tapps-agents 3.6.0__py3-none-any.whl → 3.6.1__py3-none-any.whl

tapps_agents/core/agent_learning.py

@@ -1,1865 +1,1865 @@
-"""
-Agent Learning System
-
-Enables agents to learn from past tasks and improve over time.
-"""
-
-# @ai-prime-directive: This file implements the Agent Learning System for pattern extraction, prompt optimization,
-# and feedback analysis. The system enables agents to learn from past tasks and improve over time through
-# meta-learning, pattern recognition, and adaptive learning strategies.
-
-# @ai-constraints:
-# - Learning intensity must adapt to hardware profile (NUC, Desktop, Server)
-# - Pattern extraction must include security scanning before pattern acceptance
-# - Anti-pattern extraction must learn from negative feedback and failures
-# - Prompt optimization must respect hardware constraints and token budgets
-# - Meta-learning must track learning effectiveness and adjust strategies accordingly
-# - Performance: Learning operations should not significantly impact agent response times
-
-# @note[2025-01-15]: Agent learning is an advanced feature that improves agent performance over time.
-# The system uses meta-learning to adapt learning strategies based on effectiveness tracking.
-# See docs/architecture/decisions/ for related architectural decisions.
-
-import logging
-import re
-from dataclasses import dataclass, field
-from datetime import UTC, datetime
-from typing import Any
-
-from .best_practice_consultant import BestPracticeConsultant
-from .capability_registry import CapabilityRegistry, LearningIntensity
-from .hardware_profiler import HardwareProfile, HardwareProfiler
-from .learning_confidence import LearningConfidenceCalculator
-from .learning_dashboard import LearningDashboard
-from .learning_decision import LearningDecisionEngine
-from .learning_explainability import (
-    DecisionReasoningLogger,
-    LearningImpactReporter,
-    PatternSelectionExplainer,
-)
-from .meta_learning import (
-    AdaptiveLearningRate,
-    LearningEffectivenessTracker,
-    LearningSelfAssessor,
-    LearningStrategy,
-    LearningStrategySelector,
-)
-from .security_scanner import SecurityScanner
-from .task_memory import TaskMemorySystem, TaskOutcome
-
-logger = logging.getLogger(__name__)
-
-
-@dataclass
-class CodePattern:
-    """Represents a learned code pattern."""
-
-    pattern_id: str
-    pattern_type: str  # "function", "class", "import", "structure"
-    code_snippet: str
-    context: str
-    quality_score: float
-    usage_count: int
-    success_rate: float
-    learned_from: list[str]  # Task IDs where this pattern was successful
-    metadata: dict[str, Any] = field(default_factory=dict)
-    security_score: float = 0.0  # Security score (0-10, higher is better)
-    is_anti_pattern: bool = False  # True if this is an anti-pattern to avoid
-    failure_reasons: list[str] = field(default_factory=list)  # Reasons for failure
-    rejection_count: int = 0  # Number of times this pattern was rejected
-
-
-@dataclass
-class PromptVariant:
-    """Represents a prompt variation for A/B testing."""
-
-    variant_id: str
-    prompt_template: str
-    modifications: list[str]  # List of modifications made
-    test_count: int
-    success_count: int
-    average_quality: float
-    created_at: datetime
-    last_tested: datetime | None = None
-
-
-class PatternExtractor:
-    """Extracts patterns from successful code outputs."""
-
-    def __init__(
-        self,
-        min_quality_threshold: float = 0.7,
-        security_scanner: SecurityScanner | None = None,
-        security_threshold: float = 7.0,
-    ):
-        """
-        Initialize pattern extractor.
-
-        Args:
-            min_quality_threshold: Minimum quality score to extract patterns
-            security_scanner: Optional security scanner instance
-            security_threshold: Minimum security score to extract patterns (default: 7.0)
-        """
-        self.min_quality_threshold = min_quality_threshold
-        self.security_scanner = security_scanner or SecurityScanner()
-        self.security_threshold = security_threshold
-        self.patterns: dict[str, CodePattern] = {}
-
-    def extract_patterns(
-        self,
-        code: str,
-        quality_score: float,
-        task_id: str,
-        pattern_types: list[str] | None = None,
-    ) -> list[CodePattern]:
-        """
-        Extract patterns from code.
-
-        Args:
-            code: Source code
-            quality_score: Quality score of the code
-            task_id: Task identifier
-            pattern_types: Optional list of pattern types to extract
-
-        Returns:
-            List of extracted patterns
-        """
-        if quality_score < self.min_quality_threshold:
-            return []
-
-        # Security check before extraction
-        security_result = self.security_scanner.scan_code(code=code)
-        security_score = security_result["security_score"]
-        vulnerabilities = security_result["vulnerabilities"]
-        is_safe = security_result.get("is_safe", True)
-
-        # Only extract if security score meets threshold
-        # Check both score >= threshold AND is_safe flag
-        if security_score < self.security_threshold or not is_safe:
-            logger.debug(
-                f"Skipping pattern extraction: security score {security_score:.2f} "
-                f"below threshold {self.security_threshold:.2f} or is_safe={is_safe}"
-            )
-            return []
-
-        patterns = []
-
-        # Extract function patterns
-        if not pattern_types or "function" in pattern_types:
-            func_patterns = self._extract_function_patterns(
-                code, quality_score, task_id
-            )
-            patterns.extend(func_patterns)
-
-        # Extract class patterns
-        if not pattern_types or "class" in pattern_types:
-            class_patterns = self._extract_class_patterns(code, quality_score, task_id)
-            patterns.extend(class_patterns)
-
-        # Extract import patterns
-        if not pattern_types or "import" in pattern_types:
-            import_patterns = self._extract_import_patterns(
-                code, quality_score, task_id
-            )
-            patterns.extend(import_patterns)
-
-        # Extract structural patterns
-        if not pattern_types or "structure" in pattern_types:
-            struct_patterns = self._extract_structural_patterns(
-                code, quality_score, task_id
-            )
-            patterns.extend(struct_patterns)
-
-        return patterns
-
-    def _extract_function_patterns(
-        self, code: str, quality_score: float, task_id: str
-    ) -> list[CodePattern]:
-        """Extract function patterns."""
-        patterns = []
-
-        # Match function definitions - more flexible pattern that handles docstrings and type hints
-        # Find all function definitions, then extract their bodies
-        # Pattern matches: def function_name(...) with optional type hints, handling multiline signatures
-        func_def_pattern = r"def\s+(\w+)\s*\([^)]*\)\s*(?:->\s*[^:]+)?:"
-        func_matches = list(re.finditer(func_def_pattern, code, re.MULTILINE | re.DOTALL))
-        
-        for i, match in enumerate(func_matches):
-            func_name = match.group(1)
-            start_pos = match.end()
-            
-            # Find the end of this function (next function def or end of code)
-            if i + 1 < len(func_matches):
-                end_pos = func_matches[i + 1].start()
-            else:
-                end_pos = len(code)
-            
-            func_body = code[start_pos:end_pos].strip()
-            
-            # Skip if function body is empty
-            if not func_body:
-                continue
-
-            # Get security score for this pattern
-            pattern_code = f"def {func_name}(...):\n{func_body[:200]}"
-            pattern_security = self.security_scanner.scan_code(code=pattern_code)
-            pattern_security_score = pattern_security["security_score"]
-            pattern_vulnerabilities = pattern_security["vulnerabilities"]
-
-            pattern = CodePattern(
-                pattern_id=f"func_{func_name}_{hash(func_body) % 10000}",
-                pattern_type="function",
-                code_snippet=pattern_code,
-                context=f"Function: {func_name}",
-                quality_score=quality_score,
-                usage_count=1,
-                success_rate=1.0,
-                learned_from=[task_id],
-                security_score=pattern_security_score,
-                metadata={"vulnerabilities": pattern_vulnerabilities},
-            )
-            patterns.append(pattern)
-
-        return patterns
-
-    def _extract_class_patterns(
-        self, code: str, quality_score: float, task_id: str
-    ) -> list[CodePattern]:
-        """Extract class patterns."""
-        patterns = []
-
-        # Match class definitions
-        class_pattern = r"class\s+(\w+)(?:\([^)]+\))?:\s*\n((?:\s{4}.*\n?)*)"
-        matches = re.finditer(class_pattern, code, re.MULTILINE)
-
-        for match in matches:
-            class_name = match.group(1)
-            class_body = match.group(2)
-
-            # Get security score for this pattern
-            pattern_code = f"class {class_name}:\n{class_body[:200]}"
-            pattern_security = self.security_scanner.scan_code(code=pattern_code)
-            pattern_security_score = pattern_security["security_score"]
-            pattern_vulnerabilities = pattern_security["vulnerabilities"]
-
-            pattern = CodePattern(
-                pattern_id=f"class_{class_name}_{hash(class_body) % 10000}",
-                pattern_type="class",
-                code_snippet=pattern_code,
-                context=f"Class: {class_name}",
-                quality_score=quality_score,
-                usage_count=1,
-                success_rate=1.0,
-                learned_from=[task_id],
-                security_score=pattern_security_score,
-                metadata={"vulnerabilities": pattern_vulnerabilities},
-            )
-            patterns.append(pattern)
-
-        return patterns
-
-    def _extract_import_patterns(
-        self, code: str, quality_score: float, task_id: str
-    ) -> list[CodePattern]:
-        """Extract import patterns."""
-        patterns = []
-
-        # Match import statements
-        import_pattern = r"^(?:from\s+[\w.]+|import\s+[\w.,\s]+)"
-        matches = re.finditer(import_pattern, code, re.MULTILINE)
-
-        imports = []
-        for match in matches:
-            imports.append(match.group(0).strip())
-
-        if imports:
-            # Get security score for this pattern
-            pattern_code = "\n".join(imports[:10])
-            pattern_security = self.security_scanner.scan_code(code=pattern_code)
-            pattern_security_score = pattern_security["security_score"]
-            pattern_vulnerabilities = pattern_security["vulnerabilities"]
-
-            pattern = CodePattern(
-                pattern_id=f"imports_{hash(''.join(imports)) % 10000}",
-                pattern_type="import",
-                code_snippet=pattern_code,
-                context="Import statements",
-                quality_score=quality_score,
-                usage_count=1,
-                success_rate=1.0,
-                learned_from=[task_id],
-                security_score=pattern_security_score,
-                metadata={"vulnerabilities": pattern_vulnerabilities},
-            )
-            patterns.append(pattern)
-
-        return patterns
-
-    def _extract_structural_patterns(
-        self, code: str, quality_score: float, task_id: str
-    ) -> list[CodePattern]:
-        """Extract structural patterns (decorators, context managers, etc.)."""
-        patterns = []
-
-        # Match decorators
-        decorator_pattern = r"@\w+(?:\([^)]*\))?"
-        decorators = re.findall(decorator_pattern, code)
-
-        if decorators:
-            # Get security score for this pattern
-            pattern_code = "\n".join(decorators[:5])
-            pattern_security = self.security_scanner.scan_code(code=pattern_code)
-            pattern_security_score = pattern_security["security_score"]
-            pattern_vulnerabilities = pattern_security["vulnerabilities"]
-
-            pattern = CodePattern(
-                pattern_id=f"decorators_{hash(''.join(decorators)) % 10000}",
-                pattern_type="structure",
-                code_snippet=pattern_code,
-                context="Decorators",
-                quality_score=quality_score,
-                usage_count=1,
-                success_rate=1.0,
-                learned_from=[task_id],
-                security_score=pattern_security_score,
-                metadata={"vulnerabilities": pattern_vulnerabilities},
-            )
-            patterns.append(pattern)
-
-        return patterns
-
-    def get_patterns_for_context(
-        self,
-        context: str,
-        pattern_type: str | None = None,
-        limit: int = 5,
-        exclude_anti_patterns: bool = True,
-    ) -> list[CodePattern]:
-        """
-        Get relevant patterns for a context.
-
-        Args:
-            context: Context string
-            pattern_type: Optional pattern type filter
-            limit: Maximum results
-            exclude_anti_patterns: If True, exclude anti-patterns (default: True)
-
-        Returns:
-            List of relevant patterns
-        """
-        candidates = list(self.patterns.values())
-
-        # Filter out anti-patterns if requested
-        if exclude_anti_patterns:
-            candidates = [p for p in candidates if not p.is_anti_pattern]
-
-        if pattern_type:
-            candidates = [p for p in candidates if p.pattern_type == pattern_type]
-
-        # Sort by security score, quality, and usage
-        candidates.sort(
-            key=lambda p: (p.security_score, p.quality_score, p.usage_count),
-            reverse=True,
-        )
-
-        return candidates[:limit]
-
-
-class AntiPatternExtractor:
-    """Extracts anti-patterns from failed tasks and low-quality code."""
-
-    def __init__(
-        self,
-        max_quality_threshold: float = 0.7,
-        security_scanner: SecurityScanner | None = None,
-    ):
-        """
-        Initialize anti-pattern extractor.
-
-        Args:
-            max_quality_threshold: Maximum quality score to extract anti-patterns
-            security_scanner: Optional security scanner instance
-        """
-        self.max_quality_threshold = max_quality_threshold
-        self.security_scanner = security_scanner or SecurityScanner()
-        self.anti_patterns: dict[str, CodePattern] = {}
-
-    def extract_anti_patterns(
-        self,
-        code: str,
-        quality_score: float,
-        task_id: str,
-        failure_reasons: list[str] | None = None,
-        pattern_types: list[str] | None = None,
-    ) -> list[CodePattern]:
-        """
-        Extract anti-patterns from failed or low-quality code.
-
-        Args:
-            code: Source code
-            quality_score: Quality score of the code
-            task_id: Task identifier
-            failure_reasons: Optional list of failure reasons
-            pattern_types: Optional list of pattern types to extract
-
-        Returns:
-            List of extracted anti-patterns
-        """
-        # Only extract if quality is below threshold
-        if quality_score >= self.max_quality_threshold:
-            return []
-
-        failure_reasons = failure_reasons or []
-        anti_patterns = []
-
-        # Extract function anti-patterns
-        if not pattern_types or "function" in pattern_types:
-            func_patterns = self._extract_function_anti_patterns(
-                code, quality_score, task_id, failure_reasons
-            )
-            anti_patterns.extend(func_patterns)
-
-        # Extract class anti-patterns
-        if not pattern_types or "class" in pattern_types:
-            class_patterns = self._extract_class_anti_patterns(
-                code, quality_score, task_id, failure_reasons
-            )
-            anti_patterns.extend(class_patterns)
-
-        # Extract import anti-patterns
-        if not pattern_types or "import" in pattern_types:
-            import_patterns = self._extract_import_anti_patterns(
-                code, quality_score, task_id, failure_reasons
-            )
-            anti_patterns.extend(import_patterns)
-
-        # Extract structural anti-patterns
-        if not pattern_types or "structure" in pattern_types:
-            struct_patterns = self._extract_structural_anti_patterns(
-                code, quality_score, task_id, failure_reasons
-            )
-            anti_patterns.extend(struct_patterns)
-
-        return anti_patterns
-
-    def _extract_function_anti_patterns(
-        self,
-        code: str,
-        quality_score: float,
-        task_id: str,
-        failure_reasons: list[str],
-    ) -> list[CodePattern]:
-        """Extract function anti-patterns."""
