agentdb 1.0.1 → 1.0.2

package/dist/mcp/learning/core/learning-manager.js

@@ -0,0 +1,252 @@
+"use strict";
+/**
+ * LearningManager - Main orchestration layer for MCP learning integration
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.LearningManager = void 0;
+const experience_recorder_js_1 = require("./experience-recorder.js");
+const reward_estimator_js_1 = require("./reward-estimator.js");
+const session_manager_js_1 = require("./session-manager.js");
+const policy_optimizer_js_1 = require("./policy-optimizer.js");
+class LearningManager {
+    constructor(db) {
+        this.policyOptimizers = new Map();
+        this.db = db;
+        this.experienceRecorder = new experience_recorder_js_1.ExperienceRecorder(db);
+        this.rewardEstimator = new reward_estimator_js_1.RewardEstimator();
+        this.sessionManager = new session_manager_js_1.SessionManager(db);
+    }
+    /**
+     * Start a new learning session
+     */
+    async startSession(userId, sessionType, plugin = 'q-learning', config = {}) {
+        const session = await this.sessionManager.createSession(userId, sessionType, plugin, config);
+        // Initialize policy optimizer for this session
+        const optimizer = new policy_optimizer_js_1.PolicyOptimizer(config.learningRate || 0.1, config.discountFactor || 0.95, config.bufferSize || 10000);
+        this.policyOptimizers.set(session.sessionId, optimizer);
+        return session;
+    }
+    /**
+     * End a learning session
+     */
+    async endSession(sessionId) {
+        // Export and save policy BEFORE ending session
+        const optimizer = this.policyOptimizers.get(sessionId);
+        if (optimizer) {
+            const policy = optimizer.exportPolicy();
+            await this.sessionManager.updateSessionPolicy(sessionId, policy);
+            this.policyOptimizers.delete(sessionId);
+        }
+        // Now end the session
+        const session = await this.sessionManager.endSession(sessionId);
+        return session;
+    }
+    /**
+     * Record a tool execution as learning experience
+     */
+    async recordExperience(sessionId, toolName, args, result, outcome) {
+        const session = this.sessionManager.getSession(sessionId);
+        if (!session) {
+            throw new Error(`Session ${sessionId} not found`);
+        }
+        const context = {
+            userId: session.userId,
+            sessionId: session.sessionId,
+            taskType: session.sessionType,
+            timestamp: Date.now(),
+            isTerminal: outcome.success || !!outcome.error,
+        };
+        const experience = await this.experienceRecorder.recordToolExecution(toolName, args, result, context, outcome);
+        // Update policy with new experience
+        const optimizer = this.policyOptimizers.get(sessionId);
+        if (optimizer) {
+            await optimizer.updatePolicy(experience);
+        }
+        // Increment session experience count
+        this.sessionManager.incrementExperienceCount(sessionId);
+        return experience;
+    }
+    /**
+     * Predict next best action
+     */
+    async predictAction(sessionId, currentState, availableTools) {
+        const optimizer = this.policyOptimizers.get(sessionId);
+        if (!optimizer) {
+            throw new Error(`No policy optimizer for session ${sessionId}`);
+        }
+        return await optimizer.predictAction(currentState, availableTools);
+    }
+    /**
+     * Provide user feedback on action
+     */
+    async provideFeedback(sessionId, actionId, feedback) {
+        // Calculate feedback-adjusted reward
+        const normalizedRating = feedback.rating / 5.0; // Assume 0-5 scale
+        await this.experienceRecorder.updateExperienceReward(actionId, normalizedRating);
+    }
+    /**
+     * Train policy on collected experiences
+     */
+    async train(sessionId, options = {}) {
+        const optimizer = this.policyOptimizers.get(sessionId);
+        if (!optimizer) {
+            throw new Error(`No policy optimizer for session ${sessionId}`);
+        }
+        return await optimizer.train(options);
+    }
+    /**
+     * Get learning metrics
