sentinel-agentos 0.1.0

package/dist/evaluator/exec-evaluator.js

@@ -0,0 +1,266 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.PostExecEvaluator = exports.RuntimeEvaluator = exports.PreExecEvaluator = void 0;
+/**
+ * PreExecEvaluator — captures metrics before tool execution.
+ *
+ * Watches the Guard layer output and WorkingMemory context
+ * to score parameter quality, context utilization, and risk.
+ */
+class PreExecEvaluator {
+    schemaGate;
+    riskGate;
+    workingMemory;
+    constructor(schemaGate, riskGate, workingMemory) {
+        this.schemaGate = schemaGate;
+        this.riskGate = riskGate;
+        this.workingMemory = workingMemory;
+    }
+    /**
+     * Evaluate a tool call before execution.
+     */
+    evaluate(toolName, parameters) {
+        // 1. Schema check
+        const schemaCheck = this.schemaGate.check(toolName, parameters);
+        // 2. Risk assessment
+        const riskScore = this.riskGate.evaluate(toolName, parameters);
+        // 3. Parameter quality: does the agent use context-aware params?
+        const paramQuality = this.evaluateParamQuality(toolName, parameters);
+        // 4. Context utilization: is the agent leveraging WorkingMemory?
+        const contextUtilization = this.evaluateContextUtilization(toolName, parameters);
+        return {
+            timestamp: Date.now(),
+            toolName,
+            schemaCheck,
+            riskScore,
+            paramQuality,
+            contextUtilization,
+        };
+    }
+    /**
+     * Score parameter quality based on contextual awareness.
+     *
+     * High quality: path contains session-relevant project paths,
+     * content references open files, etc.
+     * Low quality: bare strings, random-looking paths, missing files.
+     */
+    evaluateParamQuality(_toolName, parameters) {
+        let score = 0.5; // neutral start
+        const observations = [];
+        // Check if path references an open file
+        if (typeof parameters['path'] === 'string') {
+            const path = parameters['path'];
+            if (this.workingMemory.openFiles.some((f) => path.includes(f))) {
+                score += 0.3;
+                observations.push('Path references an open file');
+            }
+            if (path.startsWith('/') || path.match(/^[A-Z]:\\/)) {
+                observations.push('Absolute path used');
+            }
+        }
+        // Check if content parameter is meaningful
+        if (typeof parameters['content'] === 'string') {
+            const content = parameters['content'];
+            if (content.length > 20) {
+                score = Math.min(1.0, score + 0.1);
+            }
+            if (content.length === 0) {
+                score -= 0.2;
+                observations.push('Empty content — possible error');
+            }
+        }
+        // Check for file paths in multiple parameters
+        const filePaths = Object.values(parameters).filter((v) => typeof v === 'string' && (v.includes('.ts') || v.includes('.js') || v.includes('.json')));
+        if (filePaths.length > 1) {
+            score = Math.min(1.0, score + 0.1);
+            observations.push('Multiple file references — coordinated operation');
+        }
+        return {
+            score: Math.round(Math.max(0, Math.min(1, score)) * 100) / 100,
+            observations,
+        };
+    }
+    /**
+     * Score how well the agent uses stored context.
+     */
+    evaluateContextUtilization(_toolName, parameters) {
+        let score = 0.4;
+        const patterns = [];
+        // Check if agent references recent messages
+        if (this.workingMemory.recentMessages.length > 0) {
+            score += 0.1;
+            patterns.push(`${this.workingMemory.recentMessages.length} recent messages available`);
+        }
+        // Check if agent uses cached tool results
+        const cachedCount = this.workingMemory.recentToolResults.size;
+        if (cachedCount > 0) {
+            score += 0.1;
+            patterns.push(`${cachedCount} cached results available`);
+        }
+        // Check parameter values for context patterns
+        const allValues = Object.values(parameters).map(String).join(' ');
+        for (const msg of this.workingMemory.recentMessages.slice(-3)) {
+            const words = msg.content.split(/\s+/).filter((w) => w.length > 3);
+            for (const word of words.slice(0, 5)) {
+                if (allValues.includes(word)) {
+                    score += 0.1;
+                    patterns.push(`Parameter references recent context: "${word}"`);
+                }
+            }
+        }
+        return {
+            score: Math.round(Math.min(1.0, score) * 100) / 100,
+            patterns,
+        };
+    }
+}
+exports.PreExecEvaluator = PreExecEvaluator;
+/**
+ * RuntimeEvaluator — captures metrics during execution.
+ *
+ * Tracks retries, self-corrections, timeouts, and
+ * whether the agent selected the right tool for the job.
+ */
+class RuntimeEvaluator {
+    /** Historical tool selection patterns — toolName -> successful scenario count */
+    toolHistory = new Map();
+    /**
+     * Evaluate a completed tool execution.
+     */
+    evaluate(options) {
+        const durationMs = options.endTime - options.startTime;
+        const toolSuccess = !options.hadTimeout && options.toolResult !== undefined;
+        // Tool selection accuracy: compare against historical patterns
+        let toolSelectionMatch;
+        if (options.expectedTool) {
+            // Direct comparison if expectedTool is provided
+            toolSelectionMatch = options.toolName === options.expectedTool;
+        }
+        else {
+            // Auto-detect: is this tool historically successful for similar params?
+            const history = this.toolHistory.get(options.toolName);
+            if (history) {
+                const historicalSuccessRate = history.calls > 0
+                    ? history.successes / history.calls
+                    : 0;
+                // If this tool has >70% historical success, consider it a "good" selection
+                toolSelectionMatch = historicalSuccessRate > 0.7 ? true : undefined;
+            }
+        }
+        // Record this call in history
+        this.recordToolCall(options.toolName, toolSuccess);
+        // Adaptive score: composite of retry rate, timeout, correction
+        let adaptiveScore = 1.0;
+        adaptiveScore -= options.retryCount * 0.15; // Each retry costs 0.15
+        if (options.hadTimeout)
+            adaptiveScore -= 0.5;
+        if (options.wasSelfCorrected)
+            adaptiveScore += 0.2; // Self-correction is good!
