agent._send_to_program_manager.assert_awaited_once_with("Hello Override")
+    # Override (since we mapped it to a goal)
+    agent._send_to_bdi_belief.assert_awaited_once_with("some_goal_slug", "goal")
 
     assert agent._send_to_speech_agent.await_count == 1
     assert agent._send_to_gesture_agent.await_count == 1
-    assert agent._send_to_program_manager.await_count == 1
+    assert agent._send_to_bdi_belief.await_count == 1
 
 
 @pytest.mark.asyncio
@@ -125,7 +136,6 @@ async def test_receive_loop_unknown_type(agent):
 
     agent._send_to_speech_agent = AsyncMock()
     agent._send_to_gesture_agent = AsyncMock()
-    agent._send_to_belief_collector = AsyncMock()
 
     try:
         await agent._receive_button_event()
@@ -137,10 +147,165 @@ async def test_receive_loop_unknown_type(agent):
     # Ensure no handlers were called
     agent._send_to_speech_agent.assert_not_called()
     agent._send_to_gesture_agent.assert_not_called()
-    agent._send_to_belief_collector.assert_not_called()
 
-    agent.logger.warning.assert_called_with(
-        "Received button press with unknown type '%s' (context: '%s').",
-        "unknown_thing",
-        "some_data",
-    )
+    agent.logger.warning.assert_called()
+
+
+@pytest.mark.asyncio
+async def test_create_mapping(agent):
+    # Create a program with a trigger, goal, and conditional norm
+    import uuid
+
+    trigger_id = uuid.uuid4()
+    goal_id = uuid.uuid4()
+    norm_id = uuid.uuid4()
+
+    cond = KeywordBelief(id=uuid.uuid4(), name="k1", keyword="key")
+    plan = Plan(id=uuid.uuid4(), name="p1", steps=[])
+
+    trigger = Trigger(id=trigger_id, name="my_trigger", condition=cond, plan=plan)
+    goal = Goal(id=goal_id, name="my_goal", description="desc", plan=plan)
+
+    cn = ConditionalNorm(id=norm_id, name="my_norm", norm="be polite", condition=cond)
+
+    phase = Phase(id=uuid.uuid4(), name="phase1", norms=[cn], goals=[goal], triggers=[trigger])
+    prog = Program(phases=[phase])
+
+    # Call create_mapping via handle_message
+    msg = InternalMessage(to="me", thread="new_program", body=prog.model_dump_json())
+    await agent.handle_message(msg)
+
+    # Check maps
+    assert str(trigger_id) in agent._trigger_map
+    assert agent._trigger_map[str(trigger_id)] == "trigger_my_trigger"
+
+    assert str(goal_id) in agent._goal_map
+    assert agent._goal_map[str(goal_id)] == "my_goal"
+
+    assert str(norm_id) in agent._cond_norm_map
+    assert agent._cond_norm_map[str(norm_id)] == "norm_be_polite"
+
+
+@pytest.mark.asyncio
+async def test_create_mapping_invalid_json(agent):
+    # Pass invalid json to handle_message thread "new_program"
+    msg = InternalMessage(to="me", thread="new_program", body="invalid json")
+    await agent.handle_message(msg)
+
+    # Should log error and maps should remain empty or cleared
+    agent.logger.error.assert_called()
+
+
+@pytest.mark.asyncio
+async def test_handle_message_trigger_start(agent):
+    # Setup reverse map manually
+    agent._trigger_reverse_map["trigger_slug"] = "ui_id_123"
+
+    msg = InternalMessage(to="me", thread="trigger_start", body="trigger_slug")
+    await agent.handle_message(msg)
+
+    agent.pub_socket.send_multipart.assert_awaited_once()
+    args = agent.pub_socket.send_multipart.call_args[0][0]
+    assert args[0] == b"experiment"
+    payload = json.loads(args[1])
+    assert payload["type"] == "trigger_update"
+    assert payload["id"] == "ui_id_123"
+    assert payload["achieved"] is True
+
+
+@pytest.mark.asyncio
+async def test_handle_message_trigger_end(agent):
+    agent._trigger_reverse_map["trigger_slug"] = "ui_id_123"
+
+    msg = InternalMessage(to="me", thread="trigger_end", body="trigger_slug")
+    await agent.handle_message(msg)
+
