@oculum/scanner 1.0.11 → 1.0.13

package/dist/utils/schema-semantics.js

@@ -0,0 +1,193 @@
+"use strict";
+/**
+ * Schema Semantics Utility
+ * Understands validation schema logic, particularly Zod/Yup/Joi patterns
+ *
+ * This addresses false positives where OR validation (either A or B required)
+ * is flagged as "bypass" when individual fields are marked optional.
+ *
+ * Key insight: `.optional()` + `.refine()` = OR validation, not bypass
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.analyzeSchemaSemantics = analyzeSchemaSemantics;
+exports.isLegitimateOrValidation = isLegitimateOrValidation;
+exports.is2FAOrValidation = is2FAOrValidation;
+/**
+ * Patterns that indicate cross-field validation refinements
+ */
+const REFINEMENT_PATTERNS = [
+    // Zod .refine() with OR logic
+    /\.refine\s*\(\s*\([^)]*\)\s*=>\s*[^,]*\|\|/,
+    /\.refine\s*\(\s*\(\s*\{\s*\w+\s*,/, // Destructured refine
+    /\.refine\s*\(\s*\(?data\)?\s*=>/, // data => refine
+    // Zod .superRefine()
+    /\.superRefine\s*\(/,
+    // Yup .test() with cross-field validation
+    /\.test\s*\(\s*['"][^'"]+['"]\s*,\s*['"][^'"]+['"]\s*,\s*function/,
+    /\.test\s*\(\s*['"][^'"]+['"]\s*,\s*['"][^'"]+['"]\s*,\s*\([^)]*\)\s*=>/,
+    // Joi custom validation
+    /\.custom\s*\(/,
+    /Joi\.alternatives\s*\(\)/,
+    /\.xor\s*\(/,
+    /\.or\s*\(/,
+];
+/**
+ * Patterns indicating OR logic in refinement callbacks
+ */
+const OR_LOGIC_PATTERNS = [
+    // Direct OR in refine: (data) => data.a || data.b
+    /=>\s*[^,;]*\|\|/,
+    // Negated AND: !(a && b)
+    /=>\s*!\s*\([^)]*&&/,
+    // Ternary with OR
+    /\?\s*[^:]*\|\|/,
+];
+/**
+ * Find the start of a schema definition containing a given line
+ */
+function findSchemaStart(lines, lineNumber) {
+    // Look backwards for schema definition patterns
+    const schemaStartPatterns = [
+        /(?:const|let|var)\s+\w+\s*=\s*z\.object\s*\(/,
+        /(?:const|let|var)\s+\w+\s*=\s*Yup\.object\s*\(/,
+        /(?:const|let|var)\s+\w+\s*=\s*Joi\.object\s*\(/,
+        /export\s+(?:const|let|var)\s+\w+\s*=\s*z\.object\s*\(/,
+        /z\.object\s*\(\s*\{/,
+        /Yup\.object\s*\(\s*\{/,
+        /Joi\.object\s*\(\s*\{/,
+    ];
+    for (let i = lineNumber; i >= 0; i--) {
+        if (schemaStartPatterns.some(pattern => pattern.test(lines[i]))) {
+            return i;
+        }
+        // Don't look back more than 50 lines
+        if (lineNumber - i > 50)
+            break;
+    }
+    return Math.max(0, lineNumber - 10);
+}
+/**
+ * Find the end of a schema definition containing a given line
+ */
+function findSchemaEnd(lines, schemaStart) {
+    let braceCount = 0;
+    let inSchema = false;
+    for (let i = schemaStart; i < lines.length; i++) {
+        const line = lines[i];
+        // Count braces/parens
+        for (const char of line) {
+            if (char === '(' || char === '{') {
+                braceCount++;
+                inSchema = true;
+            }
+            else if (char === ')' || char === '}') {
+                braceCount--;
+            }
+        }
+        // Schema ends when we return to zero brace count after entering
+        if (inSchema && braceCount === 0) {
+            return i + 1;
+        }
+        // Don't look forward more than 100 lines
+        if (i - schemaStart > 100)
+            break;
+    }
+    return Math.min(lines.length, schemaStart + 50);
+}
+/**
+ * Count .optional() calls in content
+ */
+function countOptionalFields(content) {
+    const matches = content.match(/\.optional\s*\(\s*\)/g);
+    return matches ? matches.length : 0;
+}
+/**
+ * Analyze a validation schema for OR-logic refinement patterns
+ *
+ * @param content - The full file content
+ * @param lineNumber - The 0-indexed line number where the pattern was found
+ * @returns SchemaAnalysis with refinement information
+ */
+function analyzeSchemaSemantics(content, lineNumber) {
+    const lines = content.split('\n');
+    // Find the schema boundaries
+    const schemaStart = findSchemaStart(lines, lineNumber);
+    const schemaEnd = findSchemaEnd(lines, schemaStart);
+    // Extract schema content
+    const schemaLines = lines.slice(schemaStart, schemaEnd);
+    const schemaContent = schemaLines.join('\n');
+    // Check for refinement patterns
+    const hasRefine = REFINEMENT_PATTERNS.some(pattern => pattern.test(schemaContent));
+    const hasSuperRefine = /\.superRefine\s*\(/.test(schemaContent);
+    const hasOrLogic = OR_LOGIC_PATTERNS.some(pattern => pattern.test(schemaContent));
+    // Count optional fields
+    const optionalCount = countOptionalFields(schemaContent);
+    // Determine if this is OR validation
+    // OR validation pattern: multiple optional fields + refine with || logic
+    const isOrValidation = optionalCount >= 2 && hasRefine && hasOrLogic;
+    // Determine refinement type
+    let refinementType = 'none';
+    if (hasSuperRefine) {
+        refinementType = 'super_refine';
+    }
+    else if (hasOrLogic && hasRefine) {
+        refinementType = 'or_logic';
+    }
+    else if (hasRefine) {
+        refinementType = 'custom_validation';
+    }
+    return {
+        hasRefinement: hasRefine || hasSuperRefine,
+        refinementType,
+        optionalFieldCount: optionalCount,
+        isOrValidation,
+    };
+}
+/**
+ * Check if a line with .optional() is part of legitimate OR validation
+ *
+ * @param content - The full file content
+ * @param lineNumber - The 0-indexed line number where .optional() was found
+ * @returns true if this is part of OR validation (not a bypass)
+ */
+function isLegitimateOrValidation(content, lineNumber) {
+    const analysis = analyzeSchemaSemantics(content, lineNumber);
+    return analysis.isOrValidation || analysis.refinementType === 'super_refine';
+}
+/**
+ * Check if a 2FA-related field with .optional() has proper OR validation
+ * Specifically for 2FA patterns where either totpCode OR backupCode is required
+ *
+ * @param content - The full file content
+ * @param lineNumber - The 0-indexed line number
+ * @returns true if this is legitimate 2FA OR validation
+ */
+function is2FAOrValidation(content, lineNumber) {
+    const lines = content.split('\n');
+    const line = lines[lineNumber] || '';
+    // Check if this line is about 2FA fields
+    const is2FAField = /\b(totp|otp|backupCode|recoveryCode|twoFactor|2fa|mfa)\b/i.test(line);
+    if (!is2FAField) {
+        return false;
+    }
+    // Check for OR validation in the schema
+    const analysis = analyzeSchemaSemantics(content, lineNumber);
+    // If there's refinement with OR logic, this is legitimate
+    if (analysis.isOrValidation) {
+        return true;
+    }
+    // Check for specific 2FA OR patterns in nearby content
+    const schemaStart = findSchemaStart(lines, lineNumber);
+    const schemaEnd = findSchemaEnd(lines, schemaStart);
+    const schemaContent = lines.slice(schemaStart, schemaEnd).join('\n');
+    // Patterns like: "Either TOTP or backup code required"
+    const twoFAOrPatterns = [
+        /either.*or.*required/i,
+        /totp.*\|\|.*backup/i,
+        /backup.*\|\|.*totp/i,
+        /code.*\|\|.*code/i,
+        /one\s+of.*required/i,
+    ];
+    return twoFAOrPatterns.some(pattern => pattern.test(schemaContent));
+}
+//# sourceMappingURL=schema-semantics.js.map

