rl-simulator-core 1.0.3 → 1.0.5

package/dist/index.d.mts

@@ -297,7 +297,7 @@ async function runAgent({ targetUrl, taskInput, sessionId, simulatedUserKnownInf
 
     // 2. Launch Browser
     const browser = await chromium.launch({ 
-        headless: true, // Visible for demo/debug
+        headless: false, // Visible for demo/debug
         args: ['--start-maximized'] // Attempt to maximize
     });
 
@@ -433,7 +433,7 @@ function getFormattedDate() {
  * @param {string} targetServer - The URL of the backend task service
  * @param {string} taskId - The ID of the task to run
  */
-async function runTaskLoop(targetWeb, targetServer, taskId) {
+async function runTaskLoop(targetWeb, targetServer, taskId, agentRunner = runAgent) {
     const results = [];
     const datetime = getFormattedDate();
     const folderName = `${taskId}_${datetime}`;
@@ -454,7 +454,7 @@ async function runTaskLoop(targetWeb, targetServer, taskId) {
     for (let i = 1; i <= 5; i++) {
         console.log(`\n=== Starting Iteration ${i}/5 ===`);
         try {
-            const data = await executeSingleCycle(targetWeb, targetServer, taskId, i);
+            const data = await executeSingleCycle(targetWeb, targetServer, taskId, i, agentRunner);
             const { sessionId, agentMessages, ...rest } = data;
             
             // Create session subdirectory
@@ -485,8 +485,25 @@ async function runTaskLoop(targetWeb, targetServer, taskId) {
     }
 
     // Save final-results.json (Consolidated)
+    // Aggregate results
+    results.filter(r => r.verifyResult);
+    const scores = results.map(r => r.verifyResult?.score || 0);
+    const successCount = scores.filter(s => s === 1).length;
+    const total = 5; // Fixed 5 iterations
+
+    const rate = `${successCount}/${total}`;
+    const summary = {
+        rate,
+        score: scores
+    };
+
+    const finalOutput = {
+        summary,
+        results
+    };
+
     const finalResultPath = path.join(outputDir, 'final-results.json');
-    fs.writeFileSync(finalResultPath, JSON.stringify(results, null, 2));
+    fs.writeFileSync(finalResultPath, JSON.stringify(finalOutput, null, 2));
     console.log(`Saved consolidated results to ${finalResultPath}`);
 
     // Zip the results
@@ -503,10 +520,14 @@ async function runTaskLoop(targetWeb, targetServer, taskId) {
         console.error("Failed to create zip:", zipError);
     }
     
-    return zipPath;
+    return {
+        rate,
+        zipPath,
+        taskId
+    };
 }
 
-async function executeSingleCycle(targetWeb, targetServer, taskId, iteration) {
+async function executeSingleCycle(targetWeb, targetServer, taskId, iteration, agentRunner) {
     const fetchJson = async (url, opts) => {
         const res = await fetch(url, opts);
         if (!res.ok) {
@@ -537,7 +558,7 @@ async function executeSingleCycle(targetWeb, targetServer, taskId, iteration) {
 
     // 3. Run Agent
     console.log("Running Agent...");
-    const agentResult = await runAgent({
+    const agentResult = await agentRunner({
         taskInput: TaskJson.task.instruction,
         targetUrl: targetWeb,
         sessionId: sessionId,

package/dist/index.d.ts

@@ -297,7 +297,7 @@ async function runAgent({ targetUrl, taskInput, sessionId, simulatedUserKnownInf
 
     // 2. Launch Browser
     const browser = await chromium.launch({ 
-        headless: true, // Visible for demo/debug
+        headless: false, // Visible for demo/debug
         args: ['--start-maximized'] // Attempt to maximize
     });
 
@@ -433,7 +433,7 @@ function getFormattedDate() {
  * @param {string} targetServer - The URL of the backend task service
  * @param {string} taskId - The ID of the task to run
  */
-async function runTaskLoop(targetWeb, targetServer, taskId) {
+async function runTaskLoop(targetWeb, targetServer, taskId, agentRunner = runAgent) {
     const results = [];
     const datetime = getFormattedDate();
     const folderName = `${taskId}_${datetime}`;
@@ -454,7 +454,7 @@ async function runTaskLoop(targetWeb, targetServer, taskId) {
     for (let i = 1; i <= 5; i++) {
         console.log(`\n=== Starting Iteration ${i}/5 ===`);
         try {
-            const data = await executeSingleCycle(targetWeb, targetServer, taskId, i);
+            const data = await executeSingleCycle(targetWeb, targetServer, taskId, i, agentRunner);
             const { sessionId, agentMessages, ...rest } = data;
             
             // Create session subdirectory
@@ -485,8 +485,25 @@ async function runTaskLoop(targetWeb, targetServer, taskId) {
     }
 
     // Save final-results.json (Consolidated)
+    // Aggregate results
+    results.filter(r => r.verifyResult);
+    const scores = results.map(r => r.verifyResult?.score || 0);
+    const successCount = scores.filter(s => s === 1).length;
+    const total = 5; // Fixed 5 iterations
+
+    const rate = `${successCount}/${total}`;
+    const summary = {
+        rate,
+        score: scores
+    };
+
+    const finalOutput = {
+        summary,
+        results
+    };
+
     const finalResultPath = path.join(outputDir, 'final-results.json');
-    fs.writeFileSync(finalResultPath, JSON.stringify(results, null, 2));
+    fs.writeFileSync(finalResultPath, JSON.stringify(finalOutput, null, 2));
     console.log(`Saved consolidated results to ${finalResultPath}`);
 
     // Zip the results
@@ -503,10 +520,14 @@ async function runTaskLoop(targetWeb, targetServer, taskId) {
         console.error("Failed to create zip:", zipError);
     }
     
-    return zipPath;
+    return {
+        rate,
+        zipPath,
+        taskId
+    };
 }
 
-async function executeSingleCycle(targetWeb, targetServer, taskId, iteration) {
+async function executeSingleCycle(targetWeb, targetServer, taskId, iteration, agentRunner) {
     const fetchJson = async (url, opts) => {
         const res = await fetch(url, opts);
         if (!res.ok) {
@@ -537,7 +558,7 @@ async function executeSingleCycle(targetWeb, targetServer, taskId, iteration) {
 
     // 3. Run Agent
     console.log("Running Agent...");
-    const agentResult = await runAgent({
+    const agentResult = await agentRunner({
         taskInput: TaskJson.task.instruction,
         targetUrl: targetWeb,
         sessionId: sessionId,

package/dist/index.js

@@ -302,7 +302,7 @@ async function runAgent({ targetUrl, taskInput, sessionId, simulatedUserKnownInf
     console.log(`\u{1F4C1} \u521B\u5EFA\u622A\u56FE\u76EE\u5F55: ${screenshotsDir}`);
   }
   const browser = await import_playwright.chromium.launch({
-    headless: true,
+    headless: false,
     // Visible for demo/debug
     args: ["--start-maximized"]
     // Attempt to maximize
@@ -405,7 +405,7 @@ function getFormattedDate() {
   const pad = (n) => String(n).padStart(2, "0");
   return `${now.getFullYear()}${pad(now.getMonth() + 1)}${pad(now.getDate())}_${pad(now.getHours())}${pad(now.getMinutes())}${pad(now.getSeconds())}`;
 }
-async function runTaskLoop(targetWeb, targetServer, taskId) {
+async function runTaskLoop(targetWeb, targetServer, taskId, agentRunner = runAgent) {
   const results = [];
   const datetime = getFormattedDate();
   const folderName = `${taskId}_${datetime}`;
@@ -421,7 +421,7 @@ async function runTaskLoop(targetWeb, targetServer, taskId) {
     console.log(`
 === Starting Iteration ${i}/5 ===`);
     try {
-      const data = await executeSingleCycle(targetWeb, targetServer, taskId, i);
+      const data = await executeSingleCycle(targetWeb, targetServer, taskId, i, agentRunner);
       const { sessionId, agentMessages, ...rest } = data;
       const sessionDir = import_path3.default.join(outputDir, sessionId);
       if (!import_fs2.default.existsSync(sessionDir)) {
@@ -442,8 +442,24 @@ async function runTaskLoop(targetWeb, targetServer, taskId) {
       results.push({ iteration: i, error: e.message });
     }
   }
+  const validResults = results.filter((r) => r.verifyResult);
+  const scores = results.map((r) => {
+    var _a;
+    return ((_a = r.verifyResult) == null ? void 0 : _a.score) || 0;
+  });
+  const successCount = scores.filter((s) => s === 1).length;
+  const total = 5;
+  const rate = `${successCount}/${total}`;
+  const summary = {
+    rate,
+    score: scores
+  };
+  const finalOutput = {
+    summary,
+    results
+  };
   const finalResultPath = import_path3.default.join(outputDir, "final-results.json");
-  import_fs2.default.writeFileSync(finalResultPath, JSON.stringify(results, null, 2));
+  import_fs2.default.writeFileSync(finalResultPath, JSON.stringify(finalOutput, null, 2));
   console.log(`Saved consolidated results to ${finalResultPath}`);
   console.log("Packaging results into ZIP...");
   const zipName = `${folderName}.zip`;
@@ -457,9 +473,13 @@ async function runTaskLoop(targetWeb, targetServer, taskId) {
   } catch (zipError) {
     console.error("Failed to create zip:", zipError);
   }
-  return zipPath;
+  return {
+    rate,
+    zipPath,
+    taskId
+  };
 }
-async function executeSingleCycle(targetWeb, targetServer, taskId, iteration) {
+async function executeSingleCycle(targetWeb, targetServer, taskId, iteration, agentRunner) {
   const fetchJson = async (url, opts) => {
     const res = await fetch(url, opts);
     if (!res.ok) {
@@ -483,7 +503,7 @@ async function executeSingleCycle(targetWeb, targetServer, taskId, iteration) {
     body: JSON.stringify({ task_id: taskId, session_id: sessionId })
   });
   console.log("Running Agent...");
-  const agentResult = await runAgent({
+  const agentResult = await agentRunner({
     taskInput: TaskJson.task.instruction,
     targetUrl: targetWeb,
     sessionId,

package/dist/index.mjs

@@ -268,7 +268,7 @@ async function runAgent({ targetUrl, taskInput, sessionId, simulatedUserKnownInf
     console.log(`\u{1F4C1} \u521B\u5EFA\u622A\u56FE\u76EE\u5F55: ${screenshotsDir}`);
   }
   const browser = await chromium.launch({
-    headless: true,
+    headless: false,
     // Visible for demo/debug
     args: ["--start-maximized"]
     // Attempt to maximize
@@ -371,7 +371,7 @@ function getFormattedDate() {
   const pad = (n) => String(n).padStart(2, "0");
   return `${now.getFullYear()}${pad(now.getMonth() + 1)}${pad(now.getDate())}_${pad(now.getHours())}${pad(now.getMinutes())}${pad(now.getSeconds())}`;
 }
-async function runTaskLoop(targetWeb, targetServer, taskId) {
+async function runTaskLoop(targetWeb, targetServer, taskId, agentRunner = runAgent) {
   const results = [];
   const datetime = getFormattedDate();
   const folderName = `${taskId}_${datetime}`;
@@ -387,7 +387,7 @@ async function runTaskLoop(targetWeb, targetServer, taskId) {
     console.log(`
 === Starting Iteration ${i}/5 ===`);
     try {
-      const data = await executeSingleCycle(targetWeb, targetServer, taskId, i);
+      const data = await executeSingleCycle(targetWeb, targetServer, taskId, i, agentRunner);
       const { sessionId, agentMessages, ...rest } = data;
       const sessionDir = path3.join(outputDir, sessionId);
       if (!fs2.existsSync(sessionDir)) {
@@ -408,8 +408,24 @@ async function runTaskLoop(targetWeb, targetServer, taskId) {
       results.push({ iteration: i, error: e.message });
     }
   }
+  const validResults = results.filter((r) => r.verifyResult);
+  const scores = results.map((r) => {
+    var _a;
+    return ((_a = r.verifyResult) == null ? void 0 : _a.score) || 0;
+  });
+  const successCount = scores.filter((s) => s === 1).length;
+  const total = 5;
+  const rate = `${successCount}/${total}`;
+  const summary = {
+    rate,
+    score: scores
+  };
+  const finalOutput = {
+    summary,
+    results
+  };
   const finalResultPath = path3.join(outputDir, "final-results.json");
-  fs2.writeFileSync(finalResultPath, JSON.stringify(results, null, 2));
+  fs2.writeFileSync(finalResultPath, JSON.stringify(finalOutput, null, 2));
   console.log(`Saved consolidated results to ${finalResultPath}`);
   console.log("Packaging results into ZIP...");
   const zipName = `${folderName}.zip`;
@@ -423,9 +439,13 @@ async function runTaskLoop(targetWeb, targetServer, taskId) {
   } catch (zipError) {
     console.error("Failed to create zip:", zipError);
   }
-  return zipPath;
+  return {
+    rate,
+    zipPath,
+    taskId
+  };
 }
-async function executeSingleCycle(targetWeb, targetServer, taskId, iteration) {
+async function executeSingleCycle(targetWeb, targetServer, taskId, iteration, agentRunner) {
   const fetchJson = async (url, opts) => {
     const res = await fetch(url, opts);
     if (!res.ok) {
@@ -449,7 +469,7 @@ async function executeSingleCycle(targetWeb, targetServer, taskId, iteration) {
     body: JSON.stringify({ task_id: taskId, session_id: sessionId })
   });
   console.log("Running Agent...");
-  const agentResult = await runAgent({
+  const agentResult = await agentRunner({
     taskInput: TaskJson.task.instruction,
     targetUrl: targetWeb,
     sessionId,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "rl-simulator-core",
-  "version": "1.0.3",
+  "version": "1.0.5",
   "main": "./dist/index.js",
   "module": "./dist/index.mjs",
   "types": "./dist/index.d.ts",

package/src/index.js

@@ -25,7 +25,7 @@ function getFormattedDate() {
  * @param {string} targetServer - The URL of the backend task service
  * @param {string} taskId - The ID of the task to run
  */
-export async function runTaskLoop(targetWeb, targetServer, taskId) {
+export async function runTaskLoop(targetWeb, targetServer, taskId, agentRunner = runAgent) {
     const results = [];
     const datetime = getFormattedDate();
     const folderName = `${taskId}_${datetime}`;
@@ -46,7 +46,7 @@ export async function runTaskLoop(targetWeb, targetServer, taskId) {
     for (let i = 1; i <= 5; i++) {
         console.log(`\n=== Starting Iteration ${i}/5 ===`);
         try {
-            const data = await executeSingleCycle(targetWeb, targetServer, taskId, i);
+            const data = await executeSingleCycle(targetWeb, targetServer, taskId, i, agentRunner);
             const { sessionId, agentMessages, ...rest } = data;
             
             // Create session subdirectory
@@ -77,8 +77,25 @@ export async function runTaskLoop(targetWeb, targetServer, taskId) {
     }
 
     // Save final-results.json (Consolidated)
+    // Aggregate results
+    const validResults = results.filter(r => r.verifyResult);
+    const scores = results.map(r => r.verifyResult?.score || 0);
+    const successCount = scores.filter(s => s === 1).length;
+    const total = 5; // Fixed 5 iterations
+
+    const rate = `${successCount}/${total}`;
+    const summary = {
+        rate,
+        score: scores
+    };
+
+    const finalOutput = {
+        summary,
+        results
+    };
+
     const finalResultPath = path.join(outputDir, 'final-results.json');
-    fs.writeFileSync(finalResultPath, JSON.stringify(results, null, 2));
+    fs.writeFileSync(finalResultPath, JSON.stringify(finalOutput, null, 2));
     console.log(`Saved consolidated results to ${finalResultPath}`);
 
     // Zip the results
@@ -95,10 +112,14 @@ export async function runTaskLoop(targetWeb, targetServer, taskId) {
         console.error("Failed to create zip:", zipError);
     }
     
-    return zipPath;
+    return {
+        rate,
+        zipPath,
+        taskId
+    };
 }
 
-async function executeSingleCycle(targetWeb, targetServer, taskId, iteration) {
+async function executeSingleCycle(targetWeb, targetServer, taskId, iteration, agentRunner) {
     const fetchJson = async (url, opts) => {
         const res = await fetch(url, opts);
         if (!res.ok) {
@@ -129,7 +150,7 @@ async function executeSingleCycle(targetWeb, targetServer, taskId, iteration) {
 
     // 3. Run Agent
     console.log("Running Agent...");
-    const agentResult = await runAgent({
+    const agentResult = await agentRunner({
         taskInput: TaskJson.task.instruction,
         targetUrl: targetWeb,
         sessionId: sessionId,

package/src/runner.js

@@ -15,7 +15,7 @@ async function runAgent({ targetUrl, taskInput, sessionId, simulatedUserKnownInf
 
     // 2. Launch Browser
     const browser = await chromium.launch({ 
-        headless: true, // Visible for demo/debug
+        headless: false, // Visible for demo/debug
         args: ['--start-maximized'] // Attempt to maximize
     });
 

package/.env

@@ -1,3 +0,0 @@
-ARK_API_KEY=efa2ecef-c5eb-44e1-8853-86c592d7cdc2
-EP=ep-20251205145158-pxg57
-BASE_URL=https://ark.cn-beijing.volces.com/api/v3