+    agent.pub_socket.send_multipart.assert_awaited_once()
+    payload = json.loads(agent.pub_socket.send_multipart.call_args[0][0][1])
+    assert payload["type"] == "trigger_update"
+    assert payload["achieved"] is False
+
+
+@pytest.mark.asyncio
+async def test_handle_message_transition_phase(agent):
+    msg = InternalMessage(to="me", thread="transition_phase", body="phase_id_123")
+    await agent.handle_message(msg)
+
+    agent.pub_socket.send_multipart.assert_awaited_once()
+    payload = json.loads(agent.pub_socket.send_multipart.call_args[0][0][1])
+    assert payload["type"] == "phase_update"
+    assert payload["id"] == "phase_id_123"
+
+
+@pytest.mark.asyncio
+async def test_handle_message_goal_start(agent):
+    agent._goal_reverse_map["goal_slug"] = "goal_id_123"
+
+    msg = InternalMessage(to="me", thread="goal_start", body="goal_slug")
+    await agent.handle_message(msg)
+
+    agent.pub_socket.send_multipart.assert_awaited_once()
+    payload = json.loads(agent.pub_socket.send_multipart.call_args[0][0][1])
+    assert payload["type"] == "goal_update"
+    assert payload["id"] == "goal_id_123"
+    assert payload["active"] is True
+
+
+@pytest.mark.asyncio
+async def test_handle_message_active_norms_update(agent):
+    agent._cond_norm_reverse_map["norm_active"] = "id_1"
+    agent._cond_norm_reverse_map["norm_inactive"] = "id_2"
+
+    # Body is like: "('norm_active', 'other')"
+    # The split logic handles quotes etc.
+    msg = InternalMessage(to="me", thread="active_norms_update", body="'norm_active', 'other'")
+    await agent.handle_message(msg)
+
+    agent.pub_socket.send_multipart.assert_awaited_once()
+    payload = json.loads(agent.pub_socket.send_multipart.call_args[0][0][1])
+    assert payload["type"] == "cond_norms_state_update"
+    norms = {n["id"]: n["active"] for n in payload["norms"]}
+    assert norms["id_1"] is True
+    assert norms["id_2"] is False
+
+
+@pytest.mark.asyncio
+async def test_send_experiment_control(agent):
+    # Test next_phase
+    await agent._send_experiment_control_to_bdi_core("next_phase")
+    agent.send.assert_awaited()
+    msg = agent.send.call_args[0][0]
+    assert msg.thread == "force_next_phase"
+
+    # Test reset_phase
+    await agent._send_experiment_control_to_bdi_core("reset_phase")
+    msg = agent.send.call_args[0][0]
+    assert msg.thread == "reset_current_phase"
+
+    # Test reset_experiment
+    await agent._send_experiment_control_to_bdi_core("reset_experiment")
+    msg = agent.send.call_args[0][0]
+    assert msg.thread == "reset_experiment"
+
+
+@pytest.mark.asyncio
+async def test_send_pause_command(agent):
+    await agent._send_pause_command("true")
+    # Sends to RI and VAD
+    assert agent.send.await_count == 2
+    msgs = [call.args[0] for call in agent.send.call_args_list]
+
+    ri_msg = next(m for m in msgs if m.to == settings.agent_settings.ri_communication_name)
+    assert json.loads(ri_msg.body)["endpoint"] == ""  # PAUSE endpoint
+    assert json.loads(ri_msg.body)["data"] is True
+
+    vad_msg = next(m for m in msgs if m.to == settings.agent_settings.vad_name)
+    assert vad_msg.body == "PAUSE"
+
+    agent.send.reset_mock()
+    await agent._send_pause_command("false")
+    assert agent.send.await_count == 2
+    vad_msg = next(
+        m for m in agent.send.call_args_list if m.args[0].to == settings.agent_settings.vad_name
+    ).args[0]
+    assert vad_msg.body == "RESUME"

test/unit/api/v1/endpoints/test_logs_endpoint.py

@@ -1,7 +1,7 @@
-from unittest.mock import patch
+from unittest.mock import MagicMock, patch
 
 import pytest
-from fastapi import FastAPI
+from fastapi import FastAPI, HTTPException
 from fastapi.testclient import TestClient
 from starlette.responses import StreamingResponse
 