package/dist/utils/schema-semantics.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"schema-semantics.js","sourceRoot":"","sources":["../../src/utils/schema-semantics.ts"],"names":[],"mappings":";AAAA;;;;;;;;GAQG;;AAyHH,wDA0CC;AASD,4DAMC;AAUD,8CAoCC;AAnND;;GAEG;AACH,MAAM,mBAAmB,GAAG;IAC1B,8BAA8B;IAC9B,4CAA4C;IAC5C,mCAAmC,EAAG,sBAAsB;IAC5D,iCAAiC,EAAK,iBAAiB;IAEvD,qBAAqB;IACrB,oBAAoB;IAEpB,0CAA0C;IAC1C,kEAAkE;IAClE,wEAAwE;IAExE,wBAAwB;IACxB,eAAe;IACf,0BAA0B;IAC1B,YAAY;IACZ,WAAW;CACZ,CAAA;AAED;;GAEG;AACH,MAAM,iBAAiB,GAAG;IACxB,kDAAkD;IAClD,iBAAiB;IACjB,yBAAyB;IACzB,oBAAoB;IACpB,kBAAkB;IAClB,gBAAgB;CACjB,CAAA;AAED;;GAEG;AACH,SAAS,eAAe,CAAC,KAAe,EAAE,UAAkB;IAC1D,gDAAgD;IAChD,MAAM,mBAAmB,GAAG;QAC1B,8CAA8C;QAC9C,gDAAgD;QAChD,gDAAgD;QAChD,uDAAuD;QACvD,qBAAqB;QACrB,uBAAuB;QACvB,uBAAuB;KACxB,CAAA;IAED,KAAK,IAAI,CAAC,GAAG,UAAU,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QACrC,IAAI,mBAAmB,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;YAChE,OAAO,CAAC,CAAA;QACV,CAAC;QACD,qCAAqC;QACrC,IAAI,UAAU,GAAG,CAAC,GAAG,EAAE;YAAE,MAAK;IAChC,CAAC;IAED,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,UAAU,GAAG,EAAE,CAAC,CAAA;AACrC,CAAC;AAED;;GAEG;AACH,SAAS,aAAa,CAAC,KAAe,EAAE,WAAmB;IACzD,IAAI,UAAU,GAAG,CAAC,CAAA;IAClB,IAAI,QAAQ,GAAG,KAAK,CAAA;IAEpB,KAAK,IAAI,CAAC,GAAG,WAAW,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAChD,MAAM,IAAI,GAAG,KAAK,CAAC,CAAC,CAAC,CAAA;QAErB,sBAAsB;QACtB,KAAK,MAAM,IAAI,IAAI,IAAI,EAAE,CAAC;YACxB,IAAI,IAAI,KAAK,GAAG,IAAI,IAAI,KAAK,GAAG,EAAE,CAAC;gBACjC,UAAU,EAAE,CAAA;gBACZ,QAAQ,GAAG,IAAI,CAAA;YACjB,CAAC;iBAAM,IAAI,IAAI,KAAK,GAAG,IAAI,IAAI,KAAK,GAAG,EAAE,CAAC;gBACxC,UAAU,EAAE,CAAA;YACd,CAAC;QACH,CAAC;QAED,gEAAgE;QAChE,IAAI,QAAQ,IAAI,UAAU,KAAK,CAAC,EAAE,CAAC;YACjC,OAAO,CAAC,GAAG,CAAC,CAAA;QACd,CAAC;QAED,yCAAyC;QACzC,IAAI,CAAC,GAAG,WAAW,GAAG,GAAG;YAAE,MAAK;IAClC,CAAC;IAED,OAAO,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,EAAE,WAAW,GAAG,EAAE,CAAC,CAAA;AACjD,CAAC;AAED;;GAEG;AACH,SAAS,mBAAmB,CAAC,OAAe;IAC1C,MAAM,OAAO,GAAG,OAAO,CAAC,KAAK,CAAC,uBAAuB,CAAC,CAAA;IACtD,OAAO,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAA;AACrC,CAAC;AAED;;;;;;GAMG;AACH,SAAgB,sBAAsB,CACpC,OAAe,EACf,UAAkB;IAElB,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,IAAI,CAAC,CAAA;IAEjC,6BAA6B;IAC7B,MAAM,WAAW,GAAG,eAAe,CAAC,KAAK,EAAE,UAAU,CAAC,CAAA;IACtD,MAAM,SAAS,GAAG,aAAa,CAAC,KAAK,EAAE,WAAW,CAAC,CAAA;IAEnD,yBAAyB;IACzB,MAAM,WAAW,GAAG,KAAK,CAAC,KAAK,CAAC,WAAW,EAAE,SAAS,CAAC,CAAA;IACvD,MAAM,aAAa,GAAG,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;IAE5C,gCAAgC;IAChC,MAAM,SAAS,GAAG,mBAAmB,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC,CAAA;IAClF,MAAM,cAAc,GAAG,oBAAoB,CAAC,IAAI,CAAC,aAAa,CAAC,CAAA;IAC/D,MAAM,UAAU,GAAG,iBAAiB,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC,CAAA;IAEjF,wBAAwB;IACxB,MAAM,aAAa,GAAG,mBAAmB,CAAC,aAAa,CAAC,CAAA;IAExD,qCAAqC;IACrC,yEAAyE;IACzE,MAAM,cAAc,GAAG,aAAa,IAAI,CAAC,IAAI,SAAS,IAAI,UAAU,CAAA;IAEpE,4BAA4B;IAC5B,IAAI,cAAc,GAAqC,MAAM,CAAA;IAC7D,IAAI,cAAc,EAAE,CAAC;QACnB,cAAc,GAAG,cAAc,CAAA;IACjC,CAAC;SAAM,IAAI,UAAU,IAAI,SAAS,EAAE,CAAC;QACnC,cAAc,GAAG,UAAU,CAAA;IAC7B,CAAC;SAAM,IAAI,SAAS,EAAE,CAAC;QACrB,cAAc,GAAG,mBAAmB,CAAA;IACtC,CAAC;IAED,OAAO;QACL,aAAa,EAAE,SAAS,IAAI,cAAc;QAC1C,cAAc;QACd,kBAAkB,EAAE,aAAa;QACjC,cAAc;KACf,CAAA;AACH,CAAC;AAED;;;;;;GAMG;AACH,SAAgB,wBAAwB,CACtC,OAAe,EACf,UAAkB;IAElB,MAAM,QAAQ,GAAG,sBAAsB,CAAC,OAAO,EAAE,UAAU,CAAC,CAAA;IAC5D,OAAO,QAAQ,CAAC,cAAc,IAAI,QAAQ,CAAC,cAAc,KAAK,cAAc,CAAA;AAC9E,CAAC;AAED;;;;;;;GAOG;AACH,SAAgB,iBAAiB,CAC/B,OAAe,EACf,UAAkB;IAElB,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,IAAI,CAAC,CAAA;IACjC,MAAM,IAAI,GAAG,KAAK,CAAC,UAAU,CAAC,IAAI,EAAE,CAAA;IAEpC,yCAAyC;IACzC,MAAM,UAAU,GAAG,2DAA2D,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;IACzF,IAAI,CAAC,UAAU,EAAE,CAAC;QAChB,OAAO,KAAK,CAAA;IACd,CAAC;IAED,wCAAwC;IACxC,MAAM,QAAQ,GAAG,sBAAsB,CAAC,OAAO,EAAE,UAAU,CAAC,CAAA;IAE5D,0DAA0D;IAC1D,IAAI,QAAQ,CAAC,cAAc,EAAE,CAAC;QAC5B,OAAO,IAAI,CAAA;IACb,CAAC;IAED,uDAAuD;IACvD,MAAM,WAAW,GAAG,eAAe,CAAC,KAAK,EAAE,UAAU,CAAC,CAAA;IACtD,MAAM,SAAS,GAAG,aAAa,CAAC,KAAK,EAAE,WAAW,CAAC,CAAA;IACnD,MAAM,aAAa,GAAG,KAAK,CAAC,KAAK,CAAC,WAAW,EAAE,SAAS,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;IAEpE,uDAAuD;IACvD,MAAM,eAAe,GAAG;QACtB,uBAAuB;QACvB,qBAAqB;QACrB,qBAAqB;QACrB,mBAAmB;QACnB,qBAAqB;KACtB,CAAA;IAED,OAAO,eAAe,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC,CAAA;AACrE,CAAC"}