+        adaptiveScore = Math.max(0, Math.min(1, adaptiveScore));
+        return {
+            retryCount: options.retryCount,
+            selfCorrected: options.wasSelfCorrected,
+            hadTimeout: options.hadTimeout,
+            toolSuccess,
+            toolSelectionMatch,
+            adaptiveScore: Math.round(adaptiveScore * 100) / 100,
+            durationMs,
+        };
+    }
+    /** Record a tool call in the history tracker */
+    recordToolCall(toolName, success) {
+        const existing = this.toolHistory.get(toolName);
+        if (existing) {
+            existing.calls++;
+            if (success)
+                existing.successes++;
+        }
+        else {
+            this.toolHistory.set(toolName, { calls: 1, successes: success ? 1 : 0 });
+        }
+    }
+    /** Get tool selection accuracy statistics */
+    getToolAccuracy() {
+        const result = {};
+        for (const [tool, history] of this.toolHistory) {
+            result[tool] = {
+                calls: history.calls,
+                successRate: history.calls > 0
+                    ? Math.round((history.successes / history.calls) * 100) / 100
+                    : 0,
+            };
+        }
+        return result;
+    }
+}
+exports.RuntimeEvaluator = RuntimeEvaluator;
+/**
+ * PostExecEvaluator — captures metrics after execution.
+ *
+ * Scores verify results, user acceptance patterns,
+ * and checks if the agent actually used its own result later.
+ */
+class PostExecEvaluator {
+    /** Track result references for utilization scoring */
+    resultReferenceTracker = new Map();
+    /**
+     * Evaluate post-execution outcomes.
+     */
+    evaluate(options) {
+        // Verify score
+        const verifyScore = options.verifyChecks > 0
+            ? 1 - (options.verifyFailures / options.verifyChecks)
+            : 1;
+        // User acceptance
+        const acceptance = options.userAccepted ? 1.0 : options.userProvidedEdit ? 0.3 : 0.7;
+        // Composite outcome score
+        const outcomeScore = (verifyScore * 0.3 +
+            acceptance * 0.4 +
+            (options.resultWasUsed ? 0.3 : 0));
+        // Overall health flag
+        const healthy = verifyScore > 0.8 && acceptance > 0.5;
+        return {
+            verifyPassed: options.verifyPassed,
+            verifyScore: Math.round(verifyScore * 100) / 100,
+            userAccepted: options.userAccepted,
+            userEditRate: options.userProvidedEdit ? 1 : 0,
+            resultUtilized: options.resultWasUsed,
+            outcomeScore: Math.round(outcomeScore * 100) / 100,
+            healthy,
+            diffLinesChanged: options.diffLinesChanged,
+        };
+    }
+    /**
+     * Track a tool result for later utilization detection.
+     * Call this after each tool execution.
+     */
+    trackResult(operationId, result) {
+        this.resultReferenceTracker.set(operationId, { result, referenced: false });
+    }
+    /**
+     * Mark a previously-tracked result as referenced (used by the agent later).
+     */
+    markResultReferenced(operationId) {
+        const entry = this.resultReferenceTracker.get(operationId);
+        if (entry)
+            entry.referenced = true;
+    }
+    /**
+     * Check if a result has been utilized by the agent.
+     */
+    isResultReferenced(operationId) {
+        return this.resultReferenceTracker.get(operationId)?.referenced ?? false;
+    }
+    /**
+     * Get overall result utilization rate.
+     */
+    getUtilizationRate() {
+        const entries = Array.from(this.resultReferenceTracker.values());
+        if (entries.length === 0)
+            return 0;
+        const referenced = entries.filter((e) => e.referenced).length;
+        return Math.round((referenced / entries.length) * 100) / 100;
+    }
+}
+exports.PostExecEvaluator = PostExecEvaluator;
+//# sourceMappingURL=exec-evaluator.js.map

