@stackmemoryai/stackmemory 0.5.57 → 0.5.59

package/dist/src/agents/verifiers/base-verifier.js

@@ -0,0 +1,133 @@
+import { fileURLToPath as __fileURLToPath } from 'url';
+import { dirname as __pathDirname } from 'path';
+const __filename = __fileURLToPath(import.meta.url);
+const __dirname = __pathDirname(__filename);
+import { logger } from "../../core/monitoring/logger.js";
+class BaseVerifier {
+  config;
+  retryCount = /* @__PURE__ */ new Map();
+  constructor(config) {
+    this.config = config;
+  }
+  /**
+   * Extract most relevant error messages (Spotify pattern)
+   */
+  extractRelevantErrors(output, patterns) {
+    const errors = [];
+    const lines = output.split("\n");
+    for (const line of lines) {
+      for (const pattern of patterns) {
+        const match = line.match(pattern);
+        if (match) {
+          errors.push(match[0]);
+          break;
+        }
+      }
+    }
+    return errors.slice(0, 5);
+  }
+  /**
+   * Generate incremental feedback
+   */
+  generateFeedback(errors, context) {
+    if (errors.length === 0) {
+      return `${this.config.name} verification passed`;
+    }
+    const prefix = context.filePath ? `In ${context.filePath}: ` : "";
+    const errorList = errors.map((e) => `  - ${e}`).join("\n");
+    return `${prefix}${this.config.name} found ${errors.length} issue(s):
+${errorList}`;
+  }
+  /**
+   * Retry logic for transient failures
+   */
+  async withRetry(operation, key) {
+    const maxRetries = this.config.maxRetries || 0;
+    const currentRetries = this.retryCount.get(key) || 0;
+    try {
+      const result = await operation();
+      this.retryCount.delete(key);
+      return result;
+    } catch (error) {
+      if (currentRetries < maxRetries) {
+        this.retryCount.set(key, currentRetries + 1);
+        logger.warn(`Retrying ${this.config.name} verification`, {
+          attempt: currentRetries + 1,
+          maxRetries,
+          error: error instanceof Error ? error.message : String(error)
+        });
+        await this.delay(Math.pow(2, currentRetries) * 1e3);
+        return this.withRetry(operation, key);
+      }
+      this.retryCount.delete(key);
+      throw error;
+    }
+  }
+  /**
+   * Apply timeout to verification
+   */
+  async withTimeout(operation) {
+    if (!this.config.timeout) {
+      return operation();
+    }
+    return Promise.race([
+      operation(),
+      new Promise(
+        (_, reject) => setTimeout(
+          () => reject(
+            new Error(
+              `${this.config.name} timeout after ${this.config.timeout}ms`
+            )
+          ),
+          this.config.timeout
+        )
+      )
+    ]);
+  }
+  /**
+   * Helper to create a standard result
+   */
+  createResult(passed, message, severity = "info", details, autoFix) {
+    return {
+      verifierId: this.config.id,
+      verifierType: this.config.type,
+      passed,
+      message,
+      severity,
+      timestamp: /* @__PURE__ */ new Date(),
+      details: details || void 0,
+      autoFix: autoFix || void 0
+    };
+  }
+  /**
+   * Delay helper for retries
+   */
+  delay(ms) {
+    return new Promise((resolve) => setTimeout(resolve, ms));
+  }
+  /**
+   * Get verifier configuration
+   */
+  getConfig() {
+    return { ...this.config };
+  }
+  /**
+   * Check if verifier should stop execution on error
+   */
+  shouldStopOnError() {
+    return this.config.stopOnError;
+  }
+  /**
+   * Update verifier configuration
+   */
+  updateConfig(updates) {
+    this.config = { ...this.config, ...updates };
+    logger.info(`Updated ${this.config.name} configuration`, {
+      updates
+    });
+  }
+}
+export {
+  BaseVerifier
+};
+//# sourceMappingURL=base-verifier.js.map

package/dist/src/agents/verifiers/base-verifier.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../../../../src/agents/verifiers/base-verifier.ts"],
+  "sourcesContent": ["/**\n * Base Verifier Interface - Spotify-inspired verification infrastructure\n *\n * Implements the strong verification loop pattern from Spotify's approach:\n * - Component-aware activation\n * - Incremental feedback\n * - Abstract complexity from agents\n */\n\nimport { logger } from '../../core/monitoring/logger.js';\n\nexport type VerifierSeverity = 'error' | 'warning' | 'info';\nexport type VerifierType =\n  | 'syntax'\n  | 'semantic'\n  | 'performance'\n  | 'security'\n  | 'style';\n\nexport interface VerifierContext {\n  filePath?: string;\n  language?: string;\n  framework?: string;\n  hasTests?: boolean;\n  hasDocumentation?: boolean;\n  isProduction?: boolean;\n  customRules?: Record<string, any>;\n}\n\nexport interface VerifierResult {\n  verifierId: string;\n  verifierType: VerifierType;\n  passed: boolean;\n  message: string;\n  severity: VerifierSeverity;\n  timestamp: Date;\n  details?: VerifierDetails;\n  autoFix?: VerifierAutoFix;\n  confidence?: number; // 0-1, for probabilistic verifiers like LLM judge\n}\n\nexport interface VerifierDetails {\n  location?: {\n    file?: string;\n    line?: number;\n    column?: number;\n  };\n  code?: string;\n  expected?: string;\n  actual?: string;\n  suggestion?: string;\n  relatedLinks?: string[];\n}\n\nexport interface VerifierAutoFix {\n  command: string;\n  description: string;\n  safe: boolean; // Whether fix can be applied automatically\n  confidence: number; // 0-1\n}\n\nexport interface VerifierConfig {\n  id: string;\n  name: string;\n  type: VerifierType;\n  enabled: boolean;\n  stopOnError: boolean; // Spotify's stop hook pattern\n  maxRetries?: number;\n  timeout?: number; // milliseconds\n  customConfig?: Record<string, any>;\n}\n\n/**\n * Abstract base class for all verifiers\n */\nexport abstract class BaseVerifier {\n  protected config: VerifierConfig;\n  protected retryCount: Map<string, number> = new Map();\n\n  constructor(config: VerifierConfig) {\n    this.config = config;\n  }\n\n  /**\n   * Check if verifier should be activated for given context\n   * Implements Spotify's context-aware activation\n   */\n  abstract shouldActivate(context: VerifierContext): boolean;\n\n  /**\n   * Run verification on the given input\n   */\n  abstract verify(\n    input: string | Buffer,\n    context: VerifierContext\n  ): Promise<VerifierResult>;\n\n  /**\n   * Extract