package/dist/validate/clients.d.ts

@@ -0,0 +1,44 @@
+/**
+ * AI Provider Client Factories
+ *
+ * Provides lazy-initialized clients for OpenAI and Anthropic APIs.
+ */
+import Anthropic from '@anthropic-ai/sdk';
+import OpenAI from 'openai';
+/**
+ * Initialize Anthropic client
+ */
+export declare function getAnthropicClient(): Anthropic;
+/**
+ * Initialize OpenAI client (singleton)
+ */
+export declare function getOpenAIClient(): OpenAI;
+/**
+ * GPT-5-mini pricing constants (per 1M tokens)
+ */
+export declare const GPT5_MINI_PRICING: {
+    input: number;
+    cached: number;
+    output: number;
+};
+/**
+ * Claude 3.5 Haiku pricing constants (per 1M tokens)
+ */
+export declare const HAIKU_PRICING: {
+    input: number;
+    cacheWrite: number;
+    cacheRead: number;
+    output: number;
+};
+/**
+ * Number of files to include in each API call (Phase 2 optimization)
+ * Batching multiple files reduces API overhead and leverages prompt caching better
+ */
+export declare const FILES_PER_API_BATCH = 8;
+/**
+ * Number of API batches to process in parallel (Phase 3 optimization)
+ * Higher values = faster scans but more API load
+ * OpenAI/GPT-5-mini handles this well
+ */
+export declare const PARALLEL_API_BATCHES = 6;
+//# sourceMappingURL=clients.d.ts.map

package/dist/validate/clients.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"clients.d.ts","sourceRoot":"","sources":["../../src/validate/clients.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,SAAS,MAAM,mBAAmB,CAAA;AACzC,OAAO,MAAM,MAAM,QAAQ,CAAA;AAM3B;;GAEG;AACH,wBAAgB,kBAAkB,IAAI,SAAS,CAM9C;AASD;;GAEG;AACH,wBAAgB,eAAe,IAAI,MAAM,CASxC;AAMD;;GAEG;AACH,eAAO,MAAM,iBAAiB;;;;CAI7B,CAAA;AAED;;GAEG;AACH,eAAO,MAAM,aAAa;;;;;CAKzB,CAAA;AAMD;;;GAGG;AACH,eAAO,MAAM,mBAAmB,IAAI,CAAA;AAEpC;;;;GAIG;AACH,eAAO,MAAM,oBAAoB,IAAI,CAAA"}

