htmlgraph 0.20.1__py3-none-any.whl → 0.27.5__py3-none-any.whl

htmlgraph/cigs/autonomy.py

@@ -0,0 +1,385 @@
+"""
+AutonomyRecommender - Adaptive autonomy level management for CIGS.
+
+Recommends appropriate autonomy levels based on:
+- Average compliance from last 5 sessions
+- Anti-pattern count
+- Circuit breaker triggers
+
+Reference: .htmlgraph/spikes/computational-imperative-guidance-system-design.md (Part 5, Section 5.4)
+
+Decision Matrix:
+- Observer (>90% compliance): Minimal guidance, trust developer
+- Consultant (70-90% compliance): Moderate guidance, gentle nudges
+- Collaborator (50-70% compliance): High guidance, active collaboration
+- Operator (<50% compliance): Strict enforcement, mandatory acknowledgment
+"""
+
+from dataclasses import dataclass, field
+
+from htmlgraph.cigs.models import (
+    AutonomyLevel,
+    PatternRecord,
+    SessionViolationSummary,
+)
+
+
+@dataclass
+class AutonomyRecommender:
+    """
+    Recommend autonomy level based on compliance history and patterns.
+
+    Implements a four-level decision matrix that adapts guidance intensity
+    based on demonstrated delegation compliance and detected anti-patterns.
+
+    Attributes:
+        compliance_history: List of compliance rates from recent sessions (0.0-1.0)
+        anti_pattern_count: Number of distinct anti-patterns detected
+        circuit_breaker_active: Whether circuit breaker was triggered in recent sessions
+    """
+
+    compliance_history: list[float] = field(default_factory=list)
+    anti_pattern_count: int = 0
+    circuit_breaker_active: bool = False
+
+    # Decision matrix thresholds
+    OBSERVER_THRESHOLD = 0.90
+    CONSULTANT_THRESHOLD = 0.70
+    COLLABORATOR_THRESHOLD = 0.50
+
+    def recommend(
+        self,
+        violations: SessionViolationSummary,
+        patterns: list[PatternRecord] | None = None,
+        compliance_history: list[float] | None = None,
+    ) -> AutonomyLevel:
+        """
+        Recommend autonomy level for next session.
+
+        Implements decision matrix:
+        - Observer (>90% compliance, no anti-patterns)
+        - Consultant (70-90% compliance OR 1-2 anti-patterns)
+        - Collaborator (50-70% compliance OR 3+ anti-patterns)
+        - Operator (<50% compliance OR circuit breaker triggered)
+
+        Args:
+            violations: Summary of violations from current/recent session
+            patterns: Detected behavioral patterns (if any)
+            compliance_history: Optional list of compliance rates from last 5 sessions
+
+        Returns:
+            AutonomyLevel recommendation with messaging intensity and enforcement mode
+        """
+        # Calculate average compliance
+        if compliance_history:
+            avg_compliance = sum(compliance_history) / len(compliance_history)
+        else:
+            avg_compliance = violations.compliance_rate
+
+        # Count anti-patterns
+        anti_pattern_count = 0
+        if patterns:
+            anti_pattern_count = len(
+                [p for p in patterns if p.pattern_type == "anti-pattern"]
+            )
+
+        # Check circuit breaker status
+        circuit_breaker_active = violations.circuit_breaker_triggered
+
+        # Apply decision matrix
+        return self._apply_decision_matrix(
+            avg_compliance=avg_compliance,
+            anti_pattern_count=anti_pattern_count,
+            circuit_breaker_active=circuit_breaker_active,
+            violations_count=violations.total_violations,
+            patterns=patterns or [],
+        )
+
+    def _apply_decision_matrix(
+        self,
+        avg_compliance: float,
+        anti_pattern_count: int,
+        circuit_breaker_active: bool,
+        violations_count: int,
+        patterns: list[PatternRecord],
+    ) -> AutonomyLevel:
+        """
+        Apply the decision matrix to determine autonomy level.