+     */
+    async getMetrics(sessionId, period = 'session') {
+        const experiences = await this.experienceRecorder.getSessionExperiences(sessionId);
+        if (experiences.length === 0) {
+            return {
+                period,
+                totalExperiences: 0,
+                averageReward: 0,
+                successRate: 0,
+                learningProgress: {
+                    initial: 0,
+                    current: 0,
+                    improvement: '0%',
+                },
+                topActions: [],
+            };
+        }
+        // Calculate metrics
+        const rewards = experiences.map((exp) => exp.reward);
+        const avgReward = rewards.reduce((sum, r) => sum + r, 0) / rewards.length;
+        const successCount = experiences.filter((exp) => exp.reward > 0.5).length;
+        const successRate = successCount / experiences.length;
+        // Calculate learning progress (first 10 vs last 10)
+        const firstBatch = experiences.slice(0, 10);
+        const lastBatch = experiences.slice(-10);
+        const initialReward = firstBatch.reduce((sum, exp) => sum + exp.reward, 0) / firstBatch.length;
+        const currentReward = lastBatch.reduce((sum, exp) => sum + exp.reward, 0) / lastBatch.length;
+        const improvement = initialReward > 0
+            ? (((currentReward - initialReward) / initialReward) * 100).toFixed(1)
+            : '0';
+        // Calculate top actions
+        const actionStats = new Map();
+        for (const exp of experiences) {
+            const tool = exp.action.tool;
+            const stats = actionStats.get(tool) || {
+                count: 0,
+                totalReward: 0,
+                successCount: 0,
+            };
+            stats.count++;
+            stats.totalReward += exp.reward;
+            if (exp.reward > 0.5)
+                stats.successCount++;
+            actionStats.set(tool, stats);
+        }
+        const topActions = Array.from(actionStats.entries())
+            .map(([tool, stats]) => ({
+            tool,
+            successRate: stats.successCount / stats.count,
+            avgReward: stats.totalReward / stats.count,
+            count: stats.count,
+        }))
+            .sort((a, b) => b.avgReward - a.avgReward)
+            .slice(0, 5);
+        return {
+            period,
+            totalExperiences: experiences.length,
+            averageReward: avgReward,
+            successRate,
+            learningProgress: {
+                initial: initialReward,
+                current: currentReward,
+                improvement: `${improvement}%`,
+            },
+            topActions,
+        };
+    }
+    /**
+     * Transfer learning between tasks
+     */
+    async transferLearning(sourceSessionId, targetSessionId, similarity = 0.7) {
+        const sourceOptimizer = this.policyOptimizers.get(sourceSessionId);
+        const targetOptimizer = this.policyOptimizers.get(targetSessionId);
+        if (!sourceOptimizer || !targetOptimizer) {
+            throw new Error('Source or target session not found');
+        }
+        // Export source policy
+        const sourcePolicy = sourceOptimizer.exportPolicy();
+        // Import into target (with similarity-based weighting)
+        const targetPolicy = targetOptimizer.exportPolicy();
+        // Merge policies (simplified - in production would use more sophisticated transfer)
+        const mergedQTable = { ...targetPolicy.qTable };
+        for (const [stateKey, actions] of Object.entries(sourcePolicy.qTable)) {
+            if (!mergedQTable[stateKey]) {
+                mergedQTable[stateKey] = {};
+            }
+            for (const [action, value] of Object.entries(actions)) {
+                const currentValue = mergedQTable[stateKey][action] || 0;
+                // Weighted average based on similarity
+                mergedQTable[stateKey][action] =
+                    currentValue * (1 - similarity) + value * similarity;
+            }
+        }
+        targetOptimizer.importPolicy({ ...targetPolicy, qTable: mergedQTable });
+        const sourceSession = this.sessionManager.getSession(sourceSessionId);