package/dist/validate/clients.js

@@ -0,0 +1,81 @@
+"use strict";
+/**
+ * AI Provider Client Factories
+ *
+ * Provides lazy-initialized clients for OpenAI and Anthropic APIs.
+ */
+var __importDefault = (this && this.__importDefault) || function (mod) {
+    return (mod && mod.__esModule) ? mod : { "default": mod };
+};
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.PARALLEL_API_BATCHES = exports.FILES_PER_API_BATCH = exports.HAIKU_PRICING = exports.GPT5_MINI_PRICING = void 0;
+exports.getAnthropicClient = getAnthropicClient;
+exports.getOpenAIClient = getOpenAIClient;
+const sdk_1 = __importDefault(require("@anthropic-ai/sdk"));
+const openai_1 = __importDefault(require("openai"));
+// ============================================================================
+// Anthropic Client
+// ============================================================================
+/**
+ * Initialize Anthropic client
+ */
+function getAnthropicClient() {
+    const apiKey = process.env.ANTHROPIC_API_KEY;
+    if (!apiKey) {
+        throw new Error('ANTHROPIC_API_KEY environment variable is not set');
+    }
+    return new sdk_1.default({ apiKey });
+}
+// ============================================================================
+// OpenAI Client
+// ============================================================================
+// Singleton instance for connection reuse
+let openaiClient = null;
+/**
+ * Initialize OpenAI client (singleton)
+ */
+function getOpenAIClient() {
+    if (!openaiClient) {
+        const apiKey = process.env.OPENAI_API_KEY;
+        if (!apiKey) {
+            throw new Error('OPENAI_API_KEY environment variable is not set');
+        }
+        openaiClient = new openai_1.default({ apiKey });
+    }
+    return openaiClient;
+}
+// ============================================================================
+// Pricing Constants
+// ============================================================================
+/**
+ * GPT-5-mini pricing constants (per 1M tokens)
+ */
+exports.GPT5_MINI_PRICING = {
+    input: 0.25, // $0.25 per 1M tokens
+    cached: 0.025, // $0.025 per 1M tokens (10% of input)
+    output: 2.00, // $2.00 per 1M tokens
+};
+/**
+ * Claude 3.5 Haiku pricing constants (per 1M tokens)
+ */
+exports.HAIKU_PRICING = {
+    input: 0.80, // $0.80 per 1M tokens
+    cacheWrite: 1.00, // $1.00 per 1M tokens (5m cache)
+    cacheRead: 0.08, // $0.08 per 1M tokens
+    output: 4.00, // $4.00 per 1M tokens
+};
+// ============================================================================
+// Batching Configuration
+// ============================================================================
+/**
+ * Number of files to include in each API call (Phase 2 optimization)
+ * Batching multiple files reduces API overhead and leverages prompt caching better
+ */
+exports.FILES_PER_API_BATCH = 8;
+/**
+ * Number of API batches to process in parallel (Phase 3 optimization)
+ * Higher values = faster scans but more API load
+ * OpenAI/GPT-5-mini handles this well
+ */
+exports.PARALLEL_API_BATCHES = 6;
+//# sourceMappingURL=clients.js.map

package/dist/validate/clients.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"clients.js","sourceRoot":"","sources":["../../src/validate/clients.ts"],"names":[],"mappings":";AAAA;;;;GAIG;;;;;;AAYH,gDAMC;AAYD,0CASC;AArCD,4DAAyC;AACzC,oDAA2B;AAE3B,+EAA+E;AAC/E,mBAAmB;AACnB,+EAA+E;AAE/E;;GAEG;AACH,SAAgB,kBAAkB;IAChC,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAA;IAC5C,IAAI,CAAC,MAAM,EAAE,CAAC;QACZ,MAAM,IAAI,KAAK,CAAC,mDAAmD,CAAC,CAAA;IACtE,CAAC;IACD,OAAO,IAAI,aAAS,CAAC,EAAE,MAAM,EAAE,CAAC,CAAA;AAClC,CAAC;AAED,+EAA+E;AAC/E,gBAAgB;AAChB,+EAA+E;AAE/E,0CAA0C;AAC1C,IAAI,YAAY,GAAkB,IAAI,CAAA;AAEtC;;GAEG;AACH,SAAgB,eAAe;IAC7B,IAAI,CAAC,YAAY,EAAE,CAAC;QAClB,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,cAAc,CAAA;QACzC,IAAI,CAAC,MAAM,EAAE,CAAC;YACZ,MAAM,IAAI,KAAK,CAAC,gDAAgD,CAAC,CAAA;QACnE,CAAC;QACD,YAAY,GAAG,IAAI,gBAAM,CAAC,EAAE,MAAM,EAAE,CAAC,CAAA;IACvC,CAAC;IACD,OAAO,YAAY,CAAA;AACrB,CAAC;AAED,+EAA+E;AAC/E,oBAAoB;AACpB,+EAA+E;AAE/E;;GAEG;AACU,QAAA,iBAAiB,GAAG;IAC/B,KAAK,EAAE,IAAI,EAAO,sBAAsB;IACxC,MAAM,EAAE,KAAK,EAAK,sCAAsC;IACxD,MAAM,EAAE,IAAI,EAAM,sBAAsB;CACzC,CAAA;AAED;;GAEG;AACU,QAAA,aAAa,GAAG;IAC3B,KAAK,EAAE,IAAI,EAAQ,sBAAsB;IACzC,UAAU,EAAE,IAAI,EAAG,iCAAiC;IACpD,SAAS,EAAE,IAAI,EAAI,sBAAsB;IACzC,MAAM,EAAE,IAAI,EAAO,sBAAsB;CAC1C,CAAA;AAED,+EAA+E;AAC/E,yBAAyB;AACzB,+EAA+E;AAE/E;;;GAGG;AACU,QAAA,mBAAmB,GAAG,CAAC,CAAA;AAEpC;;;;GAIG;AACU,QAAA,oBAAoB,GAAG,CAAC,CAAA"}