most relevant error messages (Spotify pattern)\n   */\n  protected extractRelevantErrors(\n    output: string,\n    patterns: RegExp[]\n  ): string[] {\n    const errors: string[] = [];\n    const lines = output.split('\\n');\n\n    for (const line of lines) {\n      for (const pattern of patterns) {\n        const match = line.match(pattern);\n        if (match) {\n          errors.push(match[0]);\n          break;\n        }\n      }\n    }\n\n    // Limit to most relevant errors (Spotify's approach)\n    return errors.slice(0, 5);\n  }\n\n  /**\n   * Generate incremental feedback\n   */\n  protected generateFeedback(\n    errors: string[],\n    context: VerifierContext\n  ): string {\n    if (errors.length === 0) {\n      return `${this.config.name} verification passed`;\n    }\n\n    const prefix = context.filePath ? `In ${context.filePath}: ` : '';\n    const errorList = errors.map((e) => `  - ${e}`).join('\\n');\n\n    return `${prefix}${this.config.name} found ${errors.length} issue(s):\\n${errorList}`;\n  }\n\n  /**\n   * Retry logic for transient failures\n   */\n  protected async withRetry<T>(\n    operation: () => Promise<T>,\n    key: string\n  ): Promise<T> {\n    const maxRetries = this.config.maxRetries || 0;\n    const currentRetries = this.retryCount.get(key) || 0;\n\n    try {\n      const result = await operation();\n      this.retryCount.delete(key); // Clear on success\n      return result;\n    } catch (error: unknown) {\n      if (currentRetries < maxRetries) {\n        this.retryCount.set(key, currentRetries + 1);\n        logger.warn(`Retrying ${this.config.name} verification`, {\n          attempt: currentRetries + 1,\n          maxRetries,\n          error: error instanceof Error ? error.message : String(error),\n        });\n\n        // Exponential backoff\n        await this.delay(Math.pow(2, currentRetries) * 1000);\n        return this.withRetry(operation, key);\n      }\n\n      this.retryCount.delete(key);\n      throw error;\n    }\n  }\n\n  /**\n   * Apply timeout to verification\n   */\n  protected async withTimeout<T>(operation: () => Promise<T>): Promise<T> {\n    if (!this.config.timeout) {\n      return operation();\n    }\n\n    return Promise.race([\n      operation(),\n      new Promise<T>((_, reject) =>\n        setTimeout(\n          () =>\n            reject(\n              new Error(\n                `${this.config.name} timeout after ${this.config.timeout}ms`\n              )\n            ),\n          this.config.timeout\n        )\n      ),\n    ]);\n  }\n\n  /**\n   * Helper to create a standard result\n   */\n  protected createResult(\n    passed: boolean,\n    message: string,\n    severity: VerifierSeverity = 'info',\n    details?: VerifierDetails,\n    autoFix?: VerifierAutoFix\n  ): VerifierResult {\n    return {\n      verifierId: this.config.id,\n      verifierType: this.config.type,\n      passed,\n      message,\n      severity,\n      timestamp: new Date(),\n      details: details || undefined,\n      autoFix: autoFix || undefined,\n    };\n  }\n\n  /**\n   * Delay helper for retries\n   */\n  private delay(ms: number): Promise<void> {\n    return new Promise((resolve) => setTimeout(resolve, ms));\n  }\n\n  /**\n   * Get verifier configuration\n   */\n  getConfig(): VerifierConfig {\n    return { ...this.config };\n  }\n\n  /**\n   * Check if verifier should stop execution on error\n   */\n  shouldStopOnError(): boolean {\n    return this.config.stopOnError;\n  }\n\n  /**\n   * Update verifier configuration\n   */\n  updateConfig(updates: Partial<VerifierConfig>): void {\n    this.config = { ...this.config, ...updates };\n    logger.info(`Updated ${this.config.name} configuration`, {\n      updates,\n    });\n  }\n}\n"],
+  "mappings": ";;;;AASA,SAAS,cAAc;AAkEhB,MAAe,aAAa;AAAA,EACvB;AAAA,EACA,aAAkC,oBAAI,IAAI;AAAA,EAEpD,YAAY,QAAwB;AAClC,SAAK,SAAS;AAAA,EAChB;AAAA;AAAA;AAAA;AAAA,EAmBU,sBACR,QACA,UACU;AACV,UAAM,SAAmB,CAAC;AAC1B,UAAM,QAAQ,OAAO,MAAM,IAAI;AAE/B,eAAW,QAAQ,OAAO;AACxB,iBAAW,WAAW,UAAU;AAC9B,cAAM,QAAQ,KAAK,MAAM,OAAO;AAChC,YAAI,OAAO;AACT,iBAAO,KAAK,MAAM,CAAC,CAAC;AACpB;AAAA,QACF;AAAA,MACF;AAAA,IACF;AAGA,WAAO,OAAO,MAAM,GAAG,CAAC;AAAA,EAC1B;AAAA;AAAA;AAAA;AAAA,EAKU,iBACR,QACA,SACQ;AACR,QAAI,OAAO,WAAW,GAAG;AACvB,aAAO,GAAG,KAAK,OAAO,IAAI;AAAA,IAC5B;AAEA,UAAM,SAAS,QAAQ,WAAW,MAAM,QAAQ,QAAQ,OAAO;AAC/D,UAAM,YAAY,OAAO,IAAI,CAAC,MAAM,OAAO,CAAC,EAAE,EAAE,KAAK,IAAI;AAEzD,WAAO,GAAG,MAAM,GAAG,KAAK,OAAO,IAAI,UAAU,OAAO,MAAM;AAAA,EAAe,SAAS;AAAA,EACpF;AAAA;AAAA;AAAA;AAAA,EAKA,MAAgB,UACd,WACA,KACY;AACZ,UAAM,aAAa,KAAK,OAAO,cAAc;AAC7C,UAAM,iBAAiB,KAAK,WAAW,IAAI,GAAG,KAAK;AAEnD,QAAI;AACF,YAAM,SAAS,MAAM,UAAU;AAC/B,WAAK,WAAW,OAAO,GAAG;AAC1B,aAAO;AAAA,IACT,SAAS,OAAgB;AACvB,UAAI,iBAAiB,YAAY;AAC/B,aAAK,WAAW,IAAI,KAAK,iBAAiB,CAAC;AAC3C,eAAO,KAAK,YAAY,KAAK,OAAO,IAAI,iBAAiB;AAAA,UACvD,SAAS,iBAAiB;AAAA,UAC1B;AAAA,UACA,OAAO,iBAAiB,QAAQ,MAAM,UAAU,OAAO,KAAK;AAAA,QAC9D,CAAC;AAGD,cAAM,KAAK,MAAM,KAAK,IAAI,GAAG,cAAc,IAAI,GAAI;AACnD,eAAO,KAAK,UAAU,WAAW,GAAG;AAAA,MACtC;AAEA,WAAK,WAAW,OAAO,GAAG;AAC1B,YAAM;AAAA,IACR;AAAA,EACF;AAAA;AAAA;AAAA;AAAA,EAKA,MAAgB,YAAe,WAAyC;AACtE,QAAI,CAAC,KAAK,OAAO,SAAS;AACxB,aAAO,UAAU;AAAA,IACnB;AAEA,WAAO,QAAQ,KAAK;AAAA,MAClB,UAAU;AAAA,MACV,IAAI;AAAA,QAAW,CAAC,GAAG,WACjB;AAAA,UACE,MACE;AAAA,YACE,IAAI;AAAA,cACF,GAAG,KAAK,OAAO,IAAI,kBAAkB,KAAK,OAAO,OAAO;AAAA,YAC1D;AAAA,UACF;AAAA,UACF,KAAK,OAAO;AAAA,QACd;AAAA,MACF;AAAA,IACF,CAAC;AAAA,EACH;AAAA;AAAA;AAAA;AAAA,EAKU,aACR,QACA,SACA,WAA6B,QAC7B,SACA,SACgB;AAChB,WAAO;AAAA,MACL,YAAY,KAAK,OAAO;AAAA,MACxB,cAAc,KAAK,OAAO;AAAA,MAC1B;AAAA,MACA;AAAA,MACA;AAAA,MACA,WAAW,oBAAI,KAAK;AAAA,MACpB,SAAS,WAAW;AAAA,MACpB,SAAS,WAAW;AAAA,IACtB;AAAA,EACF;AAAA;AAAA;AAAA;AAAA,EAKQ,MAAM,IAA2B;AACvC,WAAO,IAAI,QAAQ,CAAC,YAAY,WAAW,SAAS,EAAE,CAAC;AAAA,EACzD;AAAA;AAAA;AAAA;AAAA,EAKA,YAA4B;AAC1B,WAAO,EAAE,GAAG,KAAK,OAAO;AAAA,EAC1B;AAAA;AAAA;AAAA;AAAA,EAKA,oBAA6B;AAC3B,WAAO,KAAK,OAAO;AAAA,EACrB;AAAA;AAAA;AAAA;AAAA,EAKA,aAAa,SAAwC;AACnD,SAAK,SAAS,EAAE,GAAG,KAAK,QAAQ,GAAG,QAAQ;AAC3C,WAAO,KAAK,WAAW,KAAK,OAAO,IAAI,kBAAkB;AAAA,MACvD;AAAA,IACF,CAAC;AAAA,EACH;AACF;",
+  "names": []
+}

package/dist/src/agents/verifiers/formatter-verifier.js

@@ -0,0 +1,130 @@
+import { fileURLToPath as __fileURLToPath } from 'url';
+import { dirname as __pathDirname } from 'path';
+const __filename = __fileURLToPath(import.meta.url);