-        anti_patterns = []
-
-        # Match function definitions
-        func_pattern = r"def\s+(\w+)\s*\([^)]*\):\s*\n((?:\s{4}.*\n?)*)"
-        matches = list(re.finditer(func_pattern, code, re.MULTILINE))
-
-        for match in matches:
-            func_name = match.group(1)
-            func_body = match.group(2)
-
-            # Get security score for this pattern
-            pattern_code = f"def {func_name}(...):\n{func_body[:200]}"
-            pattern_security = self.security_scanner.scan_code(code=pattern_code)
-            pattern_security_score = pattern_security["security_score"]
-            pattern_vulnerabilities = pattern_security["vulnerabilities"]
-
-            anti_pattern = CodePattern(
-                pattern_id=f"anti_func_{func_name}_{hash(func_body) % 10000}",
-                pattern_type="function",
-                code_snippet=pattern_code,
-                context=f"Anti-pattern Function: {func_name}",
-                quality_score=quality_score,
-                usage_count=1,
-                success_rate=0.0,  # Anti-patterns have 0 success rate
-                learned_from=[task_id],
-                security_score=pattern_security_score,
-                is_anti_pattern=True,
-                failure_reasons=failure_reasons.copy(),
-                rejection_count=0,
-                metadata={"vulnerabilities": pattern_vulnerabilities},
-            )
-            anti_patterns.append(anti_pattern)
-
-        return anti_patterns
-
-    def _extract_class_anti_patterns(
-        self,
-        code: str,
-        quality_score: float,
-        task_id: str,
-        failure_reasons: list[str],
-    ) -> list[CodePattern]:
-        """Extract class anti-patterns."""
-        anti_patterns = []
-
-        # Match class definitions
-        class_pattern = r"class\s+(\w+)(?:\([^)]+\))?:\s*\n((?:\s{4}.*\n?)*)"
-        matches = re.finditer(class_pattern, code, re.MULTILINE)
-
-        for match in matches:
-            class_name = match.group(1)
-            class_body = match.group(2)
-
-            # Get security score for this pattern
-            pattern_code = f"class {class_name}:\n{class_body[:200]}"
-            pattern_security = self.security_scanner.scan_code(code=pattern_code)
-            pattern_security_score = pattern_security["security_score"]
-            pattern_vulnerabilities = pattern_security["vulnerabilities"]
-
-            anti_pattern = CodePattern(
-                pattern_id=f"anti_class_{class_name}_{hash(class_body) % 10000}",
-                pattern_type="class",
-                code_snippet=pattern_code,
-                context=f"Anti-pattern Class: {class_name}",
-                quality_score=quality_score,
-                usage_count=1,
-                success_rate=0.0,
-                learned_from=[task_id],
-                security_score=pattern_security_score,
-                is_anti_pattern=True,
-                failure_reasons=failure_reasons.copy(),
-                rejection_count=0,
-                metadata={"vulnerabilities": pattern_vulnerabilities},
-            )
-            anti_patterns.append(anti_pattern)
-
-        return anti_patterns
-
-    def _extract_import_anti_patterns(
-        self,
-        code: str,
-        quality_score: float,
-        task_id: str,
-        failure_reasons: list[str],
-    ) -> list[CodePattern]:
-        """Extract import anti-patterns."""
-        anti_patterns = []
-
-        # Match import statements
-        import_pattern = r"^(?:from\s+[\w.]+|import\s+[\w.,\s]+)"
-        matches = re.finditer(import_pattern, code, re.MULTILINE)
-
-        imports = []
-        for match in matches:
-            imports.append(match.group(0).strip())
-
-        if imports:
-            # Get security score for this pattern
-            pattern_code = "\n".join(imports[:10])
-            pattern_security = self.security_scanner.scan_code(code=pattern_code)
-            pattern_security_score = pattern_security["security_score"]
-            pattern_vulnerabilities = pattern_security["vulnerabilities"]
-
-            anti_pattern = CodePattern(
-                pattern_id=f"anti_imports_{hash(''.join(imports)) % 10000}",
-                pattern_type="import",
-                code_snippet=pattern_code,
-                context="Anti-pattern Import statements",
-                quality_score=quality_score,
-                usage_count=1,
-                success_rate=0.0,
-                learned_from=[task_id],
-                security_score=pattern_security_score,
-                is_anti_pattern=True,
-                failure_reasons=failure_reasons.copy(),
-                rejection_count=0,
-                metadata={"vulnerabilities": pattern_vulnerabilities},
-            )
-            anti_patterns.append(anti_pattern)
-
-        return anti_patterns
-
-    def _extract_structural_anti_patterns(
-        self,
-        code: str,
-        quality_score: float,
-        task_id: str,
-        failure_reasons: list[str],
-    ) -> list[CodePattern]:
-        """Extract structural anti-patterns."""
-        anti_patterns = []
-
-        # Match decorators
-        decorator_pattern = r"@\w+(?:\([^)]*\))?"
-        decorators = re.findall(decorator_pattern, code)
-
-        if decorators:
-            # Get security score for this pattern
-            pattern_code = "\n".join(decorators[:5])
-            pattern_security = self.security_scanner.scan_code(code=pattern_code)
-            pattern_security_score = pattern_security["security_score"]
-            pattern_vulnerabilities = pattern_security["vulnerabilities"]
-
-            anti_pattern = CodePattern(
-                pattern_id=f"anti_decorators_{hash(''.join(decorators)) % 10000}",
-                pattern_type="structure",
-                code_snippet=pattern_code,
-                context="Anti-pattern Decorators",
-                quality_score=quality_score,
-                usage_count=1,
-                success_rate=0.0,
-                learned_from=[task_id],
-                security_score=pattern_security_score,
-                is_anti_pattern=True,
-                failure_reasons=failure_reasons.copy(),
-                rejection_count=0,
-                metadata={"vulnerabilities": pattern_vulnerabilities},
-            )
-            anti_patterns.append(anti_pattern)
-
-        return anti_patterns
-
-    def extract_from_failure(
-        self,
-        code: str,
-        task_id: str,
-        failure_reasons: list[str],
-        quality_score: float = 0.0,
-    ) -> list[CodePattern]:
-        """
-        Extract anti-patterns from a failed task.
-
-        Args:
-            code: Source code from failed task
-            task_id: Task identifier
-            failure_reasons: List of failure reasons
-            quality_score: Quality score (default: 0.0 for failures)
-
-        Returns:
-            List of extracted anti-patterns
-        """
-        return self.extract_anti_patterns(
-            code=code,
-            quality_score=quality_score,
-            task_id=task_id,
-            failure_reasons=failure_reasons,
-        )
-
-    def extract_from_rejection(
-        self,
-        code: str,
-        task_id: str,
-        rejection_reason: str,
-        quality_score: float = 0.5,
-    ) -> list[CodePattern]:
-        """
-        Extract anti-patterns from user rejection.
-
-        Args:
-            code: Rejected code
-            task_id: Task identifier
-            rejection_reason: Reason for rejection
-            quality_score: Quality score
-
-        Returns:
-            List of extracted anti-patterns
-        """
-        return self.extract_anti_patterns(
-            code=code,
-            quality_score=quality_score,
-            task_id=task_id,
-            failure_reasons=[f"User rejection: {rejection_reason}"],
-        )
-
-    def get_anti_patterns_for_context(
-        self, context: str, pattern_type: str | None = None, limit: int = 5
-    ) -> list[CodePattern]:
-        """
-        Get anti-patterns to avoid for a context.
-
-        Args:
-            context: Context string
-            pattern_type: Optional pattern type filter
-            limit: Maximum results
-
-        Returns:
-            List of anti-patterns to avoid
-        """
-        candidates = [p for p in self.anti_patterns.values() if p.is_anti_pattern]
-
-        if pattern_type:
-            candidates = [p for p in candidates if p.pattern_type == pattern_type]
-
-        # Sort by rejection count and quality (low quality = more important to avoid)
-        candidates.sort(
-            key=lambda p: (p.rejection_count, -p.quality_score), reverse=True
-        )
-
-        return candidates[:limit]
-
-
-class PromptOptimizer:
-    """Optimizes agent prompts based on outcomes."""
-
-    def __init__(self, hardware_profile: HardwareProfile):
-        """
-        Initialize prompt optimizer.
-
-        Args:
-            hardware_profile: Hardware profile for optimization
-        """
-        self.hardware_profile = hardware_profile
-        self.variants: dict[str, PromptVariant] = {}
-        self.base_prompt: str | None = None
-
-    def create_variant(
-        self,
-        base_prompt: str,
-        modifications: list[str],
-        variant_id: str | None = None,
-    ) -> PromptVariant:
-        """
-        Create a prompt variant for A/B testing.
-
-        Args:
-            base_prompt: Base prompt template
-            modifications: List of modifications to apply
-            variant_id: Optional variant identifier
-
-        Returns:
-            PromptVariant instance
-        """
-        if variant_id is None:
-            variant_id = f"variant_{hash(''.join(modifications)) % 10000}"
-
-        # Apply modifications (simplified - in production would be more sophisticated)
-        modified_prompt = base_prompt
-        for mod in modifications:
-            if mod.startswith("add:"):
-                modified_prompt += f"\n{mod[4:]}"
-            elif mod.startswith("prepend:"):
-                modified_prompt = f"{mod[8:]}\n{modified_prompt}"
-
-        variant = PromptVariant(
-            variant_id=variant_id,
-            prompt_template=modified_prompt,
-            modifications=modifications,
-            test_count=0,
-            success_count=0,
-            average_quality=0.0,
-            created_at=datetime.now(UTC),
-        )
-
-        self.variants[variant_id] = variant
-        return variant
-
-    def record_test_result(self, variant_id: str, success: bool, quality_score: float):
-        """
-        Record A/B test result for a variant.
-
-        Args:
-            variant_id: Variant identifier
-            success: Whether test succeeded
-            quality_score: Quality score
-        """
-        if variant_id not in self.variants:
-            logger.warning(f"Variant {variant_id} not found")
-            return
-
-        variant = self.variants[variant_id]
-        variant.test_count += 1
-        if success:
-            variant.success_count += 1
-
-        # Update average quality (exponential moving average)
-        alpha = 0.1
-        variant.average_quality = (
-            alpha * quality_score + (1 - alpha) * variant.average_quality
-        )
-        variant.last_tested = datetime.now(UTC)
-
-    def get_best_variant(self, min_tests: int = 5) -> PromptVariant | None:
-        """
-        Get the best performing variant.
-
-        Args:
-            min_tests: Minimum test count to consider
-
-        Returns:
-            Best PromptVariant if found, None otherwise
-        """
-        candidates = [v for v in self.variants.values() if v.test_count >= min_tests]
-
-        if not candidates:
-            return None
-
-        # Sort by success rate and average quality
-        candidates.sort(
-            key=lambda v: (v.success_count / v.test_count, v.average_quality),
-            reverse=True,
-        )
-
-        return candidates[0]
-
-    def optimize_for_hardware(self, prompt: str) -> str:
-        """
-        Optimize prompt for hardware profile.
-
-        Args:
-            prompt: Original prompt
-
-        Returns:
-            Optimized prompt
-        """
-        # Workstation-like: return full prompt (hardware taxonomy removed)
-        return prompt
-
-
-class NegativeFeedbackHandler:
-    """Handles negative feedback, rejections, and corrections."""
-
-    def __init__(self, anti_pattern_extractor: AntiPatternExtractor | None = None):
-        """
-        Initialize negative feedback handler.
-
-        Args:
-            anti_pattern_extractor: Optional anti-pattern extractor instance
-        """
-        self.anti_pattern_extractor = anti_pattern_extractor or AntiPatternExtractor()
-        self.rejections: list[dict[str, Any]] = []
-        self.corrections: list[dict[str, Any]] = []
-
-    def record_rejection(
-        self,
-        code: str,
-        task_id: str,
-        reason: str,
-        quality_score: float = 0.5,
-    ) -> list[CodePattern]:
-        """
-        Record user rejection and extract anti-patterns.
-
-        Args:
-            code: Rejected code
-            task_id: Task identifier
-            reason: Reason for rejection
-            quality_score: Quality score
-
-        Returns:
-            List of extracted anti-patterns
-        """
-        rejection = {
-            "task_id": task_id,
-            "reason": reason,
-            "quality_score": quality_score,
-            "timestamp": datetime.now(UTC).isoformat(),
-        }
-        self.rejections.append(rejection)
-
-        # Extract anti-patterns from rejected code
-        anti_patterns = self.anti_pattern_extractor.extract_from_rejection(
-            code=code,
-            task_id=task_id,
-            rejection_reason=reason,
-            quality_score=quality_score,
-        )
-
-        # Update rejection counts for existing anti-patterns
-        for anti_pattern in anti_patterns:
-            if anti_pattern.pattern_id in self.anti_pattern_extractor.anti_patterns:
-                existing = self.anti_pattern_extractor.anti_patterns[
-                    anti_pattern.pattern_id
-                ]
-                existing.rejection_count += 1
-                if reason not in existing.failure_reasons:
-                    existing.failure_reasons.append(reason)
-            else:
-                # New anti-pattern from rejection should have rejection_count=1
-                anti_pattern.rejection_count = 1
-                self.anti_pattern_extractor.anti_patterns[
-                    anti_pattern.pattern_id
-                ] = anti_pattern
-
-        return anti_patterns
-
-    def record_correction(
-        self,
-        original_code: str,
-        corrected_code: str,
-        task_id: str,
-        correction_reason: str,
-    ) -> tuple[list[CodePattern], list[CodePattern]]:
-        """
-        Record user correction and extract both anti-patterns and patterns.
-
-        Args:
-            original_code: Original (incorrect) code
-            corrected_code: Corrected code
-            task_id: Task identifier
-            correction_reason: Reason for correction
-
-        Returns:
-            Tuple of (anti-patterns from original, patterns from corrected)
-        """
-        correction = {
-            "task_id": task_id,
-            "reason": correction_reason,
-            "timestamp": datetime.now(UTC).isoformat(),
-        }
-        self.corrections.append(correction)
-
-        # Extract anti-patterns from original code
-        anti_patterns = self.anti_pattern_extractor.extract_from_rejection(
-            code=original_code,
-            task_id=task_id,
-            rejection_reason=f"Correction: {correction_reason}",
-            quality_score=0.3,  # Low quality for incorrect code
-        )
-
-        # Extract patterns from corrected code (using PatternExtractor logic)
-        # Note: This would require access to PatternExtractor, but for now
-        # we'll just return the anti-patterns
-        return anti_patterns, []
-
-    def extract_anti_patterns_from_feedback(
-        self, code: str, task_id: str, feedback: str
-    ) -> list[CodePattern]:
-        """
-        Extract anti-patterns from feedback text.
-
-        Args:
-            code: Code that received feedback
-            task_id: Task identifier
-            feedback: Feedback text
-
-        Returns:
-            List of extracted anti-patterns
-        """
-        return self.anti_pattern_extractor.extract_from_rejection(
-            code=code,
-            task_id=task_id,
-            rejection_reason=feedback,
-            quality_score=0.4,
-        )
-
-    def get_anti_patterns_for_context(
-        self, context: str, pattern_type: str | None = None, limit: int = 5
-    ) -> list[CodePattern]:
-        """
-        Get anti-patterns to avoid for a context.
