agentic-qe 3.7.9 → 3.7.11

package/assets/patterns/adr-051-reasoning-patterns.json

@@ -1,166 +1,166 @@
-{
-  "namespace": "adr-051/reasoning-patterns",
-  "description": "ReasoningBank Integration Patterns from ADR-051",
-  "created": "2026-01-20",
-  "patterns": [
-    {
-      "key": "reasoning-trajectory-tracking",
-      "pattern": "ReasoningBank Trajectory Tracking",
-      "description": "Track decision-making trajectories for learning and replay",
-      "trajectory_structure": {
-        "trajectory_id": "Unique identifier for sequence",
-        "task_description": "Original task prompt",
-        "steps": [
-          {
-            "step_number": 1,
-            "action": "Initial assessment",
-            "result": "Identified key issues",
-            "quality": 0.8
-          },
-          {
-            "step_number": 2,
-            "action": "Proposed solution",
-            "result": "Specific implementation",
-            "quality": 0.9
-          }
-        ],
-        "final_outcome": "Success/Failure with metrics",
-        "completion_time": "Wall-clock execution time"
-      },
-      "tracking_mechanism": {
-        "event_capture": "Record decision at each step",
-        "context_preservation": "Store relevant state",
-        "outcome_recording": "Final result and quality score",
-        "metadata_annotation": "Tags, categories, keywords"
-      },
-      "learning_extraction": {
-        "success_patterns": "Extract what worked well",
-        "failure_analysis": "Understand what failed and why",
-        "performance_correlation": "Link actions to outcomes",
-        "decision_confidence": "Track decision quality signals"
-      },
-      "trajectory_replay": {
-        "similar_task_detection": "Find analogous trajectories",
-        "experience_adaptation": "Adjust for current context",
-        "outcome_prediction": "Forecast success probability",
-        "strategy_transfer": "Apply learned strategies"
-      },
-      "successRate": 0.91,
-      "lastUpdated": "2026-01-20"
-    },
-    {
-      "key": "reasoning-pattern-quality-gates",
-      "pattern": "ReasoningBank Pattern Quality Gates",
-      "description": "Ensure high-quality patterns enter ReasoningBank with quality gates",
-      "quality_criteria": {
-        "success_rate": ">0.85 for pattern approval",
-        "outcome_consistency": "Low variance in results",
-        "generalizability": "Works across multiple contexts",
-        "edge_case_coverage": "Handles variations well",
-        "documentation": "Clear explanation of pattern"
-      },
-      "quality_scoring": {
-        "dimension_1_accuracy": "How often pattern produces correct results",
-        "dimension_2_robustness": "Handles edge cases and variations",
-        "dimension_3_efficiency": "Computational cost vs accuracy tradeoff",
-        "dimension_4_clarity": "How understandable and transferable",
-        "composite_score": "Weighted average of dimensions"
-      },
-      "gate_thresholds": {
-        "bronze_tier": ">0.70 composite score - Limited recommendation",
-        "silver_tier": ">0.80 composite score - Standard recommendation",
-        "gold_tier": ">0.90 composite score - Highly recommended",
-        "platinum_tier": ">0.95 composite score - Approved for critical tasks"
-      },
-      "pattern_validation": [
-        "Run against test suite",
-        "Cross-validation on held-out data",
-        "Edge case stress testing",
-        "Peer review by agents",
-        "Historical performance comparison"
-      ],
-      "rejection_reasons": [
-        "Insufficient success rate",
-        "High variance in outcomes",
-        "Narrow applicability",
-        "Poor documentation",
-        "Security or safety concerns"
-      ],
-      "successRate": 0.94,
-      "lastUpdated": "2026-01-20"
-    },
-    {
-      "key": "reasoning-experience-replay",
-      "pattern": "Experience Replay for Continuous Learning",
-      "description": "Replay past trajectories to reinforce successful patterns and prevent catastrophic forgetting",
-      "replay_mechanism": {
-        "sample_trajectory": "Randomly sample from trajectory buffer",
-        "extract_experience": "Get action-outcome pairs",
-        "analyze_decision": "Understand why that action succeeded",
-        "reinforce_pattern": "Strengthen association in memory",
-        "prevent_forgetting": "Consolidate knowledge with EWC++"
-      },
-      "buffer_management": {
-        "trajectory_buffer_size": "100-1000 trajectories",
-        "prioritized_sampling": "Over-sample high-value trajectories",
-        "diversity_sampling": "Ensure coverage of different task types",
-        "staleness_threshold": "Drop trajectories older than N days",
