@superblocksteam/vite-plugin-file-sync 1.99.99-next.2 → 2.0.0-SNAPSHOT.2143e2e95d0107ab

package/dist/ai-service/judge/tools/playwright-action.js

@@ -0,0 +1,171 @@
+/**
+ * Playwright action tool for browser automation via MCP.
+ *
+ * Provides the judge with browser automation capabilities
+ * to interact with and evaluate the generated application.
+ */
+import { z } from "zod";
+import { createToolFactory, ToolCategory } from "../../agent/tools2/types.js";
+/**
+ * Input schema for Playwright action tool.
+ */
+const playwrightActionSchema = z.object({
+    action: z
+        .enum([
+        "navigate",
+        "click",
+        "fill",
+        "screenshot", // currently disabled, leads to prompt_too_long errors
+        "getText",
+        "waitForSelector",
+        "evaluate",
+        "getUrl",
+        "reload",
+    ])
+        .describe("The Playwright action to perform"),
+    params: z.record(z.any()).describe("Action-specific parameters").optional(),
+});
+/**
+ * Creates a Playwright action tool factory.
+ *
+ * This tool allows the judge to perform browser automation
+ * actions through the Playwright MCP server.
+ */
+export const playwrightActionToolFactory = createToolFactory("playwright_action", (playwrightBridge) => ({
+    description: `Execute browser automation actions via Playwright.
+
+Available actions:
+- navigate: Go to a URL (params: { url: string })
+- click: Click an element (params: { selector: string })
+- fill: Fill an input field (params: { selector: string, value: string })
+- screenshot: Capture screenshot (params: { fullPage?: boolean })
+- getText: Get text from element (params: { selector: string })
+- waitForSelector: Wait for element (params: { selector: string, timeout?: number })
+- evaluate: Run JavaScript in browser context (params: { script: string }) - NOTE: Use plain JavaScript only, NO import/export/require statements
+- getUrl: Get current URL (no params)
+- reload: Reload page (no params)
+
+Examples:
+- Navigate: { action: "navigate", params: { url: "http://localhost:3000" } }
+- Click button: { action: "click", params: { selector: "button[type=submit]" } }
+- Fill input: { action: "fill", params: { selector: "#email", value: "test@example.com" } }
+- Get text: { action: "getText", params: { selector: "h1" } }
+- Screenshot: { action: "screenshot", params: { fullPage: true } }`,
+    category: ToolCategory.GENERAL,
+    inputSchema: playwrightActionSchema,
+    // This is a read-only tool from the filesystem perspective
+    readOnly: true,
+    toModelOutput: ({ output }) => {
+        if ("screenshot" in output) {
+            return {
+                type: "content",
+                value: [
+                    {
+                        type: "image-data",
+                        data: output.screenshot,
+                        mediaType: "image/png",
+                    },
+                ],
+            };
+        }
+        else {
+            return {
+                type: "json",
+                value: output,
+            };
+        }
+    },
+    execute: async ({ action, params = {} }) => {
+        console.log("🎭 Playwright Tool: Executing action", {
+            action,
+            hasParams: Object.keys(params).length > 0,
+            paramKeys: Object.keys(params),
+        });
+        // Pre-validate evaluate action to provide better error messages
+        if (action === "evaluate" && params.script) {
+            const script = params.script;
+            console.log("🎭 Playwright Tool: Evaluate action detected", {
+                scriptLength: script.length,
+                scriptPreview: script.slice(0, 100),
+            });
+            if (script.includes("import ") ||
+                script.includes("export ") ||
+                script.includes("require(")) {
+                console.log("🎭 Playwright Tool: Detected import/export/require in script, returning error");
+                // Return instructive error instead of failing
+                return {
+                    success: false,
+                    error: "Your JavaScript code contains import/export/require statements which cannot run in a browser console. Please rewrite using plain JavaScript. For example: instead of 'import React from \"react\"', check if React exists with 'typeof window.React !== \"undefined\"'. Use document.querySelector() to find elements.",
+                    suggestion: "Try using browser APIs directly: document.querySelector, window.location, etc.",
+                };
+            }
+        }
+        console.log("🎭 Playwright Tool: Calling playwrightBridge.executeAction");
+        const result = await playwrightBridge.executeAction(action, params);
+        if (!result.success) {
+            // Return error to model so it can retry with corrected approach
+            return {
+                success: false,
+                error: result.error || "Action failed",
+                suggestion: action === "evaluate"
+                    ? "Make sure your JavaScript code is browser-compatible (no imports/exports)"
+                    : undefined,
+            };
+        }
+        // Format response based on action type
+        switch (action) {
+            case "navigate":
+                return {
+                    success: true,
+                    message: `Navigated to ${params.url}`,
+                };
+            case "click":
+                return {
+                    success: true,
+                    message: `Clicked element: ${params.selector}`,
+                };
+            case "fill":
+                return {
+                    success: true,
+                    message: `Filled ${params.selector} with value`,
+                };
+            case "screenshot":
+                // Screenshots are currently disabled to prevent prompt_too_long errors
+                return {
+                    success: false,
+                    message: "tool is unavailable",
+                };
+            case "getText":
+                return {
+                    success: true,
+                    text: result.data,
+                };
+            case "getUrl":
+                return {
+                    success: true,
+                    url: result.data,
+                };
+            case "evaluate":
+                return {
+                    success: true,
+                    result: result.data,
+                };
+            case "waitForSelector":
+                return {
+                    success: true,
+                    message: `Element found: ${params.selector}`,
+                };
+            case "reload":
+                return {
+                    success: true,
+                    message: "Page reloaded",
+                };
+            default:
+                return {
+                    success: true,
+                    data: result.data,
+                };
+        }
+    },
+}));
+//# sourceMappingURL=playwright-action.js.map