package/dist/validate/index.d.ts

@@ -0,0 +1,41 @@
+/**
+ * Layer 3: AI Semantic Analysis
+ *
+ * Uses Claude to perform deep security analysis including:
+ * - Taint analysis (data flow from sources to sinks)
+ * - Business logic flaw detection
+ * - Missing authorization checks
+ * - Cryptography validation
+ * - Data exposure detection
+ * - Framework-specific deep analysis
+ *
+ * Also provides high-context validation for Layer 1/2 findings.
+ */
+import type { Vulnerability, ScanFile } from '../shared/types';
+import type { ContextEngineResult } from '../model/taint-types';
+import type { AIValidationResult, Layer3Context } from './types';
+export type { ValidationStats, AIValidationResult, Layer3Context } from './types';
+export { applyAutoDismissRules } from '../score/auto-dismiss';
+/**
+ * Analyze a single file using AI for deep security analysis (Layer 3)
+ */
+export declare function analyzeWithAI(file: ScanFile, context?: Layer3Context): Promise<Vulnerability[]>;
+/**
+ * Batch analyze multiple files using AI (Layer 3)
+ * Processes files in batches to avoid rate limits
+ */
+export declare function batchAnalyzeWithAI(files: ScanFile[], context?: Layer3Context, maxConcurrent?: number): Promise<Vulnerability[]>;
+/**
+ * Validate Layer 1/2 findings using AI with HIGH-CONTEXT validation
+ *
+ * Key improvements over previous version:
+ * 1. Sends FULL FILE CONTENT (not just snippets) for better context
+ * 2. Includes PROJECT CONTEXT (auth patterns, data access, etc.)
+ * 3. Uses generalised rules from Section 3 of the security model
+ */
+export declare function validateFindingsWithAI(findings: Vulnerability[], files: ScanFile[], ceResult?: ContextEngineResult, onProgress?: (progress: {
+    filesProcessed: number;
+    totalFiles: number;
+    status: string;
+}) => void): Promise<AIValidationResult>;
+//# sourceMappingURL=index.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/validate/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,OAAO,KAAK,EAAE,aAAa,EAAE,QAAQ,EAAE,MAAM,iBAAiB,CAAA;AAC9D,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,sBAAsB,CAAA;AAG/D,OAAO,KAAK,EAAmB,kBAAkB,EAAE,aAAa,EAAa,MAAM,SAAS,CAAA;AAS5F,YAAY,EAAE,eAAe,EAAE,kBAAkB,EAAE,aAAa,EAAE,MAAM,SAAS,CAAA;AACjF,OAAO,EAAE,qBAAqB,EAAE,MAAM,uBAAuB,CAAA;AAM7D;;GAEG;AACH,wBAAsB,aAAa,CACjC,IAAI,EAAE,QAAQ,EACd,OAAO,CAAC,EAAE,aAAa,GACtB,OAAO,CAAC,aAAa,EAAE,CAAC,CA+D1B;AAED;;;GAGG;AACH,wBAAsB,kBAAkB,CACtC,KAAK,EAAE,QAAQ,EAAE,EACjB,OAAO,CAAC,EAAE,aAAa,EACvB,aAAa,GAAE,MAAU,GACxB,OAAO,CAAC,aAAa,EAAE,CAAC,CAqB1B;AAMD;;;;;;;GAOG;AACH,wBAAsB,sBAAsB,CAC1C,QAAQ,EAAE,aAAa,EAAE,EACzB,KAAK,EAAE,QAAQ,EAAE,EACjB,QAAQ,CAAC,EAAE,mBAAmB,EAC9B,UAAU,CAAC,EAAE,CAAC,QAAQ,EAAE;IAAE,cAAc,EAAE,MAAM,CAAC;IAAC,UAAU,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,CAAA;CAAE,KAAK,IAAI,GAC9F,OAAO,CAAC,kBAAkB,CAAC,CAiB7B"}

package/dist/validate/index.js

@@ -0,0 +1,141 @@
+"use strict";
+/**
+ * Layer 3: AI Semantic Analysis
+ *
+ * Uses Claude to perform deep security analysis including:
+ * - Taint analysis (data flow from sources to sinks)
+ * - Business logic flaw detection
+ * - Missing authorization checks
+ * - Cryptography validation
+ * - Data exposure detection
+ * - Framework-specific deep analysis
+ *
+ * Also provides high-context validation for Layer 1/2 findings.
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.applyAutoDismissRules = void 0;
+exports.analyzeWithAI = analyzeWithAI;
+exports.batchAnalyzeWithAI = batchAnalyzeWithAI;
+exports.validateFindingsWithAI = validateFindingsWithAI;
+const types_1 = require("./types");
+const clients_1 = require("./clients");
+const response_parser_1 = require("./utils/response-parser");
+const semantic_analysis_1 = require("./prompts/semantic-analysis");
+const openai_1 = require("./providers/openai");
+const anthropic_1 = require("./providers/anthropic");
+var auto_dismiss_1 = require("../score/auto-dismiss");
+Object.defineProperty(exports, "applyAutoDismissRules", { enumerable: true, get: function () { return auto_dismiss_1.applyAutoDismissRules; } });
+// ============================================================================
+// Layer 3: Deep AI Analysis
+// ============================================================================
+/**
+ * Analyze a single file using AI for deep security analysis (Layer 3)
+ */
+async function analyzeWithAI(file, context) {
+    const client = (0, clients_1.getAnthropicClient)();
+    // Prepare the code with line numbers for reference
+    const numberedCode = file.content
+        .split('\n')
+        .map((line, i) => `${i + 1}: ${line}`)
+        .join('\n');
+    // Build auth context for the prompt
+    const authContext = (0, semantic_analysis_1.buildAuthContextForPrompt)(context);
+    const userMessage = `Analyze this ${file.language} file for security vulnerabilities:
+
+File: ${file.path}${authContext}
+
+\`\`\`${file.language}
+${numberedCode}
+\`\`\`
+
+Return ONLY a JSON array of findings.`;
+    try {
+        const response = await client.messages.create({
+            model: 'claude-3-5-haiku-20241022',
+            max_tokens: 4096,
+            system: semantic_analysis_1.SECURITY_ANALYSIS_PROMPT,
+            messages: [
+                {
+                    role: 'user',
+                    content: userMessage,
+                },
+            ],
+        });
+        // Extract text content from response
+        const textContent = response.content.find((block) => block.type === 'text');
+        if (!textContent || textContent.type !== 'text') {
+            console.error('No text content in AI response');
+            return [];
+        }
+        // Parse the JSON response
+        const findings = (0, response_parser_1.parseAIResponse)(textContent.text);
+        // Convert to Vulnerability format
+        return findings.map((finding, index) => ({
+            id: `ai-${file.path}-${finding.lineNumber}-${index}`,
+            filePath: file.path,
+            lineNumber: finding.lineNumber,
+            lineContent: (0, response_parser_1.getLineContent)(file.content, finding.lineNumber),
+            severity: finding.severity,
+            category: finding.category,
+            title: finding.title,
+            description: finding.description,
+            suggestedFix: finding.suggestedFix,
+            confidence: 'high',
+            layer: 3,
+        }));
+    }
+    catch (error) {
+        console.error('AI analysis error:', error);
+        return [];
+    }
+}
+/**
+ * Batch analyze multiple files using AI (Layer 3)
+ * Processes files in batches to avoid rate limits
+ */
+async function batchAnalyzeWithAI(files, context, maxConcurrent = 3) {
+    const vulnerabilities = [];
+    // Process files in batches to avoid rate limits
+    for (let i = 0; i < files.length; i += maxConcurrent) {
+        const batch = files.slice(i, i + maxConcurrent);
+        const results = await Promise.all(batch.map(file => analyzeWithAI(file, context).catch(err => {
+            console.error(`AI analysis failed for ${file.path}:`, err);
+            return [];
+        })));
+        vulnerabilities.push(...results.flat());
+        // Small delay between batches to avoid rate limits
+        if (i + maxConcurrent < files.length) {
+            await new Promise(resolve => setTimeout(resolve, 500));
+        }
+    }
+    return vulnerabilities;
+}
+// ============================================================================
+// Layer 2.5: High-Context Validation
+// ============================================================================
+/**
+ * Validate Layer 1/2 findings using AI with HIGH-CONTEXT validation
+ *
+ * Key improvements over previous version:
+ * 1. Sends FULL FILE CONTENT (not just snippets) for better context
+ * 2. Includes PROJECT CONTEXT (auth patterns, data access, etc.)
+ * 3. Uses generalised rules from Section 3 of the security model
+ */
+async function validateFindingsWithAI(findings, files, ceResult, onProgress) {
+    // Initialize stats tracking
+    const stats = (0, types_1.createInitialStats)(findings.length);
+    if (findings.length === 0) {
+        return { vulnerabilities: [], stats };
+    }
+    // Check for provider override (GPT-5-mini is default for 47% cost savings)
+    const aiProvider = process.env.AI_PROVIDER || 'openai';
+    if (aiProvider === 'anthropic') {
+        console.log('[AI Validation] Using Anthropic provider (Claude 3.5 Haiku)');
+        return (0, anthropic_1.validateWithAnthropic)(findings, files, ceResult, stats, onProgress);
+    }
+    else {
+        console.log('[AI Validation] Using OpenAI provider (GPT-5-mini)');
+        return (0, openai_1.validateWithOpenAI)(findings, files, ceResult, stats);
+    }
+}
+//# sourceMappingURL=index.js.map

