@midscene/core 0.26.2 → 0.26.3-beta-20250813021342.0

package/dist/es/index.d.ts

@@ -1,53 +0,0 @@
-import { E as ExecutionTask, a as ExecutionTaskProgressOptions, b as ExecutionTaskApply, c as ExecutionDump, U as UIContext, I as InsightAction, D as DumpSubscriber, d as InsightTaskInfo, e as InsightOptions, f as DetailedLocateParam, L as LocateResult, g as InsightExtractParam, h as InsightExtractOption, T as TMultimodalPrompt, A as AIUsageInfo, i as TUserPrompt, j as InsightAssertionResponse, k as AIDescribeElementResponse } from './types-8a6be57c.js';
-export { z as AIAssertionResponse, x as AIDataExtractionResponse, v as AIElementCoordinatesResponse, u as AIElementLocatorResponse, w as AIElementResponse, q as AIResponseFormat, y as AISectionLocatorResponse, t as AISingleElementResponse, r as AISingleElementResponseById, s as AISingleElementResponseByPosition, W as AgentAssertOpt, F as AgentDescribeElementAtPointResult, V as AgentWaitForOpt, aa as BaseAgentParserOpt, G as CallAIFn, aE as CodeGenerationChunk, a9 as Color, aG as DeviceAction, K as DumpMeta, Q as ElementById, H as EnsureObject, ad as ExecutionRecorderItem, av as ExecutionTaskAction, au as ExecutionTaskActionApply, ag as ExecutionTaskHitBy, at as ExecutionTaskInsightAssertion, as as ExecutionTaskInsightAssertionApply, ar as ExecutionTaskInsightAssertionParam, ak as ExecutionTaskInsightDumpLog, am as ExecutionTaskInsightLocate, al as ExecutionTaskInsightLocateApply, aj as ExecutionTaskInsightLocateOutput, ai as ExecutionTaskInsightLocateParam, aq as ExecutionTaskInsightQuery, ap as ExecutionTaskInsightQueryApply, ao as ExecutionTaskInsightQueryOutput, an as ExecutionTaskInsightQueryParam, ax as ExecutionTaskLog, aw as ExecutionTaskLogApply, az as ExecutionTaskPlanning, ay as ExecutionTaskPlanningApply, ah as ExecutionTaskReturn, ae as ExecutionTaskType, af as ExecutorContext, b2 as FreeFn, aA as GroupedActionDump, N as InsightDump, O as LiteUISection, p as LocateOption, J as LocateResultElement, C as LocateValidatorResult, B as LocatorValidatorOption, b5 as MidsceneYamlConfig, b6 as MidsceneYamlConfigOutput, o as MidsceneYamlConfigResult, m as MidsceneYamlFlowItem, aN as MidsceneYamlFlowItemAIAction, aS as MidsceneYamlFlowItemAIAsk, aO as MidsceneYamlFlowItemAIAssert, aT as MidsceneYamlFlowItemAIBoolean, aX as MidsceneYamlFlowItemAIHover, aY as MidsceneYamlFlowItemAIInput, aZ as MidsceneYamlFlowItemAIKeyboardPress, aU as MidsceneYamlFlowItemAILocate, aQ as MidsceneYamlFlowItemAINumber, aP as MidsceneYamlFlowItemAIQuery, n as MidsceneYamlFlowItemAIRightClick, a_ as MidsceneYamlFlowItemAIScroll, aR as MidsceneYamlFlowItemAIString, aW as MidsceneYamlFlowItemAITap, aV as MidsceneYamlFlowItemAIWaitFor, a$ as MidsceneYamlFlowItemEvaluateJavaScript, b1 as MidsceneYamlFlowItemLogScreenshot, b0 as MidsceneYamlFlowItemSleep, M as MidsceneYamlScript, aL as MidsceneYamlScriptAndroidEnv, aM as MidsceneYamlScriptEnv, aJ as MidsceneYamlScriptEnvBase, aK as MidsceneYamlScriptWebEnv, l as MidsceneYamlTask, S as OnTaskStartTip, aB as PageType, P as PartialInsightDumpFromSDK, Z as PlanningAIResponse, Y as PlanningAction, a7 as PlanningActionParamAndroidLongPress, a8 as PlanningActionParamAndroidPull, a3 as PlanningActionParamAssert, a5 as PlanningActionParamError, $ as PlanningActionParamHover, a1 as PlanningActionParamInputOrKeyPress, a0 as PlanningActionParamRightClick, a2 as PlanningActionParamScroll, a4 as PlanningActionParamSleep, _ as PlanningActionParamTap, a6 as PlanningActionParamWaitFor, X as PlanningLocateParam, ac as PlaywrightParserOpt, ab as PuppeteerParserOpt, aH as ReferenceImage, R as ReportDumpWithAttributes, b4 as ScriptPlayerStatusValue, b3 as ScriptPlayerTaskStatus, aF as StreamingAIResponse, aD as StreamingCallback, aC as StreamingCodeGenerationOptions, aI as scrollParam } from './types-8a6be57c.js';
-import { c as callAiFn } from './llm-planning-4e0c16fe.js';
-export { a as AiAssert, A as AiLocateElement, d as describeUserPage, p as plan } from './llm-planning-4e0c16fe.js';
-import { BaseElement, Rect } from '@midscene/shared/types';
-export { BaseElement, ElementTreeNode, Point, Rect, Size } from '@midscene/shared/types';
-export { getVersion } from './utils.js';
-export { MIDSCENE_MODEL_NAME, getAIConfig } from '@midscene/shared/env';
-import '@midscene/shared/constants';
-import 'openai/resources';
-
-declare class Executor {
-    name: string;
-    tasks: ExecutionTask[];
-    status: 'init' | 'pending' | 'running' | 'completed' | 'error';
-    onTaskStart?: ExecutionTaskProgressOptions['onTaskStart'];
-    constructor(name: string, options?: ExecutionTaskProgressOptions & {
-        tasks?: ExecutionTaskApply[];
-    });
-    private markTaskAsPending;
-    append(task: ExecutionTaskApply[] | ExecutionTaskApply): Promise<void>;
-    flush(): Promise<{
-        output: any;
-        thought?: string;
-    } | undefined>;
-    isInErrorState(): boolean;
-    latestErrorTask(): ExecutionTask | null;
-    dump(): ExecutionDump;
-}
-
-interface LocateOpts {
-    context?: UIContext<BaseElement>;
-    callAI?: typeof callAiFn<AIElementResponse>;
-}
-declare class Insight<ElementType extends BaