@midscene/core 0.21.4-beta-20250711063424.0 → 0.21.4-beta-20250715053831.0

package/dist/lib/{chunk-QT5OZCDN.js → chunk-NQHZHZRA.js}

@@ -1135,13 +1135,17 @@ pageDescription:
 
 // src/ai-model/service-caller/index.ts
 function checkAIConfig() {
-  if (_env.getAIConfig.call(void 0, _env.OPENAI_API_KEY))
+  const openaiKey = _env.getAIConfig.call(void 0, _env.OPENAI_API_KEY);
+  const azureConfig = _env.getAIConfig.call(void 0, _env.MIDSCENE_USE_AZURE_OPENAI);
+  const anthropicKey = _env.getAIConfig.call(void 0, _env.ANTHROPIC_API_KEY);
+  const initConfigJson = _env.getAIConfig.call(void 0, _env.MIDSCENE_OPENAI_INIT_CONFIG_JSON);
+  if (openaiKey)
     return true;
-  if (_env.getAIConfig.call(void 0, _env.MIDSCENE_USE_AZURE_OPENAI))
+  if (azureConfig)
     return true;
-  if (_env.getAIConfig.call(void 0, _env.ANTHROPIC_API_KEY))
+  if (anthropicKey)
     return true;
-  return Boolean(_env.getAIConfig.call(void 0, _env.MIDSCENE_OPENAI_INIT_CONFIG_JSON));
+  return Boolean(initConfigJson);
 }
 var debugConfigInitialized = false;
 function initDebugConfig() {
@@ -1191,9 +1195,12 @@ async function createChatClient({
   const socksProxy = _env.getAIConfig.call(void 0, _env.MIDSCENE_OPENAI_SOCKS_PROXY);
   const httpProxy = _env.getAIConfig.call(void 0, _env.MIDSCENE_OPENAI_HTTP_PROXY);
   let proxyAgent = void 0;
+  const debugProxy = _logger.getDebug.call(void 0, "ai:call:proxy");
   if (httpProxy) {
+    debugProxy("using http proxy", httpProxy);
     proxyAgent = new (0, _httpsproxyagent.HttpsProxyAgent)(httpProxy);
   } else if (socksProxy) {
+    debugProxy("using socks proxy", socksProxy);
     proxyAgent = new (0, _socksproxyagent.SocksProxyAgent)(socksProxy);
   }
   if (_env.getAIConfig.call(void 0, _env.OPENAI_USE_AZURE)) {
@@ -1290,7 +1297,7 @@ Please check your config.`
   }
   throw new Error("Openai SDK or Anthropic SDK is not initialized");
 }
-async function call(messages, AIActionTypeValue, responseFormat) {
+async function call(messages, AIActionTypeValue, responseFormat, options) {
   const { completion, style } = await createChatClient({
     AIActionTypeValue
   });
@@ -1300,91 +1307,217 @@ async function call(messages, AIActionTypeValue, responseFormat) {
   const debugProfileDetail = _logger.getDebug.call(void 0, "ai:profile:detail");
   const startTime = Date.now();
   const model = getModelName();
+  const isStreaming = _optionalChain([options, 'optionalAccess', _22 => _22.stream]) && _optionalChain([options, 'optionalAccess', _23 => _23.onChunk]);
   let content;
+  let accumulated = "";
   let usage;
   let timeCost;
   const commonConfig = {
     temperature: _env.vlLocateMode.call(void 0, ) === "vlm-ui-tars" ? 0 : 0.1,
-    stream: false,
+    stream: !!isStreaming,
     max_tokens: typeof maxTokens === "number" ? maxTokens : Number.parseInt(maxTokens || "2048", 10),
     ..._env.vlLocateMode.call(void 0, ) === "qwen-vl" ? {
       vl_high_resolution_images: true
     } : {}
   };
-  if (style === "openai") {
-    debugCall(`sending request to ${model}`);
-    let result;
-    try {
-      const startTime2 = Date.now();
-      result = await completion.create({
-        model,
-        messages,
-        response_format: responseFormat,
-        ...commonConfig
-      });
-      timeCost = Date.now() - startTime2;
-    } catch (e) {
-      const newError = new Error(
-        `failed to call AI model service: ${e.message}. Trouble shooting: https://midscenejs.com/model-provider.html`,
-        {
-          cause: e
-        }
+  try {
+    if (style === "openai") {
+      debugCall(
+        `sending ${isStreaming ? "streaming " : ""}request to ${model}`
       );
-      throw newError;
-    }
-    debugProfileStats(
-      `model, ${model}, mode, ${_env.vlLocateMode.call(void 0, ) || "default"}, ui-tars-version, ${_env.uiTarsModelVersion.call(void 0, )}, prompt-tokens, ${_optionalChain([result, 'access', _22 => _22.usage, 'optionalAccess', _23 => _23.prompt_tokens]) || ""}, completion-tokens, ${_optionalChain([result, 'access', _24 => _24.usage, 'optionalAccess', _25 => _25.completion_tokens]) || ""}, total-tokens, ${_optionalChain([result, 'access', _26 => _26.usage, 'optionalAccess', _27 => _27.total_tokens]) || ""}, cost-ms, ${Date.now() - startTime}, requestId, ${result._request_id || ""}`
-    );
-    debugProfileDetail(`model usage detail: ${JSON.stringify(result.usage)}`);
