agent_os_kernel 3.1.0__py3-none-any.whl

mute_agent/listener/listener.py

@@ -0,0 +1,610 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+"""
+Listener Agent - Layer 5 Reference Implementation
+
+The Listener Agent is a passive observer that monitors graph states
+without interfering until configured thresholds are exceeded.
+
+Architecture:
+- Consolidates: agent-control-plane, scak, iatp, caas
+- Pattern: Observer with threshold-based intervention
+- Principle: Monitor passively, intervene only when necessary
+
+This module is pure wiring - it delegates to lower layers:
+- Knowledge graph operations → scak (intelligence layer)
+- Security validation → iatp (security layer)  
+- Context management → caas (context layer)
+- Base orchestration → agent-control-plane
+"""
+
+from typing import Dict, Any, Optional, List, Callable
+from dataclasses import dataclass, field
+from datetime import datetime, timedelta
+from enum import Enum, auto
+import threading
+import time
+from collections import deque
+
+from ..knowledge_graph.multidimensional_graph import MultidimensionalKnowledgeGraph
+from ..core.handshake_protocol import HandshakeProtocol, HandshakeSession, HandshakeState
+from ..core.reasoning_agent import ReasoningAgent
+from ..core.execution_agent import ExecutionAgent
+from ..super_system.router import SuperSystemRouter
+
+from .threshold_config import (
+    ThresholdConfig,
+    ThresholdType,
+    InterventionLevel,
+    ThresholdRule,
+    DEFAULT_THRESHOLDS,
+)
+from .state_observer import StateObserver, ObservationResult
+
+
+class ListenerState(Enum):
+    """States of the Listener Agent."""
+    
+    # Not actively observing
+    IDLE = "idle"
+    
+    # Passively observing - no intervention
+    OBSERVING = "observing"
+    
+    # Detected threshold breach - evaluating response
+    EVALUATING = "evaluating"
+    
+    # Actively intervening
+    INTERVENING = "intervening"
+    
+    # Intervention complete, returning to observation
+    RECOVERING = "recovering"
+    
+    # Stopped - not operational
+    STOPPED = "stopped"
+
+
+@dataclass
+class InterventionEvent:
+    """
+    Record of a Listener intervention.
+    
+    This provides an audit trail of when and why the Listener
+    transitioned from passive observation to active intervention.
+    """
+    
+    event_id: str
+    timestamp: datetime
+    triggered_rules: List[ThresholdRule]
+    intervention_level: InterventionLevel
+    metrics_snapshot: Dict[str, float]
+    context: Dict[str, Any]
+    action_taken: str
+    outcome: Optional[str] = None
+    duration_ms: Optional[float] = None
+
+
+@dataclass
+class ListenerConfig:
+    """Configuration for the Listener Agent."""
+    
+    # Threshold configuration
+    thresholds: ThresholdConfig = field(default_factory=lambda: DEFAULT_THRESHOLDS)
+    
+    # Observation settings
+    observation_interval_seconds: float = 1.0
+    max_observation_history: int = 1000
+    
+    # Intervention settings
+    auto_intervention: bool = True
+    require_confirmation: bool = False
+    max_interventions_per_minute: int = 10
+    
+    # Recovery settings
+    recovery_observation_count: int = 5
+    recovery_success_threshold: float = 0.8
+
+
+class ListenerAgent:
+    """
+    Layer 5 Reference Implementation: The Listener Agent
+    
+    A passive observer that monitors graph states and only intervenes
+    when configured thresholds are exceeded.
+    
+    Design Principles:
+    1. Passive by default - observe without interference
+    2. Threshold-driven intervention - clear, configurable triggers
+    3. Minimal footprint - delegate to lower layers
+    4. Full audit trail - every intervention is logged
+    
+    Usage:
+        ```python
+        # Create core components
+        kg = MultidimensionalKnowledgeGraph()
+        protocol = HandshakeProtocol()
+        router = SuperSystemRouter(kg)
+        
+        # Create and start listener
+        listener = ListenerAgent(kg, protocol, router)
+        listener.start()
+        
+        # Listener now monitors passively...
+        # When thresholds are exceeded, it intervenes automatically
+        
+        # Stop when done
+        listener.stop()
+        ```
+    """
+    
+    def __init__(
+        self,
+        knowledge_graph: MultidimensionalKnowledgeGraph,
+        protocol: HandshakeProtocol,
+        router: SuperSystemRouter,
+        config: Optional[ListenerConfig] = None,
+        # Optional lower-layer adapters
+        security_adapter: Optional[Any] = None,  # iatp adapter
+        context_adapter: Optional[Any] = None,   # caas adapter
+    ):
+        """
+        Initialize the Listener Agent.
