@superblocksteam/vite-plugin-file-sync 1.99.99-next.2 → 2.0.0-SNAPSHOT.84af65cd215b6ca1

package/dist/ai-service/judge/judge-service.d.ts

@@ -0,0 +1,154 @@
+/**
+ * Judge service for evaluating AI-generated applications.
+ *
+ * Encapsulates the complete judge evaluation workflow including:
+ * - LLM client access for executing evaluations
+ * - Storage for persisting evaluation results
+ * - MCP server lifecycle management
+ * - Executor orchestration
+ */
+import type { LLMProviderConfig } from "@superblocksteam/library-shared/types";
+import type { Logger } from "../../util/logger.js";
+import type { LLMClient } from "../llm/client.js";
+import type { LLMProvider } from "../llm/types.js";
+import type { PlaywrightMcpServerManager } from "../mcp/playwright-server.js";
+import type { JudgeStorage, StoredEvaluation } from "./storage/index.js";
+import type { JudgeEvaluation, EvaluationCriteria, JudgeConfig } from "./types.js";
+/**
+ * Options for judge evaluation.
+ */
+export interface JudgeEvaluationOptions {
+    /** Application URL to evaluate */
+    appUrl: string;
+    /** JWT token for AI Gateway authentication */
+    jwt?: string;
+    /** LLM configuration (provider, thinking opts, disabled tools) */
+    llmConfig?: LLMProviderConfig;
+    /** Optional websocket endpoint to reuse an existing Playwright browser */
+    connectWsEndpoint?: string;
+    /** Raw storage state data to seed Playwright context */
+    storageStateData?: any;
+    /** Session storage entries to seed for the app origin */
+    sessionStorageData?: {
+        origin: string;
+        items: Array<{
+            name: string;
+            value: string;
+        }>;
+    };
+    /** Additional origins storage to seed (e.g., auth domain local/session storage) */
+    extraOrigins?: Array<{
+        origin: string;
+        localStorage: Array<{
+            name: string;
+            value: string;
+        }>;
+        sessionStorage?: Array<{
+            name: string;
+            value: string;
+        }>;
+    }>;
+    /** Optional evaluation criteria (auto-generated if not provided) */
+    criteria?: EvaluationCriteria;
+    /** Optional judge configuration overrides */
+    config?: Partial<JudgeConfig>;
+    /** Complexity level for model selection */
+    complexity?: "low" | "medium" | "high";
+    /** Optional path to Playwright storage state for authentication */
+    storageStatePath?: string;
+    /** Whether the Playwright MCP browser should run headless */
+    playwrightHeadless?: boolean;
+    /** Browser engine to use for Playwright MCP */
+    playwrightBrowser?: "chromium" | "firefox" | "webkit";
+    /** External Playwright MCP server to reuse instead of starting a new one */
+    playwrightServerUrl?: string;
+}
+/**
+ * Result of a judge evaluation with metadata.
+ */
+export interface JudgeEvaluationResult extends JudgeEvaluation {
+    /** Prompt ID that was evaluated */
+    promptId: string;
+    /** Application ID */
+    appId: string;
+}
+/**
+ * Judge service for orchestrating application evaluations.
+ *
+ * Provides a high-level API for running judge evaluations
+ * and managing their lifecycle.
+ */
+export declare class JudgeService {
+    private llmClient;
+    private llmProvider;
+    private storage;
+    private mcpServerManager;
+    private logger;
+    constructor(config: {
+        llmClient: LLMClient;
+        llmProvider: LLMProvider;
+        storage: JudgeStorage;
+        mcpServerManager: PlaywrightMcpServerManager;
+        logger: Logger;
+    });
+    /**
+     * Evaluates an application against a prompt.