package/dist/evaluator/exec-evaluator.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"exec-evaluator.js","sourceRoot":"","sources":["../../src/evaluator/exec-evaluator.ts"],"names":[],"mappings":";;;AAWA;;;;;GAKG;AACH,MAAa,gBAAgB;IACnB,UAAU,CAAa;IACvB,QAAQ,CAAW;IACnB,aAAa,CAAgB;IAErC,YACE,UAAsB,EACtB,QAAkB,EAClB,aAA4B;QAE5B,IAAI,CAAC,UAAU,GAAG,UAAU,CAAC;QAC7B,IAAI,CAAC,QAAQ,GAAG,QAAQ,CAAC;QACzB,IAAI,CAAC,aAAa,GAAG,aAAa,CAAC;IACrC,CAAC;IAED;;OAEG;IACH,QAAQ,CACN,QAAgB,EAChB,UAAmC;QAEnC,kBAAkB;QAClB,MAAM,WAAW,GAAgB,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,QAAQ,EAAE,UAAU,CAAC,CAAC;QAE7E,qBAAqB;QACrB,MAAM,SAAS,GAAc,IAAI,CAAC,QAAQ,CAAC,QAAQ,CAAC,QAAQ,EAAE,UAAU,CAAC,CAAC;QAE1E,iEAAiE;QACjE,MAAM,YAAY,GAAG,IAAI,CAAC,oBAAoB,CAAC,QAAQ,EAAE,UAAU,CAAC,CAAC;QAErE,iEAAiE;QACjE,MAAM,kBAAkB,GAAG,IAAI,CAAC,0BAA0B,CAAC,QAAQ,EAAE,UAAU,CAAC,CAAC;QAEjF,OAAO;YACL,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;YACrB,QAAQ;YACR,WAAW;YACX,SAAS;YACT,YAAY;YACZ,kBAAkB;SACnB,CAAC;IACJ,CAAC;IAED;;;;;;OAMG;IACK,oBAAoB,CAC1B,SAAiB,EACjB,UAAmC;QAEnC,IAAI,KAAK,GAAG,GAAG,CAAC,CAAC,gBAAgB;QACjC,MAAM,YAAY,GAAa,EAAE,CAAC;QAElC,wCAAwC;QACxC,IAAI,OAAO,UAAU,CAAC,MAAM,CAAC,KAAK,QAAQ,EAAE,CAAC;YAC3C,MAAM,IAAI,GAAG,UAAU,CAAC,MAAM,CAAW,CAAC;YAC1C,IAAI,IAAI,CAAC,aAAa,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;gBAC/D,KAAK,IAAI,GAAG,CAAC;gBACb,YAAY,CAAC,IAAI,CAAC,8BAA8B,CAAC,CAAC;YACpD,CAAC;YACD,IAAI,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,IAAI,CAAC,KAAK,CAAC,WAAW,CAAC,EAAE,CAAC;gBACpD,YAAY,CAAC,IAAI,CAAC,oBAAoB,CAAC,CAAC;YAC1C,CAAC;QACH,CAAC;QAED,2CAA2C;QAC3C,IAAI,OAAO,UAAU,CAAC,SAAS,CAAC,KAAK,QAAQ,EAAE,CAAC;YAC9C,MAAM,OAAO,GAAG,UAAU,CAAC,SAAS,CAAW,CAAC;YAChD,IAAI,OAAO,CAAC,MAAM,GAAG,EAAE,EAAE,CAAC;gBACxB,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,KAAK,GAAG,GAAG,CAAC,CAAC;YACrC,CAAC;YACD,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBACzB,KAAK,IAAI,GAAG,CAAC;gBACb,YAAY,CAAC,IAAI,CAAC,gCAAgC,CAAC,CAAC;YACtD,CAAC;QACH,CAAC;QAED,8CAA8C;QAC9C,MAAM,SAAS,GAAG,MAAM,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,MAAM,CAChD,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,IAAI,CAAC,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAChG,CAAC;QAEF,IAAI,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACzB,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,KAAK,GAAG,GAAG,CAAC,CAAC;YACnC,YAAY,CAAC,IAAI,CAAC,kDAAkD,CAAC,CAAC;QACxE,CAAC;QAED,OAAO;YACL,KAAK,EAAE,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC,GAAG,GAAG,CAAC,GAAG,GAAG;YAC9D,YAAY;SACb,CAAC;IACJ,CAAC;IAED;;OAEG;IACK,0BAA0B,CAChC,SAAiB,EACjB,UAAmC;QAEnC,IAAI,KAAK,GAAG,GAAG,CAAC;QAChB,MAAM,QAAQ,GAAa,EAAE,CAAC;QAE9B,4CAA4C;QAC5C,IAAI,IAAI,CAAC,aAAa,CAAC,cAAc,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACjD,KAAK,IAAI,GAAG,CAAC;YACb,QAAQ,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC,aAAa,CAAC,cAAc,CAAC,MAAM,4BAA4B,CAAC,CAAC;QACzF,CAAC;QAED,0CAA0C;QAC1C,MAAM,WAAW,GAAG,IAAI,CAAC,aAAa,CAAC,iBAAiB,CAAC,IAAI,CAAC;QAC9D,IAAI,WAAW,GAAG,CAAC,EAAE,CAAC;YACpB,KAAK,IAAI,GAAG,CAAC;YACb,QAAQ,CAAC,IAAI,CAAC,GAAG,WAAW,2BAA2B,CAAC,CAAC;QAC3D,CAAC;QAED,8CAA8C;QAC9C,MAAM,SAAS,GAAG,MAAM,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAClE,KAAK,MAAM,GAAG,IAAI,IAAI,CAAC,aAAa,CAAC,cAAc,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;YAC9D,MAAM,KAAK,GAAG,GAAG,CAAC,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;YACnE,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,CAAC;gBACrC,IAAI,SAAS,CAAC,QAAQ,CAAC,IAAI,CAAC,EAAE,CAAC;oBAC7B,KAAK,IAAI,GAAG,CAAC;oBACb,QAAQ,CAAC,IAAI,CAAC,yCAAyC,IAAI,GAAG,CAAC,CAAC;gBAClE,CAAC;YACH,CAAC;QACH,CAAC;QAED,OAAO;YACL,KAAK,EAAE,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,KAAK,CAAC,GAAG,GAAG,CAAC,GAAG,GAAG;YACnD,QAAQ;SACT,CAAC;IACJ,CAAC;CACF;AA1ID,4CA0IC;AAED;;;;;GAKG;AACH,MAAa,gBAAgB;IAC3B,iFAAiF;IACzE,WAAW,GAAsD,IAAI,GAAG,EAAE,CAAC;IAEnF;;OAEG;IACH,QAAQ,CAAC,OASR;QACC,MAAM,UAAU,GAAG,OAAO,CAAC,OAAO,GAAG,OAAO,CAAC,SAAS,CAAC;QACvD,MAAM,WAAW,GAAG,CAAC,OAAO,CAAC,UAAU,IAAI,OAAO,CAAC,UAAU,KAAK,SAAS,CAAC;QAE5E,+DAA+D;QAC/D,IAAI,kBAAuC,CAAC;QAC5C,IAAI,OAAO,CAAC,YAAY,EAAE,CAAC;YACzB,gDAAgD;YAChD,kBAAkB,GAAG,OAAO,CAAC,QAAQ,KAAK,OAAO,CAAC,YAAY,CAAC;QACjE,CAAC;aAAM,CAAC;YACN,wEAAwE;YACxE,MAAM,OAAO,GAAG,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;YACvD,IAAI,OAAO,EAAE,CAAC;gBACZ,MAAM,qBAAqB,GAAG,OAAO,CAAC,KAAK,GAAG,CAAC;oBAC7C,CAAC,CAAC,OAAO,CAAC,SAAS,GAAG,OAAO,CAAC,KAAK;oBACnC,CAAC,CAAC,CAAC,CAAC;gBACN,2EAA2E;gBAC3E,kBAAkB,GAAG,qBAAqB,GAAG,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,CAAC;YACtE,CAAC;QACH,CAAC;QAED,8BAA8B;QAC9B,IAAI,CAAC,cAAc,CAAC,OAAO,CAAC,QAAQ,EAAE,WAAW,CAAC,CAAC;QAEnD,+DAA+D;QAC/D,IAAI,aAAa,GAAG,GAAG,CAAC;QACxB,aAAa,IAAI,OAAO,CAAC,UAAU,GAAG,IAAI,CAAC,CAAC,wBAAwB;QACpE,IAAI,OAAO,CAAC,UAAU;YAAE,aAAa,IAAI,GAAG,CAAC;QAC7C,IAAI,OAAO,CAAC,gBAAgB;YAAE,aAAa,IAAI,GAAG,CAAC,CAAC,2BAA2B;QAC/E,aAAa,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,aAAa,CAAC,CAAC,CAAC;QAExD,OAAO;YACL,UAAU,EAAE,OAAO,CAAC,UAAU;YAC9B,aAAa,EAAE,OAAO,CAAC,gBAAgB;YACvC,UAAU,EAAE,OAAO,CAAC,UAAU;YAC9B,WAAW;YACX,kBAAkB;YAClB,aAAa,EAAE,IAAI,CAAC,KAAK,CAAC,aAAa,GAAG,GAAG,CAAC,GAAG,GAAG;YACpD,UAAU;SACX,CAAC;IACJ,CAAC;IAED,gDAAgD;IACxC,cAAc,CAAC,QAAgB,EAAE,OAAgB;QACvD,MAAM,QAAQ,GAAG,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;QAChD,IAAI,QAAQ,EAAE,CAAC;YACb,QAAQ,CAAC,KAAK,EAAE,CAAC;YACjB,IAAI,OAAO;gBAAE,QAAQ,CAAC,SAAS,EAAE,CAAC;QACpC,CAAC;aAAM,CAAC;YACN,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,QAAQ,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,SAAS,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;QAC3E,CAAC;IACH,CAAC;IAED,6CAA6C;IAC7C,eAAe;QACb,MAAM,MAAM,GAA2D,EAAE,CAAC;QAC1E,KAAK,MAAM,CAAC,IAAI,EAAE,OAAO,CAAC,IAAI,IAAI,CAAC,WAAW,EAAE,CAAC;YAC/C,MAAM,CAAC,IAAI,CAAC,GAAG;gBACb,KAAK,EAAE,OAAO,CAAC,KAAK;gBACpB,WAAW,EAAE,OAAO,CAAC,KAAK,GAAG,CAAC;oBAC5B,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,OAAO,CAAC,SAAS,GAAG,OAAO,CAAC,KAAK,CAAC,GAAG,GAAG,CAAC,GAAG,GAAG;oBAC7D,CAAC,CAAC,CAAC;aACN,CAAC;QACJ,CAAC;QACD,OAAO,MAAM,CAAC;IAChB,CAAC;CACF;AAlFD,4CAkFC;AAED;;;;;GAKG;AACH,MAAa,iBAAiB;IAC5B,sDAAsD;IAC9C,sBAAsB,GAA0D,IAAI,GAAG,EAAE,CAAC;IAElG;;OAEG;IACH,QAAQ,CAAC,OAQR;QACC,eAAe;QACf,MAAM,WAAW,GAAG,OAAO,CAAC,YAAY,GAAG,CAAC;YAC1C,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,cAAc,GAAG,OAAO,CAAC,YAAY,CAAC;YACrD,CAAC,CAAC,CAAC,CAAC;QAEN,kBAAkB;QAClB,MAAM,UAAU,GAAG,OAAO,CAAC,YAAY,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,gBAAgB,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC;QAErF,0BAA0B;QAC1B,MAAM,YAAY,GAAG,CACnB,WAAW,GAAG,GAAG;YACjB,UAAU,GAAG,GAAG;YAChB,CAAC,OAAO,CAAC,aAAa,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAClC,CAAC;QAEF,sBAAsB;QACtB,MAAM,OAAO,GAAG,WAAW,GAAG,GAAG,IAAI,UAAU,GAAG,GAAG,CAAC;QAEtD,OAAO;YACL,YAAY,EAAE,OAAO,CAAC,YAAY;YAClC,WAAW,EAAE,IAAI,CAAC,KAAK,CAAC,WAAW,GAAG,GAAG,CAAC,GAAG,GAAG;YAChD,YAAY,EAAE,OAAO,CAAC,YAAY;YAClC,YAAY,EAAE,OAAO,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YAC9C,cAAc,EAAE,OAAO,CAAC,aAAa;YACrC,YAAY,EAAE,IAAI,CAAC,KAAK,CAAC,YAAY,GAAG,GAAG,CAAC,GAAG,GAAG;YAClD,OAAO;YACP,gBAAgB,EAAE,OAAO,CAAC,gBAAgB;SAC3C,CAAC;IACJ,CAAC;IAED;;;OAGG;IACH,WAAW,CAAC,WAAmB,EAAE,MAAe;QAC9C,IAAI,CAAC,sBAAsB,CAAC,GAAG,CAAC,WAAW,EAAE,EAAE,MAAM,EAAE,UAAU,EAAE,KAAK,EAAE,CAAC,CAAC;IAC9E,CAAC;IAED;;OAEG;IACH,oBAAoB,CAAC,WAAmB;QACtC,MAAM,KAAK,GAAG,IAAI,CAAC,sBAAsB,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC;QAC3D,IAAI,KAAK;YAAE,KAAK,CAAC,UAAU,GAAG,IAAI,CAAC;IACrC,CAAC;IAED;;OAEG;IACH,kBAAkB,CAAC,WAAmB;QACpC,OAAO,IAAI,CAAC,sBAAsB,CAAC,GAAG,CAAC,WAAW,CAAC,EAAE,UAAU,IAAI,KAAK,CAAC;IAC3E,CAAC;IAED;;OAEG;IACH,kBAAkB;QAChB,MAAM,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,sBAAsB,CAAC,MAAM,EAAE,CAAC,CAAC;QACjE,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO,CAAC,CAAC;QACnC,MAAM,UAAU,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC;QAC9D,OAAO,IAAI,CAAC,KAAK,CAAC,CAAC,UAAU,GAAG,OAAO,CAAC,MAAM,CAAC,GAAG,GAAG,CAAC,GAAG,GAAG,CAAC;IAC/D,CAAC;CACF;AA9ED,8CA8EC"}

package/dist/evaluator/feedback.d.ts

@@ -0,0 +1,66 @@
+import { ImplicitFeedback, SignalType } from '../types';
+/**
+ * ImplicitFeedbackEngine — captures and interprets implicit user signals.