package/dist/validate/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/validate/index.ts"],"names":[],"mappings":";AAAA;;;;;;;;;;;;GAYG;;;AAyBH,sCAkEC;AAMD,gDAyBC;AAcD,wDAsBC;AAvJD,mCAA4C;AAC5C,uCAA8C;AAC9C,6DAA6G;AAC7G,mEAAiG;AACjG,+CAAuD;AACvD,qDAA6D;AAI7D,sDAA6D;AAApD,qHAAA,qBAAqB,OAAA;AAE9B,+EAA+E;AAC/E,4BAA4B;AAC5B,+EAA+E;AAE/E;;GAEG;AACI,KAAK,UAAU,aAAa,CACjC,IAAc,EACd,OAAuB;IAEvB,MAAM,MAAM,GAAG,IAAA,4BAAkB,GAAE,CAAA;IAEnC,mDAAmD;IACnD,MAAM,YAAY,GAAG,IAAI,CAAC,OAAO;SAC9B,KAAK,CAAC,IAAI,CAAC;SACX,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,GAAG,CAAC,KAAK,IAAI,EAAE,CAAC;SACrC,IAAI,CAAC,IAAI,CAAC,CAAA;IAEb,oCAAoC;IACpC,MAAM,WAAW,GAAG,IAAA,6CAAyB,EAAC,OAAO,CAAC,CAAA;IAEtD,MAAM,WAAW,GAAG,gBAAgB,IAAI,CAAC,QAAQ;;QAE3C,IAAI,CAAC,IAAI,GAAG,WAAW;;QAEvB,IAAI,CAAC,QAAQ;EACnB,YAAY;;;sCAGwB,CAAA;IAEpC,IAAI,CAAC;QACH,MAAM,QAAQ,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC;YAC5C,KAAK,EAAE,2BAA2B;YAClC,UAAU,EAAE,IAAI;YAChB,MAAM,EAAE,4CAAwB;YAChC,QAAQ,EAAE;gBACR;oBACE,IAAI,EAAE,MAAM;oBACZ,OAAO,EAAE,WAAW;iBACrB;aACF;SACF,CAAC,CAAA;QAEF,qCAAqC;QACrC,MAAM,WAAW,GAAG,QAAQ,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,KAAuB,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,KAAK,MAAM,CAAC,CAAA;QAC7F,IAAI,CAAC,WAAW,IAAI,WAAW,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;YAChD,OAAO,CAAC,KAAK,CAAC,gCAAgC,CAAC,CAAA;YAC/C,OAAO,EAAE,CAAA;QACX,CAAC;QAED,0BAA0B;QAC1B,MAAM,QAAQ,GAAG,IAAA,iCAAe,EAAC,WAAW,CAAC,IAAI,CAAC,CAAA;QAElD,kCAAkC;QAClC,OAAO,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,KAAK,EAAE,EAAE,CAAC,CAAC;YACvC,EAAE,EAAE,MAAM,IAAI,CAAC,IAAI,IAAI,OAAO,CAAC,UAAU,IAAI,KAAK,EAAE;YACpD,QAAQ,EAAE,IAAI,CAAC,IAAI;YACnB,UAAU,EAAE,OAAO,CAAC,UAAU;YAC9B,WAAW,EAAE,IAAA,gCAAc,EAAC,IAAI,CAAC,OAAO,EAAE,OAAO,CAAC,UAAU,CAAC;YAC7D,QAAQ,EAAE,OAAO,CAAC,QAAQ;YAC1B,QAAQ,EAAE,OAAO,CAAC,QAAQ;YAC1B,KAAK,EAAE,OAAO,CAAC,KAAK;YACpB,WAAW,EAAE,OAAO,CAAC,WAAW;YAChC,YAAY,EAAE,OAAO,CAAC,YAAY;YAClC,UAAU,EAAE,MAAe;YAC3B,KAAK,EAAE,CAAU;SAClB,CAAC,CAAC,CAAA;IACL,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,OAAO,CAAC,KAAK,CAAC,oBAAoB,EAAE,KAAK,CAAC,CAAA;QAC1C,OAAO,EAAE,CAAA;IACX,CAAC;AACH,CAAC;AAED;;;GAGG;AACI,KAAK,UAAU,kBAAkB,CACtC,KAAiB,EACjB,OAAuB,EACvB,gBAAwB,CAAC;IAEzB,MAAM,eAAe,GAAoB,EAAE,CAAA;IAE3C,gDAAgD;IAChD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,IAAI,aAAa,EAAE,CAAC;QACrD,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,GAAG,aAAa,CAAC,CAAA;QAC/C,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,GAAG,CAC/B,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,aAAa,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE;YACzD,OAAO,CAAC,KAAK,CAAC,0BAA0B,IAAI,CAAC,IAAI,GAAG,EAAE,GAAG,CAAC,CAAA;YAC1D,OAAO,EAAE,CAAA;QACX,CAAC,CAAC,CAAC,CACJ,CAAA;QACD,eAAe,CAAC,IAAI,CAAC,GAAG,OAAO,CAAC,IAAI,EAAE,CAAC,CAAA;QAEvC,mDAAmD;QACnD,IAAI,CAAC,GAAG,aAAa,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC;YACrC,MAAM,IAAI,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,GAAG,CAAC,CAAC,CAAA;QACxD,CAAC;IACH,CAAC;IAED,OAAO,eAAe,CAAA;AACxB,CAAC;AAED,+EAA+E;AAC/E,qCAAqC;AACrC,+EAA+E;AAE/E;;;;;;;GAOG;AACI,KAAK,UAAU,sBAAsB,CAC1C,QAAyB,EACzB,KAAiB,EACjB,QAA8B,EAC9B,UAA+F;IAE/F,4BAA4B;IAC5B,MAAM,KAAK,GAAoB,IAAA,0BAAkB,EAAC,QAAQ,CAAC,MAAM,CAAC,CAAA;IAElE,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC1B,OAAO,EAAE,eAAe,EAAE,EAAE,EAAE,KAAK,EAAE,CAAA;IACvC,CAAC;IAED,2EAA2E;IAC3E,MAAM,UAAU,GAAG,OAAO,CAAC,GAAG,CAAC,WAAW,IAAI,QAAQ,CAAA;IACtD,IAAI,UAAU,KAAK,WAAW,EAAE,CAAC;QAC/B,OAAO,CAAC,GAAG,CAAC,6DAA6D,CAAC,CAAA;QAC1E,OAAO,IAAA,iCAAqB,EAAC,QAAQ,EAAE,KAAK,EAAE,QAAQ,EAAE,KAAK,EAAE,UAAU,CAAC,CAAA;IAC5E,CAAC;SAAM,CAAC;QACN,OAAO,CAAC,GAAG,CAAC,oDAAoD,CAAC,CAAA;QACjE,OAAO,IAAA,2BAAkB,EAAC,QAAQ,EAAE,KAAK,EAAE,QAAQ,EAAE,KAAK,CAAC,CAAA;IAC7D,CAAC;AACH,CAAC"}

