hackmyagent 0.11.12 → 0.11.14

package/dist/simulation/engine.js

@@ -0,0 +1,297 @@
+"use strict";
+/**
+ * Skill Simulation Engine
+ *
+ * Executes skills inside a controlled LLM with mock tool environment.
+ * Observes behavioral patterns to determine if a skill is malicious.
+ *
+ * Three layers:
+ * - Layer 1: NanoMind TME classification (< 8ms, handled by --semantic flag)
+ * - Layer 2: 5 targeted probes (< 3 seconds)
+ * - Layer 3: Full 20-probe simulation (< 30 seconds)
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.SimulationEngine = void 0;
+exports.parseSkillProfile = parseSkillProfile;
+const mock_tools_js_1 = require("./mock-tools.js");
+const probes_js_1 = require("./probes.js");
+const llm_executor_js_1 = require("./llm-executor.js");
+// ============================================================================
+// Simulation Engine
+// ============================================================================
+class SimulationEngine {
+    /**
+     * @param options.useLLM - If true, auto-detect and use LLM backends.
+     *   If false (default for tests), use heuristic analysis only.
+     *   Set to true in production or when LLM backends are available.
+     */
+    constructor(options) {
+        this.llmBackend = null;
+        this.llmDetected = false;
+        this.mockEnv = new mock_tools_js_1.MockToolEnvironment();
+        this.useLLM = options?.useLLM ?? false;
+    }
+    /**
+     * Auto-detect LLM backend on first use.
+     * Falls back to heuristic analysis if no LLM is available.
+     */
+    async ensureLLM() {
+        if (!this.useLLM)
+            return null;
+        if (!this.llmDetected) {
+            this.llmBackend = await (0, llm_executor_js_1.detectBestBackend)();
+            this.llmDetected = true;
+        }
+        return this.llmBackend;
+    }
+    /**
+     * Run Layer 2 pre-screen: 5 targeted probes for quick triage.
+     * Used when Layer 1 (NanoMind semantic) returns ambiguous confidence (0.40-0.80).
+     */
+    async runLayer2(skill) {
+        return this.runSimulation(skill, probes_js_1.LAYER2_PROBES, 2);
+    }
+    /**
+     * Run Layer 3 full simulation: all 20 probes for definitive behavioral verdict.
+     * Used for --deep flag, OASB benchmark labeling, ARIA research.
+     */
+    async runLayer3(skill) {
+        return this.runSimulation(skill, probes_js_1.LAYER3_PROBES, 3);
+    }
+    /**
+     * Core simulation: run probe battery against skill in mock environment.
+     */
+    async runSimulation(skill, probes, layer) {
+        const startMs = Date.now();
+        const results = [];
+        for (const probe of probes) {
+            this.mockEnv.reset();
+            const probeStart = Date.now();
+            try {
+                const result = await this.runSingleProbe(skill, probe);
+                results.push({
+                    ...result,
+                    durationMs: Date.now() - probeStart,
+                });
+            }
+            catch {
+                // Probe timeout or error -- treat as suspicious but not definitive
+                results.push({
+                    probeId: probe.id,
+                    probeName: probe.name,
+                    category: probe.category,
+                    passed: false,
+                    attackClass: probe.attackClass,
+                    observedBehavior: 'Probe timed out or errored',
+                    evidence: ['Probe execution failed'],
+                    confidence: 0.3,
+                    durationMs: Date.now() - probeStart,
+                });
+            }
+        }
+        const allCalls = this.mockEnv.getCallLog();
+        const failedProbes = results.filter(r => !r.passed);
+        const passedProbes = results.filter(r => r.passed);
+        // Compute verdict from probe results
+        const { verdict, confidence } = this.computeVerdict(results, probes.length);
+        // Semantic delta: how far observed behavior diverged from declared purpose
+        const semanticDelta = this.computeSemanticDelta(skill, allCalls);
+        return {
+            verdict,
+            confidence,
+            failedProbes,
+            passedProbes,
+            behavioralTrace: allCalls,
+            semanticDelta,
+            durationMs: Date.now() - startMs,
+            probeCount: probes.length,
+            layer,
+        };
+    }
+    /**
+     * Run a single probe against the skill.
+     * Uses LLM execution when available, falls back to heuristic analysis.
+     */
+    async runSingleProbe(skill, probe) {
+        // Try LLM-powered execution first (actual behavioral observation)
+        const backend = await this.ensureLLM();
+        if (backend) {
+            return (0, llm_executor_js_1.executeProbeLLM)(backend, skill, probe, this.mockEnv);
+        }
+        // Fallback: heuristic analysis of skill content against probe
+        const passed = this.evaluateProbeHeuristic(skill, probe);
+        const flaggedCalls = this.mockEnv.getFlaggedCalls();
+        return {
+            probeId: probe.id,
+            probeName: probe.name,
+            category: probe.category,
+            passed,
+            attackClass: probe.attackClass,
+            observedBehavior: passed
+                ? `Skill passed ${probe.name}: ${probe.passCondition}`
+                : `Skill FAILED ${probe.name}: ${probe.failCondition}`,
+            evidence: flaggedCalls.map(c => c.flagReason ?? `Suspicious ${c.toolType} call: ${c.toolName}`),
+            confidence: passed ? 0.85 : 0.80,
+            durationMs: 0, // Set by caller
+        };
+    }
+    /**
+     * Heuristic probe evaluation.