-
-        Args:
-            context: Context string
-            pattern_type: Optional pattern type filter
-            limit: Maximum results
-
-        Returns:
-            List of anti-patterns to avoid
-        """
-        return self.anti_pattern_extractor.get_anti_patterns_for_context(
-            context=context, pattern_type=pattern_type, limit=limit
-        )
-
-
-class FailureModeAnalyzer:
-    """Analyzes failure patterns and categorizes failures."""
-
-    def __init__(self):
-        """Initialize failure mode analyzer."""
-        self.failure_modes: dict[str, dict[str, Any]] = {}
-
-    def analyze_failure(
-        self,
-        code: str,
-        task_id: str,
-        failure_reasons: list[str],
-        quality_scores: dict[str, float] | None = None,
-    ) -> dict[str, Any]:
-        """
-        Analyze a single failure.
-
-        Args:
-            code: Failed code
-            task_id: Task identifier
-            failure_reasons: List of failure reasons
-            quality_scores: Optional quality scores
-
-        Returns:
-            Analysis result with failure mode and suggestions
-        """
-        failure_mode = self.identify_failure_mode(failure_reasons, quality_scores)
-
-        # Track failure mode statistics
-        if failure_mode not in self.failure_modes:
-            self.failure_modes[failure_mode] = {
-                "count": 0,
-                "reasons": [],
-                "task_ids": [],
-            }
-
-        self.failure_modes[failure_mode]["count"] += 1
-        self.failure_modes[failure_mode]["reasons"].extend(failure_reasons)
-        self.failure_modes[failure_mode]["task_ids"].append(task_id)
-
-        # Generate prevention suggestions
-        suggestions = self.suggest_prevention(failure_mode, failure_reasons)
-
-        return {
-            "failure_mode": failure_mode,
-            "failure_reasons": failure_reasons,
-            "suggestions": suggestions,
-            "task_id": task_id,
-        }
-
-    def identify_failure_mode(
-        self,
-        failure_reasons: list[str],
-        quality_scores: dict[str, float] | None = None,
-    ) -> str:
-        """
-        Categorize failure into a failure mode.
-
-        Args:
-            failure_reasons: List of failure reasons
-            quality_scores: Optional quality scores
-
-        Returns:
-            Failure mode category
-        """
-        reasons_str = " ".join(failure_reasons).lower()
-
-        # Categorize based on keywords and quality scores
-        if any(
-            keyword in reasons_str
-            for keyword in ["syntax", "parse", "indentation", "syntaxerror"]
-        ):
-            return "syntax_error"
-        elif any(
-            keyword in reasons_str
-            for keyword in ["security", "vulnerability", "insecure", "bandit"]
-        ):
-            return "security_issue"
-        elif any(
-            keyword in reasons_str
-            for keyword in ["timeout", "slow", "performance", "efficiency"]
-        ):
-            return "performance_issue"
-        elif any(
-            keyword in reasons_str
-            for keyword in ["logic", "incorrect", "wrong", "bug", "error"]
-        ):
-            return "logic_error"
-        elif quality_scores:
-            # Check quality scores
-            if quality_scores.get("security_score", 10.0) < 5.0:
-                return "security_issue"
-            elif quality_scores.get("complexity_score", 0.0) > 8.0:
-                return "complexity_issue"
-            elif quality_scores.get("maintainability_score", 10.0) < 5.0:
-                return "maintainability_issue"
-            else:
-                return "quality_issue"
-        else:
-            return "unknown_failure"
-
-    def get_common_failure_modes(self, limit: int = 5) -> list[dict[str, Any]]:
-        """
-        Get most common failure modes.
-
-        Args:
-            limit: Maximum results
-
-        Returns:
-            List of failure mode statistics
-        """
-        modes = sorted(
-            self.failure_modes.items(),
-            key=lambda x: x[1]["count"],
-            reverse=True,
-        )
-        return [
-            {
-                "mode": mode,
-                "count": stats["count"],
-                "reasons": list(set(stats["reasons"]))[:5],  # Unique reasons, limit 5
-            }
-            for mode, stats in modes[:limit]
-        ]
-
-    def suggest_prevention(
-        self, failure_mode: str, failure_reasons: list[str]
-    ) -> list[str]:
-        """
-        Suggest how to prevent this type of failure.
-
-        Args:
-            failure_mode: Failure mode category
-            failure_reasons: List of failure reasons
-
-        Returns:
-            List of prevention suggestions
-        """
-        suggestions = []
-
-        if failure_mode == "syntax_error":
-            suggestions.append(
-                "Use syntax checking tools (e.g., Ruff, pylint) before code execution"
-            )
-            suggestions.append("Review Python syntax rules and indentation")
-        elif failure_mode == "security_issue":
-            suggestions.append("Run security scanning (Bandit) before learning patterns")
-            suggestions.append("Review security best practices for the language")
-            suggestions.append("Avoid insecure patterns (eval, exec, shell=True)")
-        elif failure_mode == "performance_issue":
-            suggestions.append("Profile code to identify bottlenecks")
-            suggestions.append("Review algorithm complexity and optimization opportunities")
-            suggestions.append("Consider caching or lazy evaluation")
-        elif failure_mode == "logic_error":
-            suggestions.append("Add unit tests to catch logic errors early")
-            suggestions.append("Use type checking (mypy) to catch type-related issues")
-            suggestions.append("Review code logic and edge cases")
-        elif failure_mode == "complexity_issue":
-            suggestions.append("Refactor complex code into smaller functions")
-            suggestions.append("Reduce nesting depth and cyclomatic complexity")
-            suggestions.append("Use design patterns to simplify code structure")
-        elif failure_mode == "maintainability_issue":
-            suggestions.append("Improve code documentation and naming")
-            suggestions.append("Follow consistent coding style")
-            suggestions.append("Reduce code duplication")
-        else:
-            suggestions.append("Review failure reasons and improve code quality")
-            suggestions.append("Add more comprehensive testing")
-
-        return suggestions
-
-
-class FeedbackAnalyzer:
-    """Analyzes feedback from code scoring system and user input."""
-
-    def __init__(self):
-        """Initialize feedback analyzer."""
-        self.feedback_history: list[dict[str, Any]] = []
-
-    def analyze_code_scores(
-        self, scores: dict[str, float], threshold: float = 0.7
-    ) -> dict[str, Any]:
-        """
-        Analyze code scoring results.
-
-        Args:
-            scores: Dictionary of metric scores
-            threshold: Quality threshold
-
-        Returns:
-            Analysis results
-        """
-        overall_score = scores.get("overall_score", 0.0)
-        metrics_obj: object = scores.get("metrics", {})
-        metrics: dict[str, float] = {}
-        if isinstance(metrics_obj, dict):
-            for metric, score in metrics_obj.items():
-                if isinstance(score, (int, float)):
-                    metrics[str(metric)] = float(score)
-
-        # Identify weak areas
-        weak_areas = [metric for metric, score in metrics.items() if score < threshold]
-
-        # Calculate improvement potential
-        improvement_potential = {}
-        for metric, score in metrics.items():
-            if score < threshold:
-                improvement_potential[metric] = threshold - score
-
-        analysis = {
-            "overall_score": overall_score,
-            "weak_areas": weak_areas,
-            "improvement_potential": improvement_potential,
-            "meets_threshold": overall_score >= threshold,
-            "timestamp": datetime.now(UTC).isoformat(),
-        }
-
-        self.feedback_history.append(analysis)
-        return analysis
-
-    def correlate_prompt_changes(
-        self, prompt_variants: list[str], quality_scores: list[float]
-    ) -> dict[str, float]:
-        """
-        Correlate prompt changes with quality improvements.
-
-        Args:
-            prompt_variants: List of prompt variant identifiers
-            quality_scores: Corresponding quality scores
-
-        Returns:
-            Dictionary mapping variants to quality scores
-        """
-        correlations = {}
-        for variant, score in zip(prompt_variants, quality_scores, strict=False):
-            correlations[variant] = score
-
-        return correlations
-
-    def get_improvement_suggestions(self, analysis: dict[str, Any]) -> list[str]:
-        """
-        Get improvement suggestions based on analysis.
-
-        Args:
-            analysis: Analysis results
-
-        Returns:
-            List of improvement suggestions
-        """
-        suggestions = []
-
-        weak_areas = analysis.get("weak_areas", [])
-        improvement_potential = analysis.get("improvement_potential", {})
-
-        for area in weak_areas:
-            potential = improvement_potential.get(area, 0.0)
-            if potential > 0.1:
-                suggestions.append(
-                    f"Focus on improving {area} (potential: {potential:.2f})"
-                )
-
-        if not suggestions:
-            suggestions.append("Code quality is good, maintain current patterns")
-
-        return suggestions
-
-
-class AgentLearner:
-    """
-    Core learning engine for agents.
-
-    Integrates pattern extraction, prompt optimization, and feedback analysis.
-    """
-
-    def __init__(
-        self,
-        capability_registry: CapabilityRegistry,
-        expert_registry: Any,  # ExpertRegistry (required)
-        memory_system: TaskMemorySystem | None = None,
-        hardware_profile: HardwareProfile | None = None,
-    ):
-        """
-        Initialize agent learner.
-
-        Args:
-            capability_registry: Capability registry
-            expert_registry: Expert registry for best practices consultation (required)
-            memory_system: Optional task memory system
-            hardware_profile: Hardware profile (auto-detected if None)
-        """
-        if hardware_profile is None:
-            profiler = HardwareProfiler()
-            hardware_profile = profiler.detect_profile()
-
-        self.capability_registry = capability_registry
-        self.memory_system = memory_system
-        self.hardware_profile = hardware_profile
-        self.learning_intensity = self._get_learning_intensity()
-
-        # Initialize security scanner
-        security_scanner = SecurityScanner()
-
-        self.pattern_extractor = PatternExtractor(security_scanner=security_scanner)
-        self.anti_pattern_extractor = AntiPatternExtractor(
-            security_scanner=security_scanner
-        )
-        self.negative_feedback_handler = NegativeFeedbackHandler(
-            anti_pattern_extractor=self.anti_pattern_extractor
-        )
-        self.failure_mode_analyzer = FailureModeAnalyzer()
-        self.prompt_optimizer = PromptOptimizer(hardware_profile)
-        self.feedback_analyzer = FeedbackAnalyzer()
-
-        # Initialize explainability components
-        self.decision_logger = DecisionReasoningLogger()
-        self.pattern_explainer = PatternSelectionExplainer()
-        self.impact_reporter = LearningImpactReporter()
-
-        # Initialize meta-learning components
-        self.effectiveness_tracker = LearningEffectivenessTracker()
-        self.self_assessor = LearningSelfAssessor()
-        self.adaptive_rate = AdaptiveLearningRate()
-        self.strategy_selector = LearningStrategySelector()
-        self.current_strategy = LearningStrategy.BALANCED
-
-        # Initialize decision engine (always required) with explainability
-        best_practice_consultant = BestPracticeConsultant(expert_registry)
-        confidence_calculator = LearningConfidenceCalculator()
-        self.decision_engine = LearningDecisionEngine(
-            capability_registry=self.capability_registry,
-            best_practice_consultant=best_practice_consultant,
-            confidence_calculator=confidence_calculator,
-            decision_logger=self.decision_logger,
-        )
-
-        # Initialize dashboard
-        self.dashboard = LearningDashboard(
-            capability_registry=self.capability_registry,
-            pattern_extractor=self.pattern_extractor,
-            anti_pattern_extractor=self.anti_pattern_extractor,
-            decision_logger=self.decision_logger,
-            impact_reporter=self.impact_reporter,
-        )
-
-    def _get_learning_intensity(self) -> LearningIntensity:
-        """Get learning intensity based on hardware."""
-        return self.capability_registry.get_learning_intensity()
-
-    async def learn_from_task(
-        self,
-        capability_id: str,
-        task_id: str,
-        code: str | None = None,
-        quality_scores: dict[str, float] | None = None,
-        success: bool = True,
-        duration: float = 0.0,
-    ) -> dict[str, Any]:
-        """
-        Learn from a completed task.
