@midscene/core 0.6.2 → 0.6.3-beta-20241017035917.0

package/dist/es/ai-model.js

@@ -4587,16 +4587,16 @@ Remember:
 
 If the planned tasks are sequential and tasks may appear only after the execution of previous tasks, this is considered normal. Thoughts, prompts, and error messages should all be in the same language as the user query.
 
-## Objective 2 (sub objective): Give a quick answer to the action with type "Locate" you just planned
+## Objective 2 (sub objective, only for action with type "Locate"): Give a quick answer to the action with type "Locate" you just planned, append a \`quickAnswer\` field after the \`param\` field
 
-Review the action you just planned. If the action type is 'Locate', provide a quick answer: Does any element meet the description in the prompt? If so, answer with the following format, as the \`quickAnswer\` field in the output JSON:
+If the action type is 'Locate', provide a quick answer: Does any element meet the description in the prompt? If so, answer with the following format, as the \`quickAnswer\` field in the output JSON:
 {
   "reason": "Reason for finding element 4: It is located in the upper right corner, is an image type, and according to the screenshot, it is a shopping cart icon button",
   "text": "PLACEHOLDER", // Replace PLACEHOLDER with the text of elementInfo, if none, leave empty
   "id": "wefew2222few2" // id of this element, replace with actual value in practice
 }
 
-If the action type is not 'Locate', or there is no element meets the description in the prompt (usually because it will show up after some interaction), the answer should be null.
+If there is no element meets the description in the prompt (usually because it will show up later after some interaction), the \`quickAnswer\` field should be null.
 
 ## Output JSON Format:
 
@@ -4610,7 +4610,7 @@ Please return the result in JSON format as follows:
       "param": {
         "prompt": "The search bar"
       },
-      "quickAnswer": { // since the first action is Locate, so we need to give a quick answer
+      "quickAnswer": { // since this action type is 'Locate', and we can find the element, so we need to give a quick answer
         "reason": "Reason for finding element 4: It is located in the upper right corner, is an input type, and according to the screenshot, it is a search bar",
         "text": "PLACEHOLDER", // Replace PLACEHOLDER with the text of elementInfo, if none, leave empty
         "id": "wefew2222few2" // ID of this element, replace with actual value in practice
@@ -4621,6 +4621,14 @@ Please return the result in JSON format as follows:
       "type": "Tap", // Type of action, like 'Tap' 'Hover' ...
       "param": any, // Parameter towards the task type
     },
+    {
+      "thought": "Reasons for generating this task, and why this task is feasible on this page",
+      "type": "Locate", // Type of action, like 'Tap' 'Hover' ...
+      "param": {
+        "prompt": "The search bar"
+      },
+      "quickAnswer": null,
+    },
     // ... more actions
   ],
   error?: string, // Overall error messages. If there is any error occurs during the task planning (i.e. error in previous 'actions' array), conclude the errors again, put error messages here,
@@ -4654,7 +4662,7 @@ var planSchema = {
               },
               param: {
                 type: ["object", "null"],
-                description: "Parameter towards the task type, can be null"
+                description: "Parameter towards the task type, can be null only when the type field is Tap or Hover"
               },
               quickAnswer: {
                 type: ["object", "null"],
@@ -4948,7 +4956,7 @@ import assert4 from "assert";
 async function AiInspectElement(options) {
   var _a;
   const { context, multi, targetElementDescription, callAI, useModel } = options;
-  const { screenshotBase64 } = context;
+  const { screenshotBase64, screenshotBase64WithElementMarker } = context;
   const { description, elementById } = await describeUserPage(context);
   if (((_a = options.quickAnswer) == null ? void 0 : _a.id) && elementById(options.quickAnswer.id)) {
     return {
@@ -4967,7 +4975,7 @@ async function AiInspectElement(options) {
         {
           type: "image_url",
           image_url: {
-            url: screenshotBase64
+            url: screenshotBase64WithElementMarker || screenshotBase64
           }
         },
         {

package/dist/es/index.js

@@ -4752,16 +4752,16 @@ Remember:
 
 If the planned tasks are sequential and tasks may appear only after the execution of previous tasks, this is considered normal. Thoughts, prompts, and error messages should all be in the same language as the user query.
 