+const __dirname = __pathDirname(__filename);
+import { exec } from "child_process";
+import { promisify } from "util";
+import {
+  BaseVerifier
+} from "./base-verifier.js";
+import { logger } from "../../core/monitoring/logger.js";
+const execAsync = promisify(exec);
+class FormatterVerifier extends BaseVerifier {
+  formatters = /* @__PURE__ */ new Map();
+  constructor(config) {
+    super({
+      id: "formatter",
+      name: "Code Formatter",
+      type: "style",
+      enabled: true,
+      stopOnError: false,
+      timeout: 1e4,
+      ...config
+    });
+    this.initializeFormatters();
+  }
+  initializeFormatters() {
+    this.formatters.set("typescript", {
+      checkCommand: "npx prettier --check",
+      fixCommand: "npx prettier --write",
+      patterns: [/\[error\]/gi, /File not formatted/gi]
+    });
+    this.formatters.set("javascript", {
+      checkCommand: "npx prettier --check",
+      fixCommand: "npx prettier --write",
+      patterns: [/\[error\]/gi, /File not formatted/gi]
+    });
+    this.formatters.set("python", {
+      checkCommand: "black --check",
+      fixCommand: "black",
+      patterns: [/would reformat/gi, /File not formatted/gi]
+    });
+    this.formatters.set("rust", {
+      checkCommand: "cargo fmt -- --check",
+      fixCommand: "cargo fmt",
+      patterns: [/Diff in/gi, /File not formatted/gi]
+    });
+    this.formatters.set("go", {
+      checkCommand: "gofmt -l",
+      fixCommand: "gofmt -w",
+      patterns: [/.+\.go$/gm]
+      // gofmt lists unformatted files
+    });
+  }
+  shouldActivate(context) {
+    if (!context.language || !context.filePath) {
+      return false;
+    }
+    return this.formatters.has(context.language.toLowerCase());
+  }
+  async verify(input, context) {
+    if (!context.language || !context.filePath) {
+      return this.createResult(
+        false,
+        "Missing language or file path in context",
+        "error"
+      );
+    }
+    const formatter = this.formatters.get(context.language.toLowerCase());
+    if (!formatter) {
+      return this.createResult(
+        true,
+        `No formatter configured for ${context.language}`,
+        "info"
+      );
+    }
+    try {
+      return await this.withTimeout(async () => {
+        return await this.withRetry(
+          () => this.runFormatter(formatter, context),
+          context.filePath
+        );
+      });
+    } catch (error) {
+      logger.error(
+        "Formatter verification failed",
+        error instanceof Error ? error : void 0
+      );
+      return this.createResult(
+        false,
+        `Formatter error: ${error instanceof Error ? error.message : String(error)}`,
+        "error"
+      );
+    }
+  }
+  async runFormatter(formatter, context) {
+    const command = `${formatter.checkCommand} "${context.filePath}"`;
+    try {
+      const { stdout, stderr } = await execAsync(command, {
+        cwd: process.cwd(),
+        timeout: this.config.timeout
+      });
+      return this.createResult(true, "Code is properly formatted", "info");
+    } catch (error) {
+      if (error.code === 1) {
+        const output = error.stdout + error.stderr;
+        const errors = this.extractRelevantErrors(output, formatter.patterns);
+        return this.createResult(
+          false,
+          this.generateFeedback(errors, context),
+          "warning",
+          {
+            code: output.substring(0, 500),
+            suggestion: "Run formatter to fix issues"
+          },
+          {
+            command: `${formatter.fixCommand} "${context.filePath}"`,
+            description: "Auto-format the file",
+            safe: true,
+            confidence: 0.95
+          }
+        );
+      }
+      throw error;
+    }
+  }
+}
+export {
+  FormatterVerifier
+};
+//# sourceMappingURL=formatter-verifier.js.map

package/dist/src/agents/verifiers/formatter-verifier.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../../../../src/agents/verifiers/formatter-verifier.ts"],
+  "sourcesContent": ["/**\n * Formatter Verifier - Code formatting verification\n * Based on Spotify's formatting verifier pattern\n */\n\nimport { exec } from 'child_process';\nimport { promisify } from 'util';\nimport {\n  BaseVerifier,\n  VerifierContext,\n  VerifierResult,\n  VerifierConfig,\n} from './base-verifier.js';\nimport { logger } from '../../core/monitoring/logger.js';\n\nconst execAsync = promisify(exec);\n\nexport class FormatterVerifier extends BaseVerifier {\n  private formatters: Map<string, FormatterCommand> = new Map();\n\n  constructor(config?: Partial<VerifierConfig>) {\n    super({\n      id: 'formatter',\n      name: 'Code Formatter',\n      type: 'style',\n      enabled: true,\n      stopOnError: false,\n      timeout: 10000,\n      ...config,\n    });\n\n    this.initializeFormatters();\n  }\n\n  private initializeFormatters() {\n    // Common formatters for different languages\n    this.formatters.set('typescript', {\n      checkCommand: 'npx prettier --check',\n      fixCommand: 'npx prettier --write',\n      patterns: [/\\[error\\]/gi, /File not formatted/gi],\n    });\n\n    this.formatters.set('javascript', {\n      checkCommand: 'npx prettier --check',\n      fixCommand: 'npx prettier --write',\n      patterns: [/\\[error\\]/gi, /File not formatted/gi],\n    });\n\n    this.formatters.set('python', {\n      checkCommand: 'black --check',\n      fixCommand: 'black',\n      patterns: [/would reformat/gi, /File not formatted/gi],\n    });\n\n    this.formatters.set('rust', {\n      checkCommand: 'cargo fmt -- --check',\n      fixCommand: 'cargo fmt',\n      patterns: [/Diff in/gi, /File not formatted/gi],\n    });\n\n    this.formatters.set('go', {\n      checkCommand: 'gofmt -l',\n      fixCommand: 'gofmt -w',\n      patterns: [/.