@@ -61,3 +61,67 @@ async def test_log_stream_endpoint_lines(client):
         # Optional: assert subscribe/connect were called
         assert dummy_socket.subscribed  # at least some log levels subscribed
         assert dummy_socket.connected  # connect was called
+
+
+@patch("control_backend.api.v1.endpoints.logs.LOGGING_DIR")
+def test_files_endpoint(LOGGING_DIR, client):
+    file_1, file_2 = MagicMock(), MagicMock()
+    file_1.name = "file_1"
+    file_2.name = "file_2"
+    LOGGING_DIR.glob.return_value = [file_1, file_2]
+    result = client.get("/api/logs/files")
+
+    assert result.status_code == 200
+    assert result.json() == ["file_1", "file_2"]
+
+
+@patch("control_backend.api.v1.endpoints.logs.FileResponse")
+@patch("control_backend.api.v1.endpoints.logs.LOGGING_DIR")
+def test_log_file_endpoint_success(LOGGING_DIR, MockFileResponse, client):
+    mock_file_path = MagicMock()
+    mock_file_path.is_relative_to.return_value = True
+    mock_file_path.is_file.return_value = True
+    mock_file_path.name = "test.log"
+
+    LOGGING_DIR.__truediv__ = MagicMock(return_value=mock_file_path)
+    mock_file_path.resolve.return_value = mock_file_path
+
+    MockFileResponse.return_value = MagicMock()
+
+    result = client.get("/api/logs/files/test.log")
+
+    assert result.status_code == 200
+    MockFileResponse.assert_called_once_with(mock_file_path, filename="test.log")
+
+
+@pytest.mark.asyncio
+@patch("control_backend.api.v1.endpoints.logs.LOGGING_DIR")
+async def test_log_file_endpoint_path_traversal(LOGGING_DIR):
+    from control_backend.api.v1.endpoints.logs import log_file
+
+    mock_file_path = MagicMock()
+    mock_file_path.is_relative_to.return_value = False
+
+    LOGGING_DIR.__truediv__ = MagicMock(return_value=mock_file_path)
+    mock_file_path.resolve.return_value = mock_file_path
+
+    with pytest.raises(HTTPException) as exc_info:
+        await log_file("../secret.txt")
+
+    assert exc_info.value.status_code == 400
+    assert exc_info.value.detail == "Invalid filename."
+
+
+@patch("control_backend.api.v1.endpoints.logs.LOGGING_DIR")
+def test_log_file_endpoint_file_not_found(LOGGING_DIR, client):
+    mock_file_path = MagicMock()
+    mock_file_path.is_relative_to.return_value = True
+    mock_file_path.is_file.return_value = False
+
+    LOGGING_DIR.__truediv__ = MagicMock(return_value=mock_file_path)
+    mock_file_path.resolve.return_value = mock_file_path
+
+    result = client.get("/api/logs/files/nonexistent.log")
+
+    assert result.status_code == 404
+    assert result.json()["detail"] == "File not found."

test/unit/api/v1/endpoints/test_router.py

@@ -11,6 +11,5 @@ def test_router_includes_expected_paths():
     # Ensure at least one route under each prefix exists
     assert any(p.startswith("/robot") for p in paths)
     assert any(p.startswith("/message") for p in paths)
-    assert any(p.startswith("/sse") for p in paths)
     assert any(p.startswith("/logs") for p in paths)
     assert any(p.startswith("/program") for p in paths)

test/unit/api/v1/endpoints/test_sse_endpoint.py

@@ -1,24 +0,0 @@
-import pytest
-from fastapi import FastAPI
-from fastapi.testclient import TestClient
-
-from control_backend.api.v1.endpoints import sse
-
-
-@pytest.fixture
-def app():
-    app = FastAPI()
-    app.include_router(sse.router)
-    return app
-
-
-@pytest.fixture
-def client(app):
-    return TestClient(app)
-
-
-def test_sse_route_exists(client):
-    """Minimal smoke test to ensure /sse route exists and responds."""
-    response = client.get("/sse")
-    # Since implementation is not done, we only assert it doesn't crash
-    assert response.status_code == 200