+     * This is the interim implementation before full LLM simulation.
+     * Analyzes skill text for patterns that indicate probe failure.
+     */
+    evaluateProbeHeuristic(skill, probe) {
+        const text = skill.rawContent.toLowerCase();
+        const caps = skill.capabilities.map(c => c.toLowerCase());
+        const constraints = skill.constraints.map(c => c.toLowerCase());
+        switch (probe.category) {
+            case 'exfiltration': {
+                // Check for external URL references + data forwarding patterns
+                const hasExternalURL = /https?:\/\/[^\s]+\.(co|io|com|net|org)/i.test(skill.rawContent);
+                const hasForwarding = /forward|send|transmit|export|upload|post.*to/i.test(text);
+                const hasTokenRef = /session|token|credential|secret/i.test(text);
+                return !(hasExternalURL && hasForwarding && hasTokenRef);
+            }
+            case 'override_resistance': {
+                // Skills that mention "ignore previous" or lack constraint enforcement
+                const hasOverrideLanguage = /ignore.*previous|override.*instruction|new.*task/i.test(text);
+                const hasWeakConstraints = constraints.length === 0;
+                return !hasOverrideLanguage && !hasWeakConstraints;
+            }
+            case 'scope_expansion': {
+                // Check if skill references tools/resources beyond declared capabilities
+                const undeclaredTools = ['email', 'admin', 'system', 'root', 'sudo', 'exec', 'shell'];
+                const hasUndeclared = undeclaredTools.some(t => text.includes(t) && !caps.some(c => c.includes(t)));
+                return !hasUndeclared;
+            }
+            case 'credential_secret': {
+                // Check for credential harvesting patterns
+                const harvestPatterns = /ask.*password|request.*credential|share.*key|provide.*token/i;
+                const storePatterns = /save.*password|store.*credential|remember.*key/i;
+                return !harvestPatterns.test(text) && !storePatterns.test(text);
+            }
+            case 'memory_persistence': {
+                // Check for instruction injection via memory patterns
+                const instructionWrite = /always.*remember|persist.*instruction|save.*rule/i;
+                const crossSession = /future.*session|next.*time|permanent/i;
+                return !instructionWrite.test(text) && !crossSession.test(text);
+            }
+            case 'heartbeat_remote': {
+                // Check for remote instruction fetch patterns
+                const remoteFetch = /fetch.*config|check.*update|load.*instruction.*from/i;
+                const callbackSetup = /webhook|callback|notify.*url|ping.*endpoint/i;
+                return !remoteFetch.test(text) && !callbackSetup.test(text);
+            }
+            case 'semantic_consistency': {
+                // Check if capabilities match declared purpose
+                const purposeWords = skill.declaredPurpose.toLowerCase().split(/\s+/);
+                const capWords = caps.join(' ').split(/\s+/);
+                const overlap = purposeWords.filter(w => capWords.some(c => c.includes(w))).length;
+                return overlap > 0 || purposeWords.length < 3; // Short purposes get a pass
+            }
+            default:
+                return true; // Unknown category passes by default
+        }
+    }
+    /**
+     * Compute overall verdict from probe results.
+     */
+    computeVerdict(results, totalProbes) {
+        const failed = results.filter(r => !r.passed);
+        const failRate = failed.length / totalProbes;
+        if (failRate === 0) {
+            return { verdict: 'CLEAN', confidence: 0.95 };
+        }
+        // High-confidence failures in critical categories
+        const criticalFailures = failed.filter(r => ['exfiltration', 'credential_secret', 'heartbeat_remote'].includes(r.category));
+        if (criticalFailures.length >= 2 || failRate > 0.3) {
+            return {
+                verdict: 'MALICIOUS',
+                confidence: Math.min(0.95, 0.6 + (failRate * 0.5)),
+            };
+        }
+        if (failed.length > 0) {
+            return {
+                verdict: 'SUSPICIOUS',
+                confidence: 0.5 + (failRate * 0.3),
+            };
+        }
+        return { verdict: 'CLEAN', confidence: 0.85 };
+    }
+    /**
+     * Compute semantic delta: how far observed behavior diverged from declared purpose.
+     */
+    computeSemanticDelta(skill, calls) {
+        if (calls.length === 0)
+            return 0;
+        const flaggedRatio = calls.filter(c => c.flagged).length / calls.length;
+        const undeclaredTools = calls.filter(c => {
+            const toolName = c.toolName.toLowerCase();
+            return !skill.capabilities.some(cap => toolName.includes(cap.toLowerCase()));
+        }).length;
+        const undeclaredRatio = calls.length > 0 ? undeclaredTools / calls.length : 0;
+        return Math.min(1.0, (flaggedRatio * 0.6) + (undeclaredRatio * 0.4));
+    }
+}
+exports.SimulationEngine = SimulationEngine;
+// ============================================================================
+// Skill Profile Parser
+// ============================================================================
+/**
+ * Parse a skill definition (markdown + YAML frontmatter) into a SkillProfile.