+        
+        Args:
+            knowledge_graph: The graph to monitor (via scak)
+            protocol: The handshake protocol to observe
+            router: The super system router
+            config: Listener configuration
+            security_adapter: Optional adapter to iatp security layer
+            context_adapter: Optional adapter to caas context layer
+        """
+        self.knowledge_graph = knowledge_graph
+        self.protocol = protocol
+        self.router = router
+        self.config = config or ListenerConfig()
+        
+        # Lower-layer adapters (for consolidated stack)
+        self._security_adapter = security_adapter
+        self._context_adapter = context_adapter
+        
+        # State observer
+        self.observer = StateObserver(
+            knowledge_graph=knowledge_graph,
+            protocol=protocol,
+            router=router,
+            sample_window_seconds=self.config.thresholds.window_size_seconds,
+        )
+        
+        # Current state
+        self._state = ListenerState.IDLE
+        self._state_lock = threading.Lock()
+        
+        # Observation thread
+        self._observation_thread: Optional[threading.Thread] = None
+        self._stop_event = threading.Event()
+        
+        # Intervention tracking
+        self._interventions: deque = deque(
+            maxlen=self.config.max_observation_history
+        )
+        self._intervention_count_this_minute = 0
+        self._minute_start = datetime.now()
+        self._event_counter = 0
+        
+        # Callbacks
+        self._intervention_callbacks: List[Callable[[InterventionEvent], None]] = []
+        self._state_change_callbacks: List[Callable[[ListenerState, ListenerState], None]] = []
+    
+    @property
+    def state(self) -> ListenerState:
+        """Get current listener state."""
+        with self._state_lock:
+            return self._state
+    
+    def _set_state(self, new_state: ListenerState) -> None:
+        """Set listener state with callback notification."""
+        with self._state_lock:
+            old_state = self._state
+            self._state = new_state
+        
+        # Notify callbacks outside lock
+        for callback in self._state_change_callbacks:
+            try:
+                callback(old_state, new_state)
+            except Exception:
+                pass  # Don't let callback errors affect listener
+    
+    def start(self) -> None:
+        """
+        Start the Listener Agent.
+        
+        Begins passive observation of graph states. The listener will
+        continue observing until stop() is called or an intervention
+        threshold is exceeded.
+        """
+        if self._state != ListenerState.IDLE and self._state != ListenerState.STOPPED:
+            raise RuntimeError(f"Cannot start listener in state {self._state}")
+        
+        self._stop_event.clear()
+        self._set_state(ListenerState.OBSERVING)
+        
+        # Start observation thread
+        self._observation_thread = threading.Thread(
+            target=self._observation_loop,
+            name="ListenerAgent-Observer",
+            daemon=True,
+        )
+        self._observation_thread.start()
+    
+    def stop(self) -> None:
+        """
+        Stop the Listener Agent.
+        
+        Ceases observation and any ongoing intervention.
+        """
+        self._stop_event.set()
+        self._set_state(ListenerState.STOPPED)
+        
+        if self._observation_thread and self._observation_thread.is_alive():
+            self._observation_thread.join(timeout=5.0)
+    
+    def observe_once(self, context: Optional[Dict[str, Any]] = None) -> ObservationResult:
+        """
+        Perform a single observation cycle.
+        
+        This is useful for synchronous observation without starting
+        the background observation loop.
+        
+        Args:
+            context: Optional context to include
+            
+        Returns:
+            ObservationResult from this cycle
+        """
+        return self.observer.observe(context)
+    
+    def evaluate_thresholds(
+        self,
+        observation: ObservationResult
+    ) -> tuple[List[ThresholdRule], InterventionLevel]:
+        """
+        Evaluate observation against configured thresholds.
+        
+        Args:
+            observation: The observation to evaluate
+            
+        Returns:
+            Tuple of (triggered_rules, max_intervention_level)
+        """
+        # Convert observation to threshold metrics
+        threshold_metrics = observation.to_threshold_metrics()
+        
+        # Evaluate against thresholds
+        triggered_rules = self.config.thresholds.evaluate_all(
+            threshold_metrics,
+            context={"observation": observation},
+        )
+        
+        # Get maximum intervention level
+        max_level = self.config.thresholds.get_maximum_intervention_level(
+            triggered_rules
+        )
+        
+        return triggered_rules, max_level
+    
+    def _observation_loop(self) -> None:
+        """Background observation loop."""
+        while not self._stop_event.is_set():
+            try:
+                self._run_observation_cycle()
+            except Exception as e:
+                # Log error but continue observation
+                # In production, this would integrate with logging framework
+                pass
+            
+            # Wait for next observation interval
+            self._stop_event.wait(self.config.observation_interval_seconds)
+    
+    def _run_observation_cycle(self) -> None:
+        """Run a single observation cycle."""