-## Objective 2 (sub objective): Give a quick answer to the action with type "Locate" you just planned
+## Objective 2 (sub objective, only for action with type "Locate"): Give a quick answer to the action with type "Locate" you just planned, append a \`quickAnswer\` field after the \`param\` field
 
-Review the action you just planned. If the action type is 'Locate', provide a quick answer: Does any element meet the description in the prompt? If so, answer with the following format, as the \`quickAnswer\` field in the output JSON:
+If the action type is 'Locate', provide a quick answer: Does any element meet the description in the prompt? If so, answer with the following format, as the \`quickAnswer\` field in the output JSON:
 {
   "reason": "Reason for finding element 4: It is located in the upper right corner, is an image type, and according to the screenshot, it is a shopping cart icon button",
   "text": "PLACEHOLDER", // Replace PLACEHOLDER with the text of elementInfo, if none, leave empty
   "id": "wefew2222few2" // id of this element, replace with actual value in practice
 }
 
-If the action type is not 'Locate', or there is no element meets the description in the prompt (usually because it will show up after some interaction), the answer should be null.
+If there is no element meets the description in the prompt (usually because it will show up later after some interaction), the \`quickAnswer\` field should be null.
 
 ## Output JSON Format:
 
@@ -4775,7 +4775,7 @@ Please return the result in JSON format as follows:
       "param": {
         "prompt": "The search bar"
       },
-      "quickAnswer": { // since the first action is Locate, so we need to give a quick answer
+      "quickAnswer": { // since this action type is 'Locate', and we can find the element, so we need to give a quick answer
         "reason": "Reason for finding element 4: It is located in the upper right corner, is an input type, and according to the screenshot, it is a search bar",
         "text": "PLACEHOLDER", // Replace PLACEHOLDER with the text of elementInfo, if none, leave empty
         "id": "wefew2222few2" // ID of this element, replace with actual value in practice
@@ -4786,6 +4786,14 @@ Please return the result in JSON format as follows:
       "type": "Tap", // Type of action, like 'Tap' 'Hover' ...
       "param": any, // Parameter towards the task type
     },
+    {
+      "thought": "Reasons for generating this task, and why this task is feasible on this page",
+      "type": "Locate", // Type of action, like 'Tap' 'Hover' ...
+      "param": {
+        "prompt": "The search bar"
+      },
+      "quickAnswer": null,
+    },
     // ... more actions
   ],
   error?: string, // Overall error messages. If there is any error occurs during the task planning (i.e. error in previous 'actions' array), conclude the errors again, put error messages here,
@@ -4819,7 +4827,7 @@ var planSchema = {
               },
               param: {
                 type: ["object", "null"],
-                description: "Parameter towards the task type, can be null"
+                description: "Parameter towards the task type, can be null only when the type field is Tap or Hover"
               },
               quickAnswer: {
                 type: ["object", "null"],
@@ -5150,7 +5158,7 @@ import assert5 from "assert";
 async function AiInspectElement(options) {
   var _a;
   const { context, multi, targetElementDescription, callAI: callAI2, useModel } = options;
-  const { screenshotBase64 } = context;
+  const { screenshotBase64, screenshotBase64WithElementMarker } = context;
   const { description, elementById } = await describeUserPage(context);
   if (((_a = options.quickAnswer) == null ? void 0 : _a.id) && elementById(options.quickAnswer.id)) {
     return {
@@ -5169,7 +5177,7 @@ async function AiInspectElement(options) {
         {
           type: "image_url",
           image_url: {
-            url: screenshotBase64
+            url: screenshotBase64WithElementMarker || screenshotBase64
           }
         },
         {
@@ -5399,7 +5407,9 @@ function writeDumpReport(fileName, dumpData) {
       const attributesArr = Object.keys(attributes || {}).map((key) => {
         return `${key}="${encodeURIComponent(attributes[key])}"`;
       });
-      return `<script type="midscene_web_dump" type="application/json" ${attributesArr.join(" ")}>${dumpString}</script>`;
+      return `<script type="midscene_web_dump" type="application/json" ${attributesArr.join(" ")}>
+${dumpString}
+</script>`;
     });
     reportContent = tpl.replace("{{dump}}", dumps.join("\n"));
   }

package/dist/es/utils.js

@@ -45,7 +45,9 @@ function writeDumpReport(fileName, dumpData) {
       const attributesArr = Object.keys(attributes || {}).map((key) => {
         return `${key}="${encodeURIComponent(attributes[key])}"`;
       });
-      return `<script type="midscene_web_dump" type="application/json" ${attributesArr.join(" ")}>${dumpString}</script>`;
+      return `<script type="midscene_web_dump" type="application/json" ${attributesArr.join(" ")}>
+${dumpString}
+</script>`;
     });
     reportContent = tpl.replace("{{dump}}", dumps.join("\n"));
   }

package/dist/lib/ai-model.js

@@ -4607,16 +4607,16 @@ Remember:
 
 If the planned tasks are sequential and tasks may appear only after the execution of previous tasks, this is considered normal. Thoughts, prompts, and error messages should all be in the same language as the user query.
 