+ *
+ * Instead of relying on explicit "thumbs up/down", this engine
+ * detects subtle signals from user behavior to infer satisfaction.
+ *
+ * This is the key differentiator of AgentOS: it learns from
+ * what users DO, not just what they SAY.
+ *
+ * Signal rules (based on DESIGN.md §6.3):
+ * - user_deleted_code: User deleted what agent wrote → strong negative (-0.8)
+ * - user_modified_output: User modified agent's output → moderate negative (-0.5)
+ * - user_repeated_instruction: User repeated same command → mild negative (-0.3)
+ * - user_immediate_continue: User immediately continued without edit → positive (+0.3)
+ * - user_used_result: User referenced agent's output later → strong positive (+0.7)
+ * - user_silence_followed_by_praise: Gap then "谢谢" → mild positive (+0.2)
+ * - user_interrupted: User stopped agent mid-execution → negative (-0.6)
+ * - agent_self_corrected: Agent caught its own mistake → mild positive for agent (+0.3)
+ */
+export declare class ImplicitFeedbackEngine {
+    private feedbackLog;
+    /**
+     * Record an implicit feedback signal.
+     *
+     * @param signal - Type of implicit signal detected
+     * @param sessionId - Session where signal was detected
+     * @param operationId - Related tool call or request ID
+     * @param confidence - How confident we are about this interpretation (0-1)
+     * @param source - Where the signal was detected (audit_log, message_pattern, diff)
+     */
+    record(signal: SignalType, sessionId: string, operationId?: string, confidence?: number, source?: string): ImplicitFeedback;
+    /**
+     * Get the default strength for a signal type.
+     */
+    private getSignalStrength;
+    /**
+     * Compute the aggregate satisfaction score from all feedback.
+     *
+     * Weighted by confidence and recency (newer signals matter more).
+     * Returns a score from -1.0 (very unhappy) to +1.0 (very happy).
+     */
+    getSatisfactionScore(sessionId?: string, recentHours?: number): number;
+    /**
+     * Get feedback events, optionally filtered.
+     */
+    query(filter?: {
+        signal?: SignalType;
+        sessionId?: string;
+        minStrength?: number;
+        maxStrength?: number;
+        since?: number;
+        limit?: number;
+    }): ImplicitFeedback[];
+    /**
+     * Get feedback summary statistics.
+     */
+    stats(): {
+        totalSignals: number;
+        positiveSignals: number;
+        negativeSignals: number;
+        averageStrength: number;
+        mostCommonSignal: SignalType | null;
+    };
+}
+//# sourceMappingURL=feedback.d.ts.map

package/dist/evaluator/feedback.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"feedback.d.ts","sourceRoot":"","sources":["../../src/evaluator/feedback.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,gBAAgB,EAAE,UAAU,EAAE,MAAM,UAAU,CAAC;AAUxD;;;;;;;;;;;;;;;;;;GAkBG;AACH,qBAAa,sBAAsB;IACjC,OAAO,CAAC,WAAW,CAA0B;IAE7C;;;;;;;;OAQG;IACH,MAAM,CACJ,MAAM,EAAE,UAAU,EAClB,SAAS,EAAE,MAAM,EACjB,WAAW,CAAC,EAAE,MAAM,EACpB,UAAU,SAAM,EAChB,MAAM,SAAkB,GACvB,gBAAgB;IAkBnB;;OAEG;IACH,OAAO,CAAC,iBAAiB;IAkBzB;;;;;OAKG;IACH,oBAAoB,CAAC,SAAS,CAAC,EAAE,MAAM,EAAE,WAAW,SAAK,GAAG,MAAM;IA+BlE;;OAEG;IACH,KAAK,CAAC,MAAM,GAAE;QACZ,MAAM,CAAC,EAAE,UAAU,CAAC;QACpB,SAAS,CAAC,EAAE,MAAM,CAAC;QACnB,WAAW,CAAC,EAAE,MAAM,CAAC;QACrB,WAAW,CAAC,EAAE,MAAM,CAAC;QACrB,KAAK,CAAC,EAAE,MAAM,CAAC;QACf,KAAK,CAAC,EAAE,MAAM,CAAC;KACX,GAAG,gBAAgB,EAAE;IAyB3B;;OAEG;IACH,KAAK,IAAI;QACP,YAAY,EAAE,MAAM,CAAC;QACrB,eAAe,EAAE,MAAM,CAAC;QACxB,eAAe,EAAE,MAAM,CAAC;QACxB,eAAe,EAAE,MAAM,CAAC;QACxB,gBAAgB,EAAE,UAAU,GAAG,IAAI,CAAC;KACrC;CA6BF"}

package/dist/evaluator/feedback.js

@@ -0,0 +1,195 @@
+"use strict";
+var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    var desc = Object.getOwnPropertyDescriptor(m, k);
+    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
+      desc = { enumerable: true, get: function() { return m[k]; } };
+    }
+    Object.defineProperty(o, k2, desc);
+}) : (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    o[k2] = m[k];
+}));
+var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
+    Object.defineProperty(o, "default", { enumerable: true, value: v });
+}) : function(o, v) {
+    o["default"] = v;
+});
+var __importStar = (this && this.__importStar) || (function () {
+    var ownKeys = function(o) {
+        ownKeys = Object.getOwnPropertyNames || function (o) {
+            var ar = [];
+            for (var k in o) if (Object.prototype.hasOwnProperty.call(o, k)) ar[ar.length] = k;
+            return ar;
+        };
+        return ownKeys(o);
+    };
+    return function (mod) {
+        if (mod && mod.__esModule) return mod;
+        var result = {};
+        if (mod != null) for (var k = ownKeys(mod), i = 0; i < k.length; i++) if (k[i] !== "default") __createBinding(result, mod, k[i]);
+        __setModuleDefault(result, mod);
+        return result;
+    };
+})();
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.ImplicitFeedbackEngine = void 0;
+const crypto = __importStar(require("crypto"));
+/**
+ * Generate a unique feedback ID.