package/dist/validate/prompts/index.d.ts

@@ -0,0 +1,8 @@
+/**
+ * Prompts Index
+ *
+ * Re-exports all prompt templates and helpers.
+ */
+export { SECURITY_ANALYSIS_PROMPT, buildAuthContextForPrompt, } from './semantic-analysis';
+export { HIGH_CONTEXT_VALIDATION_PROMPT, assembleValidationPrompt, getFullValidationPrompt, } from './validation';
+//# sourceMappingURL=index.d.ts.map

package/dist/validate/prompts/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/validate/prompts/index.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,EACL,wBAAwB,EACxB,yBAAyB,GAC1B,MAAM,qBAAqB,CAAA;AAE5B,OAAO,EACL,8BAA8B,EAC9B,wBAAwB,EACxB,uBAAuB,GACxB,MAAM,cAAc,CAAA"}

package/dist/validate/prompts/index.js

@@ -0,0 +1,16 @@
+"use strict";
+/**
+ * Prompts Index
+ *
+ * Re-exports all prompt templates and helpers.
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.getFullValidationPrompt = exports.assembleValidationPrompt = exports.HIGH_CONTEXT_VALIDATION_PROMPT = exports.buildAuthContextForPrompt = exports.SECURITY_ANALYSIS_PROMPT = void 0;
+var semantic_analysis_1 = require("./semantic-analysis");
+Object.defineProperty(exports, "SECURITY_ANALYSIS_PROMPT", { enumerable: true, get: function () { return semantic_analysis_1.SECURITY_ANALYSIS_PROMPT; } });
+Object.defineProperty(exports, "buildAuthContextForPrompt", { enumerable: true, get: function () { return semantic_analysis_1.buildAuthContextForPrompt; } });
+var validation_1 = require("./validation");
+Object.defineProperty(exports, "HIGH_CONTEXT_VALIDATION_PROMPT", { enumerable: true, get: function () { return validation_1.HIGH_CONTEXT_VALIDATION_PROMPT; } });
+Object.defineProperty(exports, "assembleValidationPrompt", { enumerable: true, get: function () { return validation_1.assembleValidationPrompt; } });
+Object.defineProperty(exports, "getFullValidationPrompt", { enumerable: true, get: function () { return validation_1.getFullValidationPrompt; } });
+//# sourceMappingURL=index.js.map

package/dist/validate/prompts/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/validate/prompts/index.ts"],"names":[],"mappings":";AAAA;;;;GAIG;;;AAEH,yDAG4B;AAF1B,6HAAA,wBAAwB,OAAA;AACxB,8HAAA,yBAAyB,OAAA;AAG3B,2CAIqB;AAHnB,4HAAA,8BAA8B,OAAA;AAC9B,sHAAA,wBAAwB,OAAA;AACxB,qHAAA,uBAAuB,OAAA"}