-
-        Args:
-            capability_id: Capability identifier
-            task_id: Task identifier
-            code: Optional source code
-            quality_scores: Optional code scoring results
-            success: Whether task succeeded
-            duration: Task duration
-
-        Returns:
-            Learning results
-        """
-        results: dict[str, Any] = {
-            "patterns_extracted": 0,
-            "anti_patterns_extracted": 0,
-            "prompt_optimized": False,
-            "feedback_analyzed": False,
-            "security_checked": False,
-            "security_score": 0.0,
-            "security_vulnerabilities": [],
-            "failure_analyzed": False,
-        }
-        patterns: list[CodePattern] = []
-        anti_patterns: list[CodePattern] = []
-
-        # Get before metrics for tracking
-        metric = self.capability_registry.get_capability(capability_id)
-        before_metrics = {
-            "quality_score": metric.quality_score if metric else 0.0,
-            "success_rate": metric.success_rate if metric else 0.0,
-            "usage_count": metric.usage_count if metric else 0,
-        }
-
-        # Extract quality score
-        quality_score = 0.5
-        if quality_scores:
-            quality_score = (
-                quality_scores.get("overall_score", 50.0) / 100.0
-            )  # Normalize to 0-1 (assumes 0-100 scale)
-
-        # Security check before learning
-        security_scanner = SecurityScanner()
-        security_result = None
-        if code:
-            security_result = security_scanner.scan_code(code=code)
-            results["security_checked"] = True
-            results["security_score"] = security_result["security_score"]
-            results["security_vulnerabilities"] = security_result["vulnerabilities"]
-
-            # Skip pattern extraction if security score is too low
-            # But continue to extract anti-patterns from vulnerable code
-            if not security_scanner.is_safe_for_learning(
-                code=code, threshold=self.pattern_extractor.security_threshold
-            ):
-                logger.debug(
-                    f"Skipping pattern extraction for task {task_id}: "
-                    f"security score {security_result['security_score']:.2f} "
-                    f"below threshold {self.pattern_extractor.security_threshold:.2f}"
-                )
-                # Still update capability metrics even if we skip pattern extraction
-                self.capability_registry.update_capability_metrics(
-                    capability_id=capability_id,
-                    success=success,
-                    duration=duration,
-                    quality_score=quality_score,
-                )
-                # Don't return early - continue to extract anti-patterns from vulnerable code
-                # The anti-pattern extraction code below will handle low-quality/vulnerable code
-
-        # Update capability metrics
-        self.capability_registry.update_capability_metrics(
-            capability_id=capability_id,
-            success=success,
-            duration=duration,
-            quality_score=quality_score,
-        )
-
-        # Extract patterns if code provided and quality is good
-        # Use decision engine for adaptive threshold, but ensure high-quality code is always extracted
-        should_extract_patterns = False
-        if code and self.learning_intensity != LearningIntensity.LOW:
-            # Base threshold: extract if quality >= 0.7 (normalized 0-1 scale)
-            base_threshold = 0.7
-            
-            # Use decision engine to get adaptive threshold (if available)
-            metric = self.capability_registry.get_capability(capability_id)
-            learned_data = {
-                "usage_count": metric.usage_count if metric else 0,
-                "success_rate": metric.success_rate if metric else 0.0,
-                "quality_score": quality_score,
-                "value": quality_score,  # The threshold value we're considering
-                "context_relevance": 1.0,
-            }
-            context = {
-                "hardware_profile": (
-                    self.hardware_profile.profile_type.value
-                    if hasattr(self.hardware_profile, "profile_type")
-                    else "unknown"
-                ),
-                "learning_intensity": self.learning_intensity.value,
-                "task_id": task_id,
-                "capability_id": capability_id,
-            }
-
-            # Use decision engine (always available)
-            decision = await self.decision_engine.make_decision(
-                decision_type="pattern_extraction_threshold",
-                learned_data=learned_data,
-                context=context,
-                default_value=base_threshold,
-            )
-
-            # Determine threshold value from decision result
-            threshold_value = base_threshold  # Default
-            if decision.result.value is not None:
-                if isinstance(decision.result.value, (int, float)):
-                    threshold_value = float(decision.result.value)
-                elif isinstance(decision.result.value, str):
-                    # Try to extract numeric value from string
-                    import re
-                    match = re.search(r'(\d+\.?\d*)', str(decision.result.value))
-                    if match:
-                        threshold_value = float(match.group(1))
-            
-            # Extract patterns if quality_score meets threshold
-            # For new capabilities (low learned_confidence), still extract if quality is high
-            should_extract_patterns = quality_score >= threshold_value
-            
-            # Fallback: if quality is very high (>= 0.8), always extract regardless of decision
-            if quality_score >= 0.8:
-                should_extract_patterns = True
-            
-            logger.debug(
-                f"Pattern extraction decision: quality_score={quality_score:.3f}, "
-                f"threshold={threshold_value:.3f}, should_extract={should_extract_patterns}"
-            )
-
-        if should_extract_patterns and code is not None:
-            logger.debug(f"Attempting to extract patterns from code (length={len(code)})")
-            patterns = self.pattern_extractor.extract_patterns(
-                code=code, quality_score=quality_score, task_id=task_id
-            )
-            logger.debug(f"Extracted {len(patterns)} patterns from code")
-
-            # Store patterns
-            for pattern in patterns:
-                if pattern.pattern_id not in self.pattern_extractor.patterns:
-                    self.pattern_extractor.patterns[pattern.pattern_id] = pattern
-                else:
-                    # Update existing pattern
-                    existing = self.pattern_extractor.patterns[pattern.pattern_id]
-                    existing.usage_count += 1
-                    existing.learned_from.append(task_id)
-
-            results["patterns_extracted"] = len(patterns)
-
-        # Handle failures: extract anti-patterns
-        if not success and code:
-            failure_reasons = [f"Task {task_id} failed"]
-            if quality_scores:
-                # Add quality-based failure reasons
-                if quality_scores.get("security_score", 10.0) < 5.0:
-                    failure_reasons.append("Low security score")
-                if quality_scores.get("overall_score", 100.0) < 50.0:
-                    failure_reasons.append("Low overall quality score")
-
-            # Extract anti-patterns from failed code
-            anti_patterns = self.anti_pattern_extractor.extract_from_failure(
-                code=code,
-                task_id=task_id,
-                failure_reasons=failure_reasons,
-                quality_score=quality_score,
-            )
-
-            # Store anti-patterns
-            for anti_pattern in anti_patterns:
-                if anti_pattern.pattern_id not in self.anti_pattern_extractor.anti_patterns:
-                    self.anti_pattern_extractor.anti_patterns[
-                        anti_pattern.pattern_id
-                    ] = anti_pattern
-                else:
-                    # Update existing anti-pattern
-                    existing = self.anti_pattern_extractor.anti_patterns[
-                        anti_pattern.pattern_id
-                    ]
-                    existing.usage_count += 1
-                    existing.learned_from.append(task_id)
-
-            results["anti_patterns_extracted"] = len(anti_patterns)
-
-            # Analyze failure mode
-            failure_analysis = self.failure_mode_analyzer.analyze_failure(
-                code=code,
-                task_id=task_id,
-                failure_reasons=failure_reasons,
-                quality_scores=quality_scores,
-            )
-            results["failure_analyzed"] = True
-            results["failure_analysis"] = failure_analysis
-
-        # Also extract anti-patterns from low-quality code (even if success=True)
-        logger.debug(f"Anti-pattern extraction check: code={code is not None}, quality_score={quality_score}, threshold={self.anti_pattern_extractor.max_quality_threshold}, success={success}")
-        if code and quality_score < self.anti_pattern_extractor.max_quality_threshold:
-            failure_reasons = [f"Low quality score: {quality_score:.2f}"]
-            anti_patterns = self.anti_pattern_extractor.extract_anti_patterns(
-                code=code,
-                quality_score=quality_score,
-                task_id=task_id,
-                failure_reasons=failure_reasons,
-            )
-
-            # Store anti-patterns
-            for anti_pattern in anti_patterns:
-                if anti_pattern.pattern_id not in self.anti_pattern_extractor.anti_patterns:
-                    self.anti_pattern_extractor.anti_patterns[
-                        anti_pattern.pattern_id
-                    ] = anti_pattern
-
-            results["anti_patterns_extracted"] = len(anti_patterns)
-
-        # Analyze feedback if scores provided
-        if quality_scores:
-            analysis = self.feedback_analyzer.analyze_code_scores(quality_scores)
-            results["feedback_analyzed"] = True
-            results["feedback_analysis"] = analysis
-
-            # Check if improvement is needed
-            if not analysis.get("meets_threshold", False):
-                suggestions = self.feedback_analyzer.get_improvement_suggestions(
-                    analysis
-                )
-                results["improvement_suggestions"] = suggestions
-
-        # Store in memory system if available
-        if self.memory_system and success:
-            outcome = TaskOutcome.SUCCESS if success else TaskOutcome.FAILURE
-            self.memory_system.store_memory(
-                task_id=task_id,
-                agent_id="unknown",  # Would be provided by agent
-                command=capability_id,
-                outcome=outcome,
-                quality_score=quality_score,
-                patterns_used=[p.pattern_id for p in patterns] if code else [],
-            )
-
-        # Track learning effectiveness
-        metric_after = self.capability_registry.get_capability(capability_id)
-        after_metrics = {
-            "quality_score": metric_after.quality_score if metric_after else 0.0,
-            "success_rate": metric_after.success_rate if metric_after else 0.0,
-            "usage_count": metric_after.usage_count if metric_after else 0,
-        }
-
-        # Track effectiveness
-        session = self.effectiveness_tracker.track_effectiveness(
-            capability_id=capability_id,
-            before_metrics=before_metrics,
-            after_metrics=after_metrics,
-            strategies_used=[self.current_strategy.value],
-        )
-
-        # Adjust learning rate based on effectiveness
-        rate_adjustment = self.adaptive_rate.adjust_learning_intensity(
-            session.effectiveness_score
-        )
-        results["learning_rate_adjustment"] = rate_adjustment
-
-        # Generate impact report
-        impact_report = self.impact_reporter.generate_impact_report(
-            capability_id=capability_id,
-            before_metrics=before_metrics,
-            after_metrics=after_metrics,
-            learning_session_id=session.session_id,
-        )
-        results["learning_impact"] = impact_report
-
-        return results
-
-    async def learn_from_rejection(
-        self,
-        capability_id: str,
-        task_id: str,
-        code: str,
-        rejection_reason: str,
-        quality_score: float = 0.5,
-    ) -> dict[str, Any]:
-        """
-        Learn from user rejection.
-
-        Args:
-            capability_id: Capability identifier
-            task_id: Task identifier
-            code: Rejected code
-            rejection_reason: Reason for rejection
-            quality_score: Quality score
-
-        Returns:
-            Learning results
-        """
-        results: dict[str, Any] = {
-            "anti_patterns_extracted": 0,
-            "rejection_recorded": False,
-        }
-
-        # Record rejection and extract anti-patterns
-        anti_patterns = self.negative_feedback_handler.record_rejection(
-            code=code,
-            task_id=task_id,
-            reason=rejection_reason,
-            quality_score=quality_score,
-        )
-
-        results["anti_patterns_extracted"] = len(anti_patterns)
-        results["rejection_recorded"] = True
-
-        # Update capability metrics (rejection counts as failure)
-        self.capability_registry.update_capability_metrics(
-            capability_id=capability_id,
-            success=False,
-            duration=0.0,
-            quality_score=quality_score,
-        )
-
-        return results
-
-    def get_learned_patterns(
-        self,
-        context: str,
-        pattern_type: str | None = None,
-        limit: int = 5,
-        exclude_anti_patterns: bool = True,
-    ) -> list[CodePattern]:
-        """
-        Get learned patterns for a context.
-
-        Args:
-            context: Context string
-            pattern_type: Optional pattern type filter
-            limit: Maximum results
-            exclude_anti_patterns: If True, exclude anti-patterns (default: True)
-
-        Returns:
-            List of relevant patterns
-        """
-        return self.pattern_extractor.get_patterns_for_context(
-            context=context,
-            pattern_type=pattern_type,
-            limit=limit,
-            exclude_anti_patterns=exclude_anti_patterns,
-        )
-
-    def optimize_prompt(self, base_prompt: str, context: str | None = None) -> str:
-        """
-        Get optimized prompt.
-
-        Args:
-            base_prompt: Base prompt
-            context: Optional context
-
-        Returns:
-            Optimized prompt
-        """
-        # Get best variant if available
-        best_variant = self.prompt_optimizer.get_best_variant()
-        if best_variant and best_variant.average_quality > 0.7:
-            return best_variant.prompt_template
-
-        # Otherwise, optimize for hardware
-        return self.prompt_optimizer.optimize_for_hardware(base_prompt)
-
-    def explain_learning(
-        self,
-        capability_id: str,
-        task_id: str | None = None,
-        decision_id: str | None = None,
-    ) -> dict[str, Any]:
-        """
-        Generate explanation for learning process.
-
-        Args:
-            capability_id: Capability identifier
-            task_id: Optional task identifier
-            decision_id: Optional decision identifier
-
-        Returns:
-            Explanation dictionary
-        """
-        explanation: dict[str, Any] = {
-            "capability_id": capability_id,
-            "task_id": task_id,
-        }
-
-        # Get decision explanation if decision_id provided
-        if decision_id:
-            decision_explanation = self.decision_logger.explain_decision(decision_id)
-            if decision_explanation:
-                explanation["decision"] = decision_explanation
-
-        # Get pattern selection explanation
-        patterns = self.get_learned_patterns(
-            context=capability_id, exclude_anti_patterns=True
-        )
-        if patterns:
-            pattern_explanation = self.pattern_explainer.explain_pattern_selection(
-                selected_patterns=patterns, context=capability_id
-            )
-            explanation["pattern_selection"] = pattern_explanation
-
-        # Get decision statistics
-        explanation["decision_statistics"] = self.decision_logger.get_decision_statistics()
-
-        return explanation
-
-    async def optimize_learning(
-        self, capability_id: str | None = None
-    ) -> dict[str, Any]:
-        """
-        Run meta-learning optimization.
-
-        Args:
-            capability_id: Optional filter by capability
-
-        Returns:
-            Optimization report
-        """
-        optimization: dict[str, Any] = {
-            "timestamp": datetime.now(UTC).isoformat(),
-            "capability_id": capability_id,
-        }
-
-        # Assess learning quality
-        pattern_count = len(self.pattern_extractor.patterns)
-        anti_pattern_count = len(self.anti_pattern_extractor.anti_patterns)
-
-        # Calculate average quality and security
-        avg_quality = 0.0
-        avg_security = 0.0
-        if self.pattern_extractor.patterns:
-            total_quality = sum(p.quality_score for p in self.pattern_extractor.patterns.values())
-            total_security = sum(
-                getattr(p, "security_score", 0.0)
-                for p in self.pattern_extractor.patterns.values()
-            )
-            avg_quality = total_quality / len(self.pattern_extractor.patterns)
-            avg_security = total_security / len(self.pattern_extractor.patterns)
-
-        quality_assessment = self.self_assessor.assess_learning_quality(
-            pattern_count=pattern_count,
-            anti_pattern_count=anti_pattern_count,
-            average_quality=avg_quality,
-            average_security=avg_security,
-        )
-        optimization["quality_assessment"] = quality_assessment
-
-        # Identify learning gaps
-        capability_metrics = {}
-        if capability_id:
-            metric = self.capability_registry.get_capability(capability_id)
-            if metric:
-                capability_metrics = {
-                    "success_rate": metric.success_rate,
-                    "quality_score": metric.quality_score,
-                    "usage_count": metric.usage_count,
-                }
-        pattern_stats = {
-            "total_patterns": pattern_count,
-            "average_quality": avg_quality,
-            "average_security": avg_security,
-        }
-        gaps = self.self_assessor.identify_learning_gaps(
-            capability_metrics=capability_metrics,
-            pattern_statistics=pattern_stats,
-        )
-        optimization["learning_gaps"] = gaps
-
-        # Get improvement suggestions
-        suggestions = self.self_assessor.suggest_improvements(quality_assessment)
-        optimization["improvement_suggestions"] = suggestions
-
-        # Select optimal strategy
-        if capability_id:
-            metric = self.capability_registry.get_capability(capability_id)
-            current_effectiveness = metric.success_rate if metric else 0.5
-        else:
-            # Use average effectiveness
-            roi = self.effectiveness_tracker.get_learning_roi(capability_id=capability_id)
-            current_effectiveness = roi.get("average_effectiveness", 0.5)
-
-        hardware_profile_str = (
-            self.hardware_profile.profile_type.value
-            if hasattr(self.hardware_profile, "profile_type")
-            else None
-        )
-        optimal_strategy = self.strategy_selector.select_strategy(
-            capability_id=capability_id or "global",
-            current_effectiveness=current_effectiveness,
-            hardware_profile=hardware_profile_str,
-        )
-        optimization["optimal_strategy"] = optimal_strategy.value
-        optimization["current_strategy"] = self.current_strategy.value
-
-        # Optimize thresholds
-        current_threshold = self.pattern_extractor.min_quality_threshold
-        metric = self.capability_registry.get_capability(capability_id) if capability_id else None
-        success_rate = metric.success_rate if metric else 0.5
-        optimized_threshold = self.adaptive_rate.optimize_thresholds(
-            current_threshold=current_threshold,
-            success_rate=success_rate,
-            quality_score=avg_quality,
-        )
-        optimization["optimized_threshold"] = optimized_threshold
-        optimization["current_threshold"] = current_threshold
-
-        # Update strategy if better one found
-        if optimal_strategy != self.current_strategy:
-            switch_result = self.strategy_selector.switch_strategy(
-                current_strategy=self.current_strategy,
-                new_strategy=optimal_strategy,
-            )
-            if switch_result["switched"]:
-                self.current_strategy = optimal_strategy
-                optimization["strategy_switched"] = True
-                optimization["switch_result"] = switch_result
-            else:
-                optimization["strategy_switched"] = False
-
-        # Get effectiveness metrics
-        effectiveness_metrics = self.effectiveness_tracker.get_learning_roi(
-            capability_id=capability_id
-        )
-        optimization["effectiveness_metrics"] = effectiveness_metrics
-
-        return optimization
-
-    def should_refine_capability(self, capability_id: str) -> bool:
-        """
-        Determine if a capability should be refined.