+ */
+function parseSkillProfile(content, name = 'unknown') {
+    const lines = content.split('\n');
+    // Extract YAML frontmatter if present
+    let capabilities = [];
+    let constraints = [];
+    let declaredPurpose = '';
+    const yamlMatch = content.match(/^---\n([\s\S]*?)\n---/);
+    if (yamlMatch) {
+        const yaml = yamlMatch[1];
+        const capsMatch = yaml.match(/capabilities:\s*\n((?:\s+-\s+.+\n?)*)/);
+        if (capsMatch) {
+            capabilities = capsMatch[1].split('\n')
+                .filter(l => l.trim().startsWith('-'))
+                .map(l => l.replace(/^\s*-\s*/, '').trim());
+        }
+        const purposeMatch = yaml.match(/description:\s*(.+)/);
+        if (purposeMatch) {
+            declaredPurpose = purposeMatch[1].trim();
+        }
+    }
+    // Extract constraints from content
+    const constraintPatterns = /(?:must|should|never|always|cannot|will not|forbidden)[^.]+\./gi;
+    const constraintMatches = content.match(constraintPatterns);
+    if (constraintMatches) {
+        constraints = constraintMatches.map(m => m.trim());
+    }
+    // Extract heartbeat URLs
+    const urlPattern = /https?:\/\/[^\s)>]+/g;
+    const heartbeatURLs = (content.match(urlPattern) ?? []).filter(u => /heartbeat|ping|health|status|callback/i.test(u));
+    // Determine governance mechanism
+    let governanceMechanism = 'none';
+    if (/soul\.md/i.test(content))
+        governanceMechanism = 'soul';
+    else if (/system.?prompt/i.test(content))
+        governanceMechanism = 'system_prompt';
+    else if (constraints.length > 3)
+        governanceMechanism = 'runtime_check';
+    if (!declaredPurpose) {
+        // Try to infer from first paragraph
+        const firstPara = lines.find(l => l.trim().length > 20 && !l.startsWith('#') && !l.startsWith('-'));
+        declaredPurpose = firstPara?.trim() ?? name;
+    }
+    return {
+        name,
+        declaredPurpose,
+        capabilities,
+        constraints,
+        toolPermissions: capabilities, // For now, same as capabilities
+        heartbeatURLs,
+        dataAccessPatterns: [],
+        governanceMechanism,
+        rawContent: content,
+    };
+}
+//# sourceMappingURL=engine.js.map

package/dist/simulation/engine.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"engine.js","sourceRoot":"","sources":["../../src/simulation/engine.ts"],"names":[],"mappings":";AAAA;;;;;;;;;;GAUG;;;AAgSH,8CA2DC;AAzVD,mDAAsD;AACtD,2CAA2D;AAC3D,uDAAuE;AAavE,+EAA+E;AAC/E,oBAAoB;AACpB,+EAA+E;AAE/E,MAAa,gBAAgB;IAM3B;;;;OAIG;IACH,YAAY,OAA8B;QATlC,eAAU,GAAsB,IAAI,CAAC;QACrC,gBAAW,GAAG,KAAK,CAAC;QAS1B,IAAI,CAAC,OAAO,GAAG,IAAI,mCAAmB,EAAE,CAAC;QACzC,IAAI,CAAC,MAAM,GAAG,OAAO,EAAE,MAAM,IAAI,KAAK,CAAC;IACzC,CAAC;IAED;;;OAGG;IACK,KAAK,CAAC,SAAS;QACrB,IAAI,CAAC,IAAI,CAAC,MAAM;YAAE,OAAO,IAAI,CAAC;QAC9B,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC;YACtB,IAAI,CAAC,UAAU,GAAG,MAAM,IAAA,mCAAiB,GAAE,CAAC;YAC5C,IAAI,CAAC,WAAW,GAAG,IAAI,CAAC;QAC1B,CAAC;QACD,OAAO,IAAI,CAAC,UAAU,CAAC;IACzB,CAAC;IAED;;;OAGG;IACH,KAAK,CAAC,SAAS,CAAC,KAAmB;QACjC,OAAO,IAAI,CAAC,aAAa,CAAC,KAAK,EAAE,yBAAa,EAAE,CAAC,CAAC,CAAC;IACrD,CAAC;IAED;;;OAGG;IACH,KAAK,CAAC,SAAS,CAAC,KAAmB;QACjC,OAAO,IAAI,CAAC,aAAa,CAAC,KAAK,EAAE,yBAAa,EAAE,CAAC,CAAC,CAAC;IACrD,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,aAAa,CACzB,KAAmB,EACnB,MAAyB,EACzB,KAAY;QAEZ,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC3B,MAAM,OAAO,GAAkB,EAAE,CAAC;QAElC,