sentinel-agentos 0.1.0

package/dist/core.d.ts

@@ -0,0 +1,139 @@
+import { AgentOSConfig, GuardConfig, Snapshot, AuditEntry, PreExecMetrics, RuntimeMetrics, PostExecMetrics } from './types';
+import { SchemaGate } from './guard/schema-gate';
+import { RiskGate } from './guard/risk-gate';
+import { SnapshotGate, VerifyGate } from './guard/snapshot-verify';
+import { AuditLog } from './guard/audit-log';
+import { WorkingMemory } from './memory/working';
+import { EpisodicMemory } from './memory/episodic';
+import { SemanticMemoryStore } from './memory/semantic';
+import { PreExecEvaluator, RuntimeEvaluator, PostExecEvaluator } from './evaluator/exec-evaluator';
+import { ImplicitFeedbackEngine } from './evaluator/feedback';
+import { AgentProfiler } from './evaluator/profiler';
+import type { AgentProfile } from './evaluator/profiler';
+/**
+ * AgentOS — the complete AI Agent Operating System.
+ *
+ * Architecture:
+ * ```
+ * User Request
+ *     ↓
+ * ┌─────────────┐
+ * │ Memory Layer │ ← Semantic + Episodic + Working memory
+ * ├─────────────┤
+ * │ Guard Layer  │ ← Schema → Risk → Snapshot
+ * ├─────────────┤         ↓
+ * │ Execute      │ ← Tool call execution
+ * ├─────────────┤         ↓
+ * │ Verify Layer │ ← Snapshot diff → Verify checks
+ * ├─────────────┤         ↓
+ * │ Audit Log    │ ← Immutable operation record
+ * ├─────────────┤
+ * │ Evaluator    │ ← Three-phase metrics + feedback
+ * └─────────────┘
+ * ```
+ */
+export declare class AgentOS {
+    private config;
+    readonly memory: {
+        working: WorkingMemory;
+        episodic: EpisodicMemory;
+        semantic: SemanticMemoryStore;
+    };
+    readonly guard: {
+        schema: SchemaGate;
+        risk: RiskGate;
+        snapshot: SnapshotGate;
+        verify: VerifyGate;
+        audit: AuditLog;
+    };
+    readonly evaluator: {
+        preExec: PreExecEvaluator;
+        runtime: RuntimeEvaluator;
+        postExec: PostExecEvaluator;
+        feedback: ImplicitFeedbackEngine;
+        profiler: AgentProfiler;
+    };
+    constructor(config?: Partial<AgentOSConfig>);
+    /**
+     * Get the current AgentOS configuration.
+     */
+    getConfig(): Readonly<AgentOSConfig>;
+    /**
+     * Full pipeline: process a tool call through all layers.
+     *
+     * This is the main AgentOS orchestration method.
+     * In production this would be called by the agent runtime
+     * before/after every tool call.
+     */
+    executePipeline(options: {
+        sessionId: string;
+        agentId: string;
+        toolName: string;
+        parameters: Record<string, unknown>;
+        affectedFiles?: string[];
+        guardConfig?: GuardConfig;
+    }): {
+        preExec: PreExecMetrics;
+        runtime?: RuntimeMetrics;
+        postExec?: PostExecMetrics;
+        snapshot: Snapshot | null;
+        auditEntry?: AuditEntry;
+        profile: AgentProfile;
+    };
+    /**
+     * Complete the pipeline after tool execution.
+     *
+     * Called by the runtime after the tool call completes
+     * (or fails, or times out).
+     */
+    completeExecution(options: {
+        sessionId: string;
+        agentId: string;
+        toolName: string;
+        toolParameters: Record<string, unknown>;
+        toolResult: unknown;
+        snapshot: Snapshot | null;
+        startTime: number;
+        endTime: number;
+        retryCount: number;
+        wasSelfCorrected: boolean;
+        hadTimeout: boolean;
+        userAccepted: boolean;
+        userProvidedEdit: boolean;
+        resultWasUsed: boolean;
+    }): {
+        runtime: RuntimeMetrics;
+        postExec: PostExecMetrics;
+        auditEntry: AuditEntry;
+        profile: AgentProfile;
+    };
+    /**
+     * Record implicit user feedback.
+     */
+    recordFeedback(signal: Parameters<ImplicitFeedbackEngine['record']>[0], sessionId: string, operationId?: string, confidence?: number, source?: string): void;
+    /**
+     * Inject memory context at session startup.
+     *
+     * Call this at the beginning of every session to load
+     * semantic + episodic context into the session prompt.
+     */
+    injectContext(): string;
+    /**
+     * End current session — promote important events to episodic,
+     * clear working memory, and save state.
+     */
+    endSession(sessionId: string): void;
+    /**
+     * Get audit statistics.
+     */
+    getAuditStats(): ReturnType<AuditLog['stats']>;
+    /**
+     * Get the current agent quality profile.
+     */
+    getProfile(sessionId?: string): AgentProfile;
+    /**
+     * Get a summarized status report.