test/unit/api/v1/endpoints/test_user_interact.py

@@ -0,0 +1,96 @@
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import pytest
+from fastapi import FastAPI
+from fastapi.testclient import TestClient
+
+from control_backend.api.v1.endpoints import user_interact
+
+
+@pytest.fixture
+def app():
+    app = FastAPI()
+    app.include_router(user_interact.router)
+    return app
+
+
+@pytest.fixture
+def client(app):
+    return TestClient(app)
+
+
+@pytest.mark.asyncio
+async def test_receive_button_event(client):
+    mock_pub_socket = AsyncMock()
+    client.app.state.endpoints_pub_socket = mock_pub_socket
+
+    payload = {"type": "speech", "context": "hello"}
+    response = client.post("/button_pressed", json=payload)
+
+    assert response.status_code == 202
+    assert response.json() == {"status": "Event received"}
+
+    mock_pub_socket.send_multipart.assert_awaited_once()
+    args = mock_pub_socket.send_multipart.call_args[0][0]
+    assert args[0] == b"button_pressed"
+    assert "speech" in args[1].decode()
+
+
+@pytest.mark.asyncio
+async def test_receive_button_event_invalid_payload(client):
+    mock_pub_socket = AsyncMock()
+    client.app.state.endpoints_pub_socket = mock_pub_socket
+
+    # Missing context
+    payload = {"type": "speech"}
+    response = client.post("/button_pressed", json=payload)
+
+    assert response.status_code == 422
+    mock_pub_socket.send_multipart.assert_not_called()
+
+
+@pytest.mark.asyncio
+async def test_experiment_stream_direct_call():
+    """
+    Directly calling the endpoint function to test the streaming logic
+    without dealing with TestClient streaming limitations.
+    """
+    mock_socket = AsyncMock()
+    # 1. recv data
+    # 2. recv timeout
+    # 3. disconnect (request.is_disconnected returns True)
+    mock_socket.recv_multipart.side_effect = [
+        (b"topic", b"message1"),
+        TimeoutError(),
+        (b"topic", b"message2"),  # Should not be reached if disconnect checks work
+    ]
+    mock_socket.close = MagicMock()
+    mock_socket.connect = MagicMock()
+    mock_socket.subscribe = MagicMock()
+
+    mock_context = MagicMock()
+    mock_context.socket.return_value = mock_socket
+
+    with patch(
+        "control_backend.api.v1.endpoints.user_interact.Context.instance", return_value=mock_context
+    ):
+        mock_request = AsyncMock()
+        # is_disconnected sequence:
+        # 1. False (before first recv) -> reads message1
+        # 2. False (before second recv) -> triggers TimeoutError, continues
+        # 3. True (before third recv) -> break loop
+        mock_request.is_disconnected.side_effect = [False, False, True]
+
+        response = await user_interact.experiment_stream(mock_request)
+
+        lines = []
+        # Consume the generator
+        async for line in response.body_iterator:
+            lines.append(line)
+
+        assert "data: message1\n\n" in lines
+        assert len(lines) == 1
+
+        mock_socket.connect.assert_called()
+        mock_socket.subscribe.assert_called_with(b"experiment")
+        mock_socket.close.assert_called()

test/unit/conftest.py

@@ -25,7 +25,6 @@ def mock_settings():
         mock.zmq_settings.internal_sub_address = "tcp://localhost:5561"
         mock.zmq_settings.ri_command_address = "tcp://localhost:0000"
         mock.agent_settings.bdi_core_name = "bdi_core_agent"
-        mock.agent_settings.bdi_belief_collector_name = "belief_collector_agent"
         mock.agent_settings.llm_name = "llm_agent"
         mock.agent_settings.robot_speech_name = "robot_speech_agent"
         mock.agent_settings.transcription_name = "transcription_agent"

test/unit/logging/test_dated_file_handler.py

@@ -0,0 +1,45 @@
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from control_backend.logging.dated_file_handler import DatedFileHandler
+
+
+@patch("control_backend.logging.dated_file_handler.DatedFileHandler._open")
+def test_reset(open_):
+    stream = MagicMock()
+    open_.return_value = stream
+
+    # A file should be opened when the logger is created
+    handler = DatedFileHandler(file_prefix="anything")
+    assert open_.call_count == 1
+
+    # Upon reset, the current file should be closed, and a new one should be opened
+    handler.do_rollover()
+    assert stream.close.call_count == 1
+    assert open_.call_count == 2
+
+
+@patch("control_backend.logging.dated_file_handler.Path")
+@patch("control_backend.logging.dated_file_handler.DatedFileHandler._open")
+def test_creates_dir(open_, Path_):
+    stream = MagicMock()
+    open_.return_value = stream
+
+    test_path = MagicMock()
+    test_path.parent.is_dir.return_value = False
+    Path_.return_value = test_path
+
+    DatedFileHandler(file_prefix="anything")
+
+    # The directory should've been created
+    test_path.parent.mkdir.assert_called_once()
+
+
+@patch("control_backend.logging.dated_file_handler.DatedFileHandler._open")
+def test_invalid_constructor(_):
+    with pytest.raises(ValueError):
+        DatedFileHandler(file_prefix=None)
+
+    with pytest.raises(ValueError):
+        DatedFileHandler(file_prefix="")