+ */
+function generateFeedbackId() {
+    return `fb_${Date.now()}_${crypto.randomBytes(4).toString('hex')}`;
+}
+/**
+ * ImplicitFeedbackEngine — captures and interprets implicit user signals.
+ *
+ * Instead of relying on explicit "thumbs up/down", this engine
+ * detects subtle signals from user behavior to infer satisfaction.
+ *
+ * This is the key differentiator of AgentOS: it learns from
+ * what users DO, not just what they SAY.
+ *
+ * Signal rules (based on DESIGN.md §6.3):
+ * - user_deleted_code: User deleted what agent wrote → strong negative (-0.8)
+ * - user_modified_output: User modified agent's output → moderate negative (-0.5)
+ * - user_repeated_instruction: User repeated same command → mild negative (-0.3)
+ * - user_immediate_continue: User immediately continued without edit → positive (+0.3)
+ * - user_used_result: User referenced agent's output later → strong positive (+0.7)
+ * - user_silence_followed_by_praise: Gap then "谢谢" → mild positive (+0.2)
+ * - user_interrupted: User stopped agent mid-execution → negative (-0.6)
+ * - agent_self_corrected: Agent caught its own mistake → mild positive for agent (+0.3)
+ */
+class ImplicitFeedbackEngine {
+    feedbackLog = [];
+    /**
+     * Record an implicit feedback signal.
+     *
+     * @param signal - Type of implicit signal detected
+     * @param sessionId - Session where signal was detected
+     * @param operationId - Related tool call or request ID
+     * @param confidence - How confident we are about this interpretation (0-1)
+     * @param source - Where the signal was detected (audit_log, message_pattern, diff)
+     */
+    record(signal, sessionId, operationId, confidence = 0.8, source = 'auto-detected') {
+        const strength = this.getSignalStrength(signal);
+        const feedback = {
+            id: generateFeedbackId(),
+            timestamp: Date.now(),
+            signal,
+            strength,
+            confidence,
+            sessionId,
+            operationId,
+            source,
+        };
+        this.feedbackLog.push(feedback);
+        return feedback;
+    }
+    /**
+     * Get the default strength for a signal type.
+     */
+    getSignalStrength(signal) {
+        switch (signal) {
+            case 'user_deleted_code': return -0.8;
+            case 'user_interrupted': return -0.6;
+            case 'user_provided_correction': return -0.7;
+            case 'user_modified_output': return -0.5;
+            case 'user_repeated_instruction': return -0.3;
+            case 'user_ignored_result': return -0.4;
+            case 'user_silence_then_praise': return 0.2;
+            case 'user_immediate_continue': return 0.3;
+            case 'agent_self_corrected': return 0.3;
+            case 'user_explicit_approval': return 0.6;
+            case 'user_used_result': return 0.7;
+            case 'user_shared_output': return 0.8; // User shared agent output → strong positive
+            default: return 0;
+        }
+    }
+    /**
+     * Compute the aggregate satisfaction score from all feedback.
+     *
+     * Weighted by confidence and recency (newer signals matter more).
+     * Returns a score from -1.0 (very unhappy) to +1.0 (very happy).
+     */
+    getSatisfactionScore(sessionId, recentHours = 24) {
+        let relevant = this.feedbackLog;
+        if (sessionId) {
+            relevant = relevant.filter((f) => f.sessionId === sessionId);
+        }
+        // Only consider recent signals
+        const cutoff = Date.now() - recentHours * 60 * 60 * 1000;
+        relevant = relevant.filter((f) => f.timestamp >= cutoff);
+        if (relevant.length === 0)
+            return 0;
+        let weightedSum = 0;
+        let totalWeight = 0;
+        for (const fb of relevant) {
+            // Recency weight: newer = more important
+            const ageHours = (Date.now() - fb.timestamp) / (60 * 60 * 1000);
+            const recencyWeight = Math.max(0.1, 1 - ageHours / recentHours);
+            const weight = fb.confidence * recencyWeight;
+            weightedSum += fb.strength * weight;
+            totalWeight += weight;
+        }
+        return totalWeight > 0
+            ? Math.round((weightedSum / totalWeight) * 100) / 100
+            : 0;
+    }
+    /**
+     * Get feedback events, optionally filtered.
+     */
+    query(filter = {}) {
+        let results = this.feedbackLog;
+        if (filter.signal) {
+            results = results.filter((f) => f.signal === filter.signal);
+        }
+        if (filter.sessionId) {
+            results = results.filter((f) => f.sessionId === filter.sessionId);
+        }
+        if (filter.minStrength !== undefined) {
+            results = results.filter((f) => f.strength >= filter.minStrength);
+        }
+        if (filter.maxStrength !== undefined) {
+            results = results.filter((f) => f.strength <= filter.maxStrength);
+        }
+        if (filter.since !== undefined) {
+            results = results.filter((f) => f.timestamp >= filter.since);
+        }
+        results.sort((a, b) => b.timestamp - a.timestamp);
+        const limit = filter.limit ?? 50;
+        return results.slice(0, limit);
+    }
+    /**
+     * Get feedback summary statistics.