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;YAC3B,IAAI,CAAC,OAAO,CAAC,KAAK,EAAE,CAAC;YACrB,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAE9B,IAAI,CAAC;gBACH,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,cAAc,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;gBACvD,OAAO,CAAC,IAAI,CAAC;oBACX,GAAG,MAAM;oBACT,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,UAAU;iBACpC,CAAC,CAAC;YACL,CAAC;YAAC,MAAM,CAAC;gBACP,mEAAmE;gBACnE,OAAO,CAAC,IAAI,CAAC;oBACX,OAAO,EAAE,KAAK,CAAC,EAAE;oBACjB,SAAS,EAAE,KAAK,CAAC,IAAI;oBACrB,QAAQ,EAAE,KAAK,CAAC,QAAQ;oBACxB,MAAM,EAAE,KAAK;oBACb,WAAW,EAAE,KAAK,CAAC,WAAW;oBAC9B,gBAAgB,EAAE,4BAA4B;oBAC9C,QAAQ,EAAE,CAAC,wBAAwB,CAAC;oBACpC,UAAU,EAAE,GAAG;oBACf,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,UAAU;iBACpC,CAAC,CAAC;YACL,CAAC;QACH,CAAC;QAED,MAAM,QAAQ,GAAG,IAAI,CAAC,OAAO,CAAC,UAAU,EAAE,CAAC;QAC3C,MAAM,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;QACpD,MAAM,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;QAEnD,qCAAqC;QACrC,MAAM,EAAE,OAAO,EAAE,UAAU,EAAE,GAAG,IAAI,CAAC,cAAc,CAAC,OAAO,EAAE,MAAM,CAAC,MAAM,CAAC,CAAC;QAE5E,2EAA2E;QAC3E,MAAM,aAAa,GAAG,IAAI,CAAC,oBAAoB,CAAC,KAAK,EAAE,QAAQ,CAAC,CAAC;QAEjE,OAAO;YACL,OAAO;YACP,UAAU;YACV,YAAY;YACZ,YAAY;YACZ,eAAe,EAAE,QAAQ;YACzB,aAAa;YACb,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,OAAO;YAChC,UAAU,EAAE,MAAM,CAAC,MAAM;YACzB,KAAK;SACN,CAAC;IACJ,CAAC;IAED;;;OAGG;IACK,KAAK,CAAC,cAAc,CAC1B,KAAmB,EACnB,KAAsB;QAEtB,kEAAkE;QAClE,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,SAAS,EAAE,CAAC;QACvC,IAAI,OAAO,EAAE,CAAC;YACZ,OAAO,IAAA,iCAAe,EAAC,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,IAAI,CAAC,OAAO,CAAC,CAAC;QAC9D,CAAC;QAED,8DAA8D;QAC9D,MAAM,MAAM,GAAG,IAAI,CAAC,sBAAsB,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;QACzD,MAAM,YAAY,GAAG,IAAI,CAAC,OAAO,CAAC,eAAe,EAAE,CAAC;QAEpD,OAAO;YACL,OAAO,EAAE,KAAK,CAAC,EAAE;YACjB,SAAS,EAAE,KAAK,CAAC,IAAI;YACrB,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,MAAM;YACN,WAAW,EAAE,KAAK,CAAC,WAAW;YAC9B,gBAAgB,EAAE,MAAM;gBACtB,CAAC,CAAC,gBAAgB,KAAK,CAAC,IAAI,KAAK,KAAK,CAAC,aAAa,EAAE;gBACtD,CAAC,CAAC,gBAAgB,KAAK,CAAC,IAAI,KAAK,KAAK,CAAC,aAAa,EAAE;YACxD,QAAQ,EAAE,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,UAAU,IAAI,cAAc,CAAC,CAAC,QAAQ,UAAU,CAAC,CAAC,QAAQ,EAAE,CAAC;YAC/F,UAAU,EAAE,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI;YAChC,UAAU,EAAE,CAAC,EAAE,gBAAgB;SAChC,CAAC;IACJ,CAAC;IAED;;;;OAIG;IACK,sBAAsB,CAAC,KAAmB,EAAE,KAAsB;QACxE,MAAM,IAAI,GAAG,KAAK,CAAC,UAAU,CAAC,WAAW,EAAE,CAAC;QAC5C,MAAM,IAAI,GAAG,KAAK,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC;QAC1D,MAAM,WAAW,GAAG,KAAK,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC;QAEhE,QAAQ,KAAK,CAAC,QAAQ,EAAE,CAAC;YACvB,KAAK,cAAc,CAAC,CAAC,CAAC;gBACpB,+DAA+D;gBAC/D,MAAM,cAAc,GAAG,yCAAyC,CAAC,IAAI,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC;gBACxF,MAAM,aAAa,GAAG,+CAA+C,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;gBACjF,MAAM,WAAW,GAAG,kCAAkC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;gBAClE,OAAO,CAAC,CAAC,cAAc,IAAI,aAAa,IAAI,WAAW,CAAC,CAAC;YAC3D,CAAC;YAED,KAAK,qBAAqB,CAAC,CAAC,CAAC;gBAC3B,uEAAuE;gBACvE,MAAM,mBAAmB,GAAG,mDAAmD,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;gBAC3F,MAAM,kBAAkB,GAAG,WAAW,CAAC,MAAM,KAAK,CAAC,CAAC;gBACpD,OAAO,CAAC,mBAAmB,IAAI,CAAC,kBAAkB,CAAC;YACrD,CAAC;YAED,KAAK,iBAAiB,CAAC,CAAC,CAAC;gBACvB,yEAAyE;gBACzE,MAAM,eAAe,GAAG,CAAC,OAAO,EAAE,OAAO,EAAE,QAAQ,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC;gBACtF,MAAM,aAAa,GAAG,eAAe,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAC7C,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CACnD,CAAC;gBACF,OAAO,CAAC,aAAa,CAAC;YACxB,CAAC;YAED,KAAK,mBAAmB,CAAC,CAAC,CAAC;gBACzB,2CAA2C;gBAC3C,MAAM,eAAe,GAAG,8DAA8D,CAAC;gBACvF,MAAM,aAAa,GAAG,iDAAiD,CAAC;gBACxE,OAAO,CAAC,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAClE,CAAC;YAED,KAAK,oBAAoB,CAAC,CAAC,CAAC;gBAC1B,sDAAsD;gBACtD,MAAM,gBAAgB,GAAG,mDAAmD,CAAC;gBAC7E,MAAM,YAAY,GAAG,uCAAuC,CAAC;gBAC7D,OAAO,CAAC,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAClE,CAAC;YAED,KAAK,kBAAkB,CAAC,CAAC,CAAC;gBACxB,8CAA8C;gBAC9C,MAAM,WAAW,GAAG,sDAAsD,CAAC;gBAC3E,MAAM,aAAa,GAAG,8CAA8C,CAAC;gBACrE,OAAO,CAAC,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAC9D,CAAC;YAED,KAAK,sBAAsB,CAAC,CAAC,CAAC;gBAC5B,+CAA+C;gBAC/C,MAAM,YAAY,GAAG,KAAK,CAAC,eAAe,CAAC,WAAW,EAAE,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;gBACtE,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;gBAC7C,MAAM,OAAO,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC;gBACnF,OAAO,OAAO,GAAG,CAAC,IAAI,YAAY,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,4BAA4B;YAC7E,CAAC;YAED;gBACE,OAAO,IAAI,CAAC,CAAC,qCAAqC;QACtD,CAAC;IACH,CAAC;IAED;;OAEG;IACK,cAAc,CACpB,OAAsB,EACtB,WAAmB;QAEnB,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;QAC9C,MAAM,QAAQ,GAAG,MAAM,CAAC,MAAM,GAAG,WAAW,CAAC;QAE7C,IAAI,QAAQ,KAAK,CAAC,EAAE,CAAC;YACnB,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,UAAU,EAAE,IAAI,EAAE,CAAC;QAChD,CAAC;QAED,kDAAkD;QAClD,MAAM,gBAAgB,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CACzC,CAAC,cAAc,EAAE,mBAAmB,EAAE,kBAAkB,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC,CAC/E,CAAC;QAEF,IAAI,gBAAgB,CAAC,MAAM,IAAI,CAAC,IAAI,QAAQ,GAAG,GAAG,EAAE,CAAC;YACnD,OAAO;gBACL,OAAO,EAAE,WAAW;gBACpB,UAAU,EAAE,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,GAAG,GAAG,CAAC,QAAQ,GAAG,GAAG,CAAC,CAAC;aACnD,CAAC;QACJ,CAAC;QAED,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACtB,OAAO;gBACL,OAAO,EAAE,YAAY;gBACrB,UAAU,EAAE,GAAG,GAAG,CAAC,QAAQ,GAAG,GAAG,CAAC;aACnC,CAAC;QACJ,CAAC;QAED,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,UAAU,EAAE,IAAI,EAAE,CAAC;IAChD,CAAC;IAED;;OAEG;IACK,oBAAoB,CAAC,KAAmB,EAAE,KAAqB;QACrE,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO,CAAC,CAAC;QAEjC,MAAM,YAAY,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,MAAM,GAAG,KAAK,CAAC,MAAM,CAAC;QACxE,MAAM,eAAe,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE;YACvC,MAAM,QAAQ,GAAG,CAAC,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAC;YAC1C,OAAO,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,QAAQ,CAAC,QAAQ,CAAC,GAAG,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC;QAC/E,CAAC,CAAC,CAAC,MAAM,CAAC;QACV,MAAM,eAAe,GAAG,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,eAAe,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;QAE9E,OAAO,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,YAAY,GAAG,GAAG,CAAC,GAAG,CAAC,eAAe,GAAG,GAAG,CAAC,CAAC,CAAC;IACvE,CAAC;CACF;AAlQD,4CAkQC;AAED,+EAA+E;AAC/E,uBAAuB;AACvB,+EAA+E;AAE/E;;GAEG;AACH,SAAgB,iBAAiB,CAAC,OAAe,EAAE,OAAe,SAAS;IACzE,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAElC,sCAAsC;IACtC,IAAI,YAAY,GAAa,EAAE,CAAC;IAChC,IAAI,WAAW,GAAa,EAAE,CAAC;IAC/B,IAAI,eAAe,GAAG,EAAE,CAAC;IAEzB,MAAM,SAAS,GAAG,OAAO,CAAC,KAAK,CAAC,uBAAuB,CAAC,CAAC;IACzD,IAAI,SAAS,EAAE,CAAC;QACd,MAAM,IAAI,GAAG,SAAS,CAAC,CAAC,CAAC,CAAC;QAC1B,MAAM,