+        const targetSession = this.sessionManager.getSession(targetSessionId);
+        return {
+            sourceTask: sourceSession?.sessionType || 'unknown',
+            targetTask: targetSession?.sessionType || 'unknown',
+            similarity,
+            transferSuccess: true,
+            performanceGain: similarity * 0.3, // Estimated gain
+            experiencesTransferred: Object.keys(sourcePolicy.qTable).length,
+        };
+    }
+    /**
+     * Explain a prediction
+     */
+    async explainPrediction(sessionId, state) {
+        // Get similar experiences
+        const similarExperiences = await this.experienceRecorder.retrieveSimilarExperiences(state, 5);
+        // Calculate confidence factors
+        const confidenceFactors = {
+            experienceCount: Math.min(1.0, similarExperiences.length / 10),
+            avgReward: similarExperiences.reduce((sum, exp) => sum + exp.reward, 0) /
+                (similarExperiences.length || 1),
+            consistency: this.calculateConsistency(similarExperiences),
+        };
+        const reasoning = `Based on ${similarExperiences.length} similar past experiences with average reward ${confidenceFactors.avgReward.toFixed(2)}. Action consistency: ${(confidenceFactors.consistency * 100).toFixed(0)}%.`;
+        return {
+            reasoning,
+            similarExperiences,
+            confidenceFactors,
+        };
+    }
+    /**
+     * Calculate consistency of actions in similar experiences
+     */
+    calculateConsistency(experiences) {
+        if (experiences.length === 0)
+            return 0;
+        const actionCounts = new Map();
+        for (const exp of experiences) {
+            const tool = exp.action.tool;
+            actionCounts.set(tool, (actionCounts.get(tool) || 0) + 1);
+        }
+        const maxCount = Math.max(...Array.from(actionCounts.values()));
+        return maxCount / experiences.length;
+    }
+    /**
+     * Get session info
+     */
+    getSessionInfo(sessionId) {
+        return this.sessionManager.getSession(sessionId);
+    }
+    /**
+     * Restore sessions from database
+     */
+    async restoreSessions(userId) {
+        return await this.sessionManager.restoreSessions(userId);
+    }
+}
+exports.LearningManager = LearningManager;
+//# sourceMappingURL=learning-manager.js.map

package/dist/mcp/learning/core/learning-manager.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"learning-manager.js","sourceRoot":"","sources":["../../../../src/mcp/learning/core/learning-manager.ts"],"names":[],"mappings":";AAAA;;GAEG;;;AAgBH,qEAA8D;AAC9D,+DAAwD;AACxD,6DAAsD;AACtD,+DAAwD;AAExD,MAAa,eAAe;IAO1B,YAAY,EAAkB;QAFtB,qBAAgB,GAAiC,IAAI,GAAG,EAAE,CAAC;QAGjE,IAAI,CAAC,EAAE,GAAG,EAAE,CAAC;QACb,IAAI,CAAC,kBAAkB,GAAG,IAAI,2CAAkB,CAAC,EAAE,CAAC,CAAC;QACrD,IAAI,CAAC,eAAe,GAAG,IAAI,qCAAe,EAAE,CAAC;QAC7C,IAAI,CAAC,cAAc,GAAG,IAAI,mCAAc,CAAC,EAAE,CAAC,CAAC;IAC/C,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,YAAY,CAChB,MAAc,EACd,WAA4D,EAC5D,SAAiB,YAAY,EAC7B,SAA8B,EAAE;QAEhC,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,cAAc,CAAC,aAAa,CACrD,MAAM,EACN,WAAW,EACX,MAAM,EACN,MAAM,CACP,CAAC;QAEF,+CAA+C;QAC/C,MAAM,SAAS,GAAG,IAAI,qCAAe,CACnC,MAAM,CAAC,YAAY,IAAI,GAAG,EAC1B,MAAM,CAAC,cAAc,IAAI,IAAI,EAC7B,MAAM,CAAC,UAAU,IAAI,KAAK,CAC3B,CAAC;QACF,IAAI,CAAC,gBAAgB,CAAC,GAAG,CAAC,OAAO,CAAC,SAAS,EAAE,SAAS,CAAC,CAAC;QAExD,OAAO,OAAO,CAAC;IACjB,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,UAAU,CAAC,SAAiB;QAChC,+CAA+C;QAC/C,MAAM,SAAS,GAAG,IAAI,CAAC,gBAAgB,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;QACvD,IAAI,SAAS,EAAE,CAAC;YACd,MAAM,MAAM,GAAG,SAAS,CAAC,YAAY,EAAE,CAAC;YACxC,MAAM,IAAI,CAAC,cAAc,CAAC,mBAAmB,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC;YACjE,IAAI,CAAC,gBAAgB,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC;QAC1C,CAAC;QAED,sBAAsB;QACtB,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,cAAc,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC;QAChE,OAAO,OAAO,CAAC;IACjB,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,gBAAgB,CACpB,SAAiB,EACjB,QAAgB,EAChB,IAAS,EACT,MAAW,EACX,OAAgB;QAEhB,MAAM,OAAO,GAAG,IAAI,