-        "checkpointing": "Periodically save important trajectories"
-      },
-      "learning_integration": {
-        "offline_replay": "Run during idle time",
-        "online_replay": "Interleave with new experiences",
-        "consolidation": "EWC++ protects old knowledge",
-        "transfer_learning": "Apply patterns to new tasks"
-      },
-      "metrics_tracking": {
-        "replay_frequency": "How often each trajectory replayed",
-        "learning_gain": "Improvement from replay",
-        "forgetting_rate": "How fast knowledge decays",
-        "transfer_success": "Patterns working on new tasks"
-      },
-      "scaling_strategy": "Distributed replay across agent swarm",
-      "successRate": 0.89,
-      "lastUpdated": "2026-01-20"
-    },
-    {
-      "key": "reasoning-cross-agent-sharing",
-      "pattern": "Cross-Agent Knowledge Sharing",
-      "description": "Share learned patterns and trajectories between agents for collective intelligence",
-      "sharing_channels": {
-        "pattern_broadcast": "Share high-quality patterns to all agents",
-        "trajectory_publishing": "Publish successful trajectories",
-        "failure_alerts": "Warn about patterns that failed",
-        "performance_updates": "Share metric improvements"
-      },
-      "knowledge_transfer": {
-        "sender_agent": "Identifies pattern worth sharing",
-        "quality_assessment": "Verify pattern meets quality gates",
-        "packaging": "Package with metadata and context",
-        "distribution": "Broadcast to relevant agent types",
-        "integration": "Receiving agents incorporate pattern"
-      },
-      "recipient_adaptation": {
-        "context_adjustment": "Adapt pattern to local context",
-        "validation": "Test pattern in new environment",
-        "confidence_tracking": "Monitor transfer success rate",
-        "feedback_loop": "Report back on effectiveness"
-      },
-      "knowledge_domains": [
-        "Test generation strategies",
-        "Code analysis patterns",
-        "Error diagnosis techniques",
-        "Optimization approaches",
-        "Security checking methods"
-      ],
-      "conflict_resolution": {
-        "version_mismatch": "Prefer newer patterns with higher success rate",
-        "contradictory_patterns": "Run both and compare empirically",
-        "domain_specificity": "Context-weight patterns by relevance"
-      },
-      "scaling_approach": "Gossip protocol for decentralized propagation",
-      "successRate": 0.87,
-      "lastUpdated": "2026-01-20"
-    }
-  ]
-}
+{
+  "namespace": "adr-051/reasoning-patterns",
+  "description": "ReasoningBank Integration Patterns from ADR-051",
+  "created": "2026-01-20",
+  "patterns": [
+    {
+      "key": "reasoning-trajectory-tracking",
+      "pattern": "ReasoningBank Trajectory Tracking",
+      "description": "Track decision-making trajectories for learning and replay",
+      "trajectory_structure": {
+        "trajectory_id": "Unique identifier for sequence",
+        "task_description": "Original task prompt",
+        "steps": [
+          {
+            "step_number": 1,
+            "action": "Initial assessment",
+            "result": "Identified key issues",
+            "quality": 0.8
+          },
+          {
+            "step_number": 2,
+            "action": "Proposed solution",
+            "result": "Specific implementation",
+            "quality": 0.9
+          }
+        ],
+        "final_outcome": "Success/Failure with metrics",
+        "completion_time": "Wall-clock execution time"
+      },
+      "tracking_mechanism": {
+        "event_capture": "Record decision at each step",
+        "context_preservation": "Store relevant state",
+        "outcome_recording": "Final result and quality score",
+        "metadata_annotation": "Tags, categories, keywords"
+      },
+      "learning_extraction": {
+        "success_patterns": "Extract what worked well",
+        "failure_analysis": "Understand what failed and why",
+        "performance_correlation": "Link actions to outcomes",
+        "decision_confidence": "Track decision quality signals"
+      },
+      "trajectory_replay": {
+        "similar_task_detection": "Find analogous trajectories",
+        "experience_adaptation": "Adjust for current context",
+        "outcome_prediction": "Forecast success probability",
+        "strategy_transfer": "Apply learned strategies"
+      },
+      "successRate": 0.91,
+      "lastUpdated": "2026-01-20"
+    },
+    {
+      "key": "reasoning-pattern-quality-gates",
+      "pattern": "ReasoningBank Pattern Quality Gates",
+      "description": "Ensure high-quality patterns enter ReasoningBank with quality gates",