package/dist/validate/prompts/modules/ai-patterns.d.ts

@@ -0,0 +1,19 @@
+/**
+ * AI Patterns Module
+ *
+ * Categories: ai_pattern, ai_prompt_injection, ai_unsafe_execution,
+ *             ai_overpermissive_tool, suspicious_package, ai_rag_exfiltration,
+ *             ai_endpoint_unprotected, ai_schema_mismatch, ai_package_hallucination,
+ *             ai_rag_corpus_poisoning, ai_rag_pii_leakage, ai_mcp_tool_poisoning,
+ *             ai_mcp_credential_issue, ai_mcp_confused_deputy,
+ *             ai_mcp_description_injection, ai_mcp_server_shadowing,
+ *             ai_mcp_config_secrets, ai_mcp_config_permissions,
+ *             ai_rag_query_injection, ai_rag_embedding_poisoning,
+ *             ai_rag_chunk_injection, ai_package_typosquat, ai_package_malicious,
+ *             ai_unsafe_model_load, ai_unverified_model, ai_unsafe_finetuning,
+ *             ai_excessive_agency
+ *
+ * Contains AI/LLM-specific patterns that require semantic AI reasoning.
+ */
+export declare const AI_PATTERNS_MODULE = "\n### AI/LLM-Specific Patterns\n\n**Prompt Injection (ai_prompt_injection):**\n- User input in system prompt WITHOUT delimiters (code fences, XML tags, separators) -> **HIGH** (real risk)\n- User input in system prompt WITH clear delimiters -> **INFO** (properly fenced)\n- Static prompts with no user interpolation -> **REJECT** (false positive)\n- Prompt templates using proper parameterization/placeholders -> **REJECT**\n\n**LLM Output Execution (ai_unsafe_execution):**\n- LLM output fed to eval()/Function()/exec() WITHOUT sandbox -> **CRITICAL** (arbitrary code execution)\n- LLM output to execution WITH sandbox (vm2, isolated-vm) -> **MEDIUM** (risk mitigated)\n- LLM output to execution WITH validation AND sandbox -> **LOW** (well-protected)\n- LLM output used for display only (console.log, UI) -> **REJECT** (not execution)\n- Generated SQL from LLM without parameterization -> **CRITICAL** (SQL injection)\n- Generated SQL with parameterized queries -> **MEDIUM** (logic may still be wrong)\n\n**Agent Tool Permissions (ai_overpermissive_tool):**\n- Tool with unrestricted file/network/exec access -> **HIGH** (overpermissive)\n- Tool without user context verification -> **MEDIUM** (missing authorization)\n- Tool with proper scoping, allowlists, and user verification -> **LOW** or **REJECT**\n- Test files with tool definitions -> **INFO** or **REJECT**\n\n**Hallucinated Dependencies (suspicious_package):**\n- Package not found in registry -> **CRITICAL** (likely AI-hallucinated name)\n- Very new package (less than 7 days old) with low downloads and typosquat pattern -> **HIGH**\n- Legitimate looking package with source/repo but low popularity -> **MEDIUM** (needs review)\n- Known legitimate package with unusual name (in allowlist) -> **REJECT**\n\n**CRITICAL AI PATTERN RULES**:\n- AI code generation often produces non-existent package names - flag these prominently\n- Prompt injection is NOT the same as XSS - different threat model and severity\n- Sandboxed code execution (vm2, isolated-vm) significantly reduces risk\n- Agent tools need both access restrictions AND user context verification\n\n### RAG Data Exfiltration (ai_rag_exfiltration)\nRetrieval Augmented Generation systems can leak sensitive data across tenant boundaries.\n\n**Unscoped Retrieval Queries:**\n- Vector store query WITHOUT user/tenant filter -> **HIGH** (cross-tenant data access)\n  - .query(), .search(), .similaritySearch() without filter/where/userId/tenantId parameter\n  - LangChain retriever.invoke() without metadata filter\n  - Pinecone/Chroma/Weaviate query without namespace or metadata filter\n- Query WITH proper scoping (filter by userId/tenantId) -> **REJECT** (properly scoped)\n- Query with RLS-enabled Supabase tables -> **LOW/INFO** (verify RLS policy)\n\n**Raw Context Exposure:**\n- Raw sourceDocuments/chunks returned in API response -> **MEDIUM** (data leak to client)\n- Raw context returned WITHOUT authentication -> **HIGH** (public data leak)\n- Filtered response (only IDs, titles, metadata) -> **REJECT** (properly filtered)\n- Response filtering visible nearby (.map, sanitize, redact) -> **INFO**\n\n**Context Logging:**\n- Logging retrieved documents (debug) -> **INFO** (hygiene, not direct risk)\n- Logging full prompts with context -> **LOW** (audit concern if logs are accessible)\n- Persisting prompts/context to database -> **MEDIUM** (sensitive data retention)\n\n**CRITICAL RAG RULES**:\n- Cross-tenant data access is the PRIMARY risk - always check for user/tenant scoping\n- Authenticated endpoints exposing context are MEDIUM; unauthenticated are HIGH\n- Debug logging is INFO severity - it's not a direct vulnerability\n- If RLS or middleware protection is visible, downgrade significantly\n\n### AI Endpoint Protection (ai_endpoint_unprotected)\nAI/LLM API endpoints can incur significant costs and enable data exfiltration.\n\n**No Authentication + No Rate Limiting -> HIGH:**\n- Endpoint calls OpenAI/Anthropic/etc. without any auth check or rate limit\n- Anyone on the internet can abuse the endpoint and run up API costs\n- Potential for prompt exfiltration or model abuse\n\n**Has Rate Limiting but No Authentication -> MEDIUM:**\n- Rate limit provides some protection against abuse\n- Still allows anonymous access to AI functionality\n- Suggest adding authentication\n\n**Has Authentication but No Rate Limiting -> LOW:**\n- Authenticated users could still abuse the endpoint\n- Suggest adding rate limiting for cost control\n- severity: low (suggest improvement)\n\n**Has Both Auth and Rate Limiting -> INFO/REJECT:**\n- Properly protected endpoint\n- REJECT if both are clearly present\n- INFO if you want to note the good pattern\n\n**BYOK (Bring Your Own Key) Endpoints:**\n- If user provides their own API key, risk is LOWER\n- User pays for their own usage - cost abuse is their problem\n- Downgrade severity by one level for BYOK patterns\n\n**Protected by Middleware:**\n- If project context shows auth middleware protecting the route, downgrade to INFO\n- Internal/admin routes should be INFO or REJECT\n\n**CRITICAL ENDPOINT RULES**:\n- Cost abuse is real - unprotected AI endpoints can bankrupt a startup\n- Rate limiting alone isn't enough - need auth to prevent anonymous abuse\n- BYOK endpoints have lower risk since user bears the cost\n- Check for middleware protection before flagging\n\n### Schema/Tooling Mismatch (ai_schema_mismatch)\nAI-generated structured outputs need validation before use in security-sensitive contexts.\n\n**Unvalidated AI Output Parsing:**\n- JSON.parse(response.content) without schema validation -> **MEDIUM**\n  - AI may return malformed or unexpected structures\n  - Suggest zod/ajv/joi validation\n- AI output to EXECUTION SINK (eval, exec, query) without validation -> **HIGH**\n  - Direct path to code/SQL injection\n- AI output to DISPLAY only (console.log, UI render) -> **REJECT**\n  - Not a security issue for display purposes\n- OpenAI Structured Outputs (json_schema in request) -> **REJECT**\n  - API-level validation provides guarantees\n\n**Weak Schema Patterns:**\n- response: any at API boundary -> **MEDIUM** (no type safety)\n- z.any() or z.unknown() -> **LOW** (defeats purpose of validation)\n- z.passthrough() -> **INFO** (allows extra properties, minor concern)\n- Specific schema defined and used -> **REJECT** (properly validated)\n\n**Tool Parameter Validation:**\n- Tool parameter -> file path without validation -> **HIGH** (path traversal)\n- Tool parameter -> shell command without validation -> **CRITICAL** (command injection)\n- Tool parameter -> URL without validation -> **HIGH** (SSRF)\n- Tool parameter -> DB query without validation -> **HIGH** (SQL injection)\n- Tool parameter with allowlist check visible -> **LOW/REJECT** (mitigated)\n\n**CRITICAL SCHEMA RULES**:\n- The severity depends on WHERE the AI output is used, not just that it's parsed\n- Execution sinks (eval, exec, query, fs) need HIGH severity without validation\n- Display-only usage is NOT a security issue\n- Schema validation (zod, ajv, joi) significantly reduces risk\n- OpenAI Structured Outputs provide API-level guarantees\n";
+//# sourceMappingURL=ai-patterns.d.ts.map

package/dist/validate/prompts/modules/ai-patterns.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"ai-patterns.d.ts","sourceRoot":"","sources":["../../../../src/validate/prompts/modules/ai-patterns.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAEH,eAAO,MAAM,kBAAkB,2+NAsI9B,CAAA"}