CAAC,cAAc,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC;QAC1D,IAAI,CAAC,OAAO,EAAE,CAAC;YACb,MAAM,IAAI,KAAK,CAAC,WAAW,SAAS,YAAY,CAAC,CAAC;QACpD,CAAC;QAED,MAAM,OAAO,GAAqB;YAChC,MAAM,EAAE,OAAO,CAAC,MAAM;YACtB,SAAS,EAAE,OAAO,CAAC,SAAS;YAC5B,QAAQ,EAAE,OAAO,CAAC,WAAW;YAC7B,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;YACrB,UAAU,EAAE,OAAO,CAAC,OAAO,IAAI,CAAC,CAAC,OAAO,CAAC,KAAK;SAC/C,CAAC;QAEF,MAAM,UAAU,GAAG,MAAM,IAAI,CAAC,kBAAkB,CAAC,mBAAmB,CAClE,QAAQ,EACR,IAAI,EACJ,MAAM,EACN,OAAO,EACP,OAAO,CACR,CAAC;QAEF,oCAAoC;QACpC,MAAM,SAAS,GAAG,IAAI,CAAC,gBAAgB,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;QACvD,IAAI,SAAS,EAAE,CAAC;YACd,MAAM,SAAS,CAAC,YAAY,CAAC,UAAU,CAAC,CAAC;QAC3C,CAAC;QAED,qCAAqC;QACrC,IAAI,CAAC,cAAc,CAAC,wBAAwB,CAAC,SAAS,CAAC,CAAC;QAExD,OAAO,UAAU,CAAC;IACpB,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,aAAa,CACjB,SAAiB,EACjB,YAAmB,EACnB,cAAwB;QAExB,MAAM,SAAS,GAAG,IAAI,CAAC,gBAAgB,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;QACvD,IAAI,CAAC,SAAS,EAAE,CAAC;YACf,MAAM,IAAI,KAAK,CAAC,mCAAmC,SAAS,EAAE,CAAC,CAAC;QAClE,CAAC;QAED,OAAO,MAAM,SAAS,CAAC,aAAa,CAAC,YAAY,EAAE,cAAc,CAAC,CAAC;IACrE,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,eAAe,CACnB,SAAiB,EACjB,QAAgB,EAChB,QAAuB;QAEvB,qCAAqC;QACrC,MAAM,gBAAgB,GAAG,QAAQ,CAAC,MAAM,GAAG,GAAG,CAAC,CAAC,mBAAmB;QAEnE,MAAM,IAAI,CAAC,kBAAkB,CAAC,sBAAsB,CAClD,QAAQ,EACR,gBAAgB,CACjB,CAAC;IACJ,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,KAAK,CACT,SAAiB,EACjB,UAA2B,EAAE;QAE7B,MAAM,SAAS,GAAG,IAAI,CAAC,gBAAgB,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;QACvD,IAAI,CAAC,SAAS,EAAE,CAAC;YACf,MAAM,IAAI,KAAK,CAAC,mCAAmC,SAAS,EAAE,CAAC,CAAC;QAClE,CAAC;QAED,OAAO,MAAM,SAAS,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;IACxC,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,UAAU,CACd,SAAiB,EACjB,SAAuD,SAAS;QAEhE,MAAM,WAAW,GAAG,MAAM,IAAI,CAAC,kBAAkB,CAAC,qBAAqB,CACrE,SAAS,CACV,CAAC;QAEF,IAAI,WAAW,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC7B,OAAO;gBACL,MAAM;gBACN,gBAAgB,EAAE,CAAC;gBACnB,aAAa,EAAE,CAAC;gBAChB,WAAW,EAAE,CAAC;gBACd,gBAAgB,EAAE;oBAChB,OAAO,EAAE,CAAC;oBACV,OAAO,EAAE,CAAC;oBACV,WAAW,EAAE,IAAI;iBAClB;gBACD,UAAU,EAAE,EAAE;aACf,CAAC;QACJ,CAAC;QAED,oBAAoB;QACpB,MAAM,OAAO,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;QACrD,MAAM,SAAS,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,OAAO,CAAC,MAAM,CAAC;QAE1E,MAAM,YAAY,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,MAAM,GAAG,GAAG,CAAC,CAAC,MAAM,CAAC;QAC1E,MAAM,WAAW,GAAG,YAAY,GAAG,WAAW,CAAC,MAAM,CAAC;QAEtD,oDAAoD;QACpD,MAAM,UAAU,GAAG,WAAW,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;QAC5C,MAAM,SAAS,GAAG,WAAW,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC;QACzC,MAAM,aAAa,GACjB,UAAU,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,GAAG,EAAE,EAAE,CAAC,GAAG,GAAG,GAAG,CAAC,MAAM,EAAE,CAAC,CAAC,GAAG,UAAU,CAAC,MAAM,CAAC;QAC3E,MAAM,aAAa,GACjB,SAAS,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,GAAG,EAAE,EAAE,CAAC,GAAG,GAAG,GAAG,CAAC,MAAM,EAAE,CAAC,CAAC,GAAG,SAAS,CAAC,MAAM,CAAC;QACzE,MAAM,WAAW,GACf,aAAa,GAAG,CAAC;YACf,CAAC,CAAC,CAAC,CAAC,CAAC,aAAa,GAAG,aAAa,CAAC,GAAG,aAAa,CAAC,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC;YACtE,CAAC,CAAC,GAAG,CAAC;QAEV,wBAAwB;QACxB,MAAM,WAAW,GAAG,IAAI,GAAG,EAGxB,CAAC;QAEJ,KAAK,MAAM,GAAG,IAAI,WAAW,EAAE,CAAC;YAC9B,MAAM,IAAI,GAAG,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC;YAC7B,MAAM,KAAK,GAAG,WAAW,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI;gBACrC,KAAK,EAAE,CAAC;gBACR,WAAW,EAAE,CAAC;gBACd,YAAY,EAAE,CAAC;aAChB,CAAC;YACF,KAAK,CAAC,KAAK,EAAE,CAAC;YACd,KAAK,CAAC,WAAW,IAAI,GAAG,CAAC,MAAM,CAAC;YAChC,IAAI,GAAG,CAAC,MAAM,GAAG,GAAG;gBAAE,KAAK,CAAC,YAAY,EAAE,CAAC;YAC3C,WAAW,CAAC,GAAG,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;QAC/B,CAAC;QAED,MAAM,UAAU,GAAG,KAAK,CAAC,IAAI,CAAC,WAAW,CAAC,OAAO,EAAE,CAAC;aACjD,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,CAAC;YACvB,IAAI;YACJ,WAAW,EAAE,KAAK,CAAC,YAAY,GAAG,KAAK,CAAC,KAAK;YAC7C,SAAS,EAAE,KAAK,CAAC,WAAW,GAAG,KAAK,CAAC,KAAK;YAC1C,KAAK,EAAE,KAAK,CAAC,KAAK;SACnB,CAAC,CAAC;aACF,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,GAAG,CAAC,CAAC,SAAS,CAAC;aACzC