+        # Perform observation
+        observation = self.observer.observe()
+        
+        # Evaluate thresholds
+        triggered_rules, intervention_level = self.evaluate_thresholds(observation)
+        
+        if not triggered_rules:
+            # No thresholds exceeded - continue passive observation
+            return
+        
+        # Thresholds exceeded - evaluate intervention
+        self._set_state(ListenerState.EVALUATING)
+        
+        # Check rate limiting
+        if not self._can_intervene():
+            self._set_state(ListenerState.OBSERVING)
+            return
+        
+        # Determine if intervention is needed based on level
+        if intervention_level == InterventionLevel.OBSERVE:
+            # Log only
+            self._set_state(ListenerState.OBSERVING)
+            return
+        
+        # Perform intervention
+        if self.config.auto_intervention:
+            self._perform_intervention(
+                triggered_rules,
+                intervention_level,
+                observation,
+            )
+        
+        # Return to observation (or recovery)
+        if self._state == ListenerState.INTERVENING:
+            self._set_state(ListenerState.RECOVERING)
+            # In recovery mode, continue observation with heightened awareness
+            self._recovery_check()
+    
+    def _can_intervene(self) -> bool:
+        """Check if intervention is allowed (rate limiting)."""
+        now = datetime.now()
+        
+        # Reset counter if minute has passed
+        if (now - self._minute_start).total_seconds() >= 60:
+            self._intervention_count_this_minute = 0
+            self._minute_start = now
+        
+        return self._intervention_count_this_minute < self.config.max_interventions_per_minute
+    
+    def _perform_intervention(
+        self,
+        triggered_rules: List[ThresholdRule],
+        intervention_level: InterventionLevel,
+        observation: ObservationResult,
+    ) -> InterventionEvent:
+        """
+        Perform an intervention based on triggered rules.
+        
+        This is where the Listener transitions from passive to active.
+        """
+        self._set_state(ListenerState.INTERVENING)
+        start_time = datetime.now()
+        
+        # Generate event ID
+        self._event_counter += 1
+        event_id = f"intervention_{self._event_counter}_{start_time.timestamp()}"
+        
+        # Determine action based on intervention level
+        action_taken = self._determine_action(intervention_level, triggered_rules)
+        
+        # Execute intervention action
+        outcome = self._execute_intervention_action(
+            action_taken,
+            intervention_level,
+            triggered_rules,
+        )
+        
+        # Calculate duration
+        duration_ms = (datetime.now() - start_time).total_seconds() * 1000
+        
+        # Create event record
+        event = InterventionEvent(
+            event_id=event_id,
+            timestamp=start_time,
+            triggered_rules=triggered_rules,
+            intervention_level=intervention_level,
+            metrics_snapshot=observation.derived_metrics.copy(),
+            context={
+                "anomalies": observation.anomalies_detected,
+                "graph_snapshot": observation.graph_snapshot,
+            },
+            action_taken=action_taken,
+            outcome=outcome,
+            duration_ms=duration_ms,
+        )
+        
+        # Store and notify
+        self._interventions.append(event)
+        self._intervention_count_this_minute += 1
+        
+        for callback in self._intervention_callbacks:
+            try:
+                callback(event)
+            except Exception:
+                pass
+        
+        return event
+    
+    def _determine_action(
+        self,
+        level: InterventionLevel,
+        rules: List[ThresholdRule],
+    ) -> str:
+        """Determine intervention action based on level and rules."""
+        if level == InterventionLevel.WARN:
+            return "emit_warning"
+        elif level == InterventionLevel.SOFT_BLOCK:
+            return "require_confirmation"
+        elif level == InterventionLevel.HARD_BLOCK:
+            return "block_pending_actions"
+        elif level == InterventionLevel.EMERGENCY:
+            return "emergency_halt"
+        else:
+            return "observe_only"
+    
+    def _execute_intervention_action(
+        self,
+        action: str,
+        level: InterventionLevel,
+        rules: List[ThresholdRule],
+    ) -> str:
+        """
+        Execute the determined intervention action.