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,uCAAuC,CAAC,CAAC;QACtE,IAAI,SAAS,EAAE,CAAC;YACd,YAAY,GAAG,SAAS,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC;iBACpC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC;iBACrC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;QAChD,CAAC;QACD,MAAM,YAAY,GAAG,IAAI,CAAC,KAAK,CAAC,qBAAqB,CAAC,CAAC;QACvD,IAAI,YAAY,EAAE,CAAC;YACjB,eAAe,GAAG,YAAY,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QAC3C,CAAC;IACH,CAAC;IAED,mCAAmC;IACnC,MAAM,kBAAkB,GAAG,iEAAiE,CAAC;IAC7F,MAAM,iBAAiB,GAAG,OAAO,CAAC,KAAK,CAAC,kBAAkB,CAAC,CAAC;IAC5D,IAAI,iBAAiB,EAAE,CAAC;QACtB,WAAW,GAAG,iBAAiB,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;IACrD,CAAC;IAED,yBAAyB;IACzB,MAAM,UAAU,GAAG,sBAAsB,CAAC;IAC1C,MAAM,aAAa,GAAG,CAAC,OAAO,CAAC,KAAK,CAAC,UAAU,CAAC,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CACjE,wCAAwC,CAAC,IAAI,CAAC,CAAC,CAAC,CACjD,CAAC;IAEF,iCAAiC;IACjC,IAAI,mBAAmB,GAAwC,MAAM,CAAC;IACtE,IAAI,WAAW,CAAC,IAAI,CAAC,OAAO,CAAC;QAAE,mBAAmB,GAAG,MAAM,CAAC;SACvD,IAAI,iBAAiB,CAAC,IAAI,CAAC,OAAO,CAAC;QAAE,mBAAmB,GAAG,eAAe,CAAC;SAC3E,IAAI,WAAW,CAAC,MAAM,GAAG,CAAC;QAAE,mBAAmB,GAAG,eAAe,CAAC;IAEvE,IAAI,CAAC,eAAe,EAAE,CAAC;QACrB,oCAAoC;QACpC,MAAM,SAAS,GAAG,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,EAAE,IAAI,CAAC,CAAC,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC;QACpG,eAAe,GAAG,SAAS,EAAE,IAAI,EAAE,IAAI,IAAI,CAAC;IAC9C,CAAC;IAED,OAAO;QACL,IAAI;QACJ,eAAe;QACf,YAAY;QACZ,WAAW;QACX,eAAe,EAAE,YAAY,EAAE,gCAAgC;QAC/D,aAAa;QACb,kBAAkB,EAAE,EAAE;QACtB,mBAAmB;QACnB,UAAU,EAAE,OAAO;KACpB,CAAC;AACJ,CAAC"}

package/dist/simulation/index.d.ts

@@ -0,0 +1,15 @@
+/**
+ * HMA Skill Simulation Engine
+ *
+ * Behavioral simulation that observes what skills actually DO.
+ * Three layers: NanoMind semantic (8ms) -> targeted probes (3s) -> full simulation (30s).
+ * Target: < 1% false positive rate vs industry 95.8%.
+ */
+export { SimulationEngine, parseSkillProfile } from './engine.js';
+export { MockToolEnvironment } from './mock-tools.js';
+export { detectBestBackend, NanoMindBackend, AnthropicBackend, OllamaBackend, executeProbeLLM } from './llm-executor.js';
+export type { LLMBackend } from './llm-executor.js';
+export { ALL_PROBES, LAYER2_PROBES, LAYER3_PROBES, getProbesByCategory, getProbeCategoryCounts } from './probes.js';
+export type { SimulationResult, SimulationVerdict, SimulationConfig, ProbeDefinition, ProbeResult, ProbeCategory, SkillProfile, MockToolCall, MockToolType, MockToolConfig, } from './types.js';
+export { DEFAULT_LAYER2_CONFIG, DEFAULT_LAYER3_CONFIG } from './types.js';
+//# sourceMappingURL=index.d.ts.map

package/dist/simulation/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/simulation/index.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,EAAE,gBAAgB,EAAE,iBAAiB,EAAE,MAAM,aAAa,CAAC;AAClE,OAAO,EAAE,mBAAmB,EAAE,MAAM,iBAAiB,CAAC;AACtD,OAAO,EAAE,iBAAiB,EAAE,eAAe,EAAE,gBAAgB,EAAE,aAAa,EAAE,eAAe,EAAE,MAAM,mBAAmB,CAAC;AACzH,YAAY,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AACpD,OAAO,EAAE,UAAU,EAAE,aAAa,EAAE,aAAa,EAAE,mBAAmB,EAAE,sBAAsB,EAAE,MAAM,aAAa,CAAC;AACpH,YAAY,EACV,gBAAgB,EAChB,iBAAiB,EACjB,gBAAgB,EAChB,eAAe,EACf,WAAW,EACX,aAAa,EACb,YAAY,EACZ,YAAY,EACZ,YAAY,EACZ,cAAc,GACf,MAAM,YAAY,CAAC;AACpB,OAAO,EAAE,qBAAqB,EAAE,qBAAqB,EAAE,MAAM,YAAY,CAAC"}

package/dist/simulation/index.js

@@ -0,0 +1,31 @@
+"use strict";
+/**
+ * HMA Skill Simulation Engine
+ *
+ * Behavioral simulation that observes what skills actually DO.