package/dist/ai-service/judge/tools/playwright-action.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"playwright-action.js","sourceRoot":"","sources":["../../../../src/ai-service/judge/tools/playwright-action.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,OAAO,EAAE,iBAAiB,EAAE,YAAY,EAAE,MAAM,6BAA6B,CAAC;AAG9E;;GAEG;AACH,MAAM,sBAAsB,GAAG,CAAC,CAAC,MAAM,CAAC;IACtC,MAAM,EAAE,CAAC;SACN,IAAI,CAAC;QACJ,UAAU;QACV,OAAO;QACP,MAAM;QACN,YAAY,EAAE,sDAAsD;QACpE,SAAS;QACT,iBAAiB;QACjB,UAAU;QACV,QAAQ;QACR,QAAQ;KACT,CAAC;SACD,QAAQ,CAAC,kCAAkC,CAAC;IAE/C,MAAM,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,CAAC,4BAA4B,CAAC,CAAC,QAAQ,EAAE;CAC5E,CAAC,CAAC;AAEH;;;;;GAKG;AACH,MAAM,CAAC,MAAM,2BAA2B,GAAG,iBAAiB,CAC1D,mBAAmB,EACnB,CAAC,gBAAkC,EAAE,EAAE,CAAC,CAAC;IACvC,WAAW,EAAE;;;;;;;;;;;;;;;;;;mEAkBkD;IAE/D,QAAQ,EAAE,YAAY,CAAC,OAAO;IAE9B,WAAW,EAAE,sBAAsB;IAEnC,2DAA2D;IAC3D,QAAQ,EAAE,IAAI;IACd,aAAa,EAAE,CAAC,EAAE,MAAM,EAAE,EAAE,EAAE;QAC5B,IAAI,YAAY,IAAI,MAAM,EAAE,CAAC;YAC3B,OAAO;gBACL,IAAI,EAAE,SAAS;gBACf,KAAK,EAAE;oBACL;wBACE,IAAI,EAAE,YAAY;wBAClB,IAAI,EAAE,MAAM,CAAC,UAAU;wBACvB,SAAS,EAAE,WAAW;qBACvB;iBACF;aACF,CAAC;QACJ,CAAC;aAAM,CAAC;YACN,OAAO;gBACL,IAAI,EAAE,MAAM;gBACZ,KAAK,EAAE,MAAM;aACd,CAAC;QACJ,CAAC;IACH,CAAC;IACD,OAAO,EAAE,KAAK,EAAE,EAAE,MAAM,EAAE,MAAM,GAAG,EAAE,EAAE,EAAE,EAAE;QACzC,OAAO,CAAC,GAAG,CAAC,sCAAsC,EAAE;YAClD,MAAM;YACN,SAAS,EAAE,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,MAAM,GAAG,CAAC;YACzC,SAAS,EAAE,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC;SAC/B,CAAC,CAAC;QAEH,gEAAgE;QAChE,IAAI,MAAM,KAAK,UAAU,IAAI,MAAM,CAAC,MAAM,EAAE,CAAC;YAC3C,MAAM,MAAM,GAAG,MAAM,CAAC,MAAgB,CAAC;YACvC,OAAO,CAAC,GAAG,CAAC,8CAA8C,EAAE;gBAC1D,YAAY,EAAE,MAAM,CAAC,MAAM;gBAC3B,aAAa,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC;aACpC,CAAC,CAAC;YAEH,IACE,MAAM,CAAC,QAAQ,CAAC,SAAS,CAAC;gBAC1B,MAAM,CAAC,QAAQ,CAAC,SAAS,CAAC;gBAC1B,MAAM,CAAC,QAAQ,CAAC,UAAU,CAAC,EAC3B,CAAC;gBACD,OAAO,CAAC,GAAG,CACT,+EAA+E,CAChF,CAAC;gBACF,8CAA8C;gBAC9C,OAAO;oBACL,OAAO,EAAE,KAAK;oBACd,KAAK,EACH,wTAAwT;oBAC1T,UAAU,EACR,gFAAgF;iBACnF,CAAC;YACJ,CAAC;QACH,CAAC;QAED,OAAO,CAAC,GAAG,CAAC,4DAA4D,CAAC,CAAC;QAC1E,MAAM,MAAM,GAAG,MAAM,gBAAgB,CAAC,aAAa,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;QAEpE,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;YACpB,gEAAgE;YAChE,OAAO;gBACL,OAAO,EAAE,KAAK;gBACd,KAAK,EAAE,MAAM,CAAC,KAAK,IAAI,eAAe;gBACtC,UAAU,EACR,MAAM,KAAK,UAAU;oBACnB,CAAC,CAAC,2EAA2E;oBAC7E,CAAC,CAAC,SAAS;aAChB,CAAC;QACJ,CAAC;QAED,uCAAuC;QACvC,QAAQ,MAAM,EAAE,CAAC;YACf,KAAK,UAAU;gBACb,OAAO;oBACL,OAAO,EAAE,IAAI;oBACb,OAAO,EAAE,gBAAgB,MAAM,CAAC,GAAG,EAAE;iBACtC,CAAC;YAEJ,KAAK,OAAO;gBACV,OAAO;oBACL,OAAO,EAAE,IAAI;oBACb,OAAO,EAAE,oBAAoB,MAAM,CAAC,QAAQ,EAAE;iBAC/C,CAAC;YAEJ,KAAK,MAAM;gBACT,OAAO;oBACL,OAAO,EAAE,IAAI;oBACb,OAAO,EAAE,UAAU,MAAM,CAAC,QAAQ,aAAa;iBAChD,CAAC;YAEJ,KAAK,YAAY;gBACf,uEAAuE;gBACvE,OAAO;oBACL,OAAO,EAAE,KAAK;oBACd,OAAO,EAAE,qBAAqB;iBAC/B,CAAC;YAEJ,KAAK,SAAS;gBACZ,OAAO;oBACL,OAAO,EAAE,IAAI;oBACb,IAAI,EAAE,MAAM,CAAC,IAAI;iBAClB,CAAC;YAEJ,KAAK,QAAQ;gBACX,OAAO;oBACL,OAAO,EAAE,IAAI;oBACb,GAAG,EAAE,MAAM,CAAC,IAAI;iBACjB,CAAC;YAEJ,KAAK,UAAU;gBACb,OAAO;oBACL,OAAO,EAAE,IAAI;oBACb,MAAM,EAAE,MAAM,CAAC,IAAI;iBACpB,CAAC;YAEJ,KAAK,iBAAiB;gBACpB,OAAO;oBACL,OAAO,EAAE,IAAI;oBACb,OAAO,EAAE,kBAAkB,MAAM,CAAC,QAAQ,EAAE;iBAC7C,CAAC;YAEJ,KAAK,QAAQ;gBACX,OAAO;oBACL,OAAO,EAAE,IAAI;oBACb,OAAO,EAAE,eAAe;iBACzB,CAAC;YAEJ;gBACE,OAAO;oBACL,OAAO,EAAE,IAAI;oBACb,IAAI,EAAE,MAAM,CAAC,IAAI;iBAClB,CAAC;QACN,CAAC;IACH,CAAC;CACF,CAAC,CACH,CAAC"}

package/dist/ai-service/judge/tools/submit-feedback.d.ts

@@ -0,0 +1,34 @@
+/**
+ * Feedback submission tool for judge evaluation.
+ *
+ * Allows the judge to submit structured evaluation results
+ * after assessing the generated application.
+ */
+import type { JudgeEvaluation } from "../types.js";
+/**
+ * Creates a feedback submission tool factory.
+ *
+ * This tool allows the judge to submit a comprehensive
+ * evaluation after testing the application.
+ */
+export declare const submitFeedbackToolFactory: import("../../agent/tools2/types.js").ToolFactory<unknown, import("../../agent/tools2/types.js").Tool<{
+    summary: string;
+    passed: boolean;
+    suggestions: string[];
+    overallScore: number;
+    criteriaEvaluations: {
+        reasoning: string;
+        passed: boolean;
+        criterion: string;
+        score: number;
+        evidence?: string | undefined;
+        suggestions?: string[] | undefined;
+    }[];
+    confidence?: number | undefined;
+    strengths?: string[] | undefined;
+    weaknesses?: string[] | undefined;
+}, {
+    success: boolean;
+    evaluation: JudgeEvaluation;
+}>>;
+//# sourceMappingURL=submit-feedback.d.ts.map

package/dist/ai-service/judge/tools/submit-feedback.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"submit-feedback.d.ts","sourceRoot":"","sources":["../../../../src/ai-service/judge/tools/submit-feedback.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAKH,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,aAAa,CAAC;AAwDnD;;;;;GAKG;AACH,eAAO,MAAM,yBAAyB;;;;;;;;;;;;;;;;;aAkDZ,OAAO;gBAAc,eAAe;GA0C7D,CAAC"}

package/dist/ai-service/judge/tools/submit-feedback.js

@@ -0,0 +1,137 @@
+/**
+ * Feedback submission tool for judge evaluation.
+ *
+ * Allows the judge to submit structured evaluation results
+ * after assessing the generated application.
+ */
+import { z } from "zod";
+import { createToolFactory, ToolCategory } from "../../agent/tools2/types.js";
+/**
+ * Schema for criterion evaluation input.
+ */
+const criterionEvaluationSchema = z.object({
+    criterion: z.string().describe("The criterion being evaluated"),
+    passed: z.boolean().describe("Whether the criterion passed"),
+    score: z.number().min(0).max(100).describe("Score from 0-100"),
+    reasoning: z.string().describe("Detailed reasoning for the evaluation"),
+    evidence: z
+        .string()
+        .optional()
+        .describe("Supporting evidence (e.g., screenshot URL)"),
+    suggestions: z
+        .array(z.string())
+        .optional()
+        .describe("Improvement suggestions"),
+});
+/**
+ * Schema for feedback submission input.
+ */
+const submitFeedbackSchema = z.object({
+    overallScore: z.number().min(0).max(100).describe("Overall score from 0-100"),
+    passed: z.boolean().describe("Overall pass/fail determination"),
+    criteriaEvaluations: z
+        .array(criterionEvaluationSchema)
+        .describe("Individual criterion evaluations"),
+    summary: z.string().describe("Executive summary of the evaluation"),
+    suggestions: z
+        .array(z.string())
+        .describe("High-level suggestions for improvement"),
+    strengths: z
+        .array(z.string())
+        .optional()
+        .describe("Identified strengths in the implementation"),
+    weaknesses: z
+        .array(z.string())
+        .optional()
+        .describe("Identified weaknesses in the implementation"),
+    confidence: z
+        .number()
+        .min(0)
+        .max(1)
+        .optional()
+        .describe("Confidence level of the evaluation (0-1)"),
+});
+/**
+ * Creates a feedback submission tool factory.
+ *
+ * This tool allows the judge to submit a comprehensive
+ * evaluation after testing the application.
+ */
+export const submitFeedbackToolFactory = createToolFactory("submitFeedback", () => ({
+    description: `Submit your final evaluation of the application.
+
+Call this tool once you have:
+1. Tested all functional requirements
+2. Verified UI/UX requirements
+3. Checked data integration
+4. Gathered sufficient evidence
+
+Provide a comprehensive assessment with:
+- Overall score (0-100)
+- Pass/fail determination
+- Individual criterion evaluations with scores and reasoning
+- Executive summary
+- Specific suggestions for improvement
+- Identified strengths and weaknesses
+- Confidence level in your evaluation
+
+Example:
+{
+  "overallScore": 85,
+  "passed": true,
+  "criteriaEvaluations": [
+    {
+      "criterion": "User can create new records",
+      "passed": true,
+      "score": 90,
+      "reasoning": "The create functionality works well with proper validation",
+      "evidence": "screenshot_create_form.png",
+      "suggestions": ["Add confirmation dialog for create action"]
+    }
+  ],
+  "summary": "The application successfully implements core CRUD operations with good UX",
+  "suggestions": ["Improve error handling", "Add loading states"],
+  "strengths": ["Clean UI", "Responsive design", "Good data validation"],
+  "weaknesses": ["Limited error feedback", "No undo functionality"],
+  "confidence": 0.9
+}`,
+    category: ToolCategory.GENERAL,
+    inputSchema: submitFeedbackSchema,
+    // This is a write operation in terms of evaluation
+    readOnly: false,
+    execute: async (input) => {
+        // Validate that we have at least one criterion evaluation
+        if (input.criteriaEvaluations.length === 0) {
+            throw new Error("At least one criterion evaluation is required");
+        }
+        // Calculate weighted score if not matching overall
+        const avgScore = input.criteriaEvaluations.reduce((sum, e) => sum + e.score, 0) /
+            input.criteriaEvaluations.length;
+        // Warn if overall score differs significantly from average
+        if (Math.abs(input.overallScore - avgScore) > 20) {
+            console.warn(`Overall score (${input.overallScore}) differs significantly from average criterion score (${avgScore})`);
+        }
+        // Construct the evaluation object
+        const evaluation = {
+            overallScore: input.overallScore,
+            passed: input.passed,
+            criteriaEvaluations: input.criteriaEvaluations,
+            summary: input.summary,
+            suggestions: input.suggestions,
+            strengths: input.strengths,
+            weaknesses: input.weaknesses,
+            confidence: input.confidence,
+            // Token usage will be filled by the executor
+            tokensUsed: {
+                input: 0,
+                output: 0,
+                total: 0,
+            },
+        };
+        return {
+            success: true,
+            evaluation,
+        };
+    },
+}));
+//# sourceMappingURL=submit-feedback.js.map

package/dist/ai-service/judge/tools/submit-feedback.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"submit-feedback.js","sourceRoot":"","sources":["../../../../src/ai-service/judge/tools/submit-feedback.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,OAAO,EAAE,iBAAiB,EAAE,YAAY,EAAE,MAAM,6BAA6B,CAAC;AAG9E;;GAEG;AACH,MAAM,yBAAyB,GAAG,CAAC,CAAC,MAAM,CAAC;IACzC,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,+BAA+B,CAAC;IAC/D,MAAM,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,CAAC,8BAA8B,CAAC;IAC5D,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,QAAQ,CAAC,kBAAkB,CAAC;IAC9D,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,uCAAuC,CAAC;IACvE,QAAQ,EAAE,CAAC;SACR,MAAM,EAAE;SACR,QAAQ,EAAE;SACV,QAAQ,CAAC,4CAA4C,CAAC;IACzD,WAAW,EAAE,CAAC;SACX,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC;SACjB,QAAQ,EAAE;SACV,QAAQ,CAAC,yBAAyB,CAAC;CACvC,CAAC,CAAC;AAEH;;GAEG;AACH,MAAM,oBAAoB,GAAG,CAAC,CAAC,MAAM,CAAC;IACpC,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,QAAQ,CAAC,0BAA0B,CAAC;IAE7E,MAAM,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,CAAC,iCAAiC,CAAC;IAE/D,mBAAmB,EAAE,CAAC;SACnB,KAAK,CAAC,yBAAyB,CAAC;SAChC,QAAQ,CAAC,kCAAkC,CAAC;IAE/C,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,qCAAqC,CAAC;IAEnE,WAAW,EAAE,CAAC;SACX,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC;SACjB,QAAQ,CAAC,wCAAwC,CAAC;IAErD,SAAS,EAAE,CAAC;SACT,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC;SACjB,QAAQ,EAAE;SACV,QAAQ,CAAC,4CAA4C,CAAC;IAEzD,UAAU,EAAE,CAAC;SACV,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC;SACjB,QAAQ,EAAE;SACV,QAAQ,CAAC,6CAA6C,CAAC;IAE1D,UAAU,EAAE,CAAC;SACV,MAAM,EAAE;SACR,GAAG,CAAC,CAAC,CAAC;SACN,GAAG,CAAC,CAAC,CAAC;SACN,QAAQ,EAAE;SACV,QAAQ,CAAC,0CAA0C,CAAC;CACxD,CAAC,CAAC;AAEH;;;;;GAKG;AACH,MAAM,CAAC,MAAM,yBAAyB,GAAG,iBAAiB,CACxD,gBAAgB,EAChB,GAAG,EAAE,CAAC,CAAC;IACL,WAAW,EAAE;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAoCf;IAEE,QAAQ,EAAE,YAAY,CAAC,OAAO;IAE9B,WAAW,EAAE,oBAAoB;IAEjC,mDAAmD;IACnD,QAAQ,EAAE,KAAK;IAEf,OAAO,EAAE,KAAK,EACZ,KAAK,EACuD,EAAE;QAC9D,0DAA0D;QAC1D,IAAI,KAAK,CAAC,mBAAmB,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC3C,MAAM,IAAI,KAAK,CAAC,+CAA+C,CAAC,CAAC;QACnE,CAAC;QAED,mDAAmD;QACnD,MAAM,QAAQ,GACZ,KAAK,CAAC,mBAAmB,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC;YAC9D,KAAK,CAAC,mBAAmB,CAAC,MAAM,CAAC;QAEnC,2DAA2D;QAC3D,IAAI,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,YAAY,GAAG,QAAQ,CAAC,GAAG,EAAE,EAAE,CAAC;YACjD,OAAO,CAAC,IAAI,CACV,kBAAkB,KAAK,CAAC,YAAY,yDAAyD,QAAQ,GAAG,CACzG,CAAC;QACJ,CAAC;QAED,kCAAkC;QAClC,MAAM,UAAU,GAAoB;YAClC,YAAY,EAAE,KAAK,CAAC,YAAY;YAChC,MAAM,EAAE,KAAK,CAAC,MAAM;YACpB,mBAAmB,EAAE,KAAK,CAAC,mBAAmB;YAC9C,OAAO,EAAE,KAAK,CAAC,OAAO;YACtB,WAAW,EAAE,KAAK,CAAC,WAAW;YAC9B,SAAS,EAAE,KAAK,CAAC,SAAS;YAC1B,UAAU,EAAE,KAAK,CAAC,UAAU;YAC5B,UAAU,EAAE,KAAK,CAAC,UAAU;YAC5B,6CAA6C;YAC7C,UAAU,EAAE;gBACV,KAAK,EAAE,CAAC;gBACR,MAAM,EAAE,CAAC;gBACT,KAAK,EAAE,CAAC;aACT;SACF,CAAC;QAEF,OAAO;YACL,OAAO,EAAE,IAAI;YACb,UAAU;SACX,CAAC;IACJ,CAAC;CACF,CAAC,CACH,CAAC"}

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

@@ -0,0 +1,169 @@
+/**
+ * Judge evaluation types for AI simulation testing.
+ *
+ * Provides structured types for evaluation criteria, results, and scoring
+ * used by the LLM judge to assess AI-generated applications.
+ */
+/**
+ * Evaluation criteria for the judge to assess.
+ *
+ * Defines the requirements and expectations that the generated
+ * application should meet across different dimensions.
+ */
+export interface EvaluationCriteria {
+    /** Functional requirements that must be met */
+    functionalRequirements: string[];
+    /** UI/UX requirements for interface design and usability */
+    uiRequirements: string[];
+    /** Data integration and flow requirements */
+    dataRequirements: string[];
+    /** Performance and efficiency considerations */
+    performanceRequirements?: string[];
+    /** Additional custom criteria */
+    customCriteria?: string[];
+}
+/**
+ * Judge evaluation result for a single criterion.
+ *
+ * Represents the assessment of one specific requirement
+ * with scoring, reasoning, and evidence.
+ */
+export interface CriterionEvaluation {
+    /** The criterion being evaluated */
+    criterion: string;
+    /** Whether the criterion passed evaluation */
+    passed: boolean;
+    /** Numerical score from 0-100 */
+    score: number;
+    /** Detailed reasoning for the evaluation */
+    reasoning: string;
+    /** Supporting evidence (screenshots, DOM state, etc.) */
+    evidence?: string;
+    /** Suggestions for improvement */
+    suggestions?: string[];
+}
+/**
+ * Complete judge evaluation result.
+ *
+ * Comprehensive assessment of the AI-generated application
+ * including overall scoring and detailed criterion evaluations.
+ */
+export interface JudgeEvaluation {
+    /** Overall score from 0-100 */
+    overallScore: number;
+    /** Overall pass/fail determination */
+    passed: boolean;
+    /** Individual criterion evaluations */
+    criteriaEvaluations: CriterionEvaluation[];
+    /** Executive summary of the evaluation */
+    summary: string;
+    /** High-level suggestions for improvement */
+    suggestions: string[];
+    /** Strengths identified in the implementation */
+    strengths?: string[];
+    /** Weaknesses identified in the implementation */
+    weaknesses?: string[];
+    /** Token usage statistics */
+    tokensUsed: {
+        input: number;
+        output: number;
+        cached?: number;
+        total: number;
+    };
+    /** Time taken for evaluation in milliseconds */
+    evaluationDurationMs?: number;
+    /** Confidence level of the evaluation (0-1) */
+    confidence?: number;
+    /** Error message if evaluation failed */
+    error?: string;
+}
+/**
+ * Configuration for the judge execution.
+ */
+export interface JudgeConfig {
+    /** Playwright MCP server URL */
+    playwrightMcpUrl: string;
+    /** JWT token for AI Gateway authentication */
+    jwt?: string;
+    /** Maximum steps the judge can take */
+    maxSteps?: number;
+    /** Timeout for evaluation in milliseconds */
+    timeoutMs?: number;
+    /** Minimum passing score (0-100) */
+    passingThreshold?: number;
+    /** Whether to capture screenshots as evidence */
+    captureScreenshots?: boolean;
+    /** Whether to include detailed reasoning */
+    detailedReasoning?: boolean;
+}
+/**
+ * MCP tool call parameters for Playwright actions.
+ */
+export interface PlaywrightActionParams {
+    /** The action to perform */
+    action: "navigate" | "click" | "fill" | "screenshot" | "getText" | "waitForSelector" | "evaluate" | "getUrl" | "reload";
+    /** Action-specific parameters */
+    params: Record<string, any>;
+}
+/**
+ * Result from a Playwright action execution.
+ */
+export interface PlaywrightActionResult {
+    /** Whether the action succeeded */
+    success: boolean;
+    /** Result data from the action */
+    data?: any;
+    /** Error message if action failed */
+    error?: string;
+    /** Screenshot if captured */
+    screenshot?: string;
+}
+/**
+ * Judge context passed to tools.
+ */
+export interface JudgeContext {
+    /** Application URL being evaluated */
+    appUrl: string;
+    /** Original simulation prompts */
+    prompts: string[];
+    /** Evaluation criteria */
+    criteria: EvaluationCriteria;
+    /** Current evaluation state */
+    evaluationState: {
+        criteriaChecked: string[];
+        evidence: Map<string, string>;
+        scores: Map<string, number>;
+    };
+}
+/**
+ * Extended simulation result with judge evaluation.
+ *
+ * Augments the standard simulation result with judge assessment data.
+ */
+export interface EvaluatedSimulationResult {
+    /** Base simulation result */
+    appUrl?: string;
+    promptId: string;
+    success: boolean;
+    error?: string;
+    duration: number;
+    stepResults: Array<{
+        prompt: string;
+        success: boolean;
+        duration: number;
+        error?: string;
+    }>;
+    tokens: {
+        input: number;
+        output: number;
+        cached?: number;
+        total: number;
+    };
+    /** Judge evaluation if performed */
+    judgeEvaluation?: JudgeEvaluation;
+    /** Error during evaluation if any */
+    evaluationError?: string;
+    /** Time taken for judge evaluation */
+    evaluationDuration?: number;
+}
+//# sourceMappingURL=types.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../../src/ai-service/judge/types.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH;;;;;GAKG;AACH,MAAM,WAAW,kBAAkB;IACjC,+CAA+C;IAC/C,sBAAsB,EAAE,MAAM,EAAE,CAAC;IAEjC,4DAA4D;IAC5D,cAAc,EAAE,MAAM,EAAE,CAAC;IAEzB,6CAA6C;IAC7C,gBAAgB,EAAE,MAAM,EAAE,CAAC;IAE3B,gDAAgD;IAChD,uBAAuB,CAAC,EAAE,MAAM,EAAE,CAAC;IAEnC,iCAAiC;IACjC,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;CAC3B;AAED;;;;;GAKG;AACH,MAAM,WAAW,mBAAmB;IAClC,oCAAoC;IACpC,SAAS,EAAE,MAAM,CAAC;IAElB,8CAA8C;IAC9C,MAAM,EAAE,OAAO,CAAC;IAEhB,iCAAiC;IACjC,KAAK,EAAE,MAAM,CAAC;IAEd,4CAA4C;IAC5C,SAAS,EAAE,MAAM,CAAC;IAElB,yDAAyD;IACzD,QAAQ,CAAC,EAAE,MAAM,CAAC;IAElB,kCAAkC;IAClC,WAAW,CAAC,EAAE,MAAM,EAAE,CAAC;CACxB;AAED;;;;;GAKG;AACH,MAAM,WAAW,eAAe;IAC9B,+BAA+B;IAC/B,YAAY,EAAE,MAAM,CAAC;IAErB,sCAAsC;IACtC,MAAM,EAAE,OAAO,CAAC;IAEhB,uCAAuC;IACvC,mBAAmB,EAAE,mBAAmB,EAAE,CAAC;IAE3C,0CAA0C;IAC1C,OAAO,EAAE,MAAM,CAAC;IAEhB,6CAA6C;IAC7C,WAAW,EAAE,MAAM,EAAE,CAAC;IAEtB,iDAAiD;IACjD,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC;IAErB,kDAAkD;IAClD,UAAU,CAAC,EAAE,MAAM,EAAE,CAAC;IAEtB,6BAA6B;IAC7B,UAAU,EAAE;QACV,KAAK,EAAE,MAAM,CAAC;QACd,MAAM,EAAE,MAAM,CAAC;QACf,MAAM,CAAC,EAAE,MAAM,CAAC;QAChB,KAAK,EAAE,MAAM,CAAC;KACf,CAAC;IAEF,gDAAgD;IAChD,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAE9B,+CAA+C;IAC/C,UAAU,CAAC,EAAE,MAAM,CAAC;IAEpB,yCAAyC;IACzC,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED;;GAEG;AACH,MAAM,WAAW,WAAW;IAC1B,gCAAgC;IAChC,gBAAgB,EAAE,MAAM,CAAC;IAEzB,8CAA8C;IAC9C,GAAG,CAAC,EAAE,MAAM,CAAC;IAEb,uCAAuC;IACvC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAElB,6CAA6C;IAC7C,SAAS,CAAC,EAAE,MAAM,CAAC;IAEnB,oCAAoC;IACpC,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAE1B,iDAAiD;IACjD,kBAAkB,CAAC,EAAE,OAAO,CAAC;IAE7B,4CAA4C;IAC5C,iBAAiB,CAAC,EAAE,OAAO,CAAC;CAC7B;AAED;;GAEG;AACH,MAAM,WAAW,sBAAsB;IACrC,4BAA4B;IAC5B,MAAM,EACF,UAAU,GACV,OAAO,GACP,MAAM,GACN,YAAY,GACZ,SAAS,GACT,iBAAiB,GACjB,UAAU,GACV,QAAQ,GACR,QAAQ,CAAC;IAEb,iCAAiC;IACjC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;CAC7B;AAED;;GAEG;AACH,MAAM,WAAW,sBAAsB;IACrC,mCAAmC;IACnC,OAAO,EAAE,OAAO,CAAC;IAEjB,kCAAkC;IAClC,IAAI,CAAC,EAAE,GAAG,CAAC;IAEX,qCAAqC;IACrC,KAAK,CAAC,EAAE,MAAM,CAAC;IAEf,6BAA6B;IAC7B,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED;;GAEG;AACH,MAAM,WAAW,YAAY;IAC3B,sCAAsC;IACtC,MAAM,EAAE,MAAM,CAAC;IAEf,kCAAkC;IAClC,OAAO,EAAE,MAAM,EAAE,CAAC;IAElB,0BAA0B;IAC1B,QAAQ,EAAE,kBAAkB,CAAC;IAE7B,+BAA+B;IAC/B,eAAe,EAAE;QACf,eAAe,EAAE,MAAM,EAAE,CAAC;QAC1B,QAAQ,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;QAC9B,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;KAC7B,CAAC;CACH;AAED;;;;GAIG;AACH,MAAM,WAAW,yBAAyB;IACxC,6BAA6B;IAC7B,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,OAAO,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,KAAK,CAAC;QACjB,MAAM,EAAE,MAAM,CAAC;QACf,OAAO,EAAE,OAAO,CAAC;QACjB,QAAQ,EAAE,MAAM,CAAC;QACjB,KAAK,CAAC,EAAE,MAAM,CAAC;KAChB,CAAC,CAAC;IACH,MAAM,EAAE;QACN,KAAK,EAAE,MAAM,CAAC;QACd,MAAM,EAAE,MAAM,CAAC;QACf,MAAM,CAAC,EAAE,MAAM,CAAC;QAChB,KAAK,EAAE,MAAM,CAAC;KACf,CAAC;IAEF,oCAAoC;IACpC,eAAe,CAAC,EAAE,eAAe,CAAC;IAElC,qCAAqC;IACrC,eAAe,CAAC,EAAE,MAAM,CAAC;IAEzB,sCAAsC;IACtC,kBAAkB,CAAC,EAAE,MAAM,CAAC;CAC7B"}

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

@@ -0,0 +1,8 @@
+/**
+ * Judge evaluation types for AI simulation testing.
+ *
+ * Provides structured types for evaluation criteria, results, and scoring
+ * used by the LLM judge to assess AI-generated applications.
+ */
+export {};
+//# sourceMappingURL=types.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"types.js","sourceRoot":"","sources":["../../../src/ai-service/judge/types.ts"],"names":[],"mappings":"AAAA;;;;;GAKG"}

package/dist/ai-service/knowledge/helpers.d.ts

@@ -0,0 +1,17 @@
+import type { PromptContextIntegration } from "@superblocksteam/library-shared/types";
+import { type FactDto } from "@superblocksteam/shared";
+type IntegrationFactsResult = {
+    integration: PromptContextIntegration;
+    facts: FactDto[];
+};
+export declare function formatSystemPromptFacts(orgFacts: FactDto[]): string;
+export declare function formatUserMessageFacts(userFacts: FactDto[], integrationFactsResults: IntegrationFactsResult[]): string;
+/**
+ * Generate XML index for on-demand facts that can be loaded via build_readFile.
+ * Similar to system skills index, but for org/user/integration facts.
+ *
+ * @returns XML string listing available on-demand knowledge, or null if none
+ */
+export declare function generateOnDemandFactsSkillIndex(orgFacts: FactDto[], userFacts: FactDto[], integrationFactsResults: IntegrationFactsResult[]): string | null;
+export {};
+//# sourceMappingURL=helpers.d.ts.map

package/dist/ai-service/knowledge/helpers.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"helpers.d.ts","sourceRoot":"","sources":["../../../src/ai-service/knowledge/helpers.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,wBAAwB,EAAE,MAAM,uCAAuC,CAAC;AACtF,OAAO,EAAkB,KAAK,OAAO,EAAE,MAAM,yBAAyB,CAAC;AAEvE,KAAK,sBAAsB,GAAG;IAC5B,WAAW,EAAE,wBAAwB,CAAC;IACtC,KAAK,EAAE,OAAO,EAAE,CAAC;CAClB,CAAC;AA2CF,wBAAgB,uBAAuB,CAAC,QAAQ,EAAE,OAAO,EAAE,GAAG,MAAM,CAiCnE;AAED,wBAAgB,sBAAsB,CACpC,SAAS,EAAE,OAAO,EAAE,EACpB,uBAAuB,EAAE,sBAAsB,EAAE,GAChD,MAAM,CA8CR;AAED;;;;;GAKG;AACH,wBAAgB,+BAA+B,CAC7C,QAAQ,EAAE,OAAO,EAAE,EACnB,SAAS,EAAE,OAAO,EAAE,EACpB,uBAAuB,EAAE,sBAAsB,EAAE,GAChD,MAAM,GAAG,IAAI,CA+Cf"}