-    _utils.assert.call(void 0, 
-      result.choices,
-      `invalid response from LLM service: ${JSON.stringify(result)}`
-    );
-    content = result.choices[0].message.content;
-    debugCall(`response: ${content}`);
-    _utils.assert.call(void 0, content, "empty content");
-    usage = result.usage;
-  } else if (style === "anthropic") {
-    const convertImageContent = (content2) => {
-      if (content2.type === "image_url") {
-        const imgBase64 = content2.image_url.url;
-        _utils.assert.call(void 0, imgBase64, "image_url is required");
-        return {
-          source: {
-            type: "base64",
-            media_type: imgBase64.includes("data:image/png;base64,") ? "image/png" : "image/jpeg",
-            data: imgBase64.split(",")[1]
+      if (isStreaming) {
+        const stream = await completion.create(
+          {
+            model,
+            messages,
+            response_format: responseFormat,
+            ...commonConfig
           },
-          type: "image"
-        };
+          {
+            stream: true
+          }
+        );
+        for await (const chunk of stream) {
+          const content2 = _optionalChain([chunk, 'access', _24 => _24.choices, 'optionalAccess', _25 => _25[0], 'optionalAccess', _26 => _26.delta, 'optionalAccess', _27 => _27.content]) || "";
+          const reasoning_content = _optionalChain([chunk, 'access', _28 => _28.choices, 'optionalAccess', _29 => _29[0], 'optionalAccess', _30 => _30.delta, 'optionalAccess', _31 => _31.reasoning_content]) || "";
+          if (chunk.usage) {
+            usage = chunk.usage;
+          }
+          if (content2 || reasoning_content) {
+            accumulated += content2;
+            const chunkData = {
+              content: content2,
+              reasoning_content,
+              accumulated,
+              isComplete: false,
+              usage: void 0
+            };
+            options.onChunk(chunkData);
+          }
+          if (_optionalChain([chunk, 'access', _32 => _32.choices, 'optionalAccess', _33 => _33[0], 'optionalAccess', _34 => _34.finish_reason])) {
+            timeCost = Date.now() - startTime;
+            if (!usage) {
+              const estimatedTokens = Math.max(
+                1,
+                Math.floor(accumulated.length / 4)
+              );
+              usage = {
+                prompt_tokens: estimatedTokens,
+                completion_tokens: estimatedTokens,
+                total_tokens: estimatedTokens * 2
+              };
+            }
+            const finalChunk = {
+              content: "",
+              accumulated,
+              reasoning_content: "",
+              isComplete: true,
+              usage: {
+                prompt_tokens: _nullishCoalesce(usage.prompt_tokens, () => ( 0)),
+                completion_tokens: _nullishCoalesce(usage.completion_tokens, () => ( 0)),
+                total_tokens: _nullishCoalesce(usage.total_tokens, () => ( 0)),
+                time_cost: _nullishCoalesce(timeCost, () => ( 0))
+              }
+            };
+            options.onChunk(finalChunk);
+            break;
+          }
+        }
+        content = accumulated;
+        debugProfileStats(
+          `streaming model, ${model}, mode, ${_env.vlLocateMode.call(void 0, ) || "default"}, cost-ms, ${timeCost}`
+        );
+      } else {
+        const result = await completion.create({
+          model,
+          messages,
+          response_format: responseFormat,
+          ...commonConfig
+        });
+        timeCost = Date.now() - startTime;
+        debugProfileStats(
+          `model, ${model}, mode, ${_env.vlLocateMode.call(void 0, ) || "default"}, ui-tars-version, ${_env.uiTarsModelVersion.call(void 0, )}, prompt-tokens, ${_optionalChain([result, 'access', _35 => _35.usage, 'optionalAccess', _36 => _36.prompt_tokens]) || ""}, completion-tokens, ${_optionalChain([result, 'access', _37 => _37.usage, 'optionalAccess', _38 => _38.completion_tokens]) || ""}, total-tokens, ${_optionalChain([result, 'access', _39 => _39.usage, 'optionalAccess', _40 => _40.total_tokens]) || ""}, cost-ms, ${timeCost}, requestId, ${result._request_id || ""}`
+        );
+        debugProfileDetail(
+          `model usage detail: ${JSON.stringify(result.usage)}`
+        );
+        _utils.assert.call(void 0, 