+     */
+    stats() {
+        const positive = this.feedbackLog.filter((f) => f.strength > 0);
+        const negative = this.feedbackLog.filter((f) => f.strength < 0);
+        const avgStrength = this.feedbackLog.length > 0
+            ? this.feedbackLog.reduce((s, f) => s + f.strength, 0) / this.feedbackLog.length
+            : 0;
+        // Most common signal
+        const counts = new Map();
+        for (const fb of this.feedbackLog) {
+            counts.set(fb.signal, (counts.get(fb.signal) || 0) + 1);
+        }
+        let mostCommon = null;
+        let maxCount = 0;
+        for (const [sig, count] of counts) {
+            if (count > maxCount) {
+                maxCount = count;
+                mostCommon = sig;
+            }
+        }
+        return {
+            totalSignals: this.feedbackLog.length,
+            positiveSignals: positive.length,
+            negativeSignals: negative.length,
+            averageStrength: Math.round(avgStrength * 100) / 100,
+            mostCommonSignal: mostCommon,
+        };
+    }
+}
+exports.ImplicitFeedbackEngine = ImplicitFeedbackEngine;
+//# sourceMappingURL=feedback.js.map

package/dist/evaluator/feedback.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"feedback.js","sourceRoot":"","sources":["../../src/evaluator/feedback.ts"],"names":[],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AACA,+CAAiC;AAEjC;;GAEG;AACH,SAAS,kBAAkB;IACzB,OAAO,MAAM,IAAI,CAAC,GAAG,EAAE,IAAI,MAAM,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;AACrE,CAAC;AAED;;;;;;;;;;;;;;;;;;GAkBG;AACH,MAAa,sBAAsB;IACzB,WAAW,GAAuB,EAAE,CAAC;IAE7C;;;;;;;;OAQG;IACH,MAAM,CACJ,MAAkB,EAClB,SAAiB,EACjB,WAAoB,EACpB,UAAU,GAAG,GAAG,EAChB,MAAM,GAAG,eAAe;QAExB,MAAM,QAAQ,GAAG,IAAI,CAAC,iBAAiB,CAAC,MAAM,CAAC,CAAC;QAEhD,MAAM,QAAQ,GAAqB;YACjC,EAAE,EAAE,kBAAkB,EAAE;YACxB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;YACrB,MAAM;YACN,QAAQ;YACR,UAAU;YACV,SAAS;YACT,WAAW;YACX,MAAM;SACP,CAAC;QAEF,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QAChC,OAAO,QAAQ,CAAC;IAClB,CAAC;IAED;;OAEG;IACK,iBAAiB,CAAC,MAAkB;QAC1C,QAAQ,MAAM,EAAE,CAAC;YACf,KAAK,mBAAmB,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC;YACtC,KAAK,kBAAkB,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC;YACrC,KAAK,0BAA0B,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC;YAC7C,KAAK,sBAAsB,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC;YACzC,KAAK,2BAA2B,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC;YAC9C,KAAK,qBAAqB,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC;YACxC,KAAK,0BAA0B,CAAC,CAAC,OAAO,GAAG,CAAC;YAC5C,KAAK,yBAAyB,CAAC,CAAC,OAAO,GAAG,CAAC;YAC3C,KAAK,sBAAsB,CAAC,CAAC,OAAO,GAAG,CAAC;YACxC,KAAK,wBAAwB,CAAC,CAAC,OAAO,GAAG,CAAC;YAC1C,KAAK,kBAAkB,CAAC,CAAC,OAAO,GAAG,CAAC;YACpC,KAAK,oBAAoB,CAAC,CAAC,OAAO,GAAG,CAAC,CAAC,6CAA6C;YACpF,OAAO,CAAC,CAAC,OAAO,CAAC,CAAC;QACpB,CAAC;IACH,CAAC;IAED;;;;;OAKG;IACH,oBAAoB,CAAC,SAAkB,EAAE,WAAW,GAAG,EAAE;QACvD,IAAI,QAAQ,GAAG,IAAI,CAAC,WAAW,CAAC;QAEhC,IAAI,SAAS,EAAE,CAAC;YACd,QAAQ,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,KAAK,SAAS,CAAC,CAAC;QAC/D,CAAC;QAED,+BAA+B;QAC/B,MAAM,MAAM,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,WAAW,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC;QACzD,QAAQ,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,IAAI,MAAM,CAAC,CAAC;QAEzD,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO,CAAC,CAAC;QAEpC,IAAI,WAAW,GAAG,CAAC,CAAC;QACpB,IAAI,WAAW,GAAG,CAAC,CAAC;QAEpB,KAAK,MAAM,EAAE,IAAI,QAAQ,EAAE,CAAC;YAC1B,yCAAyC;YACzC,MAAM,QAAQ,GAAG,CAAC,IAAI,CAAC,GAAG,EAAE,GAAG,EAAE,CAAC,SAAS,CAAC,GAAG,CAAC,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,CAAC;YAChE,MAAM,aAAa,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,GAAG,QAAQ,GAAG,WAAW,CAAC,CAAC;YAEhE,MAAM,MAAM,GAAG,EAAE,CAAC,UAAU,GAAG,aAAa,CAAC;YAC7C,WAAW,IAAI,EAAE,CAAC,QAAQ,GAAG,MAAM,CAAC;YACpC,WAAW,IAAI,MAAM,CAAC;QACxB,CAAC;QAED,OAAO,WAAW,GAAG,CAAC;YACpB,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,WAAW,GAAG,WAAW,CAAC,GAAG,GAAG,CAAC,GAAG,GAAG;YACrD,CAAC,CAAC,CAAC,CAAC;IACR,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,SAOF,EAAE;QACJ,IAAI,OAAO,GAAG,IAAI,CAAC,WAAW,CAAC;QAE/B,IAAI,MAAM,CAAC,MAAM,EAAE,CAAC;YAClB,OAAO,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,MAAM,CAAC,MAAM,CAAC,CAAC;QAC9D,CAAC;QACD,IAAI,MAAM,CAAC,SAAS,EAAE,CAAC;YACrB,OAAO,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,KAAK,MAAM,CAAC,SAAS,CAAC,CAAC;QACpE,CAAC;QACD,IAAI,MAAM,CAAC,WAAW,KAAK,SAAS,EAAE,CAAC;YACrC,OAAO,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,IAAI,MAAM,CAAC