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;QAEf,OAAO;YACL,MAAM;YACN,gBAAgB,EAAE,WAAW,CAAC,MAAM;YACpC,aAAa,EAAE,SAAS;YACxB,WAAW;YACX,gBAAgB,EAAE;gBAChB,OAAO,EAAE,aAAa;gBACtB,OAAO,EAAE,aAAa;gBACtB,WAAW,EAAE,GAAG,WAAW,GAAG;aAC/B;YACD,UAAU;SACX,CAAC;IACJ,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,gBAAgB,CACpB,eAAuB,EACvB,eAAuB,EACvB,aAAqB,GAAG;QAExB,MAAM,eAAe,GAAG,IAAI,CAAC,gBAAgB,CAAC,GAAG,CAAC,eAAe,CAAC,CAAC;QACnE,MAAM,eAAe,GAAG,IAAI,CAAC,gBAAgB,CAAC,GAAG,CAAC,eAAe,CAAC,CAAC;QAEnE,IAAI,CAAC,eAAe,IAAI,CAAC,eAAe,EAAE,CAAC;YACzC,MAAM,IAAI,KAAK,CAAC,oCAAoC,CAAC,CAAC;QACxD,CAAC;QAED,uBAAuB;QACvB,MAAM,YAAY,GAAG,eAAe,CAAC,YAAY,EAAE,CAAC;QAEpD,uDAAuD;QACvD,MAAM,YAAY,GAAG,eAAe,CAAC,YAAY,EAAE,CAAC;QAEpD,oFAAoF;QACpF,MAAM,YAAY,GAAQ,EAAE,GAAG,YAAY,CAAC,MAAM,EAAE,CAAC;QAErD,KAAK,MAAM,CAAC,QAAQ,EAAE,OAAO,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,YAAY,CAAC,MAAM,CAAC,EAAE,CAAC;YACtE,IAAI,CAAC,YAAY,CAAC,QAAQ,CAAC,EAAE,CAAC;gBAC5B,YAAY,CAAC,QAAQ,CAAC,GAAG,EAAE,CAAC;YAC9B,CAAC;YAED,KAAK,MAAM,CAAC,MAAM,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,OAAc,CAAC,EAAE,CAAC;gBAC7D,MAAM,YAAY,GAAG,YAAY,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;gBACzD,uCAAuC;gBACvC,YAAY,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC;oBAC5B,YAAY,GAAG,CAAC,CAAC,GAAG,UAAU,CAAC,GAAI,KAAgB,GAAG,UAAU,CAAC;YACrE,CAAC;QACH,CAAC;QAED,eAAe,CAAC,YAAY,CAAC,EAAE,GAAG,YAAY,EAAE,MAAM,EAAE,YAAY,EAAE,CAAC,CAAC;QAExE,MAAM,aAAa,GAAG,IAAI,CAAC,cAAc,CAAC,UAAU,CAAC,eAAe,CAAC,CAAC;QACtE,MAAM,aAAa,GAAG,IAAI,CAAC,cAAc,CAAC,UAAU,CAAC,eAAe,CAAC,CAAC;QAEtE,OAAO;YACL,UAAU,EAAE,aAAa,EAAE,WAAW,IAAI,SAAS;YACnD,UAAU,EAAE,aAAa,EAAE,WAAW,IAAI,SAAS;YACnD,UAAU;YACV,eAAe,EAAE,IAAI;YACrB,eAAe,EAAE,UAAU,GAAG,GAAG,EAAE,iBAAiB;YACpD,sBAAsB,EAAE,MAAM,CAAC,IAAI,CAAC,YAAY,CAAC,MAAM,CAAC,CAAC,MAAM;SAChE,CAAC;IACJ,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,iBAAiB,CACrB,SAAiB,EACjB,KAAY;QAMZ,0BAA0B;QAC1B,MAAM,kBAAkB,GACtB,MAAM,IAAI,CAAC,kBAAkB,CAAC,0BAA0B,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC;QAErE,+BAA+B;QAC/B,MAAM,iBAAiB,GAA2B;YAChD,eAAe,EAAE,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,kBAAkB,CAAC,MAAM,GAAG,EAAE,CAAC;YAC9D,SAAS,EACP,kBAAkB,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,GAAG,EAAE,EAAE,CAAC,GAAG,GAAG,GAAG,CAAC,MAAM,EAAE,CAAC,CAAC;gBAC5D,CAAC,kBAAkB,CAAC,MAAM,IAAI,CAAC,CAAC;YAClC,WAAW,EAAE,IAAI,CAAC,oBAAoB,CAAC,kBAAkB,CAAC;SAC3D,CAAC;QAEF,MAAM,SAAS,GAAG,YAAY,kBAAkB,CAAC,MAAM,iDAAiD,iBAAiB,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,yBAAyB,CAAC,iBAAiB,CAAC,WAAW,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC;QAE5N,OAAO;YACL,SAAS;YACT,kBAAkB;YAClB,iBAAiB;SAClB,CAAC;IACJ,CAAC;IAED;;OAEG;IACK,oBAAoB,CAAC,WAAyB;QACpD,IAAI,WAAW,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO,CAAC,CAAC;QAEvC,MAAM,YAAY,GAAG,IAAI,GAAG,EAAkB,CAAC;QAC/C,KAAK,MAAM,GAAG,IAAI,WAAW,EAAE,CAAC;YAC9B,MAAM,IAAI,GAAG,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC;YAC7B,YAAY,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,YAAY,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;QAC5D,CAAC;QAED,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,KAAK,CAAC,IAAI,CAAC,YAAY,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;QAChE,OAAO,QAAQ,GAAG,WAAW,CAAC,MAAM,CAAC;IACvC,CAAC;IAED;;OAEG;IACH,cAAc,CAAC,SAAiB;QAC9B,OAAO,IAAI,CAAC,cAAc,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC;IACnD,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,eAAe,CAAC,MAAe;QACnC,OAAO,MAAM,IAAI,CAAC,cAAc,CAAC,eAAe,CAAC,MAAM,CAAC,CAAC;IAC3D,CAAC;CACF;AA/VD,0CA+VC"}

package/dist/mcp/learning/core/learning-manager.mjs

@@ -0,0 +1,247 @@
+/**
+ * LearningManager - Main orchestration layer for MCP learning integration
+ */
+import { ExperienceRecorder } from './experience-recorder.mjs';
+import { RewardEstimator } from './reward-estimator.mjs';
+import { SessionManager } from './session-manager.mjs';
+import { PolicyOptimizer } from './policy-optimizer.mjs';
+export class LearningManager {
+    constructor(db) {
+        this.policyOptimizers = new Map();
+        this.db = db;
+        this.experienceRecorder = new ExperienceRecorder(db);
+        this.rewardEstimator = new RewardEstimator();