-## Objective 2 (sub objective): Give a quick answer to the action with type "Locate" you just planned
+## Objective 2 (sub objective, only for action with type "Locate"): Give a quick answer to the action with type "Locate" you just planned, append a \`quickAnswer\` field after the \`param\` field
 
-Review the action you just planned. If the action type is 'Locate', provide a quick answer: Does any element meet the description in the prompt? If so, answer with the following format, as the \`quickAnswer\` field in the output JSON:
+If the action type is 'Locate', provide a quick answer: Does any element meet the description in the prompt? If so, answer with the following format, as the \`quickAnswer\` field in the output JSON:
 {
   "reason": "Reason for finding element 4: It is located in the upper right corner, is an image type, and according to the screenshot, it is a shopping cart icon button",
   "text": "PLACEHOLDER", // Replace PLACEHOLDER with the text of elementInfo, if none, leave empty
   "id": "wefew2222few2" // id of this element, replace with actual value in practice
 }
 
-If the action type is not 'Locate', or there is no element meets the description in the prompt (usually because it will show up after some interaction), the answer should be null.
+If there is no element meets the description in the prompt (usually because it will show up later after some interaction), the \`quickAnswer\` field should be null.
 
 ## Output JSON Format:
 
@@ -4630,7 +4630,7 @@ Please return the result in JSON format as follows:
       "param": {
         "prompt": "The search bar"
       },
-      "quickAnswer": { // since the first action is Locate, so we need to give a quick answer
+      "quickAnswer": { // since this action type is 'Locate', and we can find the element, so we need to give a quick answer
         "reason": "Reason for finding element 4: It is located in the upper right corner, is an input type, and according to the screenshot, it is a search bar",
         "text": "PLACEHOLDER", // Replace PLACEHOLDER with the text of elementInfo, if none, leave empty
         "id": "wefew2222few2" // ID of this element, replace with actual value in practice
@@ -4641,6 +4641,14 @@ Please return the result in JSON format as follows:
       "type": "Tap", // Type of action, like 'Tap' 'Hover' ...
       "param": any, // Parameter towards the task type
     },
+    {
+      "thought": "Reasons for generating this task, and why this task is feasible on this page",
+      "type": "Locate", // Type of action, like 'Tap' 'Hover' ...
+      "param": {
+        "prompt": "The search bar"
+      },
+      "quickAnswer": null,
+    },
     // ... more actions
   ],
   error?: string, // Overall error messages. If there is any error occurs during the task planning (i.e. error in previous 'actions' array), conclude the errors again, put error messages here,
@@ -4674,7 +4682,7 @@ var planSchema = {
               },
               param: {
                 type: ["object", "null"],
-                description: "Parameter towards the task type, can be null"
+                description: "Parameter towards the task type, can be null only when the type field is Tap or Hover"
               },
               quickAnswer: {
                 type: ["object", "null"],
@@ -4960,7 +4968,7 @@ var import_node_assert4 = __toESM(require("assert"));
 async function AiInspectElement(options) {
   var _a;
   const { context, multi, targetElementDescription, callAI, useModel } = options;
-  const { screenshotBase64 } = context;
+  const { screenshotBase64, screenshotBase64WithElementMarker } = context;
   const { description, elementById } = await describeUserPage(context);
   if (((_a = options.quickAnswer) == null ? void 0 : _a.id) && elementById(options.quickAnswer.id)) {
     return {
@@ -4979,7 +4987,7 @@ async function AiInspectElement(options) {
         {
           type: "image_url",
           image_url: {
-            url: screenshotBase64
+            url: screenshotBase64WithElementMarker || screenshotBase64
           }
         },
         {

package/dist/lib/index.js

@@ -4774,16 +4774,16 @@ Remember:
 
 If the planned tasks are sequential and tasks may appear only after the execution of previous tasks, this is considered normal. Thoughts, prompts, and error messages should all be in the same language as the user query.
 