+
+        Decision Logic:
+        1. If circuit breaker active -> OPERATOR (strict)
+        2. If avg_compliance > 90% AND no anti-patterns -> OBSERVER (minimal)
+        3. If avg_compliance > 70% OR anti_pattern_count <= 2 -> CONSULTANT (moderate)
+        4. If avg_compliance > 50% OR anti_pattern_count <= 4 -> COLLABORATOR (high)
+        5. Otherwise -> OPERATOR (strict)
+
+        Args:
+            avg_compliance: Average compliance rate (0.0-1.0)
+            anti_pattern_count: Number of anti-patterns detected
+            circuit_breaker_active: Whether circuit breaker is active
+            violations_count: Total violation count
+            patterns: List of detected patterns
+
+        Returns:
+            AutonomyLevel recommendation
+        """
+        # Circuit breaker takes precedence
+        if circuit_breaker_active:
+            return AutonomyLevel(
+                level="operator",
+                messaging_intensity="maximal",
+                enforcement_mode="strict",
+                reason=f"Circuit breaker triggered ({violations_count} violations). Strict enforcement required.",
+                based_on_violations=violations_count,
+                based_on_patterns=[
+                    p.name for p in patterns if p.pattern_type == "anti-pattern"
+                ],
+            )
+
+        # Observer: Excellent compliance, no patterns
+        if avg_compliance >= self.OBSERVER_THRESHOLD and anti_pattern_count == 0:
+            return AutonomyLevel(
+                level="observer",
+                messaging_intensity="minimal",
+                enforcement_mode="guidance",
+                reason=f"Excellent compliance ({avg_compliance:.0%}). Minimal guidance needed.",
+                based_on_violations=violations_count,
+                based_on_patterns=[
+                    p.name for p in patterns if p.pattern_type == "anti-pattern"
+                ],
+            )
+
+        # Consultant: Good compliance (70-90%) AND few anti-patterns (<=2)
+        if (
+            self.CONSULTANT_THRESHOLD <= avg_compliance < self.OBSERVER_THRESHOLD
+        ) and anti_pattern_count <= 2:
+            anti_pattern_list = [
+                p.name for p in patterns if p.pattern_type == "anti-pattern"
+            ]
+            pattern_note = (
+                f", {anti_pattern_count} anti-pattern(s)"
+                if anti_pattern_count > 0
+                else ""
+            )
+
+            return AutonomyLevel(
+                level="consultant",
+                messaging_intensity="moderate",
+                enforcement_mode="guidance",
+                reason=f"Good compliance ({avg_compliance:.0%}){pattern_note}. Moderate guidance recommended.",
+                based_on_violations=violations_count,
+                based_on_patterns=anti_pattern_list,
+            )
+
+        # Collaborator: Moderate compliance (50-70%) AND moderate anti-patterns (3-4)
+        if (
+            self.COLLABORATOR_THRESHOLD <= avg_compliance < self.CONSULTANT_THRESHOLD
+        ) and (2 < anti_pattern_count <= 4):
+            anti_pattern_list = [
+                p.name for p in patterns if p.pattern_type == "anti-pattern"
+            ]
+            return AutonomyLevel(
+                level="collaborator",
+                messaging_intensity="high",
+                enforcement_mode="strict",
+                reason=f"Moderate compliance ({avg_compliance:.0%}), {anti_pattern_count} anti-pattern(s). Active guidance needed.",
+                based_on_violations=violations_count,
+                based_on_patterns=anti_pattern_list,
+            )
+
+        # Operator: Low compliance (<50%) OR many anti-patterns (5+)
+        anti_pattern_list = [
+            p.name for p in patterns if p.pattern_type == "anti-pattern"
+        ]
+        return AutonomyLevel(
+            level="operator",
+            messaging_intensity="maximal",
+            enforcement_mode="strict",
+            reason=f"Low compliance ({avg_compliance:.0%}), {anti_pattern_count} anti-pattern(s). Strict enforcement required.",
+            based_on_violations=violations_count,
+            based_on_patterns=anti_pattern_list,
+        )
+
+    def recommend_from_compliance_history(
+        self,
+        compliance_history: list[float],
+        anti_pattern_count: int = 0,
+        circuit_breaker_active: bool = False,
+    ) -> AutonomyLevel:
+        """
+        Recommend autonomy level from compliance history alone.
+
+        Convenience method for cross-session recommendations when
+        only compliance history is available.