+          result.choices,
+          `invalid response from LLM service: ${JSON.stringify(result)}`
+        );
+        content = result.choices[0].message.content;
+        usage = result.usage;
       }
-      return content2;
+      debugCall(`response: ${content}`);
+      _utils.assert.call(void 0, content, "empty content");
+    } else if (style === "anthropic") {
+      const convertImageContent = (content2) => {
+        if (content2.type === "image_url") {
+          const imgBase64 = content2.image_url.url;
+          _utils.assert.call(void 0, imgBase64, "image_url is required");
+          return {
+            source: {
+              type: "base64",
+              media_type: imgBase64.includes("data:image/png;base64,") ? "image/png" : "image/jpeg",
+              data: imgBase64.split(",")[1]
+            },
+            type: "image"
+          };
+        }
+        return content2;
+      };
+      if (isStreaming) {
+        const stream = await completion.create({
+          model,
+          system: "You are a versatile professional in software UI automation",
+          messages: messages.map((m) => ({
+            role: "user",
+            content: Array.isArray(m.content) ? m.content.map(convertImageContent) : m.content
+          })),
+          response_format: responseFormat,
+          ...commonConfig
+        });
+        for await (const chunk of stream) {
+          const content2 = _optionalChain([chunk, 'access', _41 => _41.delta, 'optionalAccess', _42 => _42.text]) || "";
+          if (content2) {
+            accumulated += content2;
+            const chunkData = {
+              content: content2,
+              accumulated,
+              reasoning_content: "",
+              isComplete: false,
+              usage: void 0
+            };
+            options.onChunk(chunkData);
+          }
+          if (chunk.type === "message_stop") {
+            timeCost = Date.now() - startTime;
+            const anthropicUsage = chunk.usage;
+            const finalChunk = {
+              content: "",
+              accumulated,
+              reasoning_content: "",
+              isComplete: true,
+              usage: anthropicUsage ? {
+                prompt_tokens: _nullishCoalesce(anthropicUsage.input_tokens, () => ( 0)),
+                completion_tokens: _nullishCoalesce(anthropicUsage.output_tokens, () => ( 0)),
+                total_tokens: (_nullishCoalesce(anthropicUsage.input_tokens, () => ( 0))) + (_nullishCoalesce(anthropicUsage.output_tokens, () => ( 0))),
+                time_cost: _nullishCoalesce(timeCost, () => ( 0))
+              } : void 0
+            };
+            options.onChunk(finalChunk);
+            break;
+          }
+        }
+        content = accumulated;
+      } else {
+        const result = await completion.create({
+          model,
+          system: "You are a versatile professional in software UI automation",
+          messages: messages.map((m) => ({
+            role: "user",
+            content: Array.isArray(m.content) ? m.content.map(convertImageContent) : m.content
+          })),
+          response_format: responseFormat,
+          ...commonConfig
+        });
+        timeCost = Date.now() - startTime;
+        content = result.content[0].text;
+        usage = result.usage;
+      }
+      _utils.assert.call(void 0, content, "empty content");
+    }
+    if (isStreaming && !usage) {
+      const estimatedTokens = Math.max(
+        1,
+        Math.floor((content || "").length / 4)
+      );
+      usage = {
+        prompt_tokens: estimatedTokens,
+        completion_tokens: estimatedTokens,
+        total_tokens: estimatedTokens * 2
+      };
+    }
+    return {
+      content: content || "",
+      usage: usage ? {
+        prompt_tokens: _nullishCoalesce(usage.prompt_tokens, () => ( 0)),
+        completion_tokens: _nullishCoalesce(usage.completion_tokens, () => ( 0)),
+        total_tokens: _nullishCoalesce(usage.total_tokens, () => ( 0)),
+        time_cost: _nullishCoalesce(timeCost, () => ( 0))
+      } : void 0,
+      isStreamed: !!isStreaming
     };
-    const startTime2 = Date.now();
-    const result = await completion.create({
-      model,
-      system: "You are a versatile professional in software UI automation",
-      messages: messages.map((m) => ({
-        role: "user",