test/unit/logging/test_optional_field_formatter.py

@@ -0,0 +1,218 @@
+import logging
+
+import pytest
+
+from control_backend.logging.optional_field_formatter import OptionalFieldFormatter
+
+
+@pytest.fixture
+def logger():
+    """Create a fresh logger for each test."""
+    logger = logging.getLogger(f"test_{id(object())}")
+    logger.setLevel(logging.DEBUG)
+    logger.handlers = []
+    return logger
+
+
+@pytest.fixture
+def log_output(logger):
+    """Capture log output and return a function to get it."""
+
+    class ListHandler(logging.Handler):
+        def __init__(self):
+            super().__init__()
+            self.records = []
+
+        def emit(self, record):
+            self.records.append(self.format(record))
+
+    handler = ListHandler()
+    logger.addHandler(handler)
+
+    def get_output():
+        return handler.records
+
+    return get_output
+
+
+def test_optional_field_present(logger, log_output):
+    """Optional field should appear when provided in extra."""
+    formatter = OptionalFieldFormatter("%(levelname)s - %(role?)s - %(message)s")
+    logger.handlers[0].setFormatter(formatter)
+
+    logger.info("test message", extra={"role": "user"})
+
+    assert log_output() == ["INFO - user - test message"]
+
+
+def test_optional_field_missing_no_default(logger, log_output):
+    """Missing optional field with no default should be None."""
+    formatter = OptionalFieldFormatter("%(levelname)s - %(role?)s - %(message)s")
+    logger.handlers[0].setFormatter(formatter)
+
+    logger.info("test message")
+
+    assert log_output() == ["INFO - None - test message"]
+
+
+def test_optional_field_missing_with_default(logger, log_output):
+    """Missing optional field should use provided default."""
+    formatter = OptionalFieldFormatter(
+        "%(levelname)s - %(role?)s - %(message)s", defaults={"role": "assistant"}
+    )
+    logger.handlers[0].setFormatter(formatter)
+
+    logger.info("test message")
+
+    assert log_output() == ["INFO - assistant - test message"]
+
+
+def test_optional_field_overrides_default(logger, log_output):
+    """Provided extra value should override default."""
+    formatter = OptionalFieldFormatter(
+        "%(levelname)s - %(role?)s - %(message)s", defaults={"role": "assistant"}
+    )
+    logger.handlers[0].setFormatter(formatter)
+
+    logger.info("test message", extra={"role": "user"})
+
+    assert log_output() == ["INFO - user - test message"]
+
+
+def test_multiple_optional_fields(logger, log_output):
+    """Multiple optional fields should work independently."""
+    formatter = OptionalFieldFormatter(
+        "%(levelname)s - %(role?)s - %(request_id?)s - %(message)s", defaults={"role": "assistant"}
+    )
+    logger.handlers[0].setFormatter(formatter)
+
+    logger.info("test", extra={"request_id": "123"})
+
+    assert log_output() == ["INFO - assistant - 123 - test"]
+
+
+def test_mixed_optional_and_required_fields(logger, log_output):
+    """Standard fields should work alongside optional fields."""
+    formatter = OptionalFieldFormatter("%(levelname)s %(name)s %(role?)s %(message)s")
+    logger.handlers[0].setFormatter(formatter)
+
+    logger.info("test", extra={"role": "user"})
+
+    output = log_output()[0]
+    assert "INFO" in output
+    assert "user" in output
+    assert "test" in output
+
+
+def test_no_optional_fields(logger, log_output):
+    """Formatter should work normally with no optional fields."""
+    formatter = OptionalFieldFormatter("%(levelname)s %(message)s")
+    logger.handlers[0].setFormatter(formatter)
+
+    logger.info("test message")
+
+    assert log_output() == ["INFO test message"]
+
+
+def test_integer_format_specifier(logger, log_output):
+    """Optional fields with %d specifier should work."""
+    formatter = OptionalFieldFormatter(
+        "%(levelname)s %(count?)d %(message)s", defaults={"count": 0}
+    )
+    logger.handlers[0].setFormatter(formatter)
+
+    logger.info("test", extra={"count": 42})
+
+    assert log_output() == ["INFO 42 test"]
+
+
+def test_float_format_specifier(logger, log_output):
+    """Optional fields with %f specifier should work."""
+    formatter = OptionalFieldFormatter(
+        "%(levelname)s %(duration?)f %(message)s", defaults={"duration": 0.0}
+    )
+    logger.handlers[0].setFormatter(formatter)
+
+    logger.info("test", extra={"duration": 1.5})
+
+    assert "1.5" in log_output()[0]
+
+
+def test_empty_string_default(logger, log_output):
+    """Empty string default should work."""
+    formatter = OptionalFieldFormatter("%(levelname)s %(role?)s %(message)s", defaults={"role": ""})
+    logger.handlers[0].setFormatter(formatter)
+
+    logger.info("test")
+
+    assert log_output() == ["INFO  test"]
+
+
+def test_none_format_string():
+    """None format string should not raise."""
+    formatter = OptionalFieldFormatter(fmt=None)
+    assert formatter.optional_fields == set()
+
+
+def test_optional_fields_parsed_correctly():
+    """Check that optional fields are correctly identified."""
+    formatter = OptionalFieldFormatter("%(asctime)s %(role?)s %(level?)d %(name)s")
+
+    assert formatter.optional_fields == {("role", "s"), ("level", "d")}
+
+
+def test_format_string_normalized():
+    """Check that ? is removed from format string."""
+    formatter = OptionalFieldFormatter("%(role?)s %(message)s")
+
+    assert "?" not in formatter._style._fmt
+    assert "%(role)s" in formatter._style._fmt
+
+
+def test_field_with_underscore(logger, log_output):
+    """Field names with underscores should work."""
+    formatter = OptionalFieldFormatter("%(levelname)s %(user_id?)s %(message)s")
+    logger.handlers[0].setFormatter(formatter)
+
+    logger.info("test", extra={"user_id": "abc123"})
+
+    assert log_output() == ["INFO abc123 test"]
+
+
+def test_field_with_numbers(logger, log_output):
+    """Field names with numbers should work."""
+    formatter = OptionalFieldFormatter("%(levelname)s %(field2?)s %(message)s")
+    logger.handlers[0].setFormatter(formatter)
+
+    logger.info("test", extra={"field2": "value"})
+
+    assert log_output() == ["INFO value test"]
+
+
+def test_multiple_log_calls(logger, log_output):
+    """Formatter should work correctly across multiple log calls."""
+    formatter = OptionalFieldFormatter(
+        "%(levelname)s %(role?)s %(message)s", defaults={"role": "other"}
+    )
+    logger.handlers[0].setFormatter(formatter)
+
+    logger.info("first", extra={"role": "assistant"})
+    logger.info("second")
+    logger.info("third", extra={"role": "user"})
+
+    assert log_output() == [
+        "INFO assistant first",
+        "INFO other second",
+        "INFO user third",
+    ]
+
+
+def test_default_not_mutated(logger, log_output):
+    """Original defaults dict should not be mutated."""
+    defaults = {"role": "other"}
+    formatter = OptionalFieldFormatter("%(levelname)s %(role?)s %(message)s", defaults=defaults)
+    logger.handlers[0].setFormatter(formatter)
+
+    logger.info("test")
+
+    assert defaults == {"role": "other"}