-
-        Args:
-            capability_id: Capability identifier
-
-        Returns:
-            True if refinement is recommended
-        """
-        metric = self.capability_registry.get_capability(capability_id)
-        if not metric:
-            return False
-
-        # Refine if quality is below threshold and has enough usage
-        return metric.quality_score < 0.7 and metric.usage_count >= 10
+"""
+Agent Learning System
+
+Enables agents to learn from past tasks and improve over time.
+"""
+
+# @ai-prime-directive: This file implements the Agent Learning System for pattern extraction, prompt optimization,
+# and feedback analysis. The system enables agents to learn from past tasks and improve over time through
+# meta-learning, pattern recognition, and adaptive learning strategies.
+
+# @ai-constraints:
+# - Learning intensity must adapt to hardware profile (NUC, Desktop, Server)
+# - Pattern extraction must include security scanning before pattern acceptance
+# - Anti-pattern extraction must learn from negative feedback and failures
+# - Prompt optimization must respect hardware constraints and token budgets
+# - Meta-learning must track learning effectiveness and adjust strategies accordingly
+# - Performance: Learning operations should not significantly impact agent response times
+
+# @note[2025-01-15]: Agent learning is an advanced feature that improves agent performance over time.
+# The system uses meta-learning to adapt learning strategies based on effectiveness tracking.
+# See docs/architecture/decisions/ for related architectural decisions.
+
+import logging
+import re
+from dataclasses import dataclass, field
+from datetime import UTC, datetime
+from typing import Any
+
+from .best_practice_consultant import BestPracticeConsultant
+from .capability_registry import CapabilityRegistry, LearningIntensity
+from .hardware_profiler import HardwareProfile, HardwareProfiler
+from .learning_confidence import LearningConfidenceCalculator
+from .learning_dashboard import LearningDashboard
+from .learning_decision import LearningDecisionEngine
+from .learning_explainability import (
+    DecisionReasoningLogger,
+    LearningImpactReporter,
+    PatternSelectionExplainer,
+)
+from .meta_learning import (
+    AdaptiveLearningRate,
+    LearningEffectivenessTracker,
+    LearningSelfAssessor,
+    LearningStrategy,
+    LearningStrategySelector,
+)
+from .security_scanner import SecurityScanner
+from .task_memory import TaskMemorySystem, TaskOutcome
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class CodePattern:
+    """Represents a learned code pattern."""
+
+    pattern_id: str
+    pattern_type: str  # "function", "class", "import", "structure"
+    code_snippet: str
+    context: str
+    quality_score: float
+    usage_count: int
+    success_rate: float
+    learned_from: list[str]  # Task IDs where this pattern was successful
+    metadata: dict[str, Any] = field(default_factory=dict)
+    security_score: float = 0.0  # Security score (0-10, higher is better)
+    is_anti_pattern: bool = False  # True if this is an anti-pattern to avoid
+    failure_reasons: list[str] = field(default_factory=list)  # Reasons for failure
+    rejection_count: int = 0  # Number of times this pattern was rejected
+
+
+@dataclass
+class PromptVariant:
+    """Represents a prompt variation for A/B testing."""
+
+    variant_id: str
+    prompt_template: str
+    modifications: list[str]  # List of modifications made
+    test_count: int
+    success_count: int
+    average_quality: float
+    created_at: datetime
+    last_tested: datetime | None = None
+
+
+class PatternExtractor:
+    """Extracts patterns from successful code outputs."""
+
+    def __init__(
+        self,
+        min_quality_threshold: float = 0.7,
+        security_scanner: SecurityScanner | None = None,
+        security_threshold: float = 7.0,
+    ):
+        """
+        Initialize pattern extractor.
+
+        Args:
+            min_quality_threshold: Minimum quality score to extract patterns
+            security_scanner: Optional security scanner instance
+            security_threshold: Minimum security score to extract patterns (default: 7.0)
+        """
+        self.min_quality_threshold = min_quality_threshold
+        self.security_scanner = security_scanner or SecurityScanner()
+        self.security_threshold = security_threshold
+        self.patterns: dict[str, CodePattern] = {}
+
+    def extract_patterns(
+        self,
+        code: str,
+        quality_score: float,
+        task_id: str,
+        pattern_types: list[str] | None = None,
+    ) -> list[CodePattern]:
+        """
+        Extract patterns from code.
+
+        Args:
+            code: Source code
+            quality_score: Quality score of the code
+            task_id: Task identifier
+            pattern_types: Optional list of pattern types to extract
+
+        Returns:
+            List of extracted patterns
+        """
+        if quality_score < self.min_quality_threshold:
+            return []
+
+        # Security check before extraction
+        security_result = self.security_scanner.scan_code(code=code)
+        security_score = security_result["security_score"]
+        vulnerabilities = security_result["vulnerabilities"]
+        is_safe = security_result.get("is_safe", True)
+
+        # Only extract if security score meets threshold
+        # Check both score >= threshold AND is_safe flag
+        if security_score < self.security_threshold or not is_safe:
+            logger.debug(
+                f"Skipping pattern extraction: security score {security_score:.2f} "
+                f"below threshold {self.security_threshold:.2f} or is_safe={is_safe}"
+            )
+            return []
+
+        patterns = []
+
+        # Extract function patterns
+        if not pattern_types or "function" in pattern_types:
+            func_patterns = self._extract_function_patterns(
+                code, quality_score, task_id
+            )
+            patterns.extend(func_patterns)
+
+        # Extract class patterns
+        if not pattern_types or "class" in pattern_types:
+            class_patterns = self._extract_class_patterns(code, quality_score, task_id)
+            patterns.extend(class_patterns)
+
+        # Extract import patterns
+        if not pattern_types or "import" in pattern_types:
+            import_patterns = self._extract_import_patterns(
+                code, quality_score, task_id
+            )
+            patterns.extend(import_patterns)
+
+        # Extract structural patterns
+        if not pattern_types or "structure" in pattern_types:
+            struct_patterns = self._extract_structural_patterns(
+                code, quality_score, task_id
+            )
+            patterns.extend(struct_patterns)
+
+        return patterns
+
+    def _extract_function_patterns(
+        self, code: str, quality_score: float, task_id: str
+    ) -> list[CodePattern]:
+        """Extract function patterns."""
+        patterns = []
+
+        # Match function definitions - more flexible pattern that handles docstrings and type hints
+        # Find all function definitions, then extract their bodies
+        # Pattern matches: def function_name(...) with optional type hints, handling multiline signatures
+        func_def_pattern = r"def\s+(\w+)\s*\([^)]*\)\s*(?:->\s*[^:]+)?:"
+        func_matches = list(re.finditer(func_def_pattern, code, re.MULTILINE | re.DOTALL))
+        
+        for i, match in enumerate(func_matches):
+            func_name = match.group(1)
+            start_pos = match.end()
+            
+            # Find the end of this function (next function def or end of code)
+            if i + 1 < len(func_matches):
+                end_pos = func_matches[i + 1].start()
+            else:
+                end_pos = len(code)
+            
+            func_body = code[start_pos:end_pos].strip()
+            
+            # Skip if function body is empty
+            if not func_body:
+                continue
+
+            # Get security score for this pattern
+            pattern_code = f"def {func_name}(...):\n{func_body[:200]}"
+            pattern_security = self.security_scanner.scan_code(code=pattern_code)
+            pattern_security_score = pattern_security["security_score"]
+            pattern_vulnerabilities = pattern_security["vulnerabilities"]
+
+            pattern = CodePattern(
+                pattern_id=f"func_{func_name}_{hash(func_body) % 10000}",
+                pattern_type="function",
+                code_snippet=pattern_code,
+                context=f"Function: {func_name}",
+                quality_score=quality_score,
+                usage_count=1,
+                success_rate=1.0,
+                learned_from=[task_id],
+                security_score=pattern_security_score,
+                metadata={"vulnerabilities": pattern_vulnerabilities},
+            )
+            patterns.append(pattern)
+
+        return patterns
+
+    def _extract_class_patterns(
+        self, code: str, quality_score: float, task_id: str
+    ) -> list[CodePattern]:
+        """Extract class patterns."""
+        patterns = []
+
+        # Match class definitions
+        class_pattern = r"class\s+(\w+)(?:\([^)]+\))?:\s*\n((?:\s{4}.*\n?)*)"
+        matches = re.finditer(class_pattern, code, re.MULTILINE)
+
+        for match in matches:
+            class_name = match.group(1)
+            class_body = match.group(2)
+
+            # Get security score for this pattern
+            pattern_code = f"class {class_name}:\n{class_body[:200]}"
+            pattern_security = self.security_scanner.scan_code(code=pattern_code)
+            pattern_security_score = pattern_security["security_score"]
+            pattern_vulnerabilities = pattern_security["vulnerabilities"]
+
+            pattern = CodePattern(
+                pattern_id=f"class_{class_name}_{hash(class_body) % 10000}",
+                pattern_type="class",
+                code_snippet=pattern_code,
+                context=f"Class: {class_name}",
+                quality_score=quality_score,
+                usage_count=1,
+                success_rate=1.0,
+                learned_from=[task_id],
+                security_score=pattern_security_score,
+                metadata={"vulnerabilities": pattern_vulnerabilities},
+            )
+            patterns.append(pattern)
+
+        return patterns
+
+    def _extract_import_patterns(
+        self, code: str, quality_score: float, task_id: str
+    ) -> list[CodePattern]:
+        """Extract import patterns."""
+        patterns = []
+
+        # Match import statements
+        import_pattern = r"^(?:from\s+[\w.]+|import\s+[\w.,\s]+)"
+        matches = re.finditer(import_pattern, code, re.MULTILINE)
+
+        imports = []
+        for match in matches:
+            imports.append(match.group(0).strip())
+
+        if imports:
+            # Get security score for this pattern
+            pattern_code = "\n".join(imports[:10])
+            pattern_security = self.security_scanner.scan_code(code=pattern_code)
+            pattern_security_score = pattern_security["security_score"]
+            pattern_vulnerabilities = pattern_security["vulnerabilities"]
+
+            pattern = CodePattern(
+                pattern_id=f"imports_{hash(''.join(imports)) % 10000}",
+                pattern_type="import",
+                code_snippet=pattern_code,
+                context="Import statements",
+                quality_score=quality_score,
+                usage_count=1,
+                success_rate=1.0,
+                learned_from=[task_id],
+                security_score=pattern_security_score,
+                metadata={"vulnerabilities": pattern_vulnerabilities},
+            )
+            patterns.append(pattern)
+
+        return patterns
+
+    def _extract_structural_patterns(
+        self, code: str, quality_score: float, task_id: str
+    ) -> list[CodePattern]:
+        """Extract structural patterns (decorators, context managers, etc.)."""
+        patterns = []
+
+        # Match decorators
+        decorator_pattern = r"@\w+(?:\([^)]*\))?"
+        decorators = re.findall(decorator_pattern, code)
+
+        if decorators:
+            # Get security score for this pattern
+            pattern_code = "\n".join(decorators[:5])
+            pattern_security = self.security_scanner.scan_code(code=pattern_code)
+            pattern_security_score = pattern_security["security_score"]
+            pattern_vulnerabilities = pattern_security["vulnerabilities"]
+
+            pattern = CodePattern(
+                pattern_id=f"decorators_{hash(''.join(decorators)) % 10000}",
+                pattern_type="structure",
+                code_snippet=pattern_code,
+                context="Decorators",
+                quality_score=quality_score,
+                usage_count=1,
+                success_rate=1.0,
+                learned_from=[task_id],
+                security_score=pattern_security_score,
+                metadata={"vulnerabilities": pattern_vulnerabilities},
+            )
+            patterns.append(pattern)
+
+        return patterns
+
+    def get_patterns_for_context(
+        self,
+        context: str,
+        pattern_type: str | None = None,
+        limit: int = 5,
+        exclude_anti_patterns: bool = True,
+    ) -> list[CodePattern]:
+        """
+        Get relevant patterns for a context.
+
+        Args:
+            context: Context string
+            pattern_type: Optional pattern type filter
+            limit: Maximum results
+            exclude_anti_patterns: If True, exclude anti-patterns (default: True)
+
+        Returns:
+            List of relevant patterns
+        """
+        candidates = list(self.patterns.values())
+
+        # Filter out anti-patterns if requested
+        if exclude_anti_patterns:
+            candidates = [p for p in candidates if not p.is_anti_pattern]
+
+        if pattern_type:
+            candidates = [p for p in candidates if p.pattern_type == pattern_type]
+
+        # Sort by security score, quality, and usage
+        candidates.sort(
+            key=lambda p: (p.security_score, p.quality_score, p.usage_count),
+            reverse=True,
+        )
+
+        return candidates[:limit]
+
+
+class AntiPatternExtractor:
+    """Extracts anti-patterns from failed tasks and low-quality code."""
+
+    def __init__(
+        self,
+        max_quality_threshold: float = 0.7,
+        security_scanner: SecurityScanner | None = None,
+    ):
+        """
+        Initialize anti-pattern extractor.
+
+        Args:
+            max_quality_threshold: Maximum quality score to extract anti-patterns
+            security_scanner: Optional security scanner instance
+        """
+        self.max_quality_threshold = max_quality_threshold
+        self.security_scanner = security_scanner or SecurityScanner()
+        self.anti_patterns: dict[str, CodePattern] = {}
+
+    def extract_anti_patterns(
+        self,
+        code: str,
+        quality_score: float,
+        task_id: str,
+        failure_reasons: list[str] | None = None,
+        pattern_types: list[str] | None = None,
+    ) -> list[CodePattern]:
+        """
+        Extract anti-patterns from failed or low-quality code.
+
+        Args:
+            code: Source code
+            quality_score: Quality score of the code
+            task_id: Task identifier
+            failure_reasons: Optional list of failure reasons
+            pattern_types: Optional list of pattern types to extract
+
+        Returns:
+            List of extracted anti-patterns
+        """
+        # Only extract if quality is below threshold
+        if quality_score >= self.max_quality_threshold:
+            return []
+
+        failure_reasons = failure_reasons or []
+        anti_patterns = []
+
+        # Extract function anti-patterns
+        if not pattern_types or "function" in pattern_types:
+            func_patterns = self._extract_function_anti_patterns(
+                code, quality_score, task_id, failure_reasons
+            )
+            anti_patterns.extend(func_patterns)
+
+        # Extract class anti-patterns
+        if not pattern_types or "class" in pattern_types:
+            class_patterns = self._extract_class_anti_patterns(
+                code, quality_score, task_id, failure_reasons
+            )
+            anti_patterns.extend(class_patterns)
+
+        # Extract import anti-patterns
+        if not pattern_types or "import" in pattern_types:
+            import_patterns = self._extract_import_anti_patterns(
+                code, quality_score, task_id, failure_reasons
+            )
+            anti_patterns.extend(import_patterns)
+
+        # Extract structural anti-patterns
+        if not pattern_types or "structure" in pattern_types:
+            struct_patterns = self._extract_structural_anti_patterns(
+                code, quality_score, task_id, failure_reasons
+            )
+            anti_patterns.extend(struct_patterns)
+
+        return anti_patterns
+
+    def _extract_function_anti_patterns(
+        self,
+        code: str,
+        quality_score: float,
+        task_id: str,
+        failure_reasons: list[str],
+    ) -> list[CodePattern]:
+        """Extract function anti-patterns."""