+        this.sessionManager = new SessionManager(db);
+    }
+    /**
+     * Start a new learning session
+     */
+    async startSession(userId, sessionType, plugin = 'q-learning', config = {}) {
+        const session = await this.sessionManager.createSession(userId, sessionType, plugin, config);
+        // Initialize policy optimizer for this session
+        const optimizer = new PolicyOptimizer(config.learningRate || 0.1, config.discountFactor || 0.95, config.bufferSize || 10000);
+        this.policyOptimizers.set(session.sessionId, optimizer);
+        return session;
+    }
+    /**
+     * End a learning session
+     */
+    async endSession(sessionId) {
+        // Export and save policy BEFORE ending session
+        const optimizer = this.policyOptimizers.get(sessionId);
+        if (optimizer) {
+            const policy = optimizer.exportPolicy();
+            await this.sessionManager.updateSessionPolicy(sessionId, policy);
+            this.policyOptimizers.delete(sessionId);
+        }
+        // Now end the session
+        const session = await this.sessionManager.endSession(sessionId);
+        return session;
+    }
+    /**
+     * Record a tool execution as learning experience
+     */
+    async recordExperience(sessionId, toolName, args, result, outcome) {
+        const session = this.sessionManager.getSession(sessionId);
+        if (!session) {
+            throw new Error(`Session ${sessionId} not found`);
+        }
+        const context = {
+            userId: session.userId,
+            sessionId: session.sessionId,
+            taskType: session.sessionType,
+            timestamp: Date.now(),
+            isTerminal: outcome.success || !!outcome.error,
+        };
+        const experience = await this.experienceRecorder.recordToolExecution(toolName, args, result, context, outcome);
+        // Update policy with new experience
+        const optimizer = this.policyOptimizers.get(sessionId);
+        if (optimizer) {
+            await optimizer.updatePolicy(experience);
+        }
+        // Increment session experience count
+        this.sessionManager.incrementExperienceCount(sessionId);
+        return experience;
+    }
+    /**
+     * Predict next best action
+     */
+    async predictAction(sessionId, currentState, availableTools) {
+        const optimizer = this.policyOptimizers.get(sessionId);
+        if (!optimizer) {
+            throw new Error(`No policy optimizer for session ${sessionId}`);
+        }
+        return await optimizer.predictAction(currentState, availableTools);
+    }
+    /**
+     * Provide user feedback on action
+     */
+    async provideFeedback(sessionId, actionId, feedback) {
+        // Calculate feedback-adjusted reward
+        const normalizedRating = feedback.rating / 5.0; // Assume 0-5 scale
+        await this.experienceRecorder.updateExperienceReward(actionId, normalizedRating);
+    }
+    /**
+     * Train policy on collected experiences
+     */
+    async train(sessionId, options = {}) {
+        const optimizer = this.policyOptimizers.get(sessionId);
+        if (!optimizer) {
+            throw new Error(`No policy optimizer for session ${sessionId}`);
+        }
+        return await optimizer.train(options);
+    }
+    /**
+     * Get learning metrics
+     */
+    async getMetrics(sessionId, period = 'session') {
+        const experiences = await this.experienceRecorder.getSessionExperiences(sessionId);
+        if (experiences.length === 0) {
+            return {
+                period,
+                totalExperiences: 0,
+                averageReward: 0,
+                successRate: 0,
+                learningProgress: {
+                    initial: 0,
+                    current: 0,
+                    improvement: '0%',
+                },
+                topActions: [],
+            };
+        }
+        // Calculate metrics
+        const rewards = experiences.map((exp) => exp.reward);
+        const avgReward = rewards.reduce((sum, r) => sum + r, 0) / rewards.length;
+        const successCount = experiences.filter((exp) => exp.reward > 0.5).length;
+        const successRate = successCount / experiences.length;