-## Objective 2 (sub objective): Give a quick answer to the action with type "Locate" you just planned
+## Objective 2 (sub objective, only for action with type "Locate"): Give a quick answer to the action with type "Locate" you just planned, append a \`quickAnswer\` field after the \`param\` field
 
-Review the action you just planned. If the action type is 'Locate', provide a quick answer: Does any element meet the description in the prompt? If so, answer with the following format, as the \`quickAnswer\` field in the output JSON:
+If the action type is 'Locate', provide a quick answer: Does any element meet the description in the prompt? If so, answer with the following format, as the \`quickAnswer\` field in the output JSON:
 {
   "reason": "Reason for finding element 4: It is located in the upper right corner, is an image type, and according to the screenshot, it is a shopping cart icon button",
   "text": "PLACEHOLDER", // Replace PLACEHOLDER with the text of elementInfo, if none, leave empty
   "id": "wefew2222few2" // id of this element, replace with actual value in practice
 }
 
-If the action type is not 'Locate', or there is no element meets the description in the prompt (usually because it will show up after some interaction), the answer should be null.
+If there is no element meets the description in the prompt (usually because it will show up later after some interaction), the \`quickAnswer\` field should be null.
 
 ## Output JSON Format:
 
@@ -4797,7 +4797,7 @@ Please return the result in JSON format as follows:
       "param": {
         "prompt": "The search bar"
       },
-      "quickAnswer": { // since the first action is Locate, so we need to give a quick answer
+      "quickAnswer": { // since this action type is 'Locate', and we can find the element, so we need to give a quick answer
         "reason": "Reason for finding element 4: It is located in the upper right corner, is an input type, and according to the screenshot, it is a search bar",
         "text": "PLACEHOLDER", // Replace PLACEHOLDER with the text of elementInfo, if none, leave empty
         "id": "wefew2222few2" // ID of this element, replace with actual value in practice
@@ -4808,6 +4808,14 @@ Please return the result in JSON format as follows:
       "type": "Tap", // Type of action, like 'Tap' 'Hover' ...
       "param": any, // Parameter towards the task type
     },
+    {
+      "thought": "Reasons for generating this task, and why this task is feasible on this page",
+      "type": "Locate", // Type of action, like 'Tap' 'Hover' ...
+      "param": {
+        "prompt": "The search bar"
+      },
+      "quickAnswer": null,
+    },
     // ... more actions
   ],
   error?: string, // Overall error messages. If there is any error occurs during the task planning (i.e. error in previous 'actions' array), conclude the errors again, put error messages here,
@@ -4841,7 +4849,7 @@ var planSchema = {
               },
               param: {
                 type: ["object", "null"],
-                description: "Parameter towards the task type, can be null"
+                description: "Parameter towards the task type, can be null only when the type field is Tap or Hover"
               },
               quickAnswer: {
                 type: ["object", "null"],
@@ -5164,7 +5172,7 @@ var import_node_assert5 = __toESM(require("assert"));
 async function AiInspectElement(options) {
   var _a;
   const { context, multi, targetElementDescription, callAI: callAI2, useModel } = options;
-  const { screenshotBase64 } = context;
+  const { screenshotBase64, screenshotBase64WithElementMarker } = context;
   const { description, elementById } = await describeUserPage(context);
   if (((_a = options.quickAnswer) == null ? void 0 : _a.id) && elementById(options.quickAnswer.id)) {
     return {
@@ -5183,7 +5191,7 @@ async function AiInspectElement(options) {
         {
           type: "image_url",
           image_url: {
-            url: screenshotBase64
+            url: screenshotBase64WithElementMarker || screenshotBase64
           }
         },
         {
@@ -5413,7 +5421,9 @@ function writeDumpReport(fileName, dumpData) {
       const attributesArr = Object.keys(attributes || {}).map((key) => {
         return `${key}="${encodeURIComponent(attributes[key])}"`;
       });
-      return `<script type="midscene_web_dump" type="application/json" ${attributesArr.join(" ")}>${dumpString}</script>`;
+      return `<script type="midscene_web_dump" type="application/json" ${attributesArr.join(" ")}>
+${dumpString}
+</script>`;
     });
     reportContent = tpl.replace("{{dump}}", dumps.join("\n"));
   }