+        anti_patterns = []
+
+        # Match function definitions
+        func_pattern = r"def\s+(\w+)\s*\([^)]*\):\s*\n((?:\s{4}.*\n?)*)"
+        matches = list(re.finditer(func_pattern, code, re.MULTILINE))
+
+        for match in matches:
+            func_name = match.group(1)
+            func_body = match.group(2)
+
+            # Get security score for this pattern
+            pattern_code = f"def {func_name}(...):\n{func_body[:200]}"
+            pattern_security = self.security_scanner.scan_code(code=pattern_code)
+            pattern_security_score = pattern_security["security_score"]
+            pattern_vulnerabilities = pattern_security["vulnerabilities"]
+
+            anti_pattern = CodePattern(
+                pattern_id=f"anti_func_{func_name}_{hash(func_body) % 10000}",
+                pattern_type="function",
+                code_snippet=pattern_code,
+                context=f"Anti-pattern Function: {func_name}",
+                quality_score=quality_score,
+                usage_count=1,
+                success_rate=0.0,  # Anti-patterns have 0 success rate
+                learned_from=[task_id],
+                security_score=pattern_security_score,
+                is_anti_pattern=True,
+                failure_reasons=failure_reasons.copy(),
+                rejection_count=0,
+                metadata={"vulnerabilities": pattern_vulnerabilities},
+            )
+            anti_patterns.append(anti_pattern)
+
+        return anti_patterns
+
+    def _extract_class_anti_patterns(
+        self,
+        code: str,
+        quality_score: float,
+        task_id: str,
+        failure_reasons: list[str],
+    ) -> list[CodePattern]:
+        """Extract class anti-patterns."""
+        anti_patterns = []
+
+        # Match class definitions
+        class_pattern = r"class\s+(\w+)(?:\([^)]+\))?:\s*\n((?:\s{4}.*\n?)*)"
+        matches = re.finditer(class_pattern, code, re.MULTILINE)
+
+        for match in matches:
+            class_name = match.group(1)
+            class_body = match.group(2)
+
+            # Get security score for this pattern
+            pattern_code = f"class {class_name}:\n{class_body[:200]}"
+            pattern_security = self.security_scanner.scan_code(code=pattern_code)
+            pattern_security_score = pattern_security["security_score"]
+            pattern_vulnerabilities = pattern_security["vulnerabilities"]
+
+            anti_pattern = CodePattern(
+                pattern_id=f"anti_class_{class_name}_{hash(class_body) % 10000}",
+                pattern_type="class",
+                code_snippet=pattern_code,
+                context=f"Anti-pattern Class: {class_name}",
+                quality_score=quality_score,
+                usage_count=1,
+                success_rate=0.0,
+                learned_from=[task_id],
+                security_score=pattern_security_score,
+                is_anti_pattern=True,
+                failure_reasons=failure_reasons.copy(),
+                rejection_count=0,
+                metadata={"vulnerabilities": pattern_vulnerabilities},
+            )
+            anti_patterns.append(anti_pattern)
+
+        return anti_patterns
+
+    def _extract_import_anti_patterns(
+        self,
+        code: str,
+        quality_score: float,
+        task_id: str,
+        failure_reasons: list[str],
+    ) -> list[CodePattern]:
+        """Extract import anti-patterns."""
+        anti_patterns = []
+
+        # Match import statements
+        import_pattern = r"^(?:from\s+[\w.]+|import\s+[\w.,\s]+)"
+        matches = re.finditer(import_pattern, code, re.MULTILINE)
+
+        imports = []
+        for match in matches:
+            imports.append(match.group(0).strip())
+
+        if imports:
+            # Get security score for this pattern
+            pattern_code = "\n".join(imports[:10])
+            pattern_security = self.security_scanner.scan_code(code=pattern_code)
+            pattern_security_score = pattern_security["security_score"]
+            pattern_vulnerabilities = pattern_security["vulnerabilities"]
+
+            anti_pattern = CodePattern(
+                pattern_id=f"anti_imports_{hash(''.join(imports)) % 10000}",
+                pattern_type="import",
+                code_snippet=pattern_code,
+                context="Anti-pattern Import statements",
+                quality_score=quality_score,
+                usage_count=1,
+                success_rate=0.0,
+                learned_from=[task_id],
+                security_score=pattern_security_score,
+                is_anti_pattern=True,
+                failure_reasons=failure_reasons.copy(),
+                rejection_count=0,
+                metadata={"vulnerabilities": pattern_vulnerabilities},
+            )
+            anti_patterns.append(anti_pattern)
+
+        return anti_patterns
+
+    def _extract_structural_anti_patterns(
+        self,
+        code: str,
+        quality_score: float,
+        task_id: str,
+        failure_reasons: list[str],
+    ) -> list[CodePattern]:
+        """Extract structural anti-patterns."""
+        anti_patterns = []
+
+        # Match decorators
+        decorator_pattern = r"@\w+(?:\([^)]*\))?"
+        decorators = re.findall(decorator_pattern, code)
+
+        if decorators:
+            # Get security score for this pattern
+            pattern_code = "\n".join(decorators[:5])
+            pattern_security = self.security_scanner.scan_code(code=pattern_code)
+            pattern_security_score = pattern_security["security_score"]
+            pattern_vulnerabilities = pattern_security["vulnerabilities"]
+
+            anti_pattern = CodePattern(
+                pattern_id=f"anti_decorators_{hash(''.join(decorators)) % 10000}",
+                pattern_type="structure",
+                code_snippet=pattern_code,
+                context="Anti-pattern Decorators",
+                quality_score=quality_score,
+                usage_count=1,
+                success_rate=0.0,
+                learned_from=[task_id],
+                security_score=pattern_security_score,
+                is_anti_pattern=True,
+                failure_reasons=failure_reasons.copy(),
+                rejection_count=0,
+                metadata={"vulnerabilities": pattern_vulnerabilities},
+            )
+            anti_patterns.append(anti_pattern)
+
+        return anti_patterns
+
+    def extract_from_failure(
+        self,
+        code: str,
+        task_id: str,
+        failure_reasons: list[str],
+        quality_score: float = 0.0,
+    ) -> list[CodePattern]:
+        """
+        Extract anti-patterns from a failed task.
+
+        Args:
+            code: Source code from failed task
+            task_id: Task identifier
+            failure_reasons: List of failure reasons
+            quality_score: Quality score (default: 0.0 for failures)
+
+        Returns:
+            List of extracted anti-patterns
+        """
+        return self.extract_anti_patterns(
+            code=code,
+            quality_score=quality_score,
+            task_id=task_id,
+            failure_reasons=failure_reasons,
+        )
+
+    def extract_from_rejection(
+        self,
+        code: str,
+        task_id: str,
+        rejection_reason: str,
+        quality_score: float = 0.5,
+    ) -> list[CodePattern]:
+        """
+        Extract anti-patterns from user rejection.
+
+        Args:
+            code: Rejected code
+            task_id: Task identifier
+            rejection_reason: Reason for rejection
+            quality_score: Quality score
+
+        Returns:
+            List of extracted anti-patterns
+        """
+        return self.extract_anti_patterns(
+            code=code,
+            quality_score=quality_score,
+            task_id=task_id,
+            failure_reasons=[f"User rejection: {rejection_reason}"],
+        )
+
+    def get_anti_patterns_for_context(
+        self, context: str, pattern_type: str | None = None, limit: int = 5
+    ) -> list[CodePattern]:
+        """
+        Get anti-patterns to avoid for a context.
+
+        Args:
+            context: Context string
+            pattern_type: Optional pattern type filter
+            limit: Maximum results
+
+        Returns:
+            List of anti-patterns to avoid
+        """
+        candidates = [p for p in self.anti_patterns.values() if p.is_anti_pattern]
+
+        if pattern_type:
+            candidates = [p for p in candidates if p.pattern_type == pattern_type]
+
+        # Sort by rejection count and quality (low quality = more important to avoid)
+        candidates.sort(
+            key=lambda p: (p.rejection_count, -p.quality_score), reverse=True
+        )
+
+        return candidates[:limit]
+
+
+class PromptOptimizer:
+    """Optimizes agent prompts based on outcomes."""
+
+    def __init__(self, hardware_profile: HardwareProfile):
+        """
+        Initialize prompt optimizer.
+
+        Args:
+            hardware_profile: Hardware profile for optimization
+        """
+        self.hardware_profile = hardware_profile
+        self.variants: dict[str, PromptVariant] = {}
+        self.base_prompt: str | None = None
+
+    def create_variant(
+        self,
+        base_prompt: str,
+        modifications: list[str],
+        variant_id: str | None = None,
+    ) -> PromptVariant:
+        """
+        Create a prompt variant for A/B testing.
+
+        Args:
+            base_prompt: Base prompt template
+            modifications: List of modifications to apply
+            variant_id: Optional variant identifier
+
+        Returns:
+            PromptVariant instance
+        """
+        if variant_id is None:
+            variant_id = f"variant_{hash(''.join(modifications)) % 10000}"
+
+        # Apply modifications (simplified - in production would be more sophisticated)
+        modified_prompt = base_prompt
+        for mod in modifications:
+            if mod.startswith("add:"):
+                modified_prompt += f"\n{mod[4:]}"
+            elif mod.startswith("prepend:"):
+                modified_prompt = f"{mod[8:]}\n{modified_prompt}"
+
+        variant = PromptVariant(
+            variant_id=variant_id,
+            prompt_template=modified_prompt,
+            modifications=modifications,
+            test_count=0,
+            success_count=0,
+            average_quality=0.0,
+            created_at=datetime.now(UTC),
+        )
+
+        self.variants[variant_id] = variant
+        return variant
+
+    def record_test_result(self, variant_id: str, success: bool, quality_score: float):
+        """
+        Record A/B test result for a variant.
+
+        Args:
+            variant_id: Variant identifier
+            success: Whether test succeeded
+            quality_score: Quality score
+        """
+        if variant_id not in self.variants:
+            logger.warning(f"Variant {variant_id} not found")
+            return
+
+        variant = self.variants[variant_id]
+        variant.test_count += 1
+        if success:
+            variant.success_count += 1
+
+        # Update average quality (exponential moving average)
+        alpha = 0.1
+        variant.average_quality = (
+            alpha * quality_score + (1 - alpha) * variant.average_quality
+        )
+        variant.last_tested = datetime.now(UTC)
+
+    def get_best_variant(self, min_tests: int = 5) -> PromptVariant | None:
+        """
+        Get the best performing variant.
+
+        Args:
+            min_tests: Minimum test count to consider
+
+        Returns:
+            Best PromptVariant if found, None otherwise
+        """
+        candidates = [v for v in self.variants.values() if v.test_count >= min_tests]
+
+        if not candidates:
+            return None
+
+        # Sort by success rate and average quality
+        candidates.sort(
+            key=lambda v: (v.success_count / v.test_count, v.average_quality),
+            reverse=True,
+        )
+
+        return candidates[0]
+
+    def optimize_for_hardware(self, prompt: str) -> str:
+        """
+        Optimize prompt for hardware profile.
+
+        Args:
+            prompt: Original prompt
+
+        Returns:
+            Optimized prompt
+        """
+        # Workstation-like: return full prompt (hardware taxonomy removed)
+        return prompt
+
+
+class NegativeFeedbackHandler:
+    """Handles negative feedback, rejections, and corrections."""
+
+    def __init__(self, anti_pattern_extractor: AntiPatternExtractor | None = None):
+        """
+        Initialize negative feedback handler.
+
+        Args:
+            anti_pattern_extractor: Optional anti-pattern extractor instance
+        """
+        self.anti_pattern_extractor = anti_pattern_extractor or AntiPatternExtractor()
+        self.rejections: list[dict[str, Any]] = []
+        self.corrections: list[dict[str, Any]] = []
+
+    def record_rejection(
+        self,
+        code: str,
+        task_id: str,
+        reason: str,
+        quality_score: float = 0.5,
+    ) -> list[CodePattern]:
+        """
+        Record user rejection and extract anti-patterns.
+
+        Args:
+            code: Rejected code
+            task_id: Task identifier
+            reason: Reason for rejection
+            quality_score: Quality score
+
+        Returns:
+            List of extracted anti-patterns
+        """
+        rejection = {
+            "task_id": task_id,
+            "reason": reason,
+            "quality_score": quality_score,
+            "timestamp": datetime.now(UTC).isoformat(),
+        }
+        self.rejections.append(rejection)
+
+        # Extract anti-patterns from rejected code
+        anti_patterns = self.anti_pattern_extractor.extract_from_rejection(
+            code=code,
+            task_id=task_id,
+            rejection_reason=reason,
+            quality_score=quality_score,
+        )
+
+        # Update rejection counts for existing anti-patterns
+        for anti_pattern in anti_patterns:
+            if anti_pattern.pattern_id in self.anti_pattern_extractor.anti_patterns:
+                existing = self.anti_pattern_extractor.anti_patterns[
+                    anti_pattern.pattern_id
+                ]
+                existing.rejection_count += 1
+                if reason not in existing.failure_reasons:
+                    existing.failure_reasons.append(reason)
+            else:
+                # New anti-pattern from rejection should have rejection_count=1
+                anti_pattern.rejection_count = 1
+                self.anti_pattern_extractor.anti_patterns[
+                    anti_pattern.pattern_id
+                ] = anti_pattern
+
+        return anti_patterns
+
+    def record_correction(
+        self,
+        original_code: str,
+        corrected_code: str,
+        task_id: str,
+        correction_reason: str,
+    ) -> tuple[list[CodePattern], list[CodePattern]]:
+        """
+        Record user correction and extract both anti-patterns and patterns.
+
+        Args:
+            original_code: Original (incorrect) code
+            corrected_code: Corrected code
+            task_id: Task identifier
+            correction_reason: Reason for correction
+
+        Returns:
+            Tuple of (anti-patterns from original, patterns from corrected)
+        """
+        correction = {
+            "task_id": task_id,
+            "reason": correction_reason,
+            "timestamp": datetime.now(UTC).isoformat(),
+        }
+        self.corrections.append(correction)
+
+        # Extract anti-patterns from original code
+        anti_patterns = self.anti_pattern_extractor.extract_from_rejection(
+            code=original_code,
+            task_id=task_id,
+            rejection_reason=f"Correction: {correction_reason}",
+            quality_score=0.3,  # Low quality for incorrect code
+        )
+
+        # Extract patterns from corrected code (using PatternExtractor logic)
+        # Note: This would require access to PatternExtractor, but for now
+        # we'll just return the anti-patterns
+        return anti_patterns, []
+
+    def extract_anti_patterns_from_feedback(
+        self, code: str, task_id: str, feedback: str
+    ) -> list[CodePattern]:
+        """
+        Extract anti-patterns from feedback text.
+
+        Args:
+            code: Code that received feedback
+            task_id: Task identifier
+            feedback: Feedback text
+
+        Returns:
+            List of extracted anti-patterns
+        """
+        return self.anti_pattern_extractor.extract_from_rejection(
+            code=code,
+            task_id=task_id,
+            rejection_reason=feedback,
+            quality_score=0.4,
+        )
+
+    def get_anti_patterns_for_context(
+        self, context: str, pattern_type: str | None = None, limit: int = 5
+    ) -> list[CodePattern]:
+        """
+        Get anti-patterns to avoid for a context.