+     *
+     * This is the main entry point for judge evaluations. It:
+     * 1. Ensures MCP server is running
+     * 2. Generates or uses provided criteria
+     * 3. Executes the evaluation via JudgeExecutor
+     * 4. Stores the result
+     * 5. Returns the evaluation
+     *
+     * @param promptId - Unique identifier for the prompt
+     * @param prompt - The prompt text that was used to generate the app
+     * @param appId - Application identifier
+     * @param options - Evaluation options
+     * @returns Complete evaluation result
+     * @throws Error if evaluation fails
+     */
+    evaluateApplication(promptId: string, prompt: string, appId: string, options: JudgeEvaluationOptions): Promise<JudgeEvaluationResult>;
+    /**
+     * Retrieves stored evaluations.
+     *
+     * @param promptId - Optional prompt ID to filter by
+     * @returns Array of stored evaluations
+     */
+    getEvaluations(promptId?: string): Promise<StoredEvaluation[]>;
+    /**
+     * Retrieves evaluations for a specific branch.
+     *
+     * @param branchName - Git branch name
+     * @returns Array of stored evaluations for the branch
+     */
+    getEvaluationsByBranch(branchName: string): Promise<StoredEvaluation[]>;
+    /**
+     * Ensures the MCP server is running.
+     *
+     * Starts the server if not already running.
+     *
+     * @param storageStatePath - Optional path to Playwright storage state for authentication
+     * @returns MCP server URL
+     */
+    private ensureMcpServer;
+    /**
+     * Gets the current git branch name.
+     *
+     * @returns Branch name or 'unknown' if git is not available
+     */
+    private getCurrentBranch;
+    /**
+     * Gets the current git commit SHA.
+     *
+     * @returns Commit SHA or 'unknown' if git is not available
+     */
+    private getCurrentCommitSha;
+    /**
+     * Cleans up resources used by the judge service.
+     *
+     * Stops the MCP server if running.
+     */
+    dispose(): Promise<void>;
+}
+//# sourceMappingURL=judge-service.d.ts.map

package/dist/ai-service/judge/judge-service.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"judge-service.d.ts","sourceRoot":"","sources":["../../../src/ai-service/judge/judge-service.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,uCAAuC,CAAC;AAE/E,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,sBAAsB,CAAC;AACnD,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AAClD,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,iBAAiB,CAAC;AACnD,OAAO,KAAK,EAAE,0BAA0B,EAAE,MAAM,6BAA6B,CAAC;AAG9E,OAAO,KAAK,EAAE,YAAY,EAAE,gBAAgB,EAAE,MAAM,oBAAoB,CAAC;AACzE,OAAO,KAAK,EACV,eAAe,EACf,kBAAkB,EAClB,WAAW,EACZ,MAAM,YAAY,CAAC;AAEpB;;GAEG;AACH,MAAM,WAAW,sBAAsB;IACrC,kCAAkC;IAClC,MAAM,EAAE,MAAM,CAAC;IAEf,8CAA8C;IAC9C,GAAG,CAAC,EAAE,MAAM,CAAC;IAEb,kEAAkE;IAClE,SAAS,CAAC,EAAE,iBAAiB,CAAC;IAE9B,0EAA0E;IAC1E,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAE3B,wDAAwD;IACxD,gBAAgB,CAAC,EAAE,GAAG,CAAC;IAEvB,yDAAyD;IACzD,kBAAkB,CAAC,EAAE;QACnB,MAAM,EAAE,MAAM,CAAC;QACf,KAAK,EAAE,KAAK,CAAC;YAAE,IAAI,EAAE,MAAM,CAAC;YAAC,KAAK,EAAE,MAAM,CAAA;SAAE,CAAC,CAAC;KAC/C,CAAC;IAEF,mFAAmF;IACnF,YAAY,CAAC,EAAE,KAAK,CAAC;QACnB,MAAM,EAAE,MAAM,CAAC;QACf,YAAY,EAAE,KAAK,CAAC;YAAE,IAAI,EAAE,MAAM,CAAC;YAAC,KAAK,EAAE,MAAM,CAAA;SAAE,CAAC,CAAC;QACrD,cAAc,CAAC,EAAE,KAAK,CAAC;YAAE,IAAI,EAAE,MAAM,CAAC;YAAC,KAAK,EAAE,MAAM,CAAA;SAAE,CAAC,CAAC;KACzD,CAAC,CAAC;IAEH,oEAAoE;IACpE,QAAQ,CAAC,EAAE,kBAAkB,CAAC;IAE9B,6CAA6C;IAC7C,MAAM,CAAC,EAAE,OAAO,CAAC,WAAW,CAAC,CAAC;IAE9B,2CAA2C;IAC3C,UAAU,CAAC,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,CAAC;IAEvC,mEAAmE;IACnE,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAE1B,6DAA6D;IAC7D,kBAAkB,CAAC,EAAE,OAAO,CAAC;IAE7B,+CAA+C;IAC/C,iBAAiB,CAAC,EAAE,UAAU,GAAG,SAAS,GAAG,QAAQ,CAAC;IAEtD,4EAA4E;IAC5E,mBAAmB,CAAC,EAAE,MAAM,CAAC;CAC9B;AAED;;GAEG;AACH,MAAM,WAAW,qBAAsB,SAAQ,eAAe;IAC5D,mCAAmC;IACnC,QAAQ,EAAE,MAAM,CAAC;IAEjB,qBAAqB;IACrB,KAAK,EAAE,MAAM,CAAC;CACf;AAED;;;;;GAKG;AACH,qBAAa,YAAY;IACvB,OAAO,CAAC,SAAS,CAAY;IAC7B,OAAO,CAAC,WAAW,CAAc;IACjC,OAAO,CAAC,OAAO,CAAe;IAC9B,OAAO,CAAC,gBAAgB,CAA6B;IACrD,OAAO,CAAC,MAAM,CAAS;gBAEX,MAAM,EAAE;QAClB,SAAS,EAAE,SAAS,CAAC;QACrB,WAAW,EAAE,WAAW,CAAC;QACzB,OAAO,EAAE,YAAY,CAAC;QACtB,gBAAgB,EAAE,0BAA0B,CAAC;QAC7C,MAAM,EAAE,MAAM,CAAC;KAChB;IAQD;;;;;;;;;;;;;;;;OAgBG;IACG,mBAAmB,CACvB,QAAQ,EAAE,MAAM,EAChB,MAAM,EAAE,MAAM,EACd,KAAK,EAAE,MAAM,EACb,OAAO,EAAE,sBAAsB,GAC9B,OAAO,CAAC,qBAAqB,CAAC;IA0GjC;;;;;OAKG;IACG,cAAc,CAAC,QAAQ,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC,gBAAgB,EAAE,CAAC;IAOpE;;;;;OAKG;IACG,sBAAsB,CAC1B,UAAU,EAAE,MAAM,GACjB,OAAO,CAAC,gBAAgB,EAAE,CAAC;IAI9B;;;;;;;OAOG;YACW,eAAe;IA8C7B;;;;OAIG;YACW,gBAAgB;IAa9B;;;;OAIG;YACW,mBAAmB;IAajC;;;;OAIG;IACG,OAAO,IAAI,OAAO,CAAC,IAAI,CAAC;CAO/B"}

package/dist/ai-service/judge/judge-service.js

@@ -0,0 +1,220 @@
+/**
+ * Judge service for evaluating AI-generated applications.
+ *
+ * Encapsulates the complete judge evaluation workflow including:
+ * - LLM client access for executing evaluations
+ * - Storage for persisting evaluation results
+ * - MCP server lifecycle management
+ * - Executor orchestration
+ */
+import { JudgeExecutor } from "./judge-executor.js";
+import { buildCriteriaFromPrompts } from "./prompts/system-prompt.js";
+/**
+ * Judge service for orchestrating application evaluations.
+ *
+ * Provides a high-level API for running judge evaluations
+ * and managing their lifecycle.