package/dist/lib/utils.js

@@ -91,7 +91,9 @@ function writeDumpReport(fileName, dumpData) {
       const attributesArr = Object.keys(attributes || {}).map((key) => {
         return `${key}="${encodeURIComponent(attributes[key])}"`;
       });
-      return `<script type="midscene_web_dump" type="application/json" ${attributesArr.join(" ")}>${dumpString}</script>`;
+      return `<script type="midscene_web_dump" type="application/json" ${attributesArr.join(" ")}>
+${dumpString}
+</script>`;
     });
     reportContent = tpl.replace("{{dump}}", dumps.join("\n"));
   }

package/dist/types/ai-model.d.ts

@@ -1,8 +1,8 @@
 import { ChatCompletionMessageParam } from 'openai/resources';
 export { ChatCompletionMessageParam } from 'openai/resources';
-import { c as callAiFn, A as AIActionType } from './index-52ae633b.js';
-export { d as describeUserPage, p as plan } from './index-52ae633b.js';
-import { B as BaseElement, U as UIContext, A as AISingleElementResponse, b as AIElementParseResponse, c as AISectionParseResponse, d as AIAssertionResponse } from './types-57e850bf.js';
+import { c as callAiFn, A as AIActionType } from './index-1fd0dc54.js';
+export { d as describeUserPage, p as plan } from './index-1fd0dc54.js';
+import { B as BaseElement, U as UIContext, A as AISingleElementResponse, b as AIElementParseResponse, c as AISectionParseResponse, d as AIAssertionResponse } from './types-25ace486.js';
 
 declare function AiInspectElement<ElementType extends BaseElement = BaseElement>(options: {
     context: UIContext<ElementType>;

package/dist/types/{index-52ae633b.d.ts → index-1fd0dc54.d.ts}

@@ -1,4 +1,4 @@
-import { B as BaseElement, U as UIContext, P as PlanningAction } from './types-57e850bf.js';
+import { B as BaseElement, U as UIContext, P as PlanningAction } from './types-25ace486.js';
 import { ChatCompletionSystemMessageParam, ChatCompletionUserMessageParam } from 'openai/resources';
 