+        // Calculate learning progress (first 10 vs last 10)
+        const firstBatch = experiences.slice(0, 10);
+        const lastBatch = experiences.slice(-10);
+        const initialReward = firstBatch.reduce((sum, exp) => sum + exp.reward, 0) / firstBatch.length;
+        const currentReward = lastBatch.reduce((sum, exp) => sum + exp.reward, 0) / lastBatch.length;
+        const improvement = initialReward > 0
+            ? (((currentReward - initialReward) / initialReward) * 100).toFixed(1)
+            : '0';
+        // Calculate top actions
+        const actionStats = new Map();
+        for (const exp of experiences) {
+            const tool = exp.action.tool;
+            const stats = actionStats.get(tool) || {
+                count: 0,
+                totalReward: 0,
+                successCount: 0,
+            };
+            stats.count++;
+            stats.totalReward += exp.reward;
+            if (exp.reward > 0.5)
+                stats.successCount++;
+            actionStats.set(tool, stats);
+        }
+        const topActions = Array.from(actionStats.entries())
+            .map(([tool, stats]) => ({
+            tool,
+            successRate: stats.successCount / stats.count,
+            avgReward: stats.totalReward / stats.count,
+            count: stats.count,
+        }))
+            .sort((a, b) => b.avgReward - a.avgReward)
+            .slice(0, 5);
+        return {
+            period,
+            totalExperiences: experiences.length,
+            averageReward: avgReward,
+            successRate,
+            learningProgress: {
+                initial: initialReward,
+                current: currentReward,
+                improvement: `${improvement}%`,
+            },
+            topActions,
+        };
+    }
+    /**
+     * Transfer learning between tasks
+     */
+    async transferLearning(sourceSessionId, targetSessionId, similarity = 0.7) {
+        const sourceOptimizer = this.policyOptimizers.get(sourceSessionId);
+        const targetOptimizer = this.policyOptimizers.get(targetSessionId);
+        if (!sourceOptimizer || !targetOptimizer) {
+            throw new Error('Source or target session not found');
+        }
+        // Export source policy
+        const sourcePolicy = sourceOptimizer.exportPolicy();
+        // Import into target (with similarity-based weighting)
+        const targetPolicy = targetOptimizer.exportPolicy();
+        // Merge policies (simplified - in production would use more sophisticated transfer)
+        const mergedQTable = { ...targetPolicy.qTable };
+        for (const [stateKey, actions] of Object.entries(sourcePolicy.qTable)) {
+            if (!mergedQTable[stateKey]) {
+                mergedQTable[stateKey] = {};
+            }
+            for (const [action, value] of Object.entries(actions)) {
+                const currentValue = mergedQTable[stateKey][action] || 0;
+                // Weighted average based on similarity
+                mergedQTable[stateKey][action] =
+                    currentValue * (1 - similarity) + value * similarity;
+            }
+        }
+        targetOptimizer.importPolicy({ ...targetPolicy, qTable: mergedQTable });
+        const sourceSession = this.sessionManager.getSession(sourceSessionId);
+        const targetSession = this.sessionManager.getSession(targetSessionId);
+        return {
+            sourceTask: sourceSession?.sessionType || 'unknown',
+            targetTask: targetSession?.sessionType || 'unknown',
+            similarity,
+            transferSuccess: true,
+            performanceGain: similarity * 0.3, // Estimated gain
+            experiencesTransferred: Object.keys(sourcePolicy.qTable).length,
+        };
+    }
+    /**
+     * Explain a prediction
+     */
+    async explainPrediction(sessionId, state) {
+        // Get similar experiences
+        const similarExperiences = await this.experienceRecorder.retrieveSimilarExperiences(state, 5);
+        // Calculate confidence factors
+        const confidenceFactors = {
+            experienceCount: Math.min(1.0, similarExperiences.length / 10),