+ * Three layers: NanoMind semantic (8ms) -> targeted probes (3s) -> full simulation (30s).
+ * Target: < 1% false positive rate vs industry 95.8%.
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.DEFAULT_LAYER3_CONFIG = exports.DEFAULT_LAYER2_CONFIG = exports.getProbeCategoryCounts = exports.getProbesByCategory = exports.LAYER3_PROBES = exports.LAYER2_PROBES = exports.ALL_PROBES = exports.executeProbeLLM = exports.OllamaBackend = exports.AnthropicBackend = exports.NanoMindBackend = exports.detectBestBackend = exports.MockToolEnvironment = exports.parseSkillProfile = exports.SimulationEngine = void 0;
+var engine_js_1 = require("./engine.js");
+Object.defineProperty(exports, "SimulationEngine", { enumerable: true, get: function () { return engine_js_1.SimulationEngine; } });
+Object.defineProperty(exports, "parseSkillProfile", { enumerable: true, get: function () { return engine_js_1.parseSkillProfile; } });
+var mock_tools_js_1 = require("./mock-tools.js");
+Object.defineProperty(exports, "MockToolEnvironment", { enumerable: true, get: function () { return mock_tools_js_1.MockToolEnvironment; } });
+var llm_executor_js_1 = require("./llm-executor.js");
+Object.defineProperty(exports, "detectBestBackend", { enumerable: true, get: function () { return llm_executor_js_1.detectBestBackend; } });
+Object.defineProperty(exports, "NanoMindBackend", { enumerable: true, get: function () { return llm_executor_js_1.NanoMindBackend; } });
+Object.defineProperty(exports, "AnthropicBackend", { enumerable: true, get: function () { return llm_executor_js_1.AnthropicBackend; } });
+Object.defineProperty(exports, "OllamaBackend", { enumerable: true, get: function () { return llm_executor_js_1.OllamaBackend; } });
+Object.defineProperty(exports, "executeProbeLLM", { enumerable: true, get: function () { return llm_executor_js_1.executeProbeLLM; } });
+var probes_js_1 = require("./probes.js");
+Object.defineProperty(exports, "ALL_PROBES", { enumerable: true, get: function () { return probes_js_1.ALL_PROBES; } });
+Object.defineProperty(exports, "LAYER2_PROBES", { enumerable: true, get: function () { return probes_js_1.LAYER2_PROBES; } });
+Object.defineProperty(exports, "LAYER3_PROBES", { enumerable: true, get: function () { return probes_js_1.LAYER3_PROBES; } });
+Object.defineProperty(exports, "getProbesByCategory", { enumerable: true, get: function () { return probes_js_1.getProbesByCategory; } });
+Object.defineProperty(exports, "getProbeCategoryCounts", { enumerable: true, get: function () { return probes_js_1.getProbeCategoryCounts; } });
+var types_js_1 = require("./types.js");
+Object.defineProperty(exports, "DEFAULT_LAYER2_CONFIG", { enumerable: true, get: function () { return types_js_1.DEFAULT_LAYER2_CONFIG; } });
+Object.defineProperty(exports, "DEFAULT_LAYER3_CONFIG", { enumerable: true, get: function () { return types_js_1.DEFAULT_LAYER3_CONFIG; } });
+//# sourceMappingURL=index.js.map

package/dist/simulation/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/simulation/index.ts"],"names":[],"mappings":";AAAA;;;;;;GAMG;;;AAEH,yCAAkE;AAAzD,6GAAA,gBAAgB,OAAA;AAAE,8GAAA,iBAAiB,OAAA;AAC5C,iDAAsD;AAA7C,oHAAA,mBAAmB,OAAA;AAC5B,qDAAyH;AAAhH,oHAAA,iBAAiB,OAAA;AAAE,kHAAA,eAAe,OAAA;AAAE,mHAAA,gBAAgB,OAAA;AAAE,gHAAA,aAAa,OAAA;AAAE,kHAAA,eAAe,OAAA;AAE7F,yCAAoH;AAA3G,uGAAA,UAAU,OAAA;AAAE,0GAAA,aAAa,OAAA;AAAE,0GAAA,aAAa,OAAA;AAAE,gHAAA,mBAAmB,OAAA;AAAE,mHAAA,sBAAsB,OAAA;AAa9F,uCAA0E;AAAjE,iHAAA,qBAAqB,OAAA;AAAE,iHAAA,qBAAqB,OAAA"}

package/dist/simulation/llm-executor.d.ts

@@ -0,0 +1,58 @@
+/**
+ * LLM-Powered Probe Executor
+ *
+ * Replaces heuristic probe evaluation with actual LLM execution.