test/unit/logging/test_partial_filter.py

@@ -0,0 +1,83 @@
+import logging
+
+import pytest
+
+from control_backend.logging import PartialFilter
+
+
+@pytest.fixture
+def logger():
+    """Create a fresh logger for each test."""
+    logger = logging.getLogger(f"test_{id(object())}")
+    logger.setLevel(logging.DEBUG)
+    logger.handlers = []
+    return logger
+
+
+@pytest.fixture
+def log_output(logger):
+    """Capture log output and return a function to get it."""
+
+    class ListHandler(logging.Handler):
+        def __init__(self):
+            super().__init__()
+            self.records = []
+
+        def emit(self, record):
+            self.records.append(self.format(record))
+
+    handler = ListHandler()
+    handler.addFilter(PartialFilter())
+    handler.setFormatter(logging.Formatter("%(message)s"))
+    logger.addHandler(handler)
+
+    return lambda: handler.records
+
+
+def test_no_partial_attribute(logger, log_output):
+    """Records without partial attribute should pass through."""
+    logger.info("normal message")
+
+    assert log_output() == ["normal message"]
+
+
+def test_partial_true_filtered(logger, log_output):
+    """Records with partial=True should be filtered out."""
+    logger.info("partial message", extra={"partial": True})
+
+    assert log_output() == []
+
+
+def test_partial_false_passes(logger, log_output):
+    """Records with partial=False should pass through."""
+    logger.info("complete message", extra={"partial": False})
+
+    assert log_output() == ["complete message"]
+
+
+def test_partial_none_passes(logger, log_output):
+    """Records with partial=None should pass through."""
+    logger.info("message", extra={"partial": None})
+
+    assert log_output() == ["message"]
+
+
+def test_partial_truthy_value_passes(logger, log_output):
+    """
+    Records with truthy but non-True partial should pass through, that is, only when it's exactly
+    ``True`` should it pass.
+    """
+    logger.info("message", extra={"partial": "yes"})
+
+    assert log_output() == ["message"]
+
+
+def test_multiple_records_mixed(logger, log_output):
+    """Filter should handle mixed records correctly."""
+    logger.info("first")
+    logger.info("second", extra={"partial": True})
+    logger.info("third", extra={"partial": False})
+    logger.info("fourth", extra={"partial": True})
+    logger.info("fifth")
+
+    assert log_output() == ["first", "third", "fifth"]

test/unit/test_main_sockets.py

@@ -0,0 +1,40 @@
+from unittest.mock import MagicMock, patch
+
+import zmq
+
+from control_backend.main import setup_sockets
+
+
+def test_setup_sockets_proxy():
+    mock_context = MagicMock()
+    mock_pub = MagicMock()
+    mock_sub = MagicMock()
+
+    mock_context.socket.side_effect = [mock_pub, mock_sub]
+
+    with patch("zmq.asyncio.Context.instance", return_value=mock_context):
+        with patch("zmq.proxy") as mock_proxy:
+            setup_sockets()
+
+            mock_pub.bind.assert_called()
+            mock_sub.bind.assert_called()
+            mock_proxy.assert_called_with(mock_sub, mock_pub)
+
+            # Check cleanup
+            mock_pub.close.assert_called()
+            mock_sub.close.assert_called()
+
+
+def test_setup_sockets_proxy_error():
+    mock_context = MagicMock()
+    mock_pub = MagicMock()
+    mock_sub = MagicMock()
+    mock_context.socket.side_effect = [mock_pub, mock_sub]
+
+    with patch("zmq.asyncio.Context.instance", return_value=mock_context):
+        with patch("zmq.proxy", side_effect=zmq.ZMQError):
+            with patch("control_backend.main.logger") as mock_logger:
+                setup_sockets()
+                mock_logger.warning.assert_called()
+                mock_pub.close.assert_called()
+                mock_sub.close.assert_called()