+            avgReward: similarExperiences.reduce((sum, exp) => sum + exp.reward, 0) /
+                (similarExperiences.length || 1),
+            consistency: this.calculateConsistency(similarExperiences),
+        };
+        const reasoning = `Based on ${similarExperiences.length} similar past experiences with average reward ${confidenceFactors.avgReward.toFixed(2)}. Action consistency: ${(confidenceFactors.consistency * 100).toFixed(0)}%.`;
+        return {
+            reasoning,
+            similarExperiences,
+            confidenceFactors,
+        };
+    }
+    /**
+     * Calculate consistency of actions in similar experiences
+     */
+    calculateConsistency(experiences) {
+        if (experiences.length === 0)
+            return 0;
+        const actionCounts = new Map();
+        for (const exp of experiences) {
+            const tool = exp.action.tool;
+            actionCounts.set(tool, (actionCounts.get(tool) || 0) + 1);
+        }
+        const maxCount = Math.max(...Array.from(actionCounts.values()));
+        return maxCount / experiences.length;
+    }
+    /**
+     * Get session info
+     */
+    getSessionInfo(sessionId) {
+        return this.sessionManager.getSession(sessionId);
+    }
+    /**
+     * Restore sessions from database
+     */
+    async restoreSessions(userId) {
+        return await this.sessionManager.restoreSessions(userId);
+    }
+}

package/dist/mcp/learning/core/policy-optimizer.d.ts

@@ -0,0 +1,53 @@
+/**
+ * PolicyOptimizer - Optimizes action selection policy using reinforcement learning
+ */
+import type { State, Experience, ActionPrediction, TrainingOptions, TrainingMetrics } from '../types/index.js';
+export declare class PolicyOptimizer {
+    private qTable;
+    private learningRate;
+    private discountFactor;
+    private explorationRate;
+    private experienceBuffer;
+    constructor(learningRate?: number, discountFactor?: number, bufferSize?: number);
+    /**
+     * Predict best action for current state
+     */
+    predictAction(state: State, availableActions: string[]): Promise<ActionPrediction>;
+    /**
+     * Update policy based on experience
+     */
+    updatePolicy(experience: Experience): Promise<void>;
+    /**
+     * Train policy on batch of experiences
+     */
+    train(options?: TrainingOptions): Promise<TrainingMetrics>;
+    /**
+     * Get policy statistics
+     */
+    getPolicyStats(): {
+        statesLearned: number;
+        totalExperiences: number;
+        avgQValue: number;
+    };
+    /**
+     * Export policy for persistence
+     */
+    exportPolicy(): any;
+    /**
+     * Import policy from persistence
+     */
+    importPolicy(policyData: any): void;
+    /**
+     * Encode state as string key for Q-table
+     */
+    private encodeState;
+    /**
+     * Get experience count for state
+     */
+    private getExperienceCount;
+    /**
+     * Decay exploration rate over time
+     */
+    decayExploration(decayRate?: number): void;
+}
+//# sourceMappingURL=policy-optimizer.d.ts.map

package/dist/mcp/learning/core/policy-optimizer.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"policy-optimizer.d.ts","sourceRoot":"","sources":["../../../../src/mcp/learning/core/policy-optimizer.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,EACV,KAAK,EAEL,UAAU,EACV,gBAAgB,EAChB,eAAe,EACf,eAAe,EAChB,MAAM,mBAAmB,CAAC;AAG3B,qBAAa,eAAe;IAC1B,OAAO,CAAC,MAAM,CAA+C;IAC7D,OAAO,CAAC,YAAY,CAAe;IACnC,OAAO,CAAC,cAAc,CAAgB;IACtC,OAAO,CAAC,eAAe,CAAe;IACtC,OAAO,CAAC,gBAAgB,CAAmB;gBAGzC,YAAY,GAAE,MAAY,EAC1B,cAAc,GAAE,MAAa,EAC7B,UAAU,GAAE,MAAc;IAO5B;;OAEG;IACG,aAAa,CACjB,KAAK,EAAE,KAAK,EACZ,gBAAgB,EAAE,MAAM,EAAE,GACzB,OAAO,CAAC,gBAAgB,CAAC;IAyD5B;;OAEG;IACG,YAAY,CAAC,UAAU,EAAE,UAAU,GAAG,OAAO,CAAC,IAAI,CAAC;IAoCzD;;OAEG;IACG,KAAK,CAAC,OAAO,GAAE,eAAoB,GAAG,OAAO,CAAC,eAAe,CAAC;IAoFpE;;OAEG;IACH,cAAc,IAAI;QAChB,aAAa,EAAE,MAAM,CAAC;QACtB,gBAAgB,EAAE,MAAM,CAAC;QACzB,SAAS,EAAE,MAAM,CAAC;KACnB;IAkBD;;OAEG;IACH,YAAY,IAAI,GAAG;IAgBnB;;OAEG;IACH,YAAY,CAAC,UAAU,EAAE,GAAG,GAAG,IAAI;IAoBnC;;OAEG;IACH,OAAO,CAAC,WAAW;IAUnB;;OAEG;IACH,OAAO,CAAC,kBAAkB;IAK1B;;OAEG;IACH,gBAAgB,CAAC,SAAS,GAAE,MAAc,GAAG,IAAI;CAGlD"}