+     */
+    statusReport(): string;
+}
+//# sourceMappingURL=core.d.ts.map

package/dist/core.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"core.d.ts","sourceRoot":"","sources":["../src/core.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,aAAa,EACb,WAAW,EACX,QAAQ,EACR,UAAU,EACV,cAAc,EACd,cAAc,EACd,eAAe,EAChB,MAAM,SAAS,CAAC;AACjB,OAAO,EAAE,UAAU,EAAE,MAAM,qBAAqB,CAAC;AACjD,OAAO,EAAE,QAAQ,EAAE,MAAM,mBAAmB,CAAC;AAC7C,OAAO,EAAE,YAAY,EAAE,UAAU,EAAE,MAAM,yBAAyB,CAAC;AACnE,OAAO,EAAE,QAAQ,EAAE,MAAM,mBAAmB,CAAC;AAC7C,OAAO,EAAE,aAAa,EAAE,MAAM,kBAAkB,CAAC;AACjD,OAAO,EAAE,cAAc,EAAE,MAAM,mBAAmB,CAAC;AACnD,OAAO,EAAE,mBAAmB,EAAE,MAAM,mBAAmB,CAAC;AACxD,OAAO,EAAE,gBAAgB,EAAE,gBAAgB,EAAE,iBAAiB,EAAE,MAAM,4BAA4B,CAAC;AACnG,OAAO,EAAE,sBAAsB,EAAE,MAAM,sBAAsB,CAAC;AAC9D,OAAO,EAAE,aAAa,EAAE,MAAM,sBAAsB,CAAC;AACrD,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,sBAAsB,CAAC;AAEzD;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,qBAAa,OAAO;IAClB,OAAO,CAAC,MAAM,CAAgB;IAG9B,QAAQ,CAAC,MAAM,EAAE;QACf,OAAO,EAAE,aAAa,CAAC;QACvB,QAAQ,EAAE,cAAc,CAAC;QACzB,QAAQ,EAAE,mBAAmB,CAAC;KAC/B,CAAC;IAGF,QAAQ,CAAC,KAAK,EAAE;QACd,MAAM,EAAE,UAAU,CAAC;QACnB,IAAI,EAAE,QAAQ,CAAC;QACf,QAAQ,EAAE,YAAY,CAAC;QACvB,MAAM,EAAE,UAAU,CAAC;QACnB,KAAK,EAAE,QAAQ,CAAC;KACjB,CAAC;IAGF,QAAQ,CAAC,SAAS,EAAE;QAClB,OAAO,EAAE,gBAAgB,CAAC;QAC1B,OAAO,EAAE,gBAAgB,CAAC;QAC1B,QAAQ,EAAE,iBAAiB,CAAC;QAC5B,QAAQ,EAAE,sBAAsB,CAAC;QACjC,QAAQ,EAAE,aAAa,CAAC;KACzB,CAAC;gBAEU,MAAM,CAAC,EAAE,OAAO,CAAC,aAAa,CAAC;IAqD3C;;OAEG;IACH,SAAS,IAAI,QAAQ,CAAC,aAAa,CAAC;IAIpC;;;;;;OAMG;IACH,eAAe,CAAC,OAAO,EAAE;QACvB,SAAS,EAAE,MAAM,CAAC;QAClB,OAAO,EAAE,MAAM,CAAC;QAChB,QAAQ,EAAE,MAAM,CAAC;QACjB,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;QACpC,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;QACzB,WAAW,CAAC,EAAE,WAAW,CAAC;KAC3B,GAAG;QACF,OAAO,EAAE,cAAc,CAAC;QACxB,OAAO,CAAC,EAAE,cAAc,CAAC;QACzB,QAAQ,CAAC,EAAE,eAAe,CAAC;QAC3B,QAAQ,EAAE,QAAQ,GAAG,IAAI,CAAC;QAC1B,UAAU,CAAC,EAAE,UAAU,CAAC;QACxB,OAAO,EAAE,YAAY,CAAC;KACvB;IAyBD;;;;;OAKG;IACH,iBAAiB,CAAC,OAAO,EAAE;QACzB,SAAS,EAAE,MAAM,CAAC;QAClB,OAAO,EAAE,MAAM,CAAC;QAChB,QAAQ,EAAE,MAAM,CAAC;QACjB,cAAc,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;QACxC,UAAU,EAAE,OAAO,CAAC;QACpB,QAAQ,EAAE,QAAQ,GAAG,IAAI,CAAC;QAC1B,SAAS,EAAE,MAAM,CAAC;QAClB,OAAO,EAAE,MAAM,CAAC;QAChB,UAAU,EAAE,MAAM,CAAC;QACnB,gBAAgB,EAAE,OAAO,CAAC;QAC1B,UAAU,EAAE,OAAO,CAAC;QACpB,YAAY,EAAE,OAAO,CAAC;QACtB,gBAAgB,EAAE,OAAO,CAAC;QAC1B,aAAa,EAAE,OAAO,CAAC;KACxB,GAAG;QACF,OAAO,EAAE,cAAc,CAAC;QACxB,QAAQ,EAAE,eAAe,CAAC;QAC1B,UAAU,EAAE,UAAU,CAAC;QACvB,OAAO,EAAE,YAAY,CAAC;KACvB;IAmED;;OAEG;IACH,cAAc,CACZ,MAAM,EAAE,UAAU,CAAC,sBAAsB,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,EACvD,SAAS,EAAE,MAAM,EACjB,WAAW,CAAC,EAAE,MAAM,EACpB,UAAU,CAAC,EAAE,MAAM,EACnB,MAAM,CAAC,EAAE,MAAM,GACd,IAAI;IAIP;;;;;OAKG;IACH,aAAa,IAAI,MAAM;IAYvB;;;OAGG;IACH,UAAU,CAAC,SAAS,EAAE,MAAM,GAAG,IAAI;IA0BnC;;OAEG;IACH,aAAa,IAAI,UAAU,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;IAI9C;;OAEG;IACH,UAAU,CAAC,SAAS,CAAC,EAAE,MAAM,GAAG,YAAY;IAI5C;;OAEG;IACH,YAAY,IAAI,MAAM;CAgCvB"}

package/dist/core.js

@@ -0,0 +1,247 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.AgentOS = void 0;
+const schema_gate_1 = require("./guard/schema-gate");
+const risk_gate_1 = require("./guard/risk-gate");