+
+        Args:
+            compliance_history: List of compliance rates from last 5 sessions
+            anti_pattern_count: Number of anti-patterns detected
+            circuit_breaker_active: Whether circuit breaker was triggered
+
+        Returns:
+            AutonomyLevel recommendation
+        """
+        if not compliance_history:
+            # Default to Consultant if no history
+            return AutonomyLevel(
+                level="consultant",
+                messaging_intensity="moderate",
+                enforcement_mode="guidance",
+                reason="No compliance history. Defaulting to moderate guidance.",
+                based_on_violations=0,
+                based_on_patterns=[],
+            )
+
+        avg_compliance = sum(compliance_history) / len(compliance_history)
+
+        # Create minimal summary for decision matrix
+        violations = SessionViolationSummary(
+            session_id="unknown",
+            total_violations=0,
+            violations_by_type={},
+            total_waste_tokens=0,
+            circuit_breaker_triggered=circuit_breaker_active,
+            compliance_rate=avg_compliance,
+        )
+
+        return self.recommend(
+            violations=violations,
+            patterns=None,
+            compliance_history=compliance_history,
+        )
+
+    def evaluate_autonomy_transition(
+        self,
+        current_level: str,
+        new_level: str,
+    ) -> dict[str, str | int]:
+        """
+        Evaluate transition between autonomy levels.
+
+        Returns metadata about the transition for logging/reporting.
+
+        Args:
+            current_level: Current autonomy level ("observer", "consultant", etc.)
+            new_level: New autonomy level
+
+        Returns:
+            Dictionary with transition details
+        """
+        level_order = ["observer", "consultant", "collaborator", "operator"]
+
+        current_idx = (
+            level_order.index(current_level) if current_level in level_order else -1
+        )
+        new_idx = level_order.index(new_level) if new_level in level_order else -1
+
+        if new_idx < 0 or current_idx < 0:
+            return {
+                "transition": "unknown",
+                "direction": "unknown",
+                "severity": "unknown",
+            }
+
+        if new_idx > current_idx:
+            direction = "escalated"
+            severity = "high" if new_idx - current_idx > 1 else "moderate"
+        elif new_idx < current_idx:
+            direction = "relaxed"
+            severity = "high" if current_idx - new_idx > 1 else "moderate"
+        else:
+            direction = "unchanged"
+            severity = "none"
+
+        return {
+            "transition": f"{current_level} → {new_level}",
+            "direction": direction,
+            "severity": severity,
+            "escalation_level": new_idx,
+        }
+
+    def get_messaging_config(self, level: str) -> dict[str, str | bool | int]:
+        """
+        Get messaging configuration for an autonomy level.
+
+        Returns imperative message configuration based on autonomy level.
+
+        Args:
+            level: Autonomy level ("observer", "consultant", "collaborator", "operator")
+
+        Returns:
+            Dictionary with messaging configuration
+        """
+        config: dict[str, dict[str, str | bool | int]] = {
+            "observer": {
+                "prefix": "💡 GUIDANCE",
+                "tone": "informative",
+                "includes_cost": False,
+                "includes_suggestion": True,
+                "requires_acknowledgment": False,
+                "escalation_level": 0,
+            },
+            "consultant": {
+                "prefix": "🔴 IMPERATIVE",
+                "tone": "commanding",
+                "includes_cost": True,
+                "includes_suggestion": True,
+                "requires_acknowledgment": False,
+                "escalation_level": 1,
+            },
+            "collaborator": {
+                "prefix": "⚠️ FINAL WARNING",
+                "tone": "urgent",
+                "includes_cost": True,
+                "includes_suggestion": True,
+                "includes_consequences": True,
+                "requires_acknowledgment": False,
+                "escalation_level": 2,
+            },
+            "operator": {
+                "prefix": "🚨 CIRCUIT BREAKER",
+                "tone": "blocking",
+                "includes_cost": True,
+                "includes_suggestion": True,
+                "includes_consequences": True,
+                "requires_acknowledgment": True,
+                "escalation_level": 3,
+            },
+        }
+
+        return config.get(level, config["consultant"])
+
+    def estimate_next_level(
+        self,
+        current_compliance: float,
+        projected_violations: int = 0,
+    ) -> str:
+        """
+        Estimate the autonomy level for next session.
+
+        Projects autonomy level based on current trajectory.
+
+        Args:
+            current_compliance: Current session's compliance rate
+            projected_violations: Projected violations for next session (0 = assume improvement)
+
+        Returns:
+            Estimated next autonomy level
+        """
+        # If violations decrease, estimate improvement
+        if projected_violations == 0:
+            estimated_compliance = min(1.0, current_compliance + 0.05)
+        else:
+            # Assume violations stay constant or worsen
+            estimated_compliance = max(0.0, current_compliance - 0.05)
+
+        violations = SessionViolationSummary(
+            session_id="projected",
+            total_violations=0,
+            violations_by_type={},
+            total_waste_tokens=0,
+            circuit_breaker_triggered=False,
+            compliance_rate=estimated_compliance,
+        )
+
+        recommendation = self.recommend(violations)
+        return recommendation.level