+      "quality_criteria": {
+        "success_rate": ">0.85 for pattern approval",
+        "outcome_consistency": "Low variance in results",
+        "generalizability": "Works across multiple contexts",
+        "edge_case_coverage": "Handles variations well",
+        "documentation": "Clear explanation of pattern"
+      },
+      "quality_scoring": {
+        "dimension_1_accuracy": "How often pattern produces correct results",
+        "dimension_2_robustness": "Handles edge cases and variations",
+        "dimension_3_efficiency": "Computational cost vs accuracy tradeoff",
+        "dimension_4_clarity": "How understandable and transferable",
+        "composite_score": "Weighted average of dimensions"
+      },
+      "gate_thresholds": {
+        "bronze_tier": ">0.70 composite score - Limited recommendation",
+        "silver_tier": ">0.80 composite score - Standard recommendation",
+        "gold_tier": ">0.90 composite score - Highly recommended",
+        "platinum_tier": ">0.95 composite score - Approved for critical tasks"
+      },
+      "pattern_validation": [
+        "Run against test suite",
+        "Cross-validation on held-out data",
+        "Edge case stress testing",
+        "Peer review by agents",
+        "Historical performance comparison"
+      ],
+      "rejection_reasons": [
+        "Insufficient success rate",
+        "High variance in outcomes",
+        "Narrow applicability",
+        "Poor documentation",
+        "Security or safety concerns"
+      ],
+      "successRate": 0.94,
+      "lastUpdated": "2026-01-20"
+    },
+    {
+      "key": "reasoning-experience-replay",
+      "pattern": "Experience Replay for Continuous Learning",
+      "description": "Replay past trajectories to reinforce successful patterns and prevent catastrophic forgetting",
+      "replay_mechanism": {
+        "sample_trajectory": "Randomly sample from trajectory buffer",
+        "extract_experience": "Get action-outcome pairs",
+        "analyze_decision": "Understand why that action succeeded",
+        "reinforce_pattern": "Strengthen association in memory",
+        "prevent_forgetting": "Consolidate knowledge with EWC++"
+      },
+      "buffer_management": {
+        "trajectory_buffer_size": "100-1000 trajectories",
+        "prioritized_sampling": "Over-sample high-value trajectories",
+        "diversity_sampling": "Ensure coverage of different task types",
+        "staleness_threshold": "Drop trajectories older than N days",
+        "checkpointing": "Periodically save important trajectories"
+      },
+      "learning_integration": {
+        "offline_replay": "Run during idle time",
+        "online_replay": "Interleave with new experiences",
+        "consolidation": "EWC++ protects old knowledge",
+        "transfer_learning": "Apply patterns to new tasks"
+      },
+      "metrics_tracking": {
+        "replay_frequency": "How often each trajectory replayed",
+        "learning_gain": "Improvement from replay",
+        "forgetting_rate": "How fast knowledge decays",
+        "transfer_success": "Patterns working on new tasks"
+      },
+      "scaling_strategy": "Distributed replay across agent swarm",
+      "successRate": 0.89,
+      "lastUpdated": "2026-01-20"
+    },
+    {
+      "key": "reasoning-cross-agent-sharing",
+      "pattern": "Cross-Agent Knowledge Sharing",
+      "description": "Share learned patterns and trajectories between agents for collective intelligence",
+      "sharing_channels": {
+        "pattern_broadcast": "Share high-quality patterns to all agents",
+        "trajectory_publishing": "Publish successful trajectories",
+        "failure_alerts": "Warn about patterns that failed",
+        "performance_updates": "Share metric improvements"
+      },
+      "knowledge_transfer": {
+        "sender_agent": "Identifies pattern worth sharing",
+        "quality_assessment": "Verify pattern meets quality gates",
+        "packaging": "Package with metadata and context",
+        "distribution": "Broadcast to relevant agent types",
+        "integration": "Receiving agents incorporate pattern"
+      },
+      "recipient_adaptation": {
+        "context_adjustment": "Adapt pattern to local context",
+        "validation": "Test pattern in new environment",
+        "confidence_tracking": "Monitor transfer success rate",
+        "feedback_loop": "Report back on effectiveness"
+      },
+      "knowledge_domains": [
+        "Test generation strategies",
+        "Code analysis patterns",
+        "Error diagnosis techniques",
+        "Optimization approaches",
+        "Security checking methods"
+      ],
+      "conflict_resolution": {