,WAAY,CAAC,CAAC;QACrE,CAAC;QACD,IAAI,MAAM,CAAC,WAAW,KAAK,SAAS,EAAE,CAAC;YACrC,OAAO,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,IAAI,MAAM,CAAC,WAAY,CAAC,CAAC;QACrE,CAAC;QACD,IAAI,MAAM,CAAC,KAAK,KAAK,SAAS,EAAE,CAAC;YAC/B,OAAO,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,IAAI,MAAM,CAAC,KAAM,CAAC,CAAC;QAChE,CAAC;QAED,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,GAAG,CAAC,CAAC,SAAS,CAAC,CAAC;QAElD,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,IAAI,EAAE,CAAC;QACjC,OAAO,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC;IACjC,CAAC;IAED;;OAEG;IACH,KAAK;QAOH,MAAM,QAAQ,GAAG,IAAI,CAAC,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,GAAG,CAAC,CAAC,CAAC;QAChE,MAAM,QAAQ,GAAG,IAAI,CAAC,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,GAAG,CAAC,CAAC,CAAC;QAChE,MAAM,WAAW,GAAG,IAAI,CAAC,WAAW,CAAC,MAAM,GAAG,CAAC;YAC7C,CAAC,CAAC,IAAI,CAAC,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,WAAW,CAAC,MAAM;YAChF,CAAC,CAAC,CAAC,CAAC;QAEN,qBAAqB;QACrB,MAAM,MAAM,GAAG,IAAI,GAAG,EAAsB,CAAC;QAC7C,KAAK,MAAM,EAAE,IAAI,IAAI,CAAC,WAAW,EAAE,CAAC;YAClC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;QAC1D,CAAC;QACD,IAAI,UAAU,GAAsB,IAAI,CAAC;QACzC,IAAI,QAAQ,GAAG,CAAC,CAAC;QACjB,KAAK,MAAM,CAAC,GAAG,EAAE,KAAK,CAAC,IAAI,MAAM,EAAE,CAAC;YAClC,IAAI,KAAK,GAAG,QAAQ,EAAE,CAAC;gBACrB,QAAQ,GAAG,KAAK,CAAC;gBACjB,UAAU,GAAG,GAAG,CAAC;YACnB,CAAC;QACH,CAAC;QAED,OAAO;YACL,YAAY,EAAE,IAAI,CAAC,WAAW,CAAC,MAAM;YACrC,eAAe,EAAE,QAAQ,CAAC,MAAM;YAChC,eAAe,EAAE,QAAQ,CAAC,MAAM;YAChC,eAAe,EAAE,IAAI,CAAC,KAAK,CAAC,WAAW,GAAG,GAAG,CAAC,GAAG,GAAG;YACpD,gBAAgB,EAAE,UAAU;SAC7B,CAAC;IACJ,CAAC;CACF;AAvKD,wDAuKC"}

package/dist/evaluator/profiler.d.ts

@@ -0,0 +1,53 @@
+import { PreExecMetrics, RuntimeMetrics, PostExecMetrics } from '../types';
+import { ImplicitFeedbackEngine } from './feedback';
+/**
+ * Agent quality profile — accumulated across all evaluations.
+ */
+export interface AgentProfile {
+    /** Overall quality score (0-100) */
+    overallScore: number;
+    /** Number of operations evaluated */
+    totalOps: number;
+    /** Score breakdown by metric category */
+    breakdown: {
+        preExec: number;
+        runtime: number;
+        postExec: number;
+        userSatisfaction: number;
+    };
+    /** Trend data */
+    trends: {
+        improving: boolean;
+        recentOps: number;
+        recentScore: number;
+    };
+    /** Areas needing attention */
+    warnings: string[];
+    /** Kudos for good patterns */
+    strengths: string[];
+}
+/**
+ * AgentProfiler — builds and maintains the agent's quality profile.
+ *
+ * Aggregates PreExec + Runtime + PostExec metrics and
+ * ImplicitFeedback to produce a composite quality score
+ * that improves over time through self-correction.
+ */
+export declare class AgentProfiler {
+    private feedbackEngine;
+    private preMetrics;
+    private runMetrics;
+    private postMetrics;
+    private sessionScores;
+    constructor(feedbackEngine: ImplicitFeedbackEngine);
+    /**
+     * Record a complete evaluation cycle for one tool call.
+     */
+    recordCycle(sessionId: string, pre: PreExecMetrics, run: RuntimeMetrics, post: PostExecMetrics): void;
+    /**
+     * Build the current agent profile.
+     */
+    getProfile(sessionId?: string): AgentProfile;
+    private average;
+}
+//# sourceMappingURL=profiler.d.ts.map

package/dist/evaluator/profiler.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"profiler.d.ts","sourceRoot":"","sources":["../../src/evaluator/profiler.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,cAAc,EACd,cAAc,EACd,eAAe,EAChB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,sBAAsB,EAAE,MAAM,YAAY,CAAC;AAEpD;;GAEG;AACH,MAAM,WAAW,YAAY;IAC3B,oCAAoC;IACpC,YAAY,EAAE,MAAM,CAAC;IACrB,qCAAqC;IACrC,QAAQ,EAAE,MAAM,CAAC;IACjB,yCAAyC;IACzC,SAAS,EAAE;QACT,OAAO,EAAE,MAAM,CAAC;QAChB,OAAO,EAAE,MAAM,CAAC;QAChB,QAAQ,EAAE,MAAM,CAAC;QACjB,gBAAgB,EAAE,MAAM,CAAC;KAC1B,CAAC;IACF,iBAAiB;IACjB,MAAM,EAAE;QACN,SAAS,EAAE,OAAO,CAAC;QACnB,SAAS,EAAE,MAAM,CAAC;QAClB,WAAW,EAAE,MAAM,CAAC;KACrB,CAAC;IACF,8BAA8B;IAC9B,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,8BAA8B;IAC9B,SAAS,EAAE,MAAM,EAAE,CAAC;CACrB;AAED;;;;;;GAMG;AACH,qBAAa,aAAa;IACxB,OAAO,CAAC,cAAc,CAAyB;IAE/C,OAAO,CAAC,UAAU,CAAwB;IAC1C,OAAO,CAAC,UAAU,CAAwB;IAC1C,OAAO,CAAC,WAAW,CAAyB;IAC5C,OAAO,CAAC,aAAa,CAAoC;gBAE7C,cAAc,EAAE,sBAAsB;IAIlD;;OAEG;IACH,WAAW,CAAC,SAAS,EAAE,MAAM,EAAE,GAAG,EAAE,cAAc,EAAE,GAAG,EAAE,cAAc,EAAE,IAAI,EAAE,eAAe,GAAG,IAAI;IAWrG;;OAEG;IACH,UAAU,CAAC,SAAS,CAAC,EAAE,MAAM,GAAG,YAAY;IA0F5C,OAAO,CAAC,OAAO;CAKhB"}