+
+        Args:
+            context: Context string
+            pattern_type: Optional pattern type filter
+            limit: Maximum results
+
+        Returns:
+            List of anti-patterns to avoid
+        """
+        return self.anti_pattern_extractor.get_anti_patterns_for_context(
+            context=context, pattern_type=pattern_type, limit=limit
+        )
+
+
+class FailureModeAnalyzer:
+    """Analyzes failure patterns and categorizes failures."""
+
+    def __init__(self):
+        """Initialize failure mode analyzer."""
+        self.failure_modes: dict[str, dict[str, Any]] = {}
+
+    def analyze_failure(
+        self,
+        code: str,
+        task_id: str,
+        failure_reasons: list[str],
+        quality_scores: dict[str, float] | None = None,
+    ) -> dict[str, Any]:
+        """
+        Analyze a single failure.
+
+        Args:
+            code: Failed code
+            task_id: Task identifier
+            failure_reasons: List of failure reasons
+            quality_scores: Optional quality scores
+
+        Returns:
+            Analysis result with failure mode and suggestions
+        """
+        failure_mode = self.identify_failure_mode(failure_reasons, quality_scores)
+
+        # Track failure mode statistics
+        if failure_mode not in self.failure_modes:
+            self.failure_modes[failure_mode] = {
+                "count": 0,
+                "reasons": [],
+                "task_ids": [],
+            }
+
+        self.failure_modes[failure_mode]["count"] += 1
+        self.failure_modes[failure_mode]["reasons"].extend(failure_reasons)
+        self.failure_modes[failure_mode]["task_ids"].append(task_id)
+
+        # Generate prevention suggestions
+        suggestions = self.suggest_prevention(failure_mode, failure_reasons)
+
+        return {
+            "failure_mode": failure_mode,
+            "failure_reasons": failure_reasons,
+            "suggestions": suggestions,
+            "task_id": task_id,
+        }
+
+    def identify_failure_mode(
+        self,
+        failure_reasons: list[str],
+        quality_scores: dict[str, float] | None = None,
+    ) -> str:
+        """
+        Categorize failure into a failure mode.
+
+        Args:
+            failure_reasons: List of failure reasons
+            quality_scores: Optional quality scores
+
+        Returns:
+            Failure mode category
+        """
+        reasons_str = " ".join(failure_reasons).lower()
+
+        # Categorize based on keywords and quality scores
+        if any(
+            keyword in reasons_str
+            for keyword in ["syntax", "parse", "indentation", "syntaxerror"]
+        ):
+            return "syntax_error"
+        elif any(
+            keyword in reasons_str
+            for keyword in ["security", "vulnerability", "insecure", "bandit"]
+        ):
+            return "security_issue"
+        elif any(
+            keyword in reasons_str
+            for keyword in ["timeout", "slow", "performance", "efficiency"]
+        ):
+            return "performance_issue"
+        elif any(
+            keyword in reasons_str
+            for keyword in ["logic", "incorrect", "wrong", "bug", "error"]
+        ):
+            return "logic_error"
+        elif quality_scores:
+            # Check quality scores
+            if quality_scores.get("security_score", 10.0) < 5.0:
+                return "security_issue"
+            elif quality_scores.get("complexity_score", 0.0) > 8.0:
+                return "complexity_issue"
+            elif quality_scores.get("maintainability_score", 10.0) < 5.0:
+                return "maintainability_issue"
+            else:
+                return "quality_issue"
+        else:
+            return "unknown_failure"
+
+    def get_common_failure_modes(self, limit: int = 5) -> list[dict[str, Any]]:
+        """
+        Get most common failure modes.
+
+        Args:
+            limit: Maximum results
+
+        Returns:
+            List of failure mode statistics
+        """
+        modes = sorted(
+            self.failure_modes.items(),
+            key=lambda x: x[1]["count"],
+            reverse=True,
+        )
+        return [
+            {
+                "mode": mode,
+                "count": stats["count"],
+                "reasons": list(set(stats["reasons"]))[:5],  # Unique reasons, limit 5
+            }
+            for mode, stats in modes[:limit]
+        ]
+
+    def suggest_prevention(
+        self, failure_mode: str, failure_reasons: list[str]
+    ) -> list[str]:
+        """
+        Suggest how to prevent this type of failure.
+
+        Args:
+            failure_mode: Failure mode category
+            failure_reasons: List of failure reasons
+
+        Returns:
+            List of prevention suggestions
+        """
+        suggestions = []
+
+        if failure_mode == "syntax_error":
+            suggestions.append(
+                "Use syntax checking tools (e.g., Ruff, pylint) before code execution"
+            )
+            suggestions.append("Review Python syntax rules and indentation")
+        elif failure_mode == "security_issue":
+            suggestions.append("Run security scanning (Bandit) before learning patterns")
+            suggestions.append("Review security best practices for the language")
+            suggestions.append("Avoid insecure patterns (eval, exec, shell=True)")
+        elif failure_mode == "performance_issue":
+            suggestions.append("Profile code to identify bottlenecks")
+            suggestions.append("Review algorithm complexity and optimization opportunities")
+            suggestions.append("Consider caching or lazy evaluation")
+        elif failure_mode == "logic_error":
+            suggestions.append("Add unit tests to catch logic errors early")
+            suggestions.append("Use type checking (mypy) to catch type-related issues")
+            suggestions.append("Review code logic and edge cases")
+        elif failure_mode == "complexity_issue":
+            suggestions.append("Refactor complex code into smaller functions")
+            suggestions.append("Reduce nesting depth and cyclomatic complexity")
+            suggestions.append("Use design patterns to simplify code structure")
+        elif failure_mode == "maintainability_issue":
+            suggestions.append("Improve code documentation and naming")
+            suggestions.append("Follow consistent coding style")
+            suggestions.append("Reduce code duplication")
+        else:
+            suggestions.append("Review failure reasons and improve code quality")
+            suggestions.append("Add more comprehensive testing")
+
+        return suggestions
+
+
+class FeedbackAnalyzer:
+    """Analyzes feedback from code scoring system and user input."""
+
+    def __init__(self):
+        """Initialize feedback analyzer."""
+        self.feedback_history: list[dict[str, Any]] = []
+
+    def analyze_code_scores(
+        self, scores: dict[str, float], threshold: float = 0.7
+    ) -> dict[str, Any]:
+        """
+        Analyze code scoring results.
+
+        Args:
+            scores: Dictionary of metric scores
+            threshold: Quality threshold
+
+        Returns:
+            Analysis results
+        """
+        overall_score = scores.get("overall_score", 0.0)
+        metrics_obj: object = scores.get("metrics", {})
+        metrics: dict[str, float] = {}
+        if isinstance(metrics_obj, dict):
+            for metric, score in metrics_obj.items():
+                if isinstance(score, (int, float)):
+                    metrics[str(metric)] = float(score)
+
+        # Identify weak areas
+        weak_areas = [metric for metric, score in metrics.items() if score < threshold]
+
+        # Calculate improvement potential
+        improvement_potential = {}
+        for metric, score in metrics.items():
+            if score < threshold:
+                improvement_potential[metric] = threshold - score
+
+        analysis = {
+            "overall_score": overall_score,
+            "weak_areas": weak_areas,
+            "improvement_potential": improvement_potential,
+            "meets_threshold": overall_score >= threshold,
+            "timestamp": datetime.now(UTC).isoformat(),
+        }
+
+        self.feedback_history.append(analysis)
+        return analysis
+
+    def correlate_prompt_changes(
+        self, prompt_variants: list[str], quality_scores: list[float]
+    ) -> dict[str, float]:
+        """
+        Correlate prompt changes with quality improvements.
+
+        Args:
+            prompt_variants: List of prompt variant identifiers
+            quality_scores: Corresponding quality scores
+
+        Returns:
+            Dictionary mapping variants to quality scores
+        """
+        correlations = {}
+        for variant, score in zip(prompt_variants, quality_scores, strict=False):
+            correlations[variant] = score
+
+        return correlations
+
+    def get_improvement_suggestions(self, analysis: dict[str, Any]) -> list[str]:
+        """
+        Get improvement suggestions based on analysis.
+
+        Args:
+            analysis: Analysis results
+
+        Returns:
+            List of improvement suggestions
+        """
+        suggestions = []
+
+        weak_areas = analysis.get("weak_areas", [])
+        improvement_potential = analysis.get("improvement_potential", {})
+
+        for area in weak_areas:
+            potential = improvement_potential.get(area, 0.0)
+            if potential > 0.1:
+                suggestions.append(
+                    f"Focus on improving {area} (potential: {potential:.2f})"
+                )
+
+        if not suggestions:
+            suggestions.append("Code quality is good, maintain current patterns")
+
+        return suggestions
+
+
+class AgentLearner:
+    """
+    Core learning engine for agents.
+
+    Integrates pattern extraction, prompt optimization, and feedback analysis.
+    """
+
+    def __init__(
+        self,
+        capability_registry: CapabilityRegistry,
+        expert_registry: Any,  # ExpertRegistry (required)
+        memory_system: TaskMemorySystem | None = None,
+        hardware_profile: HardwareProfile | None = None,
+    ):
+        """
+        Initialize agent learner.
+
+        Args:
+            capability_registry: Capability registry
+            expert_registry: Expert registry for best practices consultation (required)
+            memory_system: Optional task memory system
+            hardware_profile: Hardware profile (auto-detected if None)
+        """
+        if hardware_profile is None:
+            profiler = HardwareProfiler()
+            hardware_profile = profiler.detect_profile()
+
+        self.capability_registry = capability_registry
+        self.memory_system = memory_system
+        self.hardware_profile = hardware_profile
+        self.learning_intensity = self._get_learning_intensity()
+
+        # Initialize security scanner
+        security_scanner = SecurityScanner()
+
+        self.pattern_extractor = PatternExtractor(security_scanner=security_scanner)
+        self.anti_pattern_extractor = AntiPatternExtractor(
+            security_scanner=security_scanner
+        )
+        self.negative_feedback_handler = NegativeFeedbackHandler(
+            anti_pattern_extractor=self.anti_pattern_extractor
+        )
+        self.failure_mode_analyzer = FailureModeAnalyzer()
+        self.prompt_optimizer = PromptOptimizer(hardware_profile)
+        self.feedback_analyzer = FeedbackAnalyzer()
+
+        # Initialize explainability components
+        self.decision_logger = DecisionReasoningLogger()
+        self.pattern_explainer = PatternSelectionExplainer()
+        self.impact_reporter = LearningImpactReporter()
+
+        # Initialize meta-learning components
+        self.effectiveness_tracker = LearningEffectivenessTracker()
+        self.self_assessor = LearningSelfAssessor()
+        self.adaptive_rate = AdaptiveLearningRate()
+        self.strategy_selector = LearningStrategySelector()
+        self.current_strategy = LearningStrategy.BALANCED
+
+        # Initialize decision engine (always required) with explainability
+        best_practice_consultant = BestPracticeConsultant(expert_registry)
+        confidence_calculator = LearningConfidenceCalculator()
+        self.decision_engine = LearningDecisionEngine(
+            capability_registry=self.capability_registry,
+            best_practice_consultant=best_practice_consultant,
+            confidence_calculator=confidence_calculator,
+            decision_logger=self.decision_logger,
+        )
+
+        # Initialize dashboard
+        self.dashboard = LearningDashboard(
+            capability_registry=self.capability_registry,
+            pattern_extractor=self.pattern_extractor,
+            anti_pattern_extractor=self.anti_pattern_extractor,
+            decision_logger=self.decision_logger,
+            impact_reporter=self.impact_reporter,
+        )
+
+    def _get_learning_intensity(self) -> LearningIntensity:
+        """Get learning intensity based on hardware."""
+        return self.capability_registry.get_learning_intensity()
+
+    async def learn_from_task(
+        self,
+        capability_id: str,
+        task_id: str,
+        code: str | None = None,
+        quality_scores: dict[str, float] | None = None,
+        success: bool = True,
+        duration: float = 0.0,
+    ) -> dict[str, Any]:
+        """
+        Learn from a completed task.