+        "version_mismatch": "Prefer newer patterns with higher success rate",
+        "contradictory_patterns": "Run both and compare empirically",
+        "domain_specificity": "Context-weight patterns by relevance"
+      },
+      "scaling_approach": "Gossip protocol for decentralized propagation",
+      "successRate": 0.87,
+      "lastUpdated": "2026-01-20"
+    }
+  ]
+}

package/assets/patterns/adr-051-router-patterns.json

@@ -1,113 +1,113 @@
-{
-  "namespace": "adr-051/router-patterns",
-  "description": "Model Router Integration Patterns from ADR-051",
-  "created": "2026-01-20",
-  "patterns": [
-    {
-      "key": "router-5tier-complexity",
-      "pattern": "5-Tier Model Complexity Scoring System",
-      "description": "Hierarchical model routing based on task complexity and constraints",
-      "tier_hierarchy": {
-        "tier_1_booster": {
-          "cost": "$0",
-          "latency": "<1ms",
-          "use_case": "Simple transforms (var-const, add-types, remove-console)",
-          "eligibility": "Pure functions, no dependencies"
-        },
-        "tier_2_haiku": {
-          "cost": "$0.80/M input tokens",
-          "latency": "~500ms",
-          "use_case": "Bug fixes, simple tasks, low complexity",
-          "complexity_score": "1-3"
-        },
-        "tier_3_sonnet": {
-          "cost": "$3/M input tokens",
-          "latency": "~2-3s",
-          "use_case": "Medium complexity, architectural decisions",
-          "complexity_score": "4-6"
-        },
-        "tier_4_opus": {
-          "cost": "$15/M input tokens",
-          "latency": "~5-8s",
-          "use_case": "Complex reasoning, security-critical",
-          "complexity_score": "7-9"
-        },
-        "tier_5_human": {
-          "cost": "Manual review",
-          "latency": "Variable",
-          "use_case": "Final decisions, critical paths",
-          "complexity_score": "10"
-        }
-      },
-      "scoring_factors": [
-        "Number of files affected",
-        "Code complexity (AST depth)",
-        "Domain sensitivity",
-        "Business impact",
-        "Security implications",
-        "Architectural change magnitude"
-      ],
-      "auto_downgrade": "Reduce tier if budget exceeded or timeout risk",
-      "successRate": 0.91,
-      "lastUpdated": "2026-01-20"
-    },
-    {
-      "key": "router-budget-enforcement",
-      "pattern": "Budget Enforcement with Auto-Downgrade",
-      "description": "Enforce token and cost budgets with intelligent model tier downgrade",
-      "budget_constraints": {
-        "max_tokens_per_task": "100000",
-        "max_cost_per_task": "$0.50",
-        "max_latency_ms": "30000"
-      },
-      "auto_downgrade_logic": {
-        "step_1": "Estimate tokens needed for task with target model",
-        "step_2": "Compare against available budget",
-        "step_3": "If exceeded: downgrade to cheaper model",
-        "step_4": "Recalculate and repeat until fits budget",
-        "step_5": "Escalate to human if no model viable"
-      },
-      "budget_tracking": {
-        "per_task": "Track cumulative tokens/cost",
-        "per_session": "Aggregate across all tasks",
-        "per_day": "Daily quota enforcement",
-        "alerts": "Warn at 80% threshold"
-      },
-      "recovery_strategies": [
-        "Task decomposition into smaller chunks",
-        "Caching partial results to reduce retokenization",
-        "Context pruning for efficiency",
-        "Result streaming to avoid buffering"
-      ],
-      "successRate": 0.88,
-      "lastUpdated": "2026-01-20"
-    },
-    {
-      "key": "router-booster-integration",
-      "pattern": "Model Router to Agent Booster Integration",
-      "description": "Seamless integration between router and booster for optimal task routing",
-      "integration_flow": {
-        "step_1_receive_task": "Router receives task description",
-        "step_2_classify": "Classify as simple transform or complex task",
-        "step_3_check_booster": "Check Agent Booster eligibility criteria",
-        "step_4_route": "If eligible: route to booster; else: route to model tier",
-        "step_5_execute": "Execute via appropriate pathway",
-        "step_6_report": "Report cost, latency, quality metrics"
-      },
-      "eligibility_handshake": {
-        "booster_can_handle": "Query booster eligibility function",
-        "transform_type": "Categorize specific transform",
-        "fallback_model": "Designate fallback if booster fails"
-      },
-      "metrics_collection": {
-        "booster_execution": "Track WASM vs TS fallback ratio",
-        "model_tier_usage": "Distribution across Haiku/Sonnet/Opus",
-        "cost_savings": "Actual vs baseline cost",