+        
+        This is where we wire together the lower layers:
+        - Use iatp for security-related interventions
+        - Use caas to update context
+        - Use agent-control-plane for action blocking
+        """
+        if action == "emit_warning":
+            # Log warning - in production, integrate with alerting system
+            return f"Warning emitted for {len(rules)} triggered rules"
+        
+        elif action == "require_confirmation":
+            # Mark pending sessions as requiring confirmation
+            pending_blocked = 0
+            for session_id, session in self.protocol.sessions.items():
+                if session.state in [HandshakeState.VALIDATED, HandshakeState.ACCEPTED]:
+                    session.metadata["requires_confirmation"] = True
+                    session.metadata["confirmation_reason"] = (
+                        f"Threshold breach: {[r.description for r in rules]}"
+                    )
+                    pending_blocked += 1
+            return f"Soft block applied to {pending_blocked} pending sessions"
+        
+        elif action == "block_pending_actions":
+            # Reject all pending sessions
+            blocked = 0
+            for session_id, session in list(self.protocol.sessions.items()):
+                if session.state in [HandshakeState.INITIATED, HandshakeState.NEGOTIATING,
+                                    HandshakeState.VALIDATED, HandshakeState.ACCEPTED]:
+                    self.protocol.reject_proposal(
+                        session_id,
+                        reason=f"Listener intervention: {level.value}"
+                    )
+                    blocked += 1
+            return f"Hard block applied, {blocked} sessions rejected"
+        
+        elif action == "emergency_halt":
+            # Emergency halt - reject all and set protective state
+            halted = 0
+            for session_id, session in list(self.protocol.sessions.items()):
+                if session.state != HandshakeState.COMPLETED:
+                    try:
+                        self.protocol.reject_proposal(
+                            session_id,
+                            reason="EMERGENCY: System halt by Listener"
+                        )
+                        halted += 1
+                    except ValueError:
+                        pass  # Session may already be in terminal state
+            
+            # If security adapter available, notify it
+            if self._security_adapter:
+                try:
+                    self._security_adapter.emergency_alert(
+                        reason="Listener emergency halt",
+                        triggered_rules=[r.description for r in rules],
+                    )
+                except Exception:
+                    pass
+            
+            return f"Emergency halt: {halted} sessions terminated"
+        
+        return "No action taken"
+    
+    def _recovery_check(self) -> None:
+        """
+        Check if system has recovered after intervention.
+        
+        Performs additional observations to verify system stability
+        before returning to normal observation.
+        """
+        success_count = 0
+        
+        for _ in range(self.config.recovery_observation_count):
+            if self._stop_event.is_set():
+                break
+            
+            observation = self.observer.observe()
+            triggered_rules, level = self.evaluate_thresholds(observation)
+            
+            # Check if we're back to safe levels
+            if level in [InterventionLevel.OBSERVE, InterventionLevel.WARN]:
+                success_count += 1
+            
+            time.sleep(self.config.observation_interval_seconds)
+        
+        # Calculate recovery success rate
+        success_rate = success_count / self.config.recovery_observation_count
+        
+        if success_rate >= self.config.recovery_success_threshold:
+            self._set_state(ListenerState.OBSERVING)
+        else:
+            # Still unstable - may need additional intervention
+            self._set_state(ListenerState.EVALUATING)
+    
+    # === Public API for external integration ===
+    
+    def register_intervention_callback(
+        self,
+        callback: Callable[[InterventionEvent], None]
+    ) -> None:
+        """Register a callback to be notified of interventions."""
+        self._intervention_callbacks.append(callback)
+    
+    def register_state_change_callback(
+        self,
+        callback: Callable[[ListenerState, ListenerState], None]
+    ) -> None:
+        """Register a callback to be notified of state changes."""
+        self._state_change_callbacks.append(callback)
+    
+    def get_intervention_history(
+        self,
+        count: Optional[int] = None
+    ) -> List[InterventionEvent]:
+        """Get recent intervention events."""
+        events = list(self._interventions)
+        if count is not None:
+            events = events[-count:]
+        return events
+    
+    def get_statistics(self) -> Dict[str, Any]:
+        """Get listener statistics."""
+        observation_history = self.observer.get_observation_history()
+        
+        return {
+            "state": self._state.value,
+            "total_observations": len(observation_history),
+            "total_interventions": len(self._interventions),
+            "interventions_this_minute": self._intervention_count_this_minute,
+            "active_thresholds": len(self.config.thresholds.rules),
+            "enabled_thresholds": sum(
+                1 for r in self.config.thresholds.rules.values() if r.enabled
+            ),
+            "observer_baselines": len(self.observer._baselines),
+        }
+    
+    def update_threshold(
+        self,
+        threshold_type: ThresholdType,
+        new_value: float
+    ) -> None:
+        """Update a threshold value at runtime."""
+        rule = self.config.thresholds.get_rule(threshold_type)
+        if rule:
+            rule.value = new_value
+    
+    def enable_threshold(self, threshold_type: ThresholdType) -> None:
+        """Enable a threshold rule."""
+        self.config.thresholds.enable_rule(threshold_type)
+    
+    def disable_threshold(self, threshold_type: ThresholdType) -> None:
+        """Disable a threshold rule."""
+        self.config.thresholds.disable_rule(threshold_type)
+    
+    def calibrate(self, observation_count: int = 10) -> None:
+        """
+        Calibrate baselines during known-good operation.
+        
+        Call this during normal operation to establish baseline
+        metrics for anomaly detection.
+        """
+        # Collect observations
+        for _ in range(observation_count):
+            self.observer.observe()
+            time.sleep(self.config.observation_interval_seconds)
+        
+        # Calibrate observer baselines
+        self.observer.calibrate_baselines(observation_count)