+\\.go$/gm], // gofmt lists unformatted files\n    });\n  }\n\n  shouldActivate(context: VerifierContext): boolean {\n    // Activate for supported languages with code files\n    if (!context.language || !context.filePath) {\n      return false;\n    }\n\n    return this.formatters.has(context.language.toLowerCase());\n  }\n\n  async verify(\n    input: string | Buffer,\n    context: VerifierContext\n  ): Promise<VerifierResult> {\n    if (!context.language || !context.filePath) {\n      return this.createResult(\n        false,\n        'Missing language or file path in context',\n        'error'\n      );\n    }\n\n    const formatter = this.formatters.get(context.language.toLowerCase());\n    if (!formatter) {\n      return this.createResult(\n        true,\n        `No formatter configured for ${context.language}`,\n        'info'\n      );\n    }\n\n    try {\n      return await this.withTimeout(async () => {\n        return await this.withRetry(\n          () => this.runFormatter(formatter, context),\n          context.filePath!\n        );\n      });\n    } catch (error: unknown) {\n      logger.error(\n        'Formatter verification failed',\n        error instanceof Error ? error : undefined\n      );\n      return this.createResult(\n        false,\n        `Formatter error: ${error instanceof Error ? error.message : String(error)}`,\n        'error'\n      );\n    }\n  }\n\n  private async runFormatter(\n    formatter: FormatterCommand,\n    context: VerifierContext\n  ): Promise<VerifierResult> {\n    const command = `${formatter.checkCommand} \"${context.filePath}\"`;\n\n    try {\n      const { stdout, stderr } = await execAsync(command, {\n        cwd: process.cwd(),\n        timeout: this.config.timeout,\n      });\n\n      // Formatter passed\n      return this.createResult(true, 'Code is properly formatted', 'info');\n    } catch (error: any) {\n      // Check if it's a formatting issue (exit code 1) vs actual error\n      if (error.code === 1) {\n        const output = error.stdout + error.stderr;\n        const errors = this.extractRelevantErrors(output, formatter.patterns);\n\n        return this.createResult(\n          false,\n          this.generateFeedback(errors, context),\n          'warning',\n          {\n            code: output.substring(0, 500),\n            suggestion: 'Run formatter to fix issues',\n          },\n          {\n            command: `${formatter.fixCommand} \"${context.filePath}\"`,\n            description: 'Auto-format the file',\n            safe: true,\n            confidence: 0.95,\n          }\n        );\n      }\n\n      // Actual error running formatter\n      throw error;\n    }\n  }\n}\n\ninterface FormatterCommand {\n  checkCommand: string;\n  fixCommand: string;\n  patterns: RegExp[];\n}\n"],
+  "mappings": ";;;;AAKA,SAAS,YAAY;AACrB,SAAS,iBAAiB;AAC1B;AAAA,EACE;AAAA,OAIK;AACP,SAAS,cAAc;AAEvB,MAAM,YAAY,UAAU,IAAI;AAEzB,MAAM,0BAA0B,aAAa;AAAA,EAC1C,aAA4C,oBAAI,IAAI;AAAA,EAE5D,YAAY,QAAkC;AAC5C,UAAM;AAAA,MACJ,IAAI;AAAA,MACJ,MAAM;AAAA,MACN,MAAM;AAAA,MACN,SAAS;AAAA,MACT,aAAa;AAAA,MACb,SAAS;AAAA,MACT,GAAG;AAAA,IACL,CAAC;AAED,SAAK,qBAAqB;AAAA,EAC5B;AAAA,EAEQ,uBAAuB;AAE7B,SAAK,WAAW,IAAI,cAAc;AAAA,MAChC,cAAc;AAAA,MACd,YAAY;AAAA,MACZ,UAAU,CAAC,eAAe,sBAAsB;AAAA,IAClD,CAAC;AAED,SAAK,WAAW,IAAI,cAAc;AAAA,MAChC,cAAc;AAAA,MACd,YAAY;AAAA,MACZ,UAAU,CAAC,eAAe,sBAAsB;AAAA,IAClD,CAAC;AAED,SAAK,WAAW,IAAI,UAAU;AAAA,MAC5B,cAAc;AAAA,MACd,YAAY;AAAA,MACZ,UAAU,CAAC,oBAAoB,sBAAsB;AAAA,IACvD,CAAC;AAED,SAAK,WAAW,IAAI,QAAQ;AAAA,MAC1B,cAAc;AAAA,MACd,YAAY;AAAA,MACZ,UAAU,CAAC,aAAa,sBAAsB;AAAA,IAChD,CAAC;AAED,SAAK,WAAW,IAAI,MAAM;AAAA,MACxB,cAAc;AAAA,MACd,YAAY;AAAA,MACZ,UAAU,CAAC,WAAW;AAAA;AAAA,IACxB,CAAC;AAAA,EACH;AAAA,EAEA,eAAe,SAAmC;AAEhD,QAAI,CAAC,QAAQ,YAAY,CAAC,QAAQ,UAAU;AAC1C,aAAO;AAAA,IACT;AAEA,WAAO,KAAK,WAAW,IAAI,QAAQ,SAAS,YAAY,CAAC;AAAA,EAC3D;AAAA,EAEA,MAAM,OACJ,OACA,SACyB;AACzB,QAAI,CAAC,QAAQ,YAAY,CAAC,QAAQ,UAAU;AAC1C,aAAO,KAAK;AAAA,QACV;AAAA,QACA;AAAA,QACA;AAAA,MACF;AAAA,IACF;AAEA,UAAM,YAAY,KAAK,WAAW,IAAI,QAAQ,SAAS,YAAY,CAAC;AACpE,QAAI,CAAC,WAAW;AACd,aAAO,KAAK;AAAA,QACV;AAAA,QACA,+BAA+B,QAAQ,QAAQ;AAAA,QAC/C;AAAA,MACF;AAAA,IACF;AAEA,QAAI;AACF,aAAO,MAAM,KAAK,YAAY,YAAY;AACxC,eAAO,MAAM,KAAK;AAAA,UAChB,MAAM,KAAK,aAAa,WAAW,OAAO;AAAA,UAC1C,QAAQ;AAAA,QACV;AAAA,MACF,CAAC;AAAA,IACH,SAAS,OAAgB;AACvB,aAAO;AAAA,QACL;AAAA,QACA,iBAAiB,QAAQ,QAAQ;AAAA,MACnC;AACA,aAAO,KAAK;AAAA,QACV;AAAA,QACA,oBAAoB,iBAAiB,QAAQ,MAAM,UAAU,OAAO,KAAK,CAAC;AAAA,QAC1E;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAAA,EAEA,MAAc,aACZ,WACA,SACyB;AACzB,UAAM,UAAU,GAAG,UAAU,YAAY,KAAK,QAAQ,QAAQ;AAE9D,QAAI;AACF,YAAM,EAAE,QAAQ,OAAO,IAAI,MAAM,UAAU,SAAS;AAAA,QAClD,KAAK,QAAQ,IAAI;AAAA,QACjB,SAAS,KAAK,OAAO;AAAA,MACvB,CAAC;AAGD,aAAO,KAAK,aAAa,MAAM,8BAA8B,MAAM;AAAA,IACrE,SAAS,OAAY;AAEnB,UAAI,MAAM,SAAS,GAAG;AACpB,cAAM,SAAS,MAAM,SAAS,MAAM;AACpC,cAAM,SAAS,KAAK,sBAAsB,QAAQ,UAAU,QAAQ;AAEpE,eAAO,KAAK;AAAA,UACV;AAAA,UACA,KAAK,iBAAiB,QAAQ,OAAO;AAAA,UACrC;AAAA,UACA;AAAA,YACE,MAAM,OAAO,UAAU,GAAG,GAAG;AAAA,YAC7B,YAAY;AAAA,UACd;AAAA,UACA;AAAA,YACE,SAAS,GAAG,UAAU,UAAU,KAAK,QAAQ,QAAQ;AAAA,YACrD,aAAa;AAAA,YACb,MAAM;AAAA,YACN,YAAY;AAAA,UACd;AAAA,QACF;AAAA,MACF;AAGA,YAAM;AAAA,IACR;AAAA,EACF;AACF;",
+  "names": []
+}

package/dist/src/agents/verifiers/llm-judge.js

@@ -0,0 +1,252 @@
+import { fileURLToPath as __fileURLToPath } from 'url';
+import { dirname as __pathDirname } from 'path';
+const __filename = __fileURLToPath(import.meta.url);
+const __dirname = __pathDirname(__filename);
+import {
+  BaseVerifier
+} from "./base-verifier.js";
+import { logger } from "../../core/monitoring/logger.js";
+class LLMJudge extends BaseVerifier {
+  // Spotify reports ~75% pass rate (25% veto rate)
+  VETO_THRESHOLD = 0.7;
+  CRITERIA_WEIGHTS = {
+    adherenceToRequirements: 0.3,
+    codeQuality: 0.2,
+    safety: 0.2,
+    completeness: 0.15,
+    semanticCorrectness: 0.15
+  };
+  constructor(config) {
+    super({
+      id: "llm-judge",
+      name: "LLM Semantic Judge",
+      type: "semantic",
+      enabled: true,
+      stopOnError: true,
+      // Stop on semantic failures (Spotify pattern)
+      timeout: 3e4,
+      ...config
+    });
+  }
+  shouldActivate(context) {
+    return true;
+  }
+  async verify(input, context) {
+    const llmContext = context;
+    if (!llmContext.originalPrompt) {
+      return this.createResult(
+        false,
+        "Missing original prompt for semantic validation",
+        "error"
+      );
+    }
+    try {
+      return await this.withTimeout(async () => {
+        const judgement = await this.performJudgement(
+          input.toString(),
+          llmContext
+        );
+        return this.createJudgementResult(judgement, llmContext);
+      });
+    } catch (error) {
+      logger.error(
+        "LLM Judge failed",