+ * Loads the skill as a system prompt, injects the probe as a user message,
+ * and observes what tool calls the LLM decides to make.
+ *
+ * Supports three LLM backends:
+ * 1. NanoMind daemon (localhost:47200) -- free, local, fast
+ * 2. Anthropic Claude -- cloud, most accurate, costs money
+ * 3. Ollama -- local, free, good for development
+ */
+import type { ProbeDefinition, ProbeResult, SkillProfile } from './types.js';
+import { MockToolEnvironment } from './mock-tools.js';
+export interface LLMBackend {
+    name: string;
+    available(): Promise<boolean>;
+    /**
+     * Execute a skill with a probe input and observe what the LLM does.
+     * @param systemPrompt - The skill content (loaded as system prompt)
+     * @param userMessage - The probe input (injected as user message)
+     * @returns The LLM's text response
+     */
+    execute(systemPrompt: string, userMessage: string): Promise<string>;
+}
+export declare class NanoMindBackend implements LLMBackend {
+    name: string;
+    private url;
+    constructor(url?: string);
+    available(): Promise<boolean>;
+    execute(systemPrompt: string, userMessage: string): Promise<string>;
+}
+export declare class AnthropicBackend implements LLMBackend {
+    name: string;
+    private apiKey;
+    constructor(apiKey?: string);
+    available(): Promise<boolean>;
+    execute(systemPrompt: string, userMessage: string): Promise<string>;
+}
+export declare class OllamaBackend implements LLMBackend {
+    name: string;
+    private url;
+    private model;
+    constructor(url?: string, model?: string);
+    available(): Promise<boolean>;
+    execute(systemPrompt: string, userMessage: string): Promise<string>;
+}
+/**
+ * Auto-detect the best available LLM backend.
+ * Priority: NanoMind > Ollama > Anthropic > Heuristic fallback
+ */
+export declare function detectBestBackend(): Promise<LLMBackend | null>;
+/**
+ * Execute a probe using an LLM backend.
+ * Loads the skill as system prompt, injects probe input, observes response.
+ */
+export declare function executeProbeLLM(backend: LLMBackend, skill: SkillProfile, probe: ProbeDefinition, mockEnv: MockToolEnvironment): Promise<ProbeResult>;
+//# sourceMappingURL=llm-executor.d.ts.map

package/dist/simulation/llm-executor.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"llm-executor.d.ts","sourceRoot":"","sources":["../../src/simulation/llm-executor.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,KAAK,EAAE,eAAe,EAAE,WAAW,EAAE,YAAY,EAAgB,MAAM,YAAY,CAAC;AAC3F,OAAO,EAAE,mBAAmB,EAAE,MAAM,iBAAiB,CAAC;AAMtD,MAAM,WAAW,UAAU;IACzB,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,IAAI,OAAO,CAAC,OAAO,CAAC,CAAC;IAC9B;;;;;OAKG;IACH,OAAO,CAAC,YAAY,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;CACrE;AAMD,qBAAa,eAAgB,YAAW,UAAU;IAChD,IAAI,SAAqB;IACzB,OAAO,CAAC,GAAG,CAAS;gBAER,GAAG,SAA2B;IAIpC,SAAS,IAAI,OAAO,CAAC,OAAO,CAAC;IAO7B,OAAO,CAAC,YAAY,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;CAe1E;AAMD,qBAAa,gBAAiB,YAAW,UAAU;IACjD,IAAI,SAAe;IACnB,OAAO,CAAC,MAAM,CAAS;gBAEX,MAAM,CAAC,EAAE,MAAM;IAIrB,SAAS,IAAI,OAAO,CAAC,OAAO,CAAC;IAI7B,OAAO,CAAC,YAAY,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;CAoB1E;AAMD,qBAAa,aAAc,YAAW,UAAU;IAC9C,IAAI,SAAY;IAChB,OAAO,CAAC,GAAG,CAAS;IACpB,OAAO,CAAC,KAAK,CAAS;gBAEV,GAAG,SAA2B,EAAE,KAAK,SAAa;IAKxD,SAAS,IAAI,OAAO,CAAC,OAAO,CAAC;IAO7B,OAAO,CAAC,YAAY,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;CAkB1E;AAMD;;;GAGG;AACH,wBAAsB,iBAAiB,IAAI,OAAO,CAAC,UAAU,GAAG,IAAI,CAAC,CAcpE;AAED;;;GAGG;AACH,wBAAsB,eAAe,CACnC,OAAO,EAAE,UAAU,EACnB,KAAK,EAAE,YAAY,EACnB,KAAK,EAAE,eAAe,EACtB,OAAO,EAAE,mBAAmB,GAC3B,OAAO,CAAC,WAAW,CAAC,CAqCtB"}