+
+        Args:
+            capability_id: Capability identifier
+            task_id: Task identifier
+            code: Optional source code
+            quality_scores: Optional code scoring results
+            success: Whether task succeeded
+            duration: Task duration
+
+        Returns:
+            Learning results
+        """
+        results: dict[str, Any] = {
+            "patterns_extracted": 0,
+            "anti_patterns_extracted": 0,
+            "prompt_optimized": False,
+            "feedback_analyzed": False,
+            "security_checked": False,
+            "security_score": 0.0,
+            "security_vulnerabilities": [],
+            "failure_analyzed": False,
+        }
+        patterns: list[CodePattern] = []
+        anti_patterns: list[CodePattern] = []
+
+        # Get before metrics for tracking
+        metric = self.capability_registry.get_capability(capability_id)
+        before_metrics = {
+            "quality_score": metric.quality_score if metric else 0.0,
+            "success_rate": metric.success_rate if metric else 0.0,
+            "usage_count": metric.usage_count if metric else 0,
+        }
+
+        # Extract quality score
+        quality_score = 0.5
+        if quality_scores:
+            quality_score = (
+                quality_scores.get("overall_score", 50.0) / 100.0
+            )  # Normalize to 0-1 (assumes 0-100 scale)
+
+        # Security check before learning
+        security_scanner = SecurityScanner()
+        security_result = None
+        if code:
+            security_result = security_scanner.scan_code(code=code)
+            results["security_checked"] = True
+            results["security_score"] = security_result["security_score"]
+            results["security_vulnerabilities"] = security_result["vulnerabilities"]
+
+            # Skip pattern extraction if security score is too low
+            # But continue to extract anti-patterns from vulnerable code
+            if not security_scanner.is_safe_for_learning(
+                code=code, threshold=self.pattern_extractor.security_threshold
+            ):
+                logger.debug(
+                    f"Skipping pattern extraction for task {task_id}: "
+                    f"security score {security_result['security_score']:.2f} "
+                    f"below threshold {self.pattern_extractor.security_threshold:.2f}"
+                )
+                # Still update capability metrics even if we skip pattern extraction
+                self.capability_registry.update_capability_metrics(
+                    capability_id=capability_id,
+                    success=success,
+                    duration=duration,
+                    quality_score=quality_score,
+                )
+                # Don't return early - continue to extract anti-patterns from vulnerable code
+                # The anti-pattern extraction code below will handle low-quality/vulnerable code
+
+        # Update capability metrics
+        self.capability_registry.update_capability_metrics(
+            capability_id=capability_id,
+            success=success,
+            duration=duration,
+            quality_score=quality_score,
+        )
+
+        # Extract patterns if code provided and quality is good
+        # Use decision engine for adaptive threshold, but ensure high-quality code is always extracted
+        should_extract_patterns = False
+        if code and self.learning_intensity != LearningIntensity.LOW:
+            # Base threshold: extract if quality >= 0.7 (normalized 0-1 scale)
+            base_threshold = 0.7
+            
+            # Use decision engine to get adaptive threshold (if available)
+            metric = self.capability_registry.get_capability(capability_id)
+            learned_data = {
+                "usage_count": metric.usage_count if metric else 0,
+                "success_rate": metric.success_rate if metric else 0.0,
+                "quality_score": quality_score,
+                "value": quality_score,  # The threshold value we're considering
+                "context_relevance": 1.0,
+            }
+            context = {
+                "hardware_profile": (
+                    self.hardware_profile.profile_type.value
+                    if hasattr(self.hardware_profile, "profile_type")
+                    else "unknown"
+                ),
+                "learning_intensity": self.learning_intensity.value,
+                "task_id": task_id,
+                "capability_id": capability_id,
+            }
+
+            # Use decision engine (always available)
+            decision = await self.decision_engine.make_decision(
+                decision_type="pattern_extraction_threshold",
+                learned_data=learned_data,
+                context=context,
+                default_value=base_threshold,
+            )
+
+            # Determine threshold value from decision result
+            threshold_value = base_threshold  # Default
+            if decision.result.value is not None:
+                if isinstance(decision.result.value, (int, float)):
+                    threshold_value = float(decision.result.value)
+                elif isinstance(decision.result.value, str):
+                    # Try to extract numeric value from string
+                    import re
+                    match = re.search(r'(\d+\.?\d*)', str(decision.result.value))
+                    if match:
+                        threshold_value = float(match.group(1))
+            
+            # Extract patterns if quality_score meets threshold
+            # For new capabilities (low learned_confidence), still extract if quality is high
+            should_extract_patterns = quality_score >= threshold_value
+            
+            # Fallback: if quality is very high (>= 0.8), always extract regardless of decision
+            if quality_score >= 0.8:
+                should_extract_patterns = True
+            
+            logger.debug(
+                f"Pattern extraction decision: quality_score={quality_score:.3f}, "
+                f"threshold={threshold_value:.3f}, should_extract={should_extract_patterns}"
+            )
+
+        if should_extract_patterns and code is not None:
+            logger.debug(f"Attempting to extract patterns from code (length={len(code)})")
+            patterns = self.pattern_extractor.extract_patterns(
+                code=code, quality_score=quality_score, task_id=task_id
+            )
+            logger.debug(f"Extracted {len(patterns)} patterns from code")
+
+            # Store patterns
+            for pattern in patterns:
+                if pattern.pattern_id not in self.pattern_extractor.patterns:
+                    self.pattern_extractor.patterns[pattern.pattern_id] = pattern
+                else:
+                    # Update existing pattern
+                    existing = self.pattern_extractor.patterns[pattern.pattern_id]
+                    existing.usage_count += 1
+                    existing.learned_from.append(task_id)
+
+            results["patterns_extracted"] = len(patterns)
+
+        # Handle failures: extract anti-patterns
+        if not success and code:
+            failure_reasons = [f"Task {task_id} failed"]
+            if quality_scores:
+                # Add quality-based failure reasons
+                if quality_scores.get("security_score", 10.0) < 5.0:
+                    failure_reasons.append("Low security score")
+                if quality_scores.get("overall_score", 100.0) < 50.0:
+                    failure_reasons.append("Low overall quality score")
+
+            # Extract anti-patterns from failed code
+            anti_patterns = self.anti_pattern_extractor.extract_from_failure(
+                code=code,
+                task_id=task_id,
+                failure_reasons=failure_reasons,
+                quality_score=quality_score,
+            )
+
+            # Store anti-patterns
+            for anti_pattern in anti_patterns:
+                if anti_pattern.pattern_id not in self.anti_pattern_extractor.anti_patterns:
+                    self.anti_pattern_extractor.anti_patterns[
+                        anti_pattern.pattern_id
+                    ] = anti_pattern
+                else:
+                    # Update existing anti-pattern
+                    existing = self.anti_pattern_extractor.anti_patterns[
+                        anti_pattern.pattern_id
+                    ]
+                    existing.usage_count += 1
+                    existing.learned_from.append(task_id)
+
+            results["anti_patterns_extracted"] = len(anti_patterns)
+
+            # Analyze failure mode
+            failure_analysis = self.failure_mode_analyzer.analyze_failure(
+                code=code,
+                task_id=task_id,
+                failure_reasons=failure_reasons,
+                quality_scores=quality_scores,
+            )
+            results["failure_analyzed"] = True
+            results["failure_analysis"] = failure_analysis
+
+        # Also extract anti-patterns from low-quality code (even if success=True)
+        logger.debug(f"Anti-pattern extraction check: code={code is not None}, quality_score={quality_score}, threshold={self.anti_pattern_extractor.max_quality_threshold}, success={success}")
+        if code and quality_score < self.anti_pattern_extractor.max_quality_threshold:
+            failure_reasons = [f"Low quality score: {quality_score:.2f}"]
+            anti_patterns = self.anti_pattern_extractor.extract_anti_patterns(
+                code=code,
+                quality_score=quality_score,
+                task_id=task_id,
+                failure_reasons=failure_reasons,
+            )
+
+            # Store anti-patterns
+            for anti_pattern in anti_patterns:
+                if anti_pattern.pattern_id not in self.anti_pattern_extractor.anti_patterns:
+                    self.anti_pattern_extractor.anti_patterns[
+                        anti_pattern.pattern_id
+                    ] = anti_pattern
+
+            results["anti_patterns_extracted"] = len(anti_patterns)
+
+        # Analyze feedback if scores provided
+        if quality_scores:
+            analysis = self.feedback_analyzer.analyze_code_scores(quality_scores)
+            results["feedback_analyzed"] = True
+            results["feedback_analysis"] = analysis
+
+            # Check if improvement is needed
+            if not analysis.get("meets_threshold", False):
+                suggestions = self.feedback_analyzer.get_improvement_suggestions(
+                    analysis
+                )
+                results["improvement_suggestions"] = suggestions
+
+        # Store in memory system if available
+        if self.memory_system and success:
+            outcome = TaskOutcome.SUCCESS if success else TaskOutcome.FAILURE
+            self.memory_system.store_memory(
+                task_id=task_id,
+                agent_id="unknown",  # Would be provided by agent
+                command=capability_id,
+                outcome=outcome,
+                quality_score=quality_score,
+                patterns_used=[p.pattern_id for p in patterns] if code else [],
+            )
+
+        # Track learning effectiveness
+        metric_after = self.capability_registry.get_capability(capability_id)
+        after_metrics = {
+            "quality_score": metric_after.quality_score if metric_after else 0.0,
+            "success_rate": metric_after.success_rate if metric_after else 0.0,
+            "usage_count": metric_after.usage_count if metric_after else 0,
+        }
+
+        # Track effectiveness
+        session = self.effectiveness_tracker.track_effectiveness(
+            capability_id=capability_id,
+            before_metrics=before_metrics,
+            after_metrics=after_metrics,
+            strategies_used=[self.current_strategy.value],
+        )
+
+        # Adjust learning rate based on effectiveness
+        rate_adjustment = self.adaptive_rate.adjust_learning_intensity(
+            session.effectiveness_score
+        )
+        results["learning_rate_adjustment"] = rate_adjustment
+
+        # Generate impact report
+        impact_report = self.impact_reporter.generate_impact_report(
+            capability_id=capability_id,
+            before_metrics=before_metrics,
+            after_metrics=after_metrics,
+            learning_session_id=session.session_id,
+        )
+        results["learning_impact"] = impact_report
+
+        return results
+
+    async def learn_from_rejection(
+        self,
+        capability_id: str,
+        task_id: str,
+        code: str,
+        rejection_reason: str,
+        quality_score: float = 0.5,
+    ) -> dict[str, Any]:
+        """
+        Learn from user rejection.
+
+        Args:
+            capability_id: Capability identifier
+            task_id: Task identifier
+            code: Rejected code
+            rejection_reason: Reason for rejection
+            quality_score: Quality score
+
+        Returns:
+            Learning results
+        """
+        results: dict[str, Any] = {
+            "anti_patterns_extracted": 0,
+            "rejection_recorded": False,
+        }
+
+        # Record rejection and extract anti-patterns
+        anti_patterns = self.negative_feedback_handler.record_rejection(
+            code=code,
+            task_id=task_id,
+            reason=rejection_reason,
+            quality_score=quality_score,
+        )
+
+        results["anti_patterns_extracted"] = len(anti_patterns)
+        results["rejection_recorded"] = True
+
+        # Update capability metrics (rejection counts as failure)
+        self.capability_registry.update_capability_metrics(
+            capability_id=capability_id,
+            success=False,
+            duration=0.0,
+            quality_score=quality_score,
+        )
+
+        return results
+
+    def get_learned_patterns(
+        self,
+        context: str,
+        pattern_type: str | None = None,
+        limit: int = 5,
+        exclude_anti_patterns: bool = True,
+    ) -> list[CodePattern]:
+        """
+        Get learned patterns for a context.
+
+        Args:
+            context: Context string
+            pattern_type: Optional pattern type filter
+            limit: Maximum results
+            exclude_anti_patterns: If True, exclude anti-patterns (default: True)
+
+        Returns:
+            List of relevant patterns
+        """
+        return self.pattern_extractor.get_patterns_for_context(
+            context=context,
+            pattern_type=pattern_type,
+            limit=limit,
+            exclude_anti_patterns=exclude_anti_patterns,
+        )
+
+    def optimize_prompt(self, base_prompt: str, context: str | None = None) -> str:
+        """
+        Get optimized prompt.
+
+        Args:
+            base_prompt: Base prompt
+            context: Optional context
+
+        Returns:
+            Optimized prompt
+        """
+        # Get best variant if available
+        best_variant = self.prompt_optimizer.get_best_variant()
+        if best_variant and best_variant.average_quality > 0.7:
+            return best_variant.prompt_template
+
+        # Otherwise, optimize for hardware
+        return self.prompt_optimizer.optimize_for_hardware(base_prompt)
+
+    def explain_learning(
+        self,
+        capability_id: str,
+        task_id: str | None = None,
+        decision_id: str | None = None,
+    ) -> dict[str, Any]:
+        """
+        Generate explanation for learning process.
+
+        Args:
+            capability_id: Capability identifier
+            task_id: Optional task identifier
+            decision_id: Optional decision identifier
+
+        Returns:
+            Explanation dictionary
+        """
+        explanation: dict[str, Any] = {
+            "capability_id": capability_id,
+            "task_id": task_id,
+        }
+
+        # Get decision explanation if decision_id provided
+        if decision_id:
+            decision_explanation = self.decision_logger.explain_decision(decision_id)
+            if decision_explanation:
+                explanation["decision"] = decision_explanation
+
+        # Get pattern selection explanation
+        patterns = self.get_learned_patterns(
+            context=capability_id, exclude_anti_patterns=True
+        )
+        if patterns:
+            pattern_explanation = self.pattern_explainer.explain_pattern_selection(
+                selected_patterns=patterns, context=capability_id
+            )
+            explanation["pattern_selection"] = pattern_explanation
+
+        # Get decision statistics
+        explanation["decision_statistics"] = self.decision_logger.get_decision_statistics()
+
+        return explanation
+
+    async def optimize_learning(
+        self, capability_id: str | None = None
+    ) -> dict[str, Any]:
+        """
+        Run meta-learning optimization.
+
+        Args:
+            capability_id: Optional filter by capability
+
+        Returns:
+            Optimization report
+        """
+        optimization: dict[str, Any] = {
+            "timestamp": datetime.now(UTC).isoformat(),
+            "capability_id": capability_id,
+        }
+
+        # Assess learning quality
+        pattern_count = len(self.pattern_extractor.patterns)
+        anti_pattern_count = len(self.anti_pattern_extractor.anti_patterns)
+
+        # Calculate average quality and security
+        avg_quality = 0.0
+        avg_security = 0.0
+        if self.pattern_extractor.patterns:
+            total_quality = sum(p.quality_score for p in self.pattern_extractor.patterns.values())
+            total_security = sum(
+                getattr(p, "security_score", 0.0)
+                for p in self.pattern_extractor.patterns.values()
+            )
+            avg_quality = total_quality / len(self.pattern_extractor.patterns)
+            avg_security = total_security / len(self.pattern_extractor.patterns)
+
+        quality_assessment = self.self_assessor.assess_learning_quality(
+            pattern_count=pattern_count,
+            anti_pattern_count=anti_pattern_count,
+            average_quality=avg_quality,
+            average_security=avg_security,
+        )
+        optimization["quality_assessment"] = quality_assessment
+
+        # Identify learning gaps
+        capability_metrics = {}
+        if capability_id:
+            metric = self.capability_registry.get_capability(capability_id)
+            if metric:
+                capability_metrics = {
+                    "success_rate": metric.success_rate,
+                    "quality_score": metric.quality_score,
+                    "usage_count": metric.usage_count,
+                }
+        pattern_stats = {
+            "total_patterns": pattern_count,
+            "average_quality": avg_quality,
+            "average_security": avg_security,
+        }
+        gaps = self.self_assessor.identify_learning_gaps(
+            capability_metrics=capability_metrics,
+            pattern_statistics=pattern_stats,
+        )
+        optimization["learning_gaps"] = gaps
+
+        # Get improvement suggestions
+        suggestions = self.self_assessor.suggest_improvements(quality_assessment)
+        optimization["improvement_suggestions"] = suggestions
+
+        # Select optimal strategy
+        if capability_id:
+            metric = self.capability_registry.get_capability(capability_id)
+            current_effectiveness = metric.success_rate if metric else 0.5
+        else:
+            # Use average effectiveness
+            roi = self.effectiveness_tracker.get_learning_roi(capability_id=capability_id)
+            current_effectiveness = roi.get("average_effectiveness", 0.5)
+
+        hardware_profile_str = (
+            self.hardware_profile.profile_type.value
+            if hasattr(self.hardware_profile, "profile_type")
+            else None
+        )
+        optimal_strategy = self.strategy_selector.select_strategy(
+            capability_id=capability_id or "global",
+            current_effectiveness=current_effectiveness,
+            hardware_profile=hardware_profile_str,
+        )
+        optimization["optimal_strategy"] = optimal_strategy.value
+        optimization["current_strategy"] = self.current_strategy.value
+
+        # Optimize thresholds
+        current_threshold = self.pattern_extractor.min_quality_threshold
+        metric = self.capability_registry.get_capability(capability_id) if capability_id else None
+        success_rate = metric.success_rate if metric else 0.5
+        optimized_threshold = self.adaptive_rate.optimize_thresholds(
+            current_threshold=current_threshold,
+            success_rate=success_rate,
+            quality_score=avg_quality,
+        )
+        optimization["optimized_threshold"] = optimized_threshold
+        optimization["current_threshold"] = current_threshold
+
+        # Update strategy if better one found
+        if optimal_strategy != self.current_strategy:
+            switch_result = self.strategy_selector.switch_strategy(
+                current_strategy=self.current_strategy,
+                new_strategy=optimal_strategy,
+            )
+            if switch_result["switched"]:
+                self.current_strategy = optimal_strategy
+                optimization["strategy_switched"] = True
+                optimization["switch_result"] = switch_result
+            else:
+                optimization["strategy_switched"] = False
+
+        # Get effectiveness metrics
+        effectiveness_metrics = self.effectiveness_tracker.get_learning_roi(
+            capability_id=capability_id
+        )
+        optimization["effectiveness_metrics"] = effectiveness_metrics
+
+        return optimization
+
+    def should_refine_capability(self, capability_id: str) -> bool:
+        """
+        Determine if a capability should be refined.
+
+        Args:
+            capability_id: Capability identifier
+
+        Returns:
+            True if refinement is recommended
+        """
+        metric = self.capability_registry.get_capability(capability_id)
+        if not metric:
+            return False
+
+        # Refine if quality is below threshold and has enough usage
+        return metric.quality_score < 0.7 and metric.usage_count >= 10