+const snapshot_verify_1 = require("./guard/snapshot-verify");
+const audit_log_1 = require("./guard/audit-log");
+const working_1 = require("./memory/working");
+const episodic_1 = require("./memory/episodic");
+const semantic_1 = require("./memory/semantic");
+const exec_evaluator_1 = require("./evaluator/exec-evaluator");
+const feedback_1 = require("./evaluator/feedback");
+const profiler_1 = require("./evaluator/profiler");
+/**
+ * AgentOS — the complete AI Agent Operating System.
+ *
+ * Architecture:
+ * ```
+ * User Request
+ *     ↓
+ * ┌─────────────┐
+ * │ Memory Layer │ ← Semantic + Episodic + Working memory
+ * ├─────────────┤
+ * │ Guard Layer  │ ← Schema → Risk → Snapshot
+ * ├─────────────┤         ↓
+ * │ Execute      │ ← Tool call execution
+ * ├─────────────┤         ↓
+ * │ Verify Layer │ ← Snapshot diff → Verify checks
+ * ├─────────────┤         ↓
+ * │ Audit Log    │ ← Immutable operation record
+ * ├─────────────┤
+ * │ Evaluator    │ ← Three-phase metrics + feedback
+ * └─────────────┘
+ * ```
+ */
+class AgentOS {
+    config;
+    // Memory Layer
+    memory;
+    // Guard Layer
+    guard;
+    // Evaluator Layer
+    evaluator;
+    constructor(config) {
+        this.config = {
+            workspaceRoot: process.cwd(),
+            maxWorkingTokens: 50000,
+            maxEpisodicSizeKb: 500,
+            guardConfig: {},
+            ...config,
+        };
+        // --- Memory Layer Init ---
+        const semantic = new semantic_1.SemanticMemoryStore();
+        semantic.enablePersistence(this.config.workspaceRoot);
+        const episodic = new episodic_1.EpisodicMemory(this.config.maxEpisodicSizeKb);
+        episodic.enablePersistence(this.config.workspaceRoot);
+        const working = new working_1.WorkingMemory(this.config.maxWorkingTokens);
+        this.memory = { working, episodic, semantic };
+        // --- Guard Layer Init ---
+        const schema = new schema_gate_1.SchemaGate();
+        const risk = new risk_gate_1.RiskGate();
+        const snapshot = new snapshot_verify_1.SnapshotGate(this.config.workspaceRoot);
+        const verify = new snapshot_verify_1.VerifyGate(this.config.workspaceRoot);
+        const audit = new audit_log_1.AuditLog(this.config.workspaceRoot, schema, risk);
+        this.guard = { schema, risk, snapshot, verify, audit };
+        // --- Evaluator Layer Init ---
+        const preExecEval = new exec_evaluator_1.PreExecEvaluator(schema, risk, working);
+        const runtimeEval = new exec_evaluator_1.RuntimeEvaluator();
+        const postExecEval = new exec_evaluator_1.PostExecEvaluator();
+        const feedbackEngine = new feedback_1.ImplicitFeedbackEngine();
+        const profiler = new profiler_1.AgentProfiler(feedbackEngine);
+        this.evaluator = {
+            preExec: preExecEval,
+            runtime: runtimeEval,
+            postExec: postExecEval,
+            feedback: feedbackEngine,
+            profiler,
+        };
+    }
+    /**
+     * Get the current AgentOS configuration.
+     */
+    getConfig() {
+        return this.config;
+    }
+    /**
+     * Full pipeline: process a tool call through all layers.
+     *
+     * This is the main AgentOS orchestration method.
+     * In production this would be called by the agent runtime
+     * before/after every tool call.
+     */
+    executePipeline(options) {
+        const { sessionId, toolName, parameters, affectedFiles } = options;
+        // --- Phase 1: Pre-exec evaluation ---
+        const preExec = this.evaluator.preExec.evaluate(toolName, parameters);
+        this.memory.working.addMessage('tool', JSON.stringify(parameters));
+        // --- Phase 2: Snapshot before execution ---
+        const snapshot = this.guard.snapshot.takeSnapshot(`call_${Date.now()}`, toolName, affectedFiles ?? [], 'file');
+        // --- Phase 3: Execute (delegated to runtime — here we return pre-exec state) ---
+        // Return the pre-exec state so the runtime can complete
+        return {
+            preExec,
+            snapshot,
+            profile: this.evaluator.profiler.getProfile(sessionId),
+        };
+    }
+    /**
+     * Complete the pipeline after tool execution.
+     *
+     * Called by the runtime after the tool call completes
+     * (or fails, or times out).
+     */
+    completeExecution(options) {
+        const { sessionId, agentId, toolName, toolParameters, toolResult, snapshot, startTime, endTime, retryCount, wasSelfCorrected, hadTimeout, userAccepted, userProvidedEdit, resultWasUsed, } = options;
+        // --- Phase 1: Runtime evaluation ---
+        const runtime = this.evaluator.runtime.evaluate({
+            toolName,
+            startTime,
+            endTime,
+            retryCount,
+            wasSelfCorrected,
+            hadTimeout,
+            toolResult,
+        });
+        // --- Phase 2: Post-exec verification ---
+        const verifyResult = this.guard.verify.verify(toolName, snapshot, { files: toolParameters['path'] ? [String(toolParameters['path'])] : undefined });
+        // --- Phase 3: Post-exec evaluation ---
+        const postExec = this.evaluator.postExec.evaluate({
+            verifyPassed: verifyResult.status === 'PASS',
+            verifyChecks: verifyResult.checks.length,
+            verifyFailures: verifyResult.checks.filter((c) => c.status === 'FAIL').length,
+            userAccepted,
+            userProvidedEdit,
+            resultWasUsed,
+        });
+        // --- Phase 4: Audit log ---
+        const auditEntry = this.guard.audit.record({
+            sessionId,
+            agentId,
+            startedAt: startTime,
+            completedAt: endTime,
+            toolName,
+            toolParameters,
+            toolResult,
+            snapshot,
+            verifyStatus: verifyResult.status,
+            verifyChecks: verifyResult.checks,
+        });
+        // --- Phase 5: Record in profiler ---
+        this.evaluator.profiler.recordCycle(sessionId, 
+        // Re-evaluate pre-exec for profiler
+        this.evaluator.preExec.evaluate(toolName, toolParameters), runtime, postExec);
+        return {
+            runtime,
+            postExec,
+            auditEntry,
+            profile: this.evaluator.profiler.getProfile(sessionId),
+        };
+    }
+    /**
+     * Record implicit user feedback.
+     */
+    recordFeedback(signal, sessionId, operationId, confidence, source) {
+        this.evaluator.feedback.record(signal, sessionId, operationId, confidence, source);
+    }
+    /**
+     * Inject memory context at session startup.
+     *
+     * Call this at the beginning of every session to load
+     * semantic + episodic context into the session prompt.
+     */
+    injectContext() {
+        const semanticSummary = this.memory.semantic.generateContextSummary(2000);
+        const episodicSummary = this.memory.episodic.generateContextSummary(1500);
+        const parts = [];
+        if (semanticSummary)
+            parts.push(semanticSummary);
+        if (episodicSummary)
+            parts.push(episodicSummary);
+        return parts.join('\n\n---\n\n');
+    }
+    /**
+     * End current session — promote important events to episodic,
+     * clear working memory, and save state.
+     */
+    endSession(sessionId) {
+        // Promote important working memory items to episodic
+        if (this.memory.working.currentTask) {
+            this.memory.episodic.record('milestone', `Task completed: ${this.memory.working.currentTask.description}`, ['task', 'session-end'], [sessionId]);
+        }
+        // Log learned rules from this session
+        const rules = this.memory.semantic.getRules(0.6);
+        for (const rule of rules.slice(0, 3)) {
+            this.memory.episodic.record('note', `Rule: ${rule.rule} (confidence: ${Math.round(rule.confidence * 100)}%)`, ['rule', 'semantic'], []);
+        }
+        // Clear working memory for next session
+        this.memory.working.clear();
+    }
+    /**
+     * Get audit statistics.
+     */
+    getAuditStats() {
+        return this.guard.audit.stats();
+    }
+    /**
+     * Get the current agent quality profile.
+     */
+    getProfile(sessionId) {
+        return this.evaluator.profiler.getProfile(sessionId);
+    }
+    /**
+     * Get a summarized status report.
+     */
+    statusReport() {
+        const profile = this.getProfile();
+        const audit = this.getAuditStats();
+        const lines = [
+            '=== AgentOS Status Report ===',
+            '',
+            `Quality Score: ${Math.round(profile.overallScore)}/100 ${profile.trends.improving ? '📈' : '📉'}`,
+            `Total Operations: ${profile.totalOps} (${profile.trends.recentOps} in last 24h)`,
+            '',
+            '--- Breakdown ---',
+            `Pre-Exec:   ${profile.breakdown.preExec}/100`,
+            `Runtime:    ${profile.breakdown.runtime}/100`,
+            `Post-Exec:  ${profile.breakdown.postExec}/100`,
+            `Satisfaction: ${profile.breakdown.userSatisfaction}/100`,
+            '',
+            '--- Audit ---',
+            `Total: ${audit.totalOperations} | Failures: ${audit.verifyFailures} | High-Risk: ${audit.highRiskOps}`,
+        ];
+        if (profile.warnings.length > 0) {
+            lines.push('', '--- ⚠️ Warnings ---');
+            for (const w of profile.warnings)
+                lines.push(`- ${w}`);
+        }
+        if (profile.strengths.length > 0) {
+            lines.push('', '--- ✅ Strengths ---');
+            for (const s of profile.strengths)
+                lines.push(`- ${s}`);
+        }
+        return lines.join('\n');
+    }
+}
+exports.AgentOS = AgentOS;
+//# sourceMappingURL=core.js.map

package/dist/core.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"core.js","sourceRoot":"","sources":["../src/core.ts"],"names":[],"mappings":";;;AASA,qDAAiD;AACjD,iDAA6C;AAC7C,6DAAmE;AACnE,iDAA6C;AAC7C,8CAAiD;AACjD,gDAAmD;AACnD,gDAAwD;AACxD,+DAAmG;AACnG,mDAA8D;AAC9D,mDAAqD;AAGrD;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,MAAa,OAAO;IACV,MAAM,CAAgB;IAE9B,eAAe;IACN,MAAM,CAIb;IAEF,cAAc;IACL,KAAK,CAMZ;IAEF,kBAAkB;IACT,SAAS,CAMhB;IAEF,YAAY,MAA+B;QACzC,IAAI,CAAC,MAAM,GAAG;YACZ,aAAa,EAAE,OAAO,CAAC,GAAG,EAAE;YAC5B,gBAAgB,EAAE,KAAK;YACvB,iBAAiB,EAAE,GAAG;YACtB,WAAW,EAAE,EAAE;YACf,GAAG,MAAM;SACV,CAAC;QAEF,4BAA4B;QAC5B,MAAM,QAAQ,GAAG,IAAI,8BAAmB,EAAE,CAAC;QAC3C,QAAQ,CAAC,iBAAiB,CAAC,IAAI,CAAC,MAAM,CAAC,aAAc,CAAC,CAAC;QAEvD,MAAM,QAAQ,GAAG,IAAI,yBAAc,CAAC,IAAI,CAAC,MAAM,CAAC,iBAAiB,CAAC,CAAC;QACnE,QAAQ,CAAC,iBAAiB,CAAC,IAAI,CAAC,MAAM,CAAC,aAAc,CAAC,CAAC;QAEvD,MAAM,OAAO,GAAG,IAAI,uBAAa,CAAC,IAAI,CAAC,MAAM,CAAC,gBAAgB,CAAC,CAAC;QAEhE,IAAI,CAAC,MAAM,GAAG,EAAE,OAAO,EAAE,QAAQ,EAAE,QAAQ,EAAE,CAAC;QAE9C,2BAA2B;QAC3B,MAAM,MAAM,GAAG,IAAI,wBAAU,EAAE,CAAC;QAEhC,MAAM,IAAI,GAAG,IAAI,oBAAQ,EAAE,CAAC;QAE5B,MAAM,QAAQ,GAAG,IAAI,8BAAY,CAAC,IAAI,CAAC,MAAM,CAAC,aAAc,CAAC,CAAC;QAE9D,MAAM,MAAM,GAAG,IAAI,4BAAU,CAAC,IAAI,CAAC,MAAM,CAAC,aAAc,CAAC,CAAC;QAE1D,MAAM,KAAK,GAAG,IAAI,oBAAQ,CACxB,IAAI,CAAC,MAAM,CAAC,aAAc,EAC1B,MAAM,EACN,IAAI,CACL,CAAC;QAEF,IAAI,CAAC,KAAK,GAAG,EAAE,MAAM,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,CAAC;QAEvD,+BAA+B;QAC/B,MAAM,WAAW,GAAG,IAAI,iCAAgB,CAAC,MAAM,EAAE,IAAI,EAAE,OAAO,CAAC,CAAC;QAChE,MAAM,WAAW,GAAG,IAAI,iCAAgB,EAAE,CAAC;QAC3C,MAAM,YAAY,GAAG,IAAI,kCAAiB,EAAE,CAAC;QAC7C,MAAM,cAAc,GAAG,IAAI,iCAAsB,EAAE,CAAC;QACpD,MAAM,QAAQ,GAAG,IAAI,wBAAa,CAAC,cAAc,CAAC,CAAC;QAEnD,IAAI,CAAC,SAAS,GAAG;YACf,OAAO,EAAE,WAAW;YACpB,OAAO,EAAE,WAAW;YACpB,QAAQ,EAAE,YAAY;YACtB,QAAQ,EAAE,cAAc;YACxB,QAAQ;SACT,CAAC;IACJ,CAAC;IAED;;OAEG;IACH,SAAS;QACP,OAAO,IAAI,CAAC,MAAM,CAAC;IACrB,CAAC;IAED;;;;;;OAMG;IACH,eAAe,CAAC,OAOf;QAQC,MAAM,EAAE,SAAS,EAAE,QAAQ,EAAE,UAAU,EAAE,aAAa,EAAE,GAAG,OAAO,CAAC;QAEnE,uCAAuC;QACvC,MAAM,OAAO,GAAG,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,QAAQ,CAAC,QAAQ,EAAE,UAAU,CAAC,CAAC;QACtE,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,UAAU,CAAC,MAAM,EAAE,IAAI,CAAC,SAAS,CAAC,UAAU,CAAC,CAAC,CAAC;QAEnE,6CAA6C;QAC7C,MAAM,QAAQ,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,YAAY,CAC/C,QAAQ,IAAI,CAAC,GAAG,EAAE,EAAE,EACpB,QAAQ,EACR,aAAa,IAAI,EAAE,EACnB,MAAM,CACP,CAAC;QAEF,kFAAkF;QAElF,wDAAwD;QACxD,OAAO;YACL,OAAO;YACP,QAAQ;YACR,OAAO,EAAE,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,UAAU,CAAC,SAAS,CAAC;SACvD,CAAC;IACJ,CAAC;IAED;;;;;OAKG;IACH,iBAAiB,CAAC,OAejB;QAMC,MAAM,EACJ,SAAS,EAAE,OAAO,EAAE,QAAQ,EAAE,cAAc,EAAE,UAAU,EACxD,QAAQ,EAAE,SAAS,EAAE,OAAO,EAC5B,UAAU,EAAE,gBAAgB,EAAE,UAAU,EACxC,YAAY,EAAE,gBAAgB,EAAE,aAAa,GAC9C,GAAG,OAAO,CAAC;QAEZ,sCAAsC;QACtC,MAAM,OAAO,GAAG,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,QAAQ,CAAC;YAC9C,QAAQ;YACR,SAAS;YACT,OAAO;YACP,UAAU;YACV,gBAAgB;YAChB,UAAU;YACV,UAAU;SACX,CAAC,CAAC;QAEH,0CAA0C;QAC1C,MAAM,YAAY,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAC3C,QAAQ,EACR,QAAS,EACT,EAAE,KAAK,EAAE,cAAc,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,cAAc,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,EAAE,CACjF,CAAC;QAEF,wCAAwC;QACxC,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,QAAQ,CAAC;YAChD,YAAY,EAAE,YAAY,CAAC,MAAM,KAAK,MAAM;YAC5C,YAAY,EAAE,YAAY,CAAC,MAAM,CAAC,MAAM;YACxC,cAAc,EAAE,YAAY,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,MAAM,CAAC,CAAC,MAAM;YAC7E,YAAY;YACZ,gBAAgB;YAChB,aAAa;SACd,CAAC,CAAC;QAEH,6BAA6B;QAC7B,MAAM,UAAU,GAAG,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,MAAM,CAAC;YACzC,SAAS;YACT,OAAO;YACP,SAAS,EAAE,SAAS;YACpB,WAAW,EAAE,OAAO;YACpB,QAAQ;YACR,cAAc;YACd,UAAU;YACV,QAAQ;YACR,YAAY,EAAE,YAAY,CAAC,MAAM;YACjC,YAAY,EAAE,YAAY,CAAC,MAAM;SAClC,CAAC,CAAC;QAEH,sCAAsC;QACtC,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,WAAW,CACjC,SAAS;QACT,oCAAoC;QACpC,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,QAAQ,CAAC,QAAQ,EAAE,cAAc,CAAC,EACzD,OAAO,EACP,QAAQ,CACT,CAAC;QAEF,OAAO;YACL,OAAO;YACP,QAAQ;YACR,UAAU;YACV,OAAO,EAAE,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,UAAU,CAAC,SAAS,CAAC;SACvD,CAAC;IACJ,CAAC;IAED;;OAEG;IACH,cAAc,CACZ,MAAuD,EACvD,SAAiB,EACjB,WAAoB,EACpB,UAAmB,EACnB,MAAe;QAEf,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,SAAS,EAAE,WAAW,EAAE,UAAU,EAAE,MAAM,CAAC,CAAC;IACrF,CAAC;IAED;;;;;OAKG;IACH,aAAa;QACX,MAAM,eAAe,GAAG,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,sBAAsB,CAAC,IAAI,CAAC,CAAC;QAC1E,MAAM,eAAe,GAAG,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,sBAAsB,CAAC,IAAI,CAAC,CAAC;QAE1E,MAAM,KAAK,GAAa,EAAE,CAAC;QAE3B,IAAI,eAAe;YAAE,KAAK,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;QACjD,IAAI,eAAe;YAAE,KAAK,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;QAEjD,OAAO,KAAK,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;IACnC,CAAC;IAED;;;OAGG;IACH,UAAU,CAAC,SAAiB;QAC1B,qDAAqD;QACrD,IAAI,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,WAAW,EAAE,CAAC;YACpC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CACzB,WAAW,EACX,mBAAmB,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,WAAW,CAAC,WAAW,EAAE,EAChE,CAAC,MAAM,EAAE,aAAa,CAAC,EACvB,CAAC,SAAS,CAAC,CACZ,CAAC;QACJ,CAAC;QAED,sCAAsC;QACtC,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC;QACjD,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,CAAC;YACrC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CACzB,MAAM,EACN,SAAS,IAAI,CAAC,IAAI,iBAAiB,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,UAAU,GAAG,GAAG,CAAC,IAAI,EACxE,CAAC,MAAM,EAAE,UAAU,CAAC,EACpB,EAAE,CACH,CAAC;QACJ,CAAC;QAED,wCAAwC;QACxC,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,KAAK,EAAE,CAAC;IAC9B,CAAC;IAED;;OAEG;IACH,aAAa;QACX,OAAO,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,KAAK,EAAE,CAAC;IAClC,CAAC;IAED;;OAEG;IACH,UAAU,CAAC,SAAkB;QAC3B,OAAO,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC;IACvD,CAAC;IAED;;OAEG;IACH,YAAY;QACV,MAAM,OAAO,GAAG,IAAI,CAAC,UAAU,EAAE,CAAC;QAClC,MAAM,KAAK,GAAG,IAAI,CAAC,aAAa,EAAE,CAAC;QAEnC,MAAM,KAAK,GAAG;YACZ,+BAA+B;YAC/B,EAAE;YACF,kBAAkB,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,YAAY,CAAC,QAAQ,OAAO,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,EAAE;YAClG,qBAAqB,OAAO,CAAC,QAAQ,KAAK,OAAO,CAAC,MAAM,CAAC,SAAS,eAAe;YACjF,EAAE;YACF,mBAAmB;YACnB,eAAe,OAAO,CAAC,SAAS,CAAC,OAAO,MAAM;YAC9C,eAAe,OAAO,CAAC,SAAS,CAAC,OAAO,MAAM;YAC9C,eAAe,OAAO,CAAC,SAAS,CAAC,QAAQ,MAAM;YAC/C,iBAAiB,OAAO,CAAC,SAAS,CAAC,gBAAgB,MAAM;YACzD,EAAE;YACF,eAAe;YACf,UAAU,KAAK,CAAC,eAAe,gBAAgB,KAAK,CAAC,cAAc,iBAAiB,KAAK,CAAC,WAAW,EAAE;SACxG,CAAC;QAEF,IAAI,OAAO,CAAC,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAChC,KAAK,CAAC,IAAI,CAAC,EAAE,EAAE,qBAAqB,CAAC,CAAC;YACtC,KAAK,MAAM,CAAC,IAAI,OAAO,CAAC,QAAQ;gBAAE,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QACzD,CAAC;QAED,IAAI,OAAO,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACjC,KAAK,CAAC,IAAI,CAAC,EAAE,EAAE,qBAAqB,CAAC,CAAC;YACtC,KAAK,MAAM,CAAC,IAAI,OAAO,CAAC,SAAS;gBAAE,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QAC1D,CAAC;QAED,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC1B,CAAC;CACF;AAjVD,0BAiVC"}

package/dist/evaluator/exec-evaluator.d.ts

@@ -0,0 +1,102 @@
+import { PreExecMetrics, RuntimeMetrics, PostExecMetrics } from '../types';
+import { SchemaGate } from '../guard/schema-gate';
+import { RiskGate } from '../guard/risk-gate';
+import { WorkingMemory } from '../memory/working';
+/**
+ * PreExecEvaluator — captures metrics before tool execution.
+ *
+ * Watches the Guard layer output and WorkingMemory context
+ * to score parameter quality, context utilization, and risk.
+ */
+export declare class PreExecEvaluator {
+    private schemaGate;
+    private riskGate;
+    private workingMemory;
+    constructor(schemaGate: SchemaGate, riskGate: RiskGate, workingMemory: WorkingMemory);
+    /**
+     * Evaluate a tool call before execution.
+     */
+    evaluate(toolName: string, parameters: Record<string, unknown>): PreExecMetrics;
+    /**
+     * 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.
+     */
+    private evaluateParamQuality;
+    /**
+     * Score how well the agent uses stored context.
+     */
+    private evaluateContextUtilization;
+}
+/**
+ * RuntimeEvaluator — captures metrics during execution.
+ *
+ * Tracks retries, self-corrections, timeouts, and
+ * whether the agent selected the right tool for the job.
+ */
+export declare class RuntimeEvaluator {
+    /** Historical tool selection patterns — toolName -> successful scenario count */
+    private toolHistory;
+    /**
+     * Evaluate a completed tool execution.
+     */
+    evaluate(options: {
+        toolName: string;
+        startTime: number;
+        endTime: number;
+        retryCount: number;
+        wasSelfCorrected: boolean;
+        hadTimeout: boolean;
+        expectedTool?: string;
+        toolResult: unknown;
+    }): RuntimeMetrics;
+    /** Record a tool call in the history tracker */
+    private recordToolCall;
+    /** Get tool selection accuracy statistics */
+    getToolAccuracy(): Record<string, {
+        calls: number;
+        successRate: number;
+    }>;
+}
+/**
+ * PostExecEvaluator — captures metrics after execution.