-        content: Array.isArray(m.content) ? m.content.map(convertImageContent) : m.content
-      })),
-      response_format: responseFormat,
-      ...commonConfig
-    });
-    timeCost = Date.now() - startTime2;
-    content = result.content[0].text;
-    _utils.assert.call(void 0, content, "empty content");
-    usage = result.usage;
+  } catch (e) {
+    console.error(" call AI error", e);
+    const newError = new Error(
+      `failed to call ${isStreaming ? "streaming " : ""}AI model service: ${e.message}. Trouble shooting: https://midscenejs.com/model-provider.html`,
+      {
+        cause: e
+      }
+    );
+    throw newError;
   }
-  return {
-    content: content || "",
-    usage: {
-      prompt_tokens: _nullishCoalesce(_optionalChain([usage, 'optionalAccess', _28 => _28.prompt_tokens]), () => ( 0)),
-      completion_tokens: _nullishCoalesce(_optionalChain([usage, 'optionalAccess', _29 => _29.completion_tokens]), () => ( 0)),
-      total_tokens: _nullishCoalesce(_optionalChain([usage, 'optionalAccess', _30 => _30.total_tokens]), () => ( 0)),
-      time_cost: _nullishCoalesce(timeCost, () => ( 0))
-    }
-  };
 }
 async function callToGetJSONObject(messages, AIActionTypeValue) {
   let responseFormat;
@@ -1444,8 +1577,8 @@ function preprocessDoubaoBboxJson(input) {
 }
 function safeParseJson(input) {
   const cleanJsonString = extractJSONFromCodeBlock(input);
-  if (_optionalChain([cleanJsonString, 'optionalAccess', _31 => _31.match, 'call', _32 => _32(/\((\d+),(\d+)\)/)])) {
-    return _optionalChain([cleanJsonString, 'access', _33 => _33.match, 'call', _34 => _34(/\((\d+),(\d+)\)/), 'optionalAccess', _35 => _35.slice, 'call', _36 => _36(1), 'access', _37 => _37.map, 'call', _38 => _38(Number)]);
+  if (_optionalChain([cleanJsonString, 'optionalAccess', _43 => _43.match, 'call', _44 => _44(/\((\d+),(\d+)\)/)])) {
+    return _optionalChain([cleanJsonString, 'access', _45 => _45.match, 'call', _46 => _46(/\((\d+),(\d+)\)/), 'optionalAccess', _47 => _47.slice, 'call', _48 => _48(1), 'access', _49 => _49.map, 'call', _50 => _50(Number)]);
   }
   try {
     return JSON.parse(cleanJsonString);
@@ -1464,6 +1597,9 @@ function safeParseJson(input) {
 
 // src/ai-model/prompt/playwright-generator.ts
 
+
+// src/ai-model/prompt/yaml-generator.ts
+
 var getScreenshotsForLLM = (events, maxScreenshots = 1) => {
   const eventsWithScreenshots = events.filter(
     (event) => event.screenshotBefore || event.screenshotAfter || event.screenshotWithBox
@@ -1529,7 +1665,6 @@ var prepareEventSummary = (events, options = {}) => {
   const filteredEvents = filterEventsByType(events);
   const eventCounts = createEventCounts(filteredEvents, events.length);
   const startUrl = filteredEvents.navigationEvents.length > 0 ? filteredEvents.navigationEvents[0].url || "" : "";
-  const pageTitles = filteredEvents.navigationEvents.map((event) => event.title).filter((title) => Boolean(title)).slice(0, 5);
   const clickDescriptions = filteredEvents.clickEvents.map((event) => event.elementDescription).filter((desc) => Boolean(desc)).slice(0, 10);
   const inputDescriptions = extractInputDescriptions(
     filteredEvents.inputEvents
@@ -1540,7 +1675,6 @@ var prepareEventSummary = (events, options = {}) => {
     testName: options.testName || "Automated test from recorded events",
     startUrl,
     eventCounts,
-    pageTitles,
     urls,
     clickDescriptions,
     inputDescriptions,
@@ -1575,6 +1709,141 @@ var validateEvents = (events) => {
     throw new Error("No events provided for test generation");
   }
 };
+var generateYamlTest = async (events, options = {}) => {
+  try {
+    validateEvents(events);
+    const summary = prepareEventSummary(events, {
+      testName: options.testName,
+      maxScreenshots: options.maxScreenshots || 3
+    });
+    const yamlSummary = {
+      ...summary,
+      includeTimestamps: options.includeTimestamps || false
+    };
+    const screenshots = getScreenshotsForLLM(
+      events,
+      options.maxScreenshots || 3
+    );
+    const prompt = [
+      {
+        role: "system",
+        content: `You are an expert in Midscene.js YAML test generation. Generate clean, accurate YAML following these rules: ${_constants.YAML_EXAMPLE_CODE}`
+      },
+      {
+        role: "user",
+        content: `Generate YAML test for Midscene.js automation from recorded browser events.
+
+Event Summary:
+${JSON.stringify(yamlSummary, null, 2)}
+
+Convert events:
+- navigation → target.url
+- click → aiTap with element description
+- input → aiInput with value and locate
+- scroll → aiScroll with appropriate direction
+- Add aiAssert for important state changes
+
+Respond with YAML only, no explanations.`
+      }
+    ];
+    if (screenshots.length > 0) {
+      prompt.push({
+        role: "user",
+        content: "Here are screenshots from the recording session to help you understand the context:"
+      });
+      prompt.push({
+        role: "user",
+        content: screenshots.map((screenshot) => ({
+          type: "image_url",
+          image_url: {
+            url: screenshot
+          }
+        }))
+      });
+    }
+    const response = await call(prompt, 2 /* EXTRACT_DATA */);
+    if (_optionalChain([response, 'optionalAccess', _51 => _51.content]) && typeof response.content === "string") {
+      return response.content;
+    }
+    throw new Error("Failed to generate YAML test configuration");
+  } catch (error) {
+    throw new Error(`Failed to generate YAML test: ${error}`);
+  }
+};
+var generateYamlTestStream = async (events, options = {}) => {
+  try {
+    validateEvents(events);
+    const summary = prepareEventSummary(events, {
+      testName: options.testName,
+      maxScreenshots: options.maxScreenshots || 3
+    });
+    const yamlSummary = {
+      ...summary,
+      includeTimestamps: options.includeTimestamps || false
+    };
+    const screenshots = getScreenshotsForLLM(
+      events,
+      options.maxScreenshots || 3
+    );
+    const prompt = [
+      {
+        role: "system",
+        content: `You are an expert in Midscene.js YAML test generation. Generate clean, accurate YAML following these rules: ${_constants.YAML_EXAMPLE_CODE}`
+      },
+      {
+        role: "user",
+        content: `Generate YAML test for Midscene.js automation from recorded browser events.
+
+Event Summary:
+${JSON.stringify(yamlSummary, null, 2)}
+
+Convert events:
+- navigation → target.url
+- click → aiTap with element description
+- input → aiInput with value and locate
+- scroll → aiScroll with appropriate direction
+- Add aiAssert for important state changes
+
+Respond with YAML only, no explanations.`
+      }
+    ];
+    if (screenshots.length > 0) {
+      prompt.push({
+        role: "user",
+        content: "Here are screenshots from the recording session to help you understand the context:"
+      });
+      prompt.push({
+        role: "user",
+        content: screenshots.map((screenshot) => ({
+          type: "image_url",
+          image_url: {
+            url: screenshot
+          }
+        }))
+      });
+    }
+    if (options.stream && options.onChunk) {
+      return await call(prompt, 2 /* EXTRACT_DATA */, void 0, {
+        stream: true,
+        onChunk: options.onChunk
+      });
+    } else {
+      const response = await call(prompt, 2 /* EXTRACT_DATA */);
+      if (_optionalChain([response, 'optionalAccess', _52 => _52.content]) && typeof response.content === "string") {
+        return {
+          content: response.content,
+          usage: response.usage,
+          isStreamed: false
+        };
+      }
+      throw new Error("Failed to generate YAML test configuration");
+    }
+  } catch (error) {
+    throw new Error(`Failed to generate YAML test: ${error}`);
+  }
+};
+
+// src/ai-model/prompt/playwright-generator.ts
 var generatePlaywrightTest = async (events, options = {}) => {
   validateEvents(events);
   const summary = prepareEventSummary(events, {
@@ -1623,161 +1892,74 @@ ${_constants.PLAYWRIGHT_EXAMPLE_CODE}`;
     }
   ];
   const response = await call(prompt, 2 /* EXTRACT_DATA */);
-  if (_optionalChain([response, 'optionalAccess', _39 => _39.content]) && typeof response.content === "string") {
+  if (_optionalChain([response, 'optionalAccess', _53 => _53.content]) && typeof response.content === "string") {
     return response.content;
   }
   throw new Error("Failed to generate Playwright test code");
 };
-
-// src/ai-model/prompt/yaml-generator.ts
-
-var getScreenshotsForLLM2 = (events, maxScreenshots = 1) => {
-  const eventsWithScreenshots = events.filter(
-    (event) => event.screenshotBefore || event.screenshotAfter || event.screenshotWithBox
-  );
-  const sortedEvents = [...eventsWithScreenshots].sort((a, b) => {
-    if (a.type === "navigation" && b.type !== "navigation")
-      return -1;
-    if (a.type !== "navigation" && b.type === "navigation")
-      return 1;
-    if (a.type === "click" && b.type !== "click")
-      return -1;
-    if (a.type !== "click" && b.type === "click")
-      return 1;
-    return 0;
+var generatePlaywrightTestStream = async (events, options = {}) => {
+  validateEvents(events);
+  const summary = prepareEventSummary(events, {
+    testName: options.testName,
+    maxScreenshots: options.maxScreenshots || 3
   });
-  const screenshots = [];
-  for (const event of sortedEvents) {
-    const screenshot = event.screenshotWithBox || event.screenshotAfter || event.screenshotBefore;
-    if (screenshot && !screenshots.includes(screenshot)) {
-      screenshots.push(screenshot);
-      if (screenshots.length >= maxScreenshots)
-        break;
-    }
-  }
-  return screenshots;
-};
-var filterEventsByType2 = (events) => {
-  return {
-    navigationEvents: events.filter((event) => event.type === "navigation"),
-    clickEvents: events.filter((event) => event.type === "click"),
-    inputEvents: events.filter((event) => event.type === "input"),
-    scrollEvents: events.filter((event) => event.type === "scroll")
-  };
-};
-var createEventCounts2 = (filteredEvents, totalEvents) => {
-  return {
-    navigation: filteredEvents.navigationEvents.length,
-    click: filteredEvents.clickEvents.length,
-    input: filteredEvents.inputEvents.length,
-    scroll: filteredEvents.scrollEvents.length,
-    total: totalEvents
-  };
-};
-var extractInputDescriptions2 = (inputEvents) => {
-  return inputEvents.map((event) => ({
-    description: event.elementDescription || "",
-    value: event.value || ""
-  })).filter((item) => item.description && item.value);
-};
-var processEventsForLLM2 = (events) => {
-  return events.map((event) => ({
-    type: event.type,
-    timestamp: event.timestamp,
-    url: event.url,
-    title: event.title,
-    elementDescription: event.elementDescription,
-    value: event.value,
-    pageInfo: event.pageInfo,
-    elementRect: event.elementRect
-  }));
-};
-var prepareEventSummary2 = (events, options = {}) => {
-  const filteredEvents = filterEventsByType2(events);
-  const eventCounts = createEventCounts2(filteredEvents, events.length);
-  const startUrl = filteredEvents.navigationEvents.length > 0 ? filteredEvents.navigationEvents[0].url || "" : "";
-  const pageTitles = filteredEvents.navigationEvents.map((event) => event.title).filter((title) => Boolean(title)).slice(0, 5);
-  const clickDescriptions = filteredEvents.clickEvents.map((event) => event.elementDescription).filter((desc) => Boolean(desc)).slice(0, 10);
-  const inputDescriptions = extractInputDescriptions2(
-    filteredEvents.inputEvents
-  ).slice(0, 10);
-  const urls = filteredEvents.navigationEvents.map((e) => e.url).filter((url) => Boolean(url)).slice(0, 5);
-  const processedEvents = processEventsForLLM2(events);
-  return {
-    testName: options.testName || "Automated test from recorded events",
-    startUrl,
-    eventCounts,
-    pageTitles,
-    urls,
-    clickDescriptions,
-    inputDescriptions,
-    events: processedEvents
+  const playwrightSummary = {
+    ...summary,
+    waitForNetworkIdle: options.waitForNetworkIdle !== false,
+    waitForNetworkIdleTimeout: options.waitForNetworkIdleTimeout || 2e3,
+    viewportSize: options.viewportSize || { width: 1280, height: 800 }
   };
-};
-var validateEvents2 = (events) => {
-  if (!events.length) {
-    throw new Error("No events provided for test generation");
-  }
-};
-var generateYamlTest = async (events, options = {}) => {
-  try {
-    validateEvents2(events);
-    const summary = prepareEventSummary2(events, {
-      testName: options.testName,
-      maxScreenshots: options.maxScreenshots || 3
-    });
-    const yamlSummary = {
-      ...summary,
-      includeTimestamps: options.includeTimestamps || false
-    };
-    const screenshots = getScreenshotsForLLM2(
-      events,
-      options.maxScreenshots || 3
-    );
-    const prompt = [
-      {
-        role: "system",
-        content: `You are an expert in Midscene.js YAML test generation. Generate clean, accurate YAML following these rules: ${_constants.YAML_EXAMPLE_CODE}`
-      },
-      {
-        role: "user",
-        content: `Generate YAML test for Midscene.js automation from recorded browser events.
+  const screenshots = getScreenshotsForLLM(events, options.maxScreenshots || 3);
+  const promptText = `Generate a Playwright test using @midscene/web/playwright that reproduces this recorded browser session. The test should be based on the following events and follow the structure of the example provided. Make the test descriptive with appropriate assertions and validations.
 
 Event Summary:
-${JSON.stringify(yamlSummary, null, 2)}
+${JSON.stringify(playwrightSummary, null, 2)}
 
-Convert events:
-- navigation → target.url
-- click → aiTap with element description
-- input → aiInput with value and locate
-- scroll → aiScroll with appropriate direction
-- Add aiAssert for important state changes
+Generated code should:
+1. Import required dependencies
+2. Set up the test with proper configuration
+3. Include a beforeEach hook to navigate to the starting URL
+4. Implement a test that uses Midscene AI methods (aiTap, aiInput, aiAssert, etc.)
+5. Include appropriate assertions and validations
+6. Follow best practices for Playwright tests
+7. Be ready to execute without further modification
+8. can't wrap this test code in markdown code block
 
-Respond with YAML only, no explanations.`
-      }
-    ];
-    if (screenshots.length > 0) {
-      prompt.push({
-        role: "user",
-        content: "Here are screenshots from the recording session to help you understand the context:"
-      });
-      prompt.push({
-        role: "user",
-        content: screenshots.map((screenshot) => ({
-          type: "image_url",
-          image_url: {
-            url: screenshot
-          }
-        }))
-      });
+Respond ONLY with the complete Playwright test code, no explanations.`;
+  const messageContent = createMessageContent(
+    promptText,
+    screenshots,
+    options.includeScreenshots !== false
+  );
+  const systemPrompt = `You are an expert test automation engineer specializing in Playwright and Midscene. 
+Your task is to generate a complete, executable Playwright test using @midscene/web/playwright that reproduces a recorded browser session.
+
+${_constants.PLAYWRIGHT_EXAMPLE_CODE}`;
+  const prompt = [
+    {
+      role: "system",
+      content: systemPrompt
+    },
+    {
+      role: "user",
+      content: messageContent
     }
+  ];
+  if (options.stream && options.onChunk) {
+    return await call(prompt, 2 /* EXTRACT_DATA */, void 0, {
+      stream: true,
+      onChunk: options.onChunk
+    });
+  } else {
     const response = await call(prompt, 2 /* EXTRACT_DATA */);
-    if (_optionalChain([response, 'optionalAccess', _40 => _40.content]) && typeof response.content === "string") {
-      return response.content;
+    if (_optionalChain([response, 'optionalAccess', _54 => _54.content]) && typeof response.content === "string") {
+      return {
+        content: response.content,
+        usage: response.usage,
+        isStreamed: false
+      };
     }
-    throw new Error("Failed to generate YAML test configuration");
-  } catch (error) {
-    throw new Error(`Failed to generate YAML test: ${error}`);
+    throw new Error("Failed to generate Playwright test code");
   }
 };
 
@@ -1976,7 +2158,7 @@ async function AiLocateElement(options) {
     );
   }
   let referenceImagePayload;
-  if (_optionalChain([options, 'access', _41 => _41.referenceImage, 'optionalAccess', _42 => _42.rect]) && options.referenceImage.base64) {
+  if (_optionalChain([options, 'access', _55 => _55.referenceImage, 'optionalAccess', _56 => _56.rect]) && options.referenceImage.base64) {
     referenceImagePayload = await _img.cropByRect.call(void 0, 
       options.referenceImage.base64,
       options.referenceImage.rect,
@@ -2012,10 +2194,10 @@ async function AiLocateElement(options) {
     if ("bbox" in res.content && Array.isArray(res.content.bbox)) {
       resRect = adaptBboxToRect(
         res.content.bbox,
-        _optionalChain([options, 'access', _43 => _43.searchConfig, 'optionalAccess', _44 => _44.rect, 'optionalAccess', _45 => _45.width]) || context.size.width,
-        _optionalChain([options, 'access', _46 => _46.searchConfig, 'optionalAccess', _47 => _47.rect, 'optionalAccess', _48 => _48.height]) || context.size.height,
-        _optionalChain([options, 'access', _49 => _49.searchConfig, 'optionalAccess', _50 => _50.rect, 'optionalAccess', _51 => _51.left]),
-        _optionalChain([options, 'access', _52 => _52.searchConfig, 'optionalAccess', _53 => _53.rect, 'optionalAccess', _54 => _54.top])
+        _optionalChain([options, 'access', _57 => _57.searchConfig, 'optionalAccess', _58 => _58.rect, 'optionalAccess', _59 => _59.width]) || context.size.width,
+        _optionalChain([options, 'access', _60 => _60.searchConfig, 'optionalAccess', _61 => _61.rect, 'optionalAccess', _62 => _62.height]) || context.size.height,
+        _optionalChain([options, 'access', _63 => _63.searchConfig, 'optionalAccess', _64 => _64.rect, 'optionalAccess', _65 => _65.left]),
+        _optionalChain([options, 'access', _66 => _66.searchConfig, 'optionalAccess', _67 => _67.rect, 'optionalAccess', _68 => _68.top])
       );
       debugInspect("resRect", resRect);
       const rectCenter = {
@@ -2034,7 +2216,7 @@ async function AiLocateElement(options) {
     }
   } catch (e) {
     const msg = e instanceof Error ? `Failed to parse bbox: ${e.message}` : "unknown error in locate";
-    if (!errors || _optionalChain([errors, 'optionalAccess', _55 => _55.length]) === 0) {
+    if (!errors || _optionalChain([errors, 'optionalAccess', _69 => _69.length]) === 0) {
       errors = [msg];
     } else {
       errors.push(`(${msg})`);
@@ -2125,14 +2307,14 @@ async function AiExtractElementInfo(options) {
     truncateTextLength: 200,
     filterNonTextContent: false,
     visibleOnly: false,
-    domIncluded: _optionalChain([extractOption, 'optionalAccess', _56 => _56.domIncluded])
+    domIncluded: _optionalChain([extractOption, 'optionalAccess', _70 => _70.domIncluded])
   });
   const extractDataPromptText = await extractDataQueryPrompt(
     description,
     dataQuery
   );
   const userContent = [];
-  if (_optionalChain([extractOption, 'optionalAccess', _57 => _57.screenshotIncluded]) !== false) {
+  if (_optionalChain([extractOption, 'optionalAccess', _71 => _71.screenshotIncluded]) !== false) {
     userContent.push({
       type: "image_url",
       image_url: {
@@ -2260,7 +2442,7 @@ async function plan(userInstruction, opts) {
   const { content, usage } = await call2(msgs, 3 /* PLAN */);
   const rawResponse = JSON.stringify(content, void 0, 2);
   const planFromAI = content;
-  const actions = (_optionalChain([planFromAI, 'access', _58 => _58.action, 'optionalAccess', _59 => _59.type]) ? [planFromAI.action] : planFromAI.actions) || [];
+  const actions = (_optionalChain([planFromAI, 'access', _72 => _72.action, 'optionalAccess', _73 => _73.type]) ? [planFromAI.action] : planFromAI.actions) || [];
   const returnValue = {
     ...planFromAI,
     actions,
@@ -2287,7 +2469,7 @@ async function plan(userInstruction, opts) {
     _utils.assert.call(void 0, !planFromAI.error, `Failed to plan actions: ${planFromAI.error}`);
   } else {
     actions.forEach((action) => {
-      if (_optionalChain([action, 'access', _60 => _60.locate, 'optionalAccess', _61 => _61.id])) {
+      if (_optionalChain([action, 'access', _74 => _74.locate, 'optionalAccess', _75 => _75.id])) {
         const element = elementById(action.locate.id);
         if (element) {
           action.locate.id = element.id;
@@ -2576,6 +2758,8 @@ async function resizeImageForUiTars(imageBase64, size) {
 
 
 
-exports.systemPromptToLocateElement = systemPromptToLocateElement; exports.elementByPositionWithElementInfo = elementByPositionWithElementInfo; exports.describeUserPage = describeUserPage; exports.call = call; exports.callToGetJSONObject = callToGetJSONObject; exports.AIActionType = AIActionType; exports.callAiFn = callAiFn; exports.adaptBboxToRect = adaptBboxToRect; exports.expandSearchArea = expandSearchArea; exports.generatePlaywrightTest = generatePlaywrightTest; exports.generateYamlTest = generateYamlTest; exports.AiLocateElement = AiLocateElement; exports.AiLocateSection = AiLocateSection; exports.AiExtractElementInfo = AiExtractElementInfo; exports.AiAssert = AiAssert; exports.plan = plan; exports.vlmPlanning = vlmPlanning; exports.resizeImageForUiTars = resizeImageForUiTars;
 
-//# sourceMappingURL=chunk-QT5OZCDN.js.map
+
+exports.systemPromptToLocateElement = systemPromptToLocateElement; exports.elementByPositionWithElementInfo = elementByPositionWithElementInfo; exports.describeUserPage = describeUserPage; exports.call = call; exports.callToGetJSONObject = callToGetJSONObject; exports.AIActionType = AIActionType; exports.callAiFn = callAiFn; exports.adaptBboxToRect = adaptBboxToRect; exports.expandSearchArea = expandSearchArea; exports.generateYamlTest = generateYamlTest; exports.generateYamlTestStream = generateYamlTestStream; exports.generatePlaywrightTest = generatePlaywrightTest; exports.generatePlaywrightTestStream = generatePlaywrightTestStream; exports.AiLocateElement = AiLocateElement; exports.AiLocateSection = AiLocateSection; exports.AiExtractElementInfo = AiExtractElementInfo; exports.AiAssert = AiAssert; exports.plan = plan; exports.vlmPlanning = vlmPlanning; exports.resizeImageForUiTars = resizeImageForUiTars;
+
+//# sourceMappingURL=chunk-NQHZHZRA.js.map