+ */
+export class JudgeService {
+    llmClient;
+    llmProvider;
+    storage;
+    mcpServerManager;
+    logger;
+    constructor(config) {
+        this.llmClient = config.llmClient;
+        this.llmProvider = config.llmProvider;
+        this.storage = config.storage;
+        this.mcpServerManager = config.mcpServerManager;
+        this.logger = config.logger;
+    }
+    /**
+     * Evaluates an application against a prompt.
+     *
+     * This is the main entry point for judge evaluations. It:
+     * 1. Ensures MCP server is running
+     * 2. Generates or uses provided criteria
+     * 3. Executes the evaluation via JudgeExecutor
+     * 4. Stores the result
+     * 5. Returns the evaluation
+     *
+     * @param promptId - Unique identifier for the prompt
+     * @param prompt - The prompt text that was used to generate the app
+     * @param appId - Application identifier
+     * @param options - Evaluation options
+     * @returns Complete evaluation result
+     * @throws Error if evaluation fails
+     */
+    async evaluateApplication(promptId, prompt, appId, options) {
+        this.logger.info(`Starting judge evaluation: promptId=${promptId}, appId=${appId}, appUrl=${options.appUrl}`);
+        try {
+            const _useExternalServer = !!options.playwrightServerUrl;
+            let mcpUrl = options.playwrightServerUrl;
+            if (!mcpUrl) {
+                mcpUrl = await this.ensureMcpServer(options.storageStatePath, options.playwrightBrowser, options.playwrightHeadless, options.jwt, options.connectWsEndpoint, options.storageStateData, options.appUrl);
+            }
+            // Generate criteria if not provided
+            const criteria = options.criteria ||
+                buildCriteriaFromPrompts([prompt], options.complexity || "medium");
+            // Select model based on complexity - create LanguageModel instance from provider
+            const { model } = this.llmProvider.modelForTask("generateBroadEdit");
+            // Build judge config
+            const judgeConfig = {
+                ...options.config,
+                playwrightMcpUrl: mcpUrl,
+                jwt: options.jwt,
+            };
+            // Create executor
+            const executor = new JudgeExecutor(this.llmClient, model, this.logger, judgeConfig);
+            // Initialize executor
+            await executor.initialize();
+            try {
+                // Execute evaluation
+                // Note: JudgeExecutor expects a simulation result, but we'll adapt it
+                const simulationResult = {
+                    promptId,
+                    success: true, // Assume success since we're evaluating the final result
+                    stepResults: [{ prompt }],
+                    tokens: { input: 0, output: 0, total: 0 },
+                    duration: 0,
+                };
+                const evaluation = await executor.evaluateSimulation(simulationResult, criteria, options.appUrl);
+                // Get current git metadata for tracking
+                const branchName = await this.getCurrentBranch();
+                const commitSha = await this.getCurrentCommitSha();
+                // Store evaluation
+                const storedEvaluation = {
+                    timestamp: new Date().toISOString(),
+                    promptId,
+                    branchName,
+                    commitSha,
+                    appId,
+                    prompt,
+                    evaluation,
+                };
+                await this.storage.saveEvaluation(storedEvaluation);
+                this.logger.info(`Judge evaluation completed and stored: promptId=${promptId}, appId=${appId}, score=${evaluation.overallScore}, passed=${evaluation.passed}`);
+                // Return result with metadata
+                return {
+                    ...evaluation,
+                    promptId,
+                    appId,
+                };
+            }
+            finally {
+                // Always cleanup executor
+                await executor.cleanup();
+            }
+        }
+        catch (error) {
+            this.logger.error(`Judge evaluation failed: ${String(error)}`, {
+                error: {
+                    kind: "JudgeEvaluationError",
+                    message: String(error),
+                    stack: error instanceof Error ? error.stack : undefined,
+                },
+            });
+            throw error;
+        }
+    }
+    /**
+     * Retrieves stored evaluations.
+     *
+     * @param promptId - Optional prompt ID to filter by
+     * @returns Array of stored evaluations
+     */
+    async getEvaluations(promptId) {
+        if (promptId) {
+            return this.storage.getEvaluationsByPrompt(promptId);
+        }
+        return this.storage.getEvaluations();
+    }
+    /**
+     * Retrieves evaluations for a specific branch.
+     *
+     * @param branchName - Git branch name
+     * @returns Array of stored evaluations for the branch
+     */
+    async getEvaluationsByBranch(branchName) {
+        return this.storage.getEvaluationsByBranch(branchName);
+    }
+    /**
+     * Ensures the MCP server is running.
+     *
+     * Starts the server if not already running.
+     *
+     * @param storageStatePath - Optional path to Playwright storage state for authentication
+     * @returns MCP server URL
+     */
+    async ensureMcpServer(storageStatePath, browser = "chromium", headless = true, jwt, connectWsEndpoint, storageStateData, appUrl, sessionStorageData, extraOrigins) {
+        if (!this.mcpServerManager.isRunning()) {
+            this.logger.info(`Starting Playwright MCP server for judge`, storageStatePath
+                ? `with storage state: ${storageStatePath}`
+                : "without storage state");
+            return await this.mcpServerManager.start({
+                browser,
+                headless,
+                storageStatePath,
+                jwt,
+                connectWsEndpoint,
+                storageStateData,
+                appUrl,
+                sessionStorageData,
+                extraOrigins,
+            });
+        }
+        const url = this.mcpServerManager.getUrl();
+        if (!url) {
+            throw new Error("MCP server is running but URL is not available");
+        }
+        return url;
+    }
+    /**
+     * Gets the current git branch name.
+     *
+     * @returns Branch name or 'unknown' if git is not available
+     */
+    async getCurrentBranch() {
+        try {
+            const { execSync } = await import("child_process");
+            const branch = execSync("git rev-parse --abbrev-ref HEAD", {
+                encoding: "utf-8",
+                cwd: process.cwd(),
+            }).trim();
+            return branch;
+        }
+        catch {
+            return "unknown";
+        }
+    }
+    /**
+     * Gets the current git commit SHA.
+     *
+     * @returns Commit SHA or 'unknown' if git is not available
+     */
+    async getCurrentCommitSha() {
+        try {
+            const { execSync } = await import("child_process");
+            const sha = execSync("git rev-parse HEAD", {
+                encoding: "utf-8",
+                cwd: process.cwd(),
+            }).trim();
+            return sha;
+        }
+        catch {
+            return "unknown";
+        }
+    }
+    /**
+     * Cleans up resources used by the judge service.
+     *
+     * Stops the MCP server if running.
+     */
+    async dispose() {
+        this.logger.info("Disposing judge service");
+        if (this.mcpServerManager.isRunning()) {
+            await this.mcpServerManager.stop();
+        }
+    }
+}
+//# sourceMappingURL=judge-service.js.map

package/dist/ai-service/judge/judge-service.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"judge-service.js","sourceRoot":"","sources":["../../../src/ai-service/judge/judge-service.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAQH,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACpD,OAAO,EAAE,wBAAwB,EAAE,MAAM,4BAA4B,CAAC;AAyEtE;;;;;GAKG;AACH,MAAM,OAAO,YAAY;IACf,SAAS,CAAY;IACrB,WAAW,CAAc;IACzB,OAAO,CAAe;IACtB,gBAAgB,CAA6B;IAC7C,MAAM,CAAS;IAEvB,YAAY,MAMX;QACC,IAAI,CAAC,SAAS,GAAG,MAAM,CAAC,SAAS,CAAC;QAClC,IAAI,CAAC,WAAW,GAAG,MAAM,CAAC,WAAW,CAAC;QACtC,IAAI,CAAC,OAAO,GAAG,MAAM,CAAC,OAAO,CAAC;QAC9B,IAAI,CAAC,gBAAgB,GAAG,MAAM,CAAC,gBAAgB,CAAC;QAChD,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC;IAC9B,CAAC;IAED;;;;;;;;;;;;;;;;OAgBG;IACH,KAAK,CAAC,mBAAmB,CACvB,QAAgB,EAChB,MAAc,EACd,KAAa,EACb,OAA+B;QAE/B,IAAI,CAAC,MAAM,CAAC,IAAI,CACd,uCAAuC,QAAQ,WAAW,KAAK,YAAY,OAAO,CAAC,MAAM,EAAE,CAC5F,CAAC;QAEF,IAAI,CAAC;YACH,MAAM,kBAAkB,GAAG,CAAC,CAAC,OAAO,CAAC,mBAAmB,CAAC;YACzD,IAAI,MAAM,GAAG,OAAO,CAAC,mBAAmB,CAAC;YACzC,IAAI,CAAC,MAAM,EAAE,CAAC;gBACZ,MAAM,GAAG,MAAM,IAAI,CAAC,eAAe,CACjC,OAAO,CAAC,gBAAgB,EACxB,OAAO,CAAC,iBAAiB,EACzB,OAAO,CAAC,kBAAkB,EAC1B,OAAO,CAAC,GAAG,EACX,OAAO,CAAC,iBAAiB,EACzB,OAAO,CAAC,gBAAgB,EACxB,OAAO,CAAC,MAAM,CACf,CAAC;YACJ,CAAC;YAED,oCAAoC;YACpC,MAAM,QAAQ,GACZ,OAAO,CAAC,QAAQ;gBAChB,wBAAwB,CAAC,CAAC,MAAM,CAAC,EAAE,OAAO,CAAC,UAAU,IAAI,QAAQ,CAAC,CAAC;YAErE,iFAAiF;YACjF,MAAM,EAAE,KAAK,EAAE,GAAG,IAAI,CAAC,WAAW,CAAC,YAAY,CAAC,mBAAmB,CAAC,CAAC;YAErE,qBAAqB;YACrB,MAAM,WAAW,GAAyB;gBACxC,GAAG,OAAO,CAAC,MAAM;gBACjB,gBAAgB,EAAE,MAAM;gBACxB,GAAG,EAAE,OAAO,CAAC,GAAG;aACjB,CAAC;YAEF,kBAAkB;YAClB,MAAM,QAAQ,GAAG,IAAI,aAAa,CAChC,IAAI,CAAC,SAAS,EACd,KAAK,EACL,IAAI,CAAC,MAAM,EACX,WAAW,CACZ,CAAC;YAEF,sBAAsB;YACtB,MAAM,QAAQ,CAAC,UAAU,EAAE,CAAC;YAE5B,IAAI,CAAC;gBACH,qBAAqB;gBACrB,sEAAsE;gBACtE,MAAM,gBAAgB,GAAG;oBACvB,QAAQ;oBACR,OAAO,EAAE,IAAI,EAAE,yDAAyD;oBACxE,WAAW,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC;oBACzB,MAAM,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE;oBACzC,QAAQ,EAAE,CAAC;iBACZ,CAAC;gBAEF,MAAM,UAAU,GAAG,MAAM,QAAQ,CAAC,kBAAkB,CAClD,gBAAgB,EAChB,QAAQ,EACR,OAAO,CAAC,MAAM,CACf,CAAC;gBAEF,wCAAwC;gBACxC,MAAM,UAAU,GAAG,MAAM,IAAI,CAAC,gBAAgB,EAAE,CAAC;gBACjD,MAAM,SAAS,GAAG,MAAM,IAAI,CAAC,mBAAmB,EAAE,CAAC;gBAEnD,mBAAmB;gBACnB,MAAM,gBAAgB,GAAqB;oBACzC,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;oBACnC,QAAQ;oBACR,UAAU;oBACV,SAAS;oBACT,KAAK;oBACL,MAAM;oBACN,UAAU;iBACX,CAAC;gBAEF,MAAM,IAAI,CAAC,OAAO,CAAC,cAAc,CAAC,gBAAgB,CAAC,CAAC;gBAEpD,IAAI,CAAC,MAAM,CAAC,IAAI,CACd,mDAAmD,QAAQ,WAAW,KAAK,WAAW,UAAU,CAAC,YAAY,YAAY,UAAU,CAAC,MAAM,EAAE,CAC7I,CAAC;gBAEF,8BAA8B;gBAC9B,OAAO;oBACL,GAAG,UAAU;oBACb,QAAQ;oBACR,KAAK;iBACN,CAAC;YACJ,CAAC;oBAAS,CAAC;gBACT,0BAA0B;gBAC1B,MAAM,QAAQ,CAAC,OAAO,EAAE,CAAC;YAC3B,CAAC;QACH,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,4BAA4B,MAAM,CAAC,KAAK,CAAC,EAAE,EAAE;gBAC7D,KAAK,EAAE;oBACL,IAAI,EAAE,sBAAsB;oBAC5B,OAAO,EAAE,MAAM,CAAC,KAAK,CAAC;oBACtB,KAAK,EAAE,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS;iBACxD;aACF,CAAC,CAAC;YACH,MAAM,KAAK,CAAC;QACd,CAAC;IACH,CAAC;IAED;;;;;OAKG;IACH,KAAK,CAAC,cAAc,CAAC,QAAiB;QACpC,IAAI,QAAQ,EAAE,CAAC;YACb,OAAO,IAAI,CAAC,OAAO,CAAC,sBAAsB,CAAC,QAAQ,CAAC,CAAC;QACvD,CAAC;QACD,OAAO,IAAI,CAAC,OAAO,CAAC,cAAc,EAAE,CAAC;IACvC,CAAC;IAED;;;;;OAKG;IACH,KAAK,CAAC,sBAAsB,CAC1B,UAAkB;QAElB,OAAO,IAAI,CAAC,OAAO,CAAC,sBAAsB,CAAC,UAAU,CAAC,CAAC;IACzD,CAAC;IAED;;;;;;;OAOG;IACK,KAAK,CAAC,eAAe,CAC3B,gBAAyB,EACzB,UAA6C,UAAU,EACvD,WAAoB,IAAI,EACxB,GAAY,EACZ,iBAA0B,EAC1B,gBAAsB,EACtB,MAAe,EACf,kBAGC,EACD,YAIE;QAEF,IAAI,CAAC,IAAI,CAAC,gBAAgB,CAAC,SAAS,EAAE,EAAE,CAAC;YACvC,IAAI,CAAC,MAAM,CAAC,IAAI,CACd,0CAA0C,EAC1C,gBAAgB;gBACd,CAAC,CAAC,uBAAuB,gBAAgB,EAAE;gBAC3C,CAAC,CAAC,uBAAuB,CAC5B,CAAC;YACF,OAAO,MAAM,IAAI,CAAC,gBAAgB,CAAC,KAAK,CAAC;gBACvC,OAAO;gBACP,QAAQ;gBACR,gBAAgB;gBAChB,GAAG;gBACH,iBAAiB;gBACjB,gBAAgB;gBAChB,MAAM;gBACN,kBAAkB;gBAClB,YAAY;aACb,CAAC,CAAC;QACL,CAAC;QAED,MAAM,GAAG,GAAG,IAAI,CAAC,gBAAgB,CAAC,MAAM,EAAE,CAAC;QAC3C,IAAI,CAAC,GAAG,EAAE,CAAC;YACT,MAAM,IAAI,KAAK,CAAC,gDAAgD,CAAC,CAAC;QACpE,CAAC;QAED,OAAO,GAAG,CAAC;IACb,CAAC;IAED;;;;OAIG;IACK,KAAK,CAAC,gBAAgB;QAC5B,IAAI,CAAC;YACH,MAAM,EAAE,QAAQ,EAAE,GAAG,MAAM,MAAM,CAAC,eAAe,CAAC,CAAC;YACnD,MAAM,MAAM,GAAG,QAAQ,CAAC,iCAAiC,EAAE;gBACzD,QAAQ,EAAE,OAAO;gBACjB,GAAG,EAAE,OAAO,CAAC,GAAG,EAAE;aACnB,CAAC,CAAC,IAAI,EAAE,CAAC;YACV,OAAO,MAAM,CAAC;QAChB,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,SAAS,CAAC;QACnB,CAAC;IACH,CAAC;IAED;;;;OAIG;IACK,KAAK,CAAC,mBAAmB;QAC/B,IAAI,CAAC;YACH,MAAM,EAAE,QAAQ,EAAE,GAAG,MAAM,MAAM,CAAC,eAAe,CAAC,CAAC;YACnD,MAAM,GAAG,GAAG,QAAQ,CAAC,oBAAoB,EAAE;gBACzC,QAAQ,EAAE,OAAO;gBACjB,GAAG,EAAE,OAAO,CAAC,GAAG,EAAE;aACnB,CAAC,CAAC,IAAI,EAAE,CAAC;YACV,OAAO,GAAG,CAAC;QACb,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,SAAS,CAAC;QACnB,CAAC;IACH,CAAC;IAED;;;;OAIG;IACH,KAAK,CAAC,OAAO;QACX,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,yBAAyB,CAAC,CAAC;QAE5C,IAAI,IAAI,CAAC,gBAAgB,CAAC,SAAS,EAAE,EAAE,CAAC;YACtC,MAAM,IAAI,CAAC,gBAAgB,CAAC,IAAI,EAAE,CAAC;QACrC,CAAC;IACH,CAAC;CACF"}

package/dist/ai-service/judge/prompts/evaluation-criteria.d.ts

@@ -0,0 +1,37 @@
+/**
+ * Evaluation criteria builder for judge assessments.
+ *
+ * Provides utilities for creating structured evaluation criteria
+ * based on simulation prompts and complexity levels.
+ */
+import type { EvaluationCriteria } from "../types.js";
+/**
+ * Simulation prompt interface matching the test infrastructure.
+ */
+interface SimulationPrompt {
+    id: string;
+    name: string;
+    description: string;
+    prompts: string[];
+    complexity: "low" | "medium" | "high";
+}
+/**
+ * Builds evaluation criteria from a simulation prompt.
+ *
+ * Analyzes the prompt structure and content to generate
+ * appropriate evaluation criteria for the judge.
+ *
+ * @param prompt - Simulation prompt object
+ * @returns Structured evaluation criteria
+ */
+export declare function buildCriteriaFromPrompt(prompt: SimulationPrompt): EvaluationCriteria;
+/**
+ * Creates minimal criteria for basic testing.
+ *
+ * Used when specific criteria cannot be extracted.
+ *
+ * @returns Minimal evaluation criteria
+ */
+export declare function createMinimalCriteria(): EvaluationCriteria;
+export {};
+//# sourceMappingURL=evaluation-criteria.d.ts.map

package/dist/ai-service/judge/prompts/evaluation-criteria.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluation-criteria.d.ts","sourceRoot":"","sources":["../../../../src/ai-service/judge/prompts/evaluation-criteria.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,aAAa,CAAC;AAEtD;;GAEG;AACH,UAAU,gBAAgB;IACxB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,UAAU,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,CAAC;CACvC;AAED;;;;;;;;GAQG;AACH,wBAAgB,uBAAuB,CACrC,MAAM,EAAE,gBAAgB,GACvB,kBAAkB,CAwCpB;AAkTD;;;;;;GAMG;AACH,wBAAgB,qBAAqB,IAAI,kBAAkB,CAS1D"}