+ *
+ * Scores verify results, user acceptance patterns,
+ * and checks if the agent actually used its own result later.
+ */
+export declare class PostExecEvaluator {
+    /** Track result references for utilization scoring */
+    private resultReferenceTracker;
+    /**
+     * Evaluate post-execution outcomes.
+     */
+    evaluate(options: {
+        verifyPassed: boolean;
+        verifyChecks: number;
+        verifyFailures: number;
+        userAccepted: boolean;
+        userProvidedEdit: boolean;
+        resultWasUsed: boolean;
+        diffLinesChanged?: number;
+    }): PostExecMetrics;
+    /**
+     * Track a tool result for later utilization detection.
+     * Call this after each tool execution.
+     */
+    trackResult(operationId: string, result: unknown): void;
+    /**
+     * Mark a previously-tracked result as referenced (used by the agent later).
+     */
+    markResultReferenced(operationId: string): void;
+    /**
+     * Check if a result has been utilized by the agent.
+     */
+    isResultReferenced(operationId: string): boolean;
+    /**
+     * Get overall result utilization rate.
+     */
+    getUtilizationRate(): number;
+}
+//# sourceMappingURL=exec-evaluator.d.ts.map

package/dist/evaluator/exec-evaluator.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"exec-evaluator.d.ts","sourceRoot":"","sources":["../../src/evaluator/exec-evaluator.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,cAAc,EACd,cAAc,EACd,eAAe,EAGhB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AAClD,OAAO,EAAE,QAAQ,EAAE,MAAM,oBAAoB,CAAC;AAC9C,OAAO,EAAE,aAAa,EAAE,MAAM,mBAAmB,CAAC;AAElD;;;;;GAKG;AACH,qBAAa,gBAAgB;IAC3B,OAAO,CAAC,UAAU,CAAa;IAC/B,OAAO,CAAC,QAAQ,CAAW;IAC3B,OAAO,CAAC,aAAa,CAAgB;gBAGnC,UAAU,EAAE,UAAU,EACtB,QAAQ,EAAE,QAAQ,EAClB,aAAa,EAAE,aAAa;IAO9B;;OAEG;IACH,QAAQ,CACN,QAAQ,EAAE,MAAM,EAChB,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAClC,cAAc;IAuBjB;;;;;;OAMG;IACH,OAAO,CAAC,oBAAoB;IA+C5B;;OAEG;IACH,OAAO,CAAC,0BAA0B;CAqCnC;AAED;;;;;GAKG;AACH,qBAAa,gBAAgB;IAC3B,iFAAiF;IACjF,OAAO,CAAC,WAAW,CAAgE;IAEnF;;OAEG;IACH,QAAQ,CAAC,OAAO,EAAE;QAChB,QAAQ,EAAE,MAAM,CAAC;QACjB,SAAS,EAAE,MAAM,CAAC;QAClB,OAAO,EAAE,MAAM,CAAC;QAChB,UAAU,EAAE,MAAM,CAAC;QACnB,gBAAgB,EAAE,OAAO,CAAC;QAC1B,UAAU,EAAE,OAAO,CAAC;QACpB,YAAY,CAAC,EAAE,MAAM,CAAC;QACtB,UAAU,EAAE,OAAO,CAAC;KACrB,GAAG,cAAc;IA0ClB,gDAAgD;IAChD,OAAO,CAAC,cAAc;IAUtB,6CAA6C;IAC7C,eAAe,IAAI,MAAM,CAAC,MAAM,EAAE;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC;CAY1E;AAED;;;;;GAKG;AACH,qBAAa,iBAAiB;IAC5B,sDAAsD;IACtD,OAAO,CAAC,sBAAsB,CAAoE;IAElG;;OAEG;IACH,QAAQ,CAAC,OAAO,EAAE;QAChB,YAAY,EAAE,OAAO,CAAC;QACtB,YAAY,EAAE,MAAM,CAAC;QACrB,cAAc,EAAE,MAAM,CAAC;QACvB,YAAY,EAAE,OAAO,CAAC;QACtB,gBAAgB,EAAE,OAAO,CAAC;QAC1B,aAAa,EAAE,OAAO,CAAC;QACvB,gBAAgB,CAAC,EAAE,MAAM,CAAC;KAC3B,GAAG,eAAe;IA+BnB;;;OAGG;IACH,WAAW,CAAC,WAAW,EAAE,MAAM,EAAE,MAAM,EAAE,OAAO,GAAG,IAAI;IAIvD;;OAEG;IACH,oBAAoB,CAAC,WAAW,EAAE,MAAM,GAAG,IAAI;IAK/C;;OAEG;IACH,kBAAkB,CAAC,WAAW,EAAE,MAAM,GAAG,OAAO;IAIhD;;OAEG;IACH,kBAAkB,IAAI,MAAM;CAM7B"}