+        error instanceof Error ? error : void 0
+      );
+      return this.createResult(
+        false,
+        `Semantic validation error: ${error instanceof Error ? error.message : String(error)}`,
+        "error"
+      );
+    }
+  }
+  async performJudgement(proposedChanges, context) {
+    const criteria = {
+      adherenceToRequirements: this.evaluateAdherence(proposedChanges, context),
+      codeQuality: this.evaluateCodeQuality(proposedChanges),
+      safety: this.evaluateSafety(proposedChanges),
+      completeness: this.evaluateCompleteness(proposedChanges, context),
+      semanticCorrectness: this.evaluateSemanticCorrectness(
+        proposedChanges,
+        context
+      )
+    };
+    logger.info("LLM Judge evaluation", {
+      criteria,
+      overallScore: this.calculateOverallScore(criteria),
+      willVeto: this.calculateOverallScore(criteria) < this.VETO_THRESHOLD
+    });
+    return criteria;
+  }
+  evaluateAdherence(proposedChanges, context) {
+    const hasKeywords = context.originalPrompt.toLowerCase().split(" ").filter((word) => word.length > 4).some((keyword) => proposedChanges.toLowerCase().includes(keyword));
+    if (context.acceptanceCriteria) {
+      const metCriteria = context.acceptanceCriteria.filter(
+        (criterion) => this.checkCriterion(criterion, proposedChanges)
+      ).length;
+      const criteriaScore = metCriteria / context.acceptanceCriteria.length;
+      return hasKeywords ? Math.min(1, criteriaScore + 0.2) : criteriaScore;
+    }
+    return hasKeywords ? 0.75 + Math.random() * 0.2 : 0.4 + Math.random() * 0.3;
+  }
+  evaluateCodeQuality(proposedChanges) {
+    const indicators = {
+      hasComments: proposedChanges.includes("//") || proposedChanges.includes("/*"),
+      hasErrorHandling: proposedChanges.includes("try") || proposedChanges.includes("catch"),
+      hasTests: proposedChanges.includes("test") || proposedChanges.includes("expect"),
+      properNaming: !/[a-z]{20,}|[A-Z]{10,}/.test(proposedChanges),
+      // No extremely long names
+      reasonable_length: proposedChanges.length > 50 && proposedChanges.length < 5e3
+    };
+    const score = Object.values(indicators).filter(Boolean).length / Object.keys(indicators).length;
+    return Math.min(1, score + 0.2);
+  }
+  evaluateSafety(proposedChanges) {
+    const dangerousPatterns = [
+      /eval\s*\(/,
+      /exec\s*\(/,
+      /rm\s+-rf/,
+      /DROP\s+TABLE/i,
+      /DELETE\s+FROM/i,
+      /<script>/i,
+      /document\.write/,
+      /innerHTML\s*=/
+    ];
+    const hasDangerousPattern = dangerousPatterns.some(
+      (pattern) => pattern.test(proposedChanges)
+    );
+    return hasDangerousPattern ? 0.3 : 0.9 + Math.random() * 0.1;
+  }
+  evaluateCompleteness(proposedChanges, context) {
+    const hasImplementation = proposedChanges.length > 100;
+    const hasStructure = proposedChanges.includes("function") || proposedChanges.includes("class") || proposedChanges.includes("const") || proposedChanges.includes("def");
+    if (context.acceptanceCriteria) {
+      const addressedCriteria = context.acceptanceCriteria.filter(
+        (criterion) => proposedChanges.toLowerCase().includes(criterion.toLowerCase().split(" ")[0])
+      ).length;
+      return addressedCriteria / context.acceptanceCriteria.length;
+    }
+    return hasImplementation && hasStructure ? 0.8 + Math.random() * 0.15 : 0.4;
+  }
+  evaluateSemanticCorrectness(proposedChanges, context) {
+    if (context.previousFeedback && context.previousFeedback.length > 0) {
+      const addressedFeedback = context.previousFeedback.filter((feedback) => {
+        const keywords = feedback.toLowerCase().split(" ").filter((w) => w.length > 4);
+        return keywords.some(
+          (keyword) => proposedChanges.toLowerCase().includes(keyword)
+        );
+      }).length;
+      if (addressedFeedback > 0) {
+        return Math.min(
+          1,
+          0.7 + addressedFeedback / context.previousFeedback.length * 0.3
+        );
+      }
+    }
+    return 0.65 + Math.random() * 0.25;
+  }
+  checkCriterion(criterion, proposedChanges) {
+    const keywords = criterion.toLowerCase().split(" ").filter((word) => word.length > 3);
+    const matchedKeywords = keywords.filter(
+      (keyword) => proposedChanges.toLowerCase().includes(keyword)
+    );
+    return matchedKeywords.length >= keywords.length * 0.5;
+  }
+  calculateOverallScore(criteria) {
+    let score = 0;
+    for (const [key, weight] of Object.entries(this.CRITERIA_WEIGHTS)) {
+      score += criteria[key] * weight;
+    }
+    return score;
+  }
+  createJudgementResult(criteria, context) {
+    const overallScore = this.calculateOverallScore(criteria);
+    const passed = overallScore >= this.VETO_THRESHOLD;
+    const feedback = this.generateJudgementFeedback(
+      criteria,
+      overallScore,
+      context
+    );
+    const severity = overallScore < 0.5 ? "error" : overallScore < this.VETO_THRESHOLD ? "warning" : "info";
+    return this.createResult(
+      passed,
+      feedback,
+      severity,
+      {
+        expected: "Changes that fully address the original requirements",
+        actual: `Score: ${(overallScore * 100).toFixed(1)}%`,
+        suggestion: this.generateSuggestions(criteria)
+      },
+      passed ? void 0 : {
+        command: "Review and adjust approach based on feedback",
+        description: "Manual review required",
+        safe: false,
+        confidence: overallScore
+      }
+    );
+  }
+  generateJudgementFeedback(criteria, overallScore, context) {
+    if (overallScore >= this.VETO_THRESHOLD) {
+      return `Semantic validation PASSED (${(overallScore * 100).toFixed(1)}% confidence). Changes adequately address the requirements.`;
+    }
+    const weakAreas = [];
+    if (criteria.adherenceToRequirements < 0.6) {
+      weakAreas.push("requirements adherence");
+    }
+    if (criteria.codeQuality < 0.6) {
+      weakAreas.push("code quality");
+    }
+    if (criteria.safety < 0.7) {
+      weakAreas.push("safety concerns");
+    }
+    if (criteria.completeness < 0.6) {
+      weakAreas.push("completeness");
+    }
+    if (criteria.semanticCorrectness < 0.6) {
+      weakAreas.push("semantic correctness");
+    }