-        "quality_correlation": "Task complexity vs outcome quality"
-      },
-      "feedback_loop": "Use execution patterns to refine routing heuristics",
-      "successRate": 0.89,
-      "lastUpdated": "2026-01-20"
-    }
-  ]
-}
+{
+  "namespace": "adr-051/router-patterns",
+  "description": "Model Router Integration Patterns from ADR-051",
+  "created": "2026-01-20",
+  "patterns": [
+    {
+      "key": "router-5tier-complexity",
+      "pattern": "5-Tier Model Complexity Scoring System",
+      "description": "Hierarchical model routing based on task complexity and constraints",
+      "tier_hierarchy": {
+        "tier_1_booster": {
+          "cost": "$0",
+          "latency": "<1ms",
+          "use_case": "Simple transforms (var-const, add-types, remove-console)",
+          "eligibility": "Pure functions, no dependencies"
+        },
+        "tier_2_haiku": {
+          "cost": "$0.80/M input tokens",
+          "latency": "~500ms",
+          "use_case": "Bug fixes, simple tasks, low complexity",
+          "complexity_score": "1-3"
+        },
+        "tier_3_sonnet": {
+          "cost": "$3/M input tokens",
+          "latency": "~2-3s",
+          "use_case": "Medium complexity, architectural decisions",
+          "complexity_score": "4-6"
+        },
+        "tier_4_opus": {
+          "cost": "$15/M input tokens",
+          "latency": "~5-8s",
+          "use_case": "Complex reasoning, security-critical",
+          "complexity_score": "7-9"
+        },
+        "tier_5_human": {
+          "cost": "Manual review",
+          "latency": "Variable",
+          "use_case": "Final decisions, critical paths",
+          "complexity_score": "10"
+        }
+      },
+      "scoring_factors": [
+        "Number of files affected",
+        "Code complexity (AST depth)",
+        "Domain sensitivity",
+        "Business impact",
+        "Security implications",
+        "Architectural change magnitude"
+      ],
+      "auto_downgrade": "Reduce tier if budget exceeded or timeout risk",
+      "successRate": 0.91,
+      "lastUpdated": "2026-01-20"
+    },
+    {
+      "key": "router-budget-enforcement",
+      "pattern": "Budget Enforcement with Auto-Downgrade",
+      "description": "Enforce token and cost budgets with intelligent model tier downgrade",
+      "budget_constraints": {
+        "max_tokens_per_task": "100000",
+        "max_cost_per_task": "$0.50",
+        "max_latency_ms": "30000"
+      },
+      "auto_downgrade_logic": {
+        "step_1": "Estimate tokens needed for task with target model",
+        "step_2": "Compare against available budget",
+        "step_3": "If exceeded: downgrade to cheaper model",
+        "step_4": "Recalculate and repeat until fits budget",
+        "step_5": "Escalate to human if no model viable"
+      },
+      "budget_tracking": {
+        "per_task": "Track cumulative tokens/cost",
+        "per_session": "Aggregate across all tasks",
+        "per_day": "Daily quota enforcement",
+        "alerts": "Warn at 80% threshold"
+      },
+      "recovery_strategies": [
+        "Task decomposition into smaller chunks",
+        "Caching partial results to reduce retokenization",
+        "Context pruning for efficiency",
+        "Result streaming to avoid buffering"
+      ],
+      "successRate": 0.88,
+      "lastUpdated": "2026-01-20"
+    },
+    {
+      "key": "router-booster-integration",
+      "pattern": "Model Router to Agent Booster Integration",
+      "description": "Seamless integration between router and booster for optimal task routing",
+      "integration_flow": {
+        "step_1_receive_task": "Router receives task description",
+        "step_2_classify": "Classify as simple transform or complex task",
+        "step_3_check_booster": "Check Agent Booster eligibility criteria",
+        "step_4_route": "If eligible: route to booster; else: route to model tier",
+        "step_5_execute": "Execute via appropriate pathway",
+        "step_6_report": "Report cost, latency, quality metrics"
+      },
+      "eligibility_handshake": {
+        "booster_can_handle": "Query booster eligibility function",
+        "transform_type": "Categorize specific transform",
+        "fallback_model": "Designate fallback if booster fails"
+      },
+      "metrics_collection": {
+        "booster_execution": "Track WASM vs TS fallback ratio",
+        "model_tier_usage": "Distribution across Haiku/Sonnet/Opus",
+        "cost_savings": "Actual vs baseline cost",
+        "quality_correlation": "Task complexity vs outcome quality"
+      },
+      "feedback_loop": "Use execution patterns to refine routing heuristics",
+      "successRate": 0.89,
+      "lastUpdated": "2026-01-20"
+    }
+  ]
+}