 type AIArgs = [

package/dist/types/index.d.ts

@@ -1,7 +1,7 @@
-import { E as ExecutionTask, e as ExecutionTaskApply, f as ExecutionDump, B as BaseElement, U as UIContext, D as DumpSubscriber, I as InsightTaskInfo, g as InsightOptions, A as AISingleElementResponse, h as InsightAssertionResponse } from './types-57e850bf.js';
-export { d as AIAssertionResponse, b as AIElementParseResponse, j as AIResponseFormat, c as AISectionParseResponse, s as AgentWaitForOpt, J as BaseAgentParserOpt, m as BasicSectionQuery, C as CallAIFn, H as Color, o as DumpMeta, r as ElementById, l as EnsureObject, N as ExecutionRecorderItem, a7 as ExecutionTaskAction, a6 as ExecutionTaskActionApply, a5 as ExecutionTaskInsightAssertion, a4 as ExecutionTaskInsightAssertionApply, a3 as ExecutionTaskInsightAssertionParam, Y as ExecutionTaskInsightDumpLog, _ as ExecutionTaskInsightLocate, Z as ExecutionTaskInsightLocateApply, X as ExecutionTaskInsightLocateOutput, W as ExecutionTaskInsightLocateParam, a2 as ExecutionTaskInsightQuery, a1 as ExecutionTaskInsightQueryApply, a0 as ExecutionTaskInsightQueryOutput, $ as ExecutionTaskInsightQueryParam, a9 as ExecutionTaskPlanning, a8 as ExecutionTaskPlanningApply, V as ExecutionTaskReturn, O as ExecutionTaskType, Q as ExecutorContext, aa as GroupedActionDump, p as InsightDump, n as InsightExtractParam, L as LiteUISection, q as PartialInsightDumpFromSDK, t as PlanningAIResponse, P as PlanningAction, y as PlanningActionParamAssert, F as PlanningActionParamError, v as PlanningActionParamHover, w as PlanningActionParamInputOrKeyPress, x as PlanningActionParamScroll, z as PlanningActionParamSleep, u as PlanningActionParamTap, G as PlanningActionParamWaitFor, M as PlaywrightParserOpt, i as Point, K as PuppeteerParserOpt, a as Rect, R as ReportDumpWithAttributes, S as Size, T as TaskCacheInfo, k as UISection } from './types-57e850bf.js';
-import { c as callAiFn, r as retrieveElement, a as retrieveSection } from './index-52ae633b.js';
-export { p as plan } from './index-52ae633b.js';
+import { E as ExecutionTask, e as ExecutionTaskApply, f as ExecutionDump, B as BaseElement, U as UIContext, D as DumpSubscriber, I as InsightTaskInfo, g as InsightOptions, A as AISingleElementResponse, h as InsightAssertionResponse } from './types-25ace486.js';
+export { d as AIAssertionResponse, b as AIElementParseResponse, j as AIResponseFormat, c as AISectionParseResponse, s as AgentWaitForOpt, J as BaseAgentParserOpt, m as BasicSectionQuery, C as CallAIFn, H as Color, o as DumpMeta, r as ElementById, l as EnsureObject, N as ExecutionRecorderItem, a7 as ExecutionTaskAction, a6 as ExecutionTaskActionApply, a5 as ExecutionTaskInsightAssertion, a4 as ExecutionTaskInsightAssertionApply, a3 as ExecutionTaskInsightAssertionParam, Y as ExecutionTaskInsightDumpLog, _ as ExecutionTaskInsightLocate, Z as ExecutionTaskInsightLocateApply, X as ExecutionTaskInsightLocateOutput, W as ExecutionTaskInsightLocateParam, a2 as ExecutionTaskInsightQuery, a1 as ExecutionTaskInsightQueryApply, a0 as ExecutionTaskInsightQueryOutput, $ as ExecutionTaskInsightQueryParam, a9 as ExecutionTaskPlanning, a8 as ExecutionTaskPlanningApply, V as ExecutionTaskReturn, O as ExecutionTaskType, Q as ExecutorContext, aa as GroupedActionDump, p as InsightDump, n as InsightExtractParam, L as LiteUISection, q as PartialInsightDumpFromSDK, t as PlanningAIResponse, P as PlanningAction, y as PlanningActionParamAssert, F as PlanningActionParamError, v as PlanningActionParamHover, w as PlanningActionParamInputOrKeyPress, x as PlanningActionParamScroll, z as PlanningActionParamSleep, u as PlanningActionParamTap, G as PlanningActionParamWaitFor, M as PlaywrightParserOpt, i as Point, K as PuppeteerParserOpt, a as Rect, R as ReportDumpWithAttributes, S as Size, T as TaskCacheInfo, k as UISection } from './types-25ace486.js';
+import { c as callAiFn, r as retrieveElement, a as retrieveSection } from './index-1fd0dc54.js';
+export { p as plan } from './index-1fd0dc54.js';
 export { setLogDir } from './utils.js';
 import 'openai/resources';
 

package/dist/types/{types-57e850bf.d.ts → types-25ace486.d.ts}

@@ -58,6 +58,7 @@ interface AIAssertionResponse {
  */
 declare abstract class UIContext<ElementType extends BaseElement = BaseElement> {
     abstract screenshotBase64: string;
+    abstract screenshotBase64WithElementMarker?: string;
     abstract content: ElementType[];
     abstract size: Size;
 }

package/dist/types/utils.d.ts

@@ -1,4 +1,4 @@
-import { R as ReportDumpWithAttributes, a as Rect } from './types-57e850bf.js';
+import { R as ReportDumpWithAttributes, a as Rect } from './types-25ace486.js';
 import 'openai/resources';
 
 declare const insightDumpFileExt = "insight-dump.json";

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@midscene/core",
   "description": "An AI-powered automation SDK can control the page, perform assertions, and extract data in JSON format using natural language. See https://midscenejs.com/ for details.",
-  "version": "0.6.2",
+  "version": "0.6.3-beta-20241017035917.0",
   "repository": "https://github.com/web-infra-dev/midscene",
   "homepage": "https://midscenejs.com/",
   "jsnext:source": "./src/index.ts",
@@ -40,7 +40,7 @@
     "node-fetch": "2.6.7",
     "openai": "4.57.1",
     "optional": "0.1.4",
-    "@midscene/shared": "0.6.2"
+    "@midscene/shared": "0.6.3-beta-20241017035917.0"
   },
   "devDependencies": {
     "@modern-js/module-tools": "2.58.2",