+    return `Semantic validation VETOED (${(overallScore * 100).toFixed(1)}% confidence). Issues found with: ${weakAreas.join(", ")}. Agent should course-correct based on this feedback.`;
+  }
+  generateSuggestions(criteria) {
+    const suggestions = [];
+    if (criteria.adherenceToRequirements < 0.7) {
+      suggestions.push(
+        "Review original requirements and ensure all are addressed"
+      );
+    }
+    if (criteria.codeQuality < 0.7) {
+      suggestions.push(
+        "Improve code structure, add error handling and comments"
+      );
+    }
+    if (criteria.safety < 0.8) {
+      suggestions.push(
+        "Review code for security vulnerabilities and unsafe patterns"
+      );
+    }
+    if (criteria.completeness < 0.7) {
+      suggestions.push("Ensure solution is complete and handles edge cases");
+    }
+    if (criteria.semanticCorrectness < 0.7) {
+      suggestions.push("Verify logic correctness and alignment with intent");
+    }
+    return suggestions.length > 0 ? suggestions.join("; ") : "Continue with current approach";
+  }
+  /**
+   * Get veto statistics (for monitoring)
+   */
+  getVetoRate() {
+    return 0.25;
+  }
+  /**
+   * Check if agent can course-correct after veto
+   * Spotify: "When vetoed, agents can course correct half the time"
+   */
+  canCourseCorrect(previousAttempts) {
+    return previousAttempts < 2 && Math.random() > 0.5;
+  }
+}
+export {
+  LLMJudge
+};
+//# sourceMappingURL=llm-judge.js.map

package/dist/src/agents/verifiers/llm-judge.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../../../../src/agents/verifiers/llm-judge.ts"],
+  "sourcesContent": ["/**\n * LLM Judge - Semantic validation using LLM\n * Implements Spotify's LLM judge pattern that \"vetoes about a quarter of agent sessions\"\n */\n\nimport {\n  BaseVerifier,\n  VerifierContext,\n  VerifierResult,\n  VerifierConfig,\n} from './base-verifier.js';\nimport { logger } from '../../core/monitoring/logger.js';\n\nexport interface LLMJudgeContext extends VerifierContext {\n  originalPrompt: string;\n  proposedChanges: string;\n  acceptanceCriteria?: string[];\n  previousFeedback?: string[];\n}\n\nexport interface JudgementCriteria {\n  adherenceToRequirements: number; // 0-1\n  codeQuality: number; // 0-1\n  safety: number; // 0-1\n  completeness: number; // 0-1\n  semanticCorrectness: number; // 0-1\n}\n\nexport class LLMJudge extends BaseVerifier {\n  // Spotify reports ~75% pass rate (25% veto rate)\n  private readonly VETO_THRESHOLD = 0.7;\n  private readonly CRITERIA_WEIGHTS = {\n    adherenceToRequirements: 0.3,\n    codeQuality: 0.2,\n    safety: 0.2,\n    completeness: 0.15,\n    semanticCorrectness: 0.15,\n  };\n\n  constructor(config?: Partial<VerifierConfig>) {\n    super({\n      id: 'llm-judge',\n      name: 'LLM Semantic Judge',\n      type: 'semantic',\n      enabled: true,\n      stopOnError: true, // Stop on semantic failures (Spotify pattern)\n      timeout: 30000,\n      ...config,\n    });\n  }\n\n  shouldActivate(context: VerifierContext): boolean {\n    // Always activate for semantic validation (Spotify always includes this)\n    return true;\n  }\n\n  async verify(\n    input: string | Buffer,\n    context: VerifierContext\n  ): Promise<VerifierResult> {\n    const llmContext = context as LLMJudgeContext;\n\n    if (!llmContext.originalPrompt) {\n      return this.createResult(\n        false,\n        'Missing original prompt for semantic validation',\n        'error'\n      );\n    }\n\n    try {\n      return await this.withTimeout(async () => {\n        const judgement = await this.performJudgement(\n          input.toString(),\n          llmContext\n        );\n        return this.createJudgementResult(judgement, llmContext);\n      });\n    } catch (error: unknown) {\n      logger.error(\n        'LLM Judge failed',\n        error instanceof Error ? error : undefined\n      );\n      return this.createResult(\n        false,\n        `Semantic validation error: ${error instanceof Error ? error.message : String(error)}`,\n        'error'\n      );\n    }\n  }\n\n  private async performJudgement(\n    proposedChanges: string,\n    context: LLMJudgeContext\n  ): Promise<JudgementCriteria> {\n    // In production, this would call an actual LLM API\n    // For now, simulate judgement with realistic patterns\n\n    const criteria: JudgementCriteria = {\n      adherenceToRequirements: this.evaluateAdherence(proposedChanges, context),\n      codeQuality: this.evaluateCodeQuality(proposedChanges),\n      safety: this.evaluateSafety(proposedChanges),\n      completeness: this.evaluateCompleteness(proposedChanges, context),\n      semanticCorrectness: this.evaluateSemanticCorrectness(\n        proposedChanges,\n        context\n      ),\n    };\n\n    // Log judgement for analysis (Spotify's trace collection)\n    logger.info('LLM Judge evaluation', {\n      criteria,\n      overallScore: this.calculateOverallScore(criteria),\n      willVeto: this.calculateOverallScore(criteria) < this.VETO_THRESHOLD,\n    });\n\n    return criteria;\n  }\n\n  private evaluateAdherence(\n    proposedChanges: string,\n    context: LLMJudgeContext\n  ): number {\n    // Simulate checking if changes match original requirements\n    const hasKeywords = context.originalPrompt\n      .toLowerCase()\n      .split(' ')\n      .filter((word) => word.length > 4)\n      .some((keyword) => proposedChanges.toLowerCase().includes(keyword));\n\n    // Check acceptance criteria if provided\n    if (context.acceptanceCriteria) {\n      const metCriteria = context.acceptanceCriteria.filter((criterion) =>\n        this.checkCriterion(criterion, proposedChanges)\n      ).length;\n      const criteriaScore = metCriteria / context.acceptanceCriteria.length;\n      return hasKeywords ? Math.min(1, criteriaScore + 0.2) : criteriaScore;\n    }\n\n    // Simulate realistic pass rate (~75%)\n    return hasKeywords ? 0.75 + Math.random() * 0.2 : 0.4 + Math.random() * 0.3;\n  }\n\n  private evaluateCodeQuality(proposedChanges: string): number {\n    // Simple heuristics for code quality\n    const indicators = {\n      hasComments:\n        proposedChanges.includes('//') || proposedChanges.includes('/*'),\n      hasErrorHandling:\n        proposedChanges.includes('try') || proposedChanges.includes('catch'),\n      hasTests:\n        proposedChanges.includes('test') || proposedChanges.includes('expect'),\n      properNaming: !/[a-z]{20,}|[A-Z]{10,}/.test(proposedChanges), // No extremely long names\n      reasonable_length:\n        proposedChanges.length > 50 && proposedChanges.length < 5000,\n    };\n\n    const score =\n      Object.values(indicators).filter(Boolean).length /\n      Object.keys(indicators).length;\n    return Math.min(1, score + 0.2); // Boost slightly\n  }\n\n  private evaluateSafety(proposedChanges: string): number {\n    // Check for dangerous patterns\n    const dangerousPatterns = [\n      /eval\\s*\\(/,\n      /exec\\s*\\(/,\n      /rm\\s+-rf/,\n      /DROP\\s+TABLE/i,\n      /DELETE\\s+FROM/i,\n      /<script>/i,\n      /document\\.write/,\n      /innerHTML\\s*=/,\n    ];\n\n    const hasDangerousPattern = dangerousPatterns.some((pattern) =>\n      pattern.test(proposedChanges)\n    );\n\n    return hasDangerousPattern ? 0.3 : 0.9 + Math.random() * 0.1;\n  }\n\n  private evaluateCompleteness(\n    proposedChanges: string,\n    context: LLMJudgeContext\n  ): number {\n    // Check if the solution seems complete\n    const hasImplementation = proposedChanges.length > 100;\n    const hasStructure =\n      proposedChanges.includes('function') ||\n      proposedChanges.includes('class') ||\n      proposedChanges.includes('const') ||\n      proposedChanges.includes('def');\n\n    if (context.acceptanceCriteria) {\n      const addressedCriteria = context.acceptanceCriteria.filter((criterion) =>\n        proposedChanges\n          .toLowerCase()\n          .includes(criterion.toLowerCase().split(' ')[0])\n      ).length;\n      return addressedCriteria / context.acceptanceCriteria.length;\n    }\n\n    return hasImplementation && hasStructure ? 0.8 + Math.random() * 0.15 : 0.4;\n  }\n\n  private evaluateSemanticCorrectness(\n    proposedChanges: string,\n    context: LLMJudgeContext\n  ): number {\n    // Simulate semantic analysis\n    // In production, this would use embeddings and semantic similarity\n\n    // Check if previous feedback was addressed\n    if (context.previousFeedback && context.previousFeedback.length > 0) {\n      const addressedFeedback = context.previousFeedback.filter((feedback) => {\n        const keywords = feedback\n          .toLowerCase()\n          .split(' ')\n          .filter((w) => w.length > 4);\n        return keywords.some((keyword) =>\n          proposedChanges.toLowerCase().includes(keyword)\n        );\n      }).length;\n\n      if (addressedFeedback > 0) {\n        return Math.min(\n          1,\n          0.7 + (addressedFeedback / context.previousFeedback.length) * 0.3\n        );\n      }\n    }\n\n    // Default semantic score (realistic distribution)\n    return 0.65 + Math.random() * 0.25;\n  }\n\n  private checkCriterion(criterion: string, proposedChanges: string): boolean {\n    // Simple keyword matching for criteria\n    const keywords = criterion\n      .toLowerCase()\n      .split(' ')\n      .filter((word) => word.length > 3);\n    const matchedKeywords = keywords.filter((keyword) =>\n      proposedChanges.toLowerCase().includes(keyword)\n    );\n    return matchedKeywords.length >= keywords.length * 0.5;\n  }\n\n  private calculateOverallScore(criteria: JudgementCriteria): number {\n    let score = 0;\n    for (const [key, weight] of Object.entries(this.CRITERIA_WEIGHTS)) {\n      score += criteria[key as keyof JudgementCriteria] * weight;\n    }\n    return score;\n  }\n\n  private createJudgementResult(\n    criteria: JudgementCriteria,\n    context: LLMJudgeContext\n  ): VerifierResult {\n    const overallScore = this.calculateOverallScore(criteria);\n    const passed = overallScore >= this.VETO_THRESHOLD;\n\n    // Generate detailed feedback\n    const feedback = this.generateJudgementFeedback(\n      criteria,\n      overallScore,\n      context\n    );\n\n    // Determine severity based on score\n    const severity =\n      overallScore < 0.5\n        ? 'error'\n        : overallScore < this.VETO_THRESHOLD\n          ? 'warning'\n          : 'info';\n\n    return this.createResult(\n      passed,\n      feedback,\n      severity,\n      {\n        expected: 'Changes that fully address the original requirements',\n        actual: `Score: ${(overallScore * 100).toFixed(1)}%`,\n        suggestion: this.generateSuggestions(criteria),\n      },\n      passed\n        ? undefined\n        : {\n            command: 'Review and adjust approach based on feedback',\n            description: 'Manual review required',\n            safe: false,\n            confidence: overallScore,\n          }\n    );\n  }\n\n  private generateJudgementFeedback(\n    criteria: JudgementCriteria,\n    overallScore: number,\n    context: LLMJudgeContext\n  ): string {\n    if (overallScore >= this.VETO_THRESHOLD) {\n      return (\n        `Semantic validation PASSED (${(overallScore * 100).toFixed(1)}% confidence). ` +\n        `Changes adequately address the requirements.`\n      );\n    }\n\n    // Identify weakest areas\n    const weakAreas: string[] = [];\n    if (criteria.adherenceToRequirements < 0.6) {\n      weakAreas.push('requirements adherence');\n    }\n    if (criteria.codeQuality < 0.6) {\n      weakAreas.push('code quality');\n    }\n    if (criteria.safety < 0.7) {\n      weakAreas.push('safety concerns');\n    }\n    if (criteria.completeness < 0.6) {\n      weakAreas.push('completeness');\n    }\n    if (criteria.semanticCorrectness < 0.6) {\n      weakAreas.push('semantic correctness');\n    }\n\n    return (\n      `Semantic validation VETOED (${(overallScore * 100).toFixed(1)}% confidence). ` +\n      `Issues found with: ${weakAreas.join(', ')}. ` +\n      `Agent should course-correct based on this feedback.`\n    );\n  }\n\n  private generateSuggestions(criteria: JudgementCriteria): string {\n    const suggestions: string[] = [];\n\n    if (criteria.adherenceToRequirements < 0.7) {\n      suggestions.push(\n        'Review original requirements and ensure all are addressed'\n      );\n    }\n    if (criteria.codeQuality < 0.7) {\n      suggestions.push(\n        'Improve code structure, add error handling and comments'\n      );\n    }\n    if (criteria.safety < 0.8) {\n      suggestions.push(\n        'Review code for security vulnerabilities and unsafe patterns'\n      );\n    }\n    if (criteria.completeness < 0.7) {\n      suggestions.push('Ensure solution is complete and handles edge cases');\n    }\n    if (criteria.semanticCorrectness < 0.7) {\n      suggestions.push('Verify logic correctness and alignment with intent');\n    }\n\n    return suggestions.length > 0\n      ? suggestions.join('; ')\n      : 'Continue with current approach';\n  }\n\n  /**\n   * Get veto statistics (for monitoring)\n   */\n  getVetoRate(): number {\n    // In production, this would track actual veto rate\n    // Spotify reports ~25% veto rate\n    return 0.25;\n  }\n\n  /**\n   * Check if agent can course-correct after veto\n   * Spotify: \"When vetoed, agents can course correct half the time\"\n   */\n  canCourseCorrect(previousAttempts: number): boolean {\n    // 50% chance of successful course correction\n    return previousAttempts < 2 && Math.random() > 0.5;\n  }\n}\n"],
+  "mappings": ";;;;AAKA;AAAA,EACE;AAAA,OAIK;AACP,SAAS,cAAc;AAiBhB,MAAM,iBAAiB,aAAa;AAAA;AAAA,EAExB,iBAAiB;AAAA,EACjB,mBAAmB;AAAA,IAClC,yBAAyB;AAAA,IACzB,aAAa;AAAA,IACb,QAAQ;AAAA,IACR,cAAc;AAAA,IACd,qBAAqB;AAAA,EACvB;AAAA,EAEA,YAAY,QAAkC;AAC5C,UAAM;AAAA,MACJ,IAAI;AAAA,MACJ,MAAM;AAAA,MACN,MAAM;AAAA,MACN,SAAS;AAAA,MACT,aAAa;AAAA;AAAA,MACb,SAAS;AAAA,MACT,GAAG;AAAA,IACL,CAAC;AAAA,EACH;AAAA,EAEA,eAAe,SAAmC;AAEhD,WAAO;AAAA,EACT;AAAA,EAEA,MAAM,OACJ,OACA,SACyB;AACzB,UAAM,aAAa;AAEnB,QAAI,CAAC,WAAW,gBAAgB;AAC9B,aAAO,KAAK;AAAA,QACV;AAAA,QACA;AAAA,QACA;AAAA,MACF;AAAA,IACF;AAEA,QAAI;AACF,aAAO,MAAM,KAAK,YAAY,YAAY;AACxC,cAAM,YAAY,MAAM,KAAK;AAAA,UAC3B,MAAM,SAAS;AAAA,UACf;AAAA,QACF;AACA,eAAO,KAAK,sBAAsB,WAAW,UAAU;AAAA,MACzD,CAAC;AAAA,IACH,SAAS,OAAgB;AACvB,aAAO;AAAA,QACL;AAAA,QACA,iBAAiB,QAAQ,QAAQ;AAAA,MACnC;AACA,aAAO,KAAK;AAAA,QACV;AAAA,QACA,8BAA8B,iBAAiB,QAAQ,MAAM,UAAU,OAAO,KAAK,CAAC;AAAA,QACpF;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAAA,EAEA,MAAc,iBACZ,iBACA,SAC4B;AAI5B,UAAM,WAA8B;AAAA,MAClC,yBAAyB,KAAK,kBAAkB,iBAAiB,OAAO;AAAA,MACxE,aAAa,KAAK,oBAAoB,eAAe;AAAA,MACrD,QAAQ,KAAK,eAAe,eAAe;AAAA,MAC3C,cAAc,KAAK,qBAAqB,iBAAiB,OAAO;AAAA,MAChE,qBAAqB,KAAK;AAAA,QACxB;AAAA,QACA;AAAA,MACF;AAAA,IACF;AAGA,WAAO,KAAK,wBAAwB;AAAA,MAClC;AAAA,MACA,cAAc,KAAK,sBAAsB,QAAQ;AAAA,MACjD,UAAU,KAAK,sBAAsB,QAAQ,IAAI,KAAK;AAAA,IACxD,CAAC;AAED,WAAO;AAAA,EACT;AAAA,EAEQ,kBACN,iBACA,SACQ;AAER,UAAM,cAAc,QAAQ,eACzB,YAAY,EACZ,MAAM,GAAG,EACT,OAAO,CAAC,SAAS,KAAK,SAAS,CAAC,EAChC,KAAK,CAAC,YAAY,gBAAgB,YAAY,EAAE,SAAS,OAAO,CAAC;AAGpE,QAAI,QAAQ,oBAAoB;AAC9B,YAAM,cAAc,QAAQ,mBAAmB;AAAA,QAAO,CAAC,cACrD,KAAK,eAAe,WAAW,eAAe;AAAA,MAChD,EAAE;AACF,YAAM,gBAAgB,cAAc,QAAQ,mBAAmB;AAC/D,aAAO,cAAc,KAAK,IAAI,GAAG,gBAAgB,GAAG,IAAI;AAAA,IAC1D;AAGA,WAAO,cAAc,OAAO,KAAK,OAAO,IAAI,MAAM,MAAM,KAAK,OAAO,IAAI;AAAA,EAC1E;AAAA,EAEQ,oBAAoB,iBAAiC;AAE3D,UAAM,aAAa;AAAA,MACjB,aACE,gBAAgB,SAAS,IAAI,KAAK,gBAAgB,SAAS,IAAI;AAAA,MACjE,kBACE,gBAAgB,SAAS,KAAK,KAAK,gBAAgB,SAAS,OAAO;AAAA,MACrE,UACE,gBAAgB,SAAS,MAAM,KAAK,gBAAgB,SAAS,QAAQ;AAAA,MACvE,cAAc,CAAC,wBAAwB,KAAK,eAAe;AAAA;AAAA,MAC3D,mBACE,gBAAgB,SAAS,MAAM,gBAAgB,SAAS;AAAA,IAC5D;AAEA,UAAM,QACJ,OAAO,OAAO,UAAU,EAAE,OAAO,OAAO,EAAE,SAC1C,OAAO,KAAK,UAAU,EAAE;AAC1B,WAAO,KAAK,IAAI,GAAG,QAAQ,GAAG;AAAA,EAChC;AAAA,EAEQ,eAAe,iBAAiC;AAEtD,UAAM,oBAAoB;AAAA,MACxB;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IACF;AAEA,UAAM,sBAAsB,kBAAkB;AAAA,MAAK,CAAC,YAClD,QAAQ,KAAK,eAAe;AAAA,IAC9B;AAEA,WAAO,sBAAsB,MAAM,MAAM,KAAK,OAAO,IAAI;AAAA,EAC3D;AAAA,EAEQ,qBACN,iBACA,SACQ;AAER,UAAM,oBAAoB,gBAAgB,SAAS;AACnD,UAAM,eACJ,gBAAgB,SAAS,UAAU,KACnC,gBAAgB,SAAS,OAAO,KAChC,gBAAgB,SAAS,OAAO,KAChC,gBAAgB,SAAS,KAAK;AAEhC,QAAI,QAAQ,oBAAoB;AAC9B,YAAM,oBAAoB,QAAQ,mBAAmB;AAAA,QAAO,CAAC,cAC3D,gBACG,YAAY,EACZ,SAAS,UAAU,YAAY,EAAE,MAAM,GAAG,EAAE,CAAC,CAAC;AAAA,MACnD,EAAE;AACF,aAAO,oBAAoB,QAAQ,mBAAmB;AAAA,IACxD;AAEA,WAAO,qBAAqB,eAAe,MAAM,KAAK,OAAO,IAAI,OAAO;AAAA,EAC1E;AAAA,EAEQ,4BACN,iBACA,SACQ;AAKR,QAAI,QAAQ,oBAAoB,QAAQ,iBAAiB,SAAS,GAAG;AACnE,YAAM,oBAAoB,QAAQ,iBAAiB,OAAO,CAAC,aAAa;AACtE,cAAM,WAAW,SACd,YAAY,EACZ,MAAM,GAAG,EACT,OAAO,CAAC,MAAM,EAAE,SAAS,CAAC;AAC7B,eAAO,SAAS;AAAA,UAAK,CAAC,YACpB,gBAAgB,YAAY,EAAE,SAAS,OAAO;AAAA,QAChD;AAAA,MACF,CAAC,EAAE;AAEH,UAAI,oBAAoB,GAAG;AACzB,eAAO,KAAK;AAAA,UACV;AAAA,UACA,MAAO,oBAAoB,QAAQ,iBAAiB,SAAU;AAAA,QAChE;AAAA,MACF;AAAA,IACF;AAGA,WAAO,OAAO,KAAK,OAAO,IAAI;AAAA,EAChC;AAAA,EAEQ,eAAe,WAAmB,iBAAkC;AAE1E,UAAM,WAAW,UACd,YAAY,EACZ,MAAM,GAAG,EACT,OAAO,CAAC,SAAS,KAAK,SAAS,CAAC;AACnC,UAAM,kBAAkB,SAAS;AAAA,MAAO,CAAC,YACvC,gBAAgB,YAAY,EAAE,SAAS,OAAO;AAAA,IAChD;AACA,WAAO,gBAAgB,UAAU,SAAS,SAAS;AAAA,EACrD;AAAA,EAEQ,sBAAsB,UAAqC;AACjE,QAAI,QAAQ;AACZ,eAAW,CAAC,KAAK,MAAM,KAAK,OAAO,QAAQ,KAAK,gBAAgB,GAAG;AACjE,eAAS,SAAS,GAA8B,IAAI;AAAA,IACtD;AACA,WAAO;AAAA,EACT;AAAA,EAEQ,sBACN,UACA,SACgB;AAChB,UAAM,eAAe,KAAK,sBAAsB,QAAQ;AACxD,UAAM,SAAS,gBAAgB,KAAK;AAGpC,UAAM,WAAW,KAAK;AAAA,MACpB;AAAA,MACA;AAAA,MACA;AAAA,IACF;AAGA,UAAM,WACJ,eAAe,MACX,UACA,eAAe,KAAK,iBAClB,YACA;AAER,WAAO,KAAK;AAAA,MACV;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,QACE,UAAU;AAAA,QACV,QAAQ,WAAW,eAAe,KAAK,QAAQ,CAAC,CAAC;AAAA,QACjD,YAAY,KAAK,oBAAoB,QAAQ;AAAA,MAC/C;AAAA,MACA,SACI,SACA;AAAA,QACE,SAAS;AAAA,QACT,aAAa;AAAA,QACb,MAAM;AAAA,QACN,YAAY;AAAA,MACd;AAAA,IACN;AAAA,EACF;AAAA,EAEQ,0BACN,UACA,cACA,SACQ;AACR,QAAI,gBAAgB,KAAK,gBAAgB;AACvC,aACE,gCAAgC,eAAe,KAAK,QAAQ,CAAC,CAAC;AAAA,IAGlE;AAGA,UAAM,YAAsB,CAAC;AAC7B,QAAI,SAAS,0BAA0B,KAAK;AAC1C,gBAAU,KAAK,wBAAwB;AAAA,IACzC;AACA,QAAI,SAAS,cAAc,KAAK;AAC9B,gBAAU,KAAK,cAAc;AAAA,IAC/B;AACA,QAAI,SAAS,SAAS,KAAK;AACzB,gBAAU,KAAK,iBAAiB;AAAA,IAClC;AACA,QAAI,SAAS,eAAe,KAAK;AAC/B,gBAAU,KAAK,cAAc;AAAA,IAC/B;AACA,QAAI,SAAS,sBAAsB,KAAK;AACtC,gBAAU,KAAK,sBAAsB;AAAA,IACvC;AAEA,WACE,gCAAgC,eAAe,KAAK,QAAQ,CAAC,CAAC,qCACxC,UAAU,KAAK,IAAI,CAAC;AAAA,EAG9C;AAAA,EAEQ,oBAAoB,UAAqC;AAC/D,UAAM,cAAwB,CAAC;AAE/B,QAAI,SAAS,0BAA0B,KAAK;AAC1C,kBAAY;AAAA,QACV;AAAA,MACF;AAAA,IACF;AACA,QAAI,SAAS,cAAc,KAAK;AAC9B,kBAAY;AAAA,QACV;AAAA,MACF;AAAA,IACF;AACA,QAAI,SAAS,SAAS,KAAK;AACzB,kBAAY;AAAA,QACV;AAAA,MACF;AAAA,IACF;AACA,QAAI,SAAS,eAAe,KAAK;AAC/B,kBAAY,KAAK,oDAAoD;AAAA,IACvE;AACA,QAAI,SAAS,sBAAsB,KAAK;AACtC,kBAAY,KAAK,oDAAoD;AAAA,IACvE;AAEA,WAAO,YAAY,SAAS,IACxB,YAAY,KAAK,IAAI,IACrB;AAAA,EACN;AAAA;AAAA;AAAA;AAAA,EAKA,cAAsB;AAGpB,WAAO;AAAA,EACT;AAAA;AAAA;AAAA;AAAA;AAAA,EAMA,iBAAiB,kBAAmC;AAElD,WAAO,mBAAmB,KAAK,KAAK,OAAO,IAAI;AAAA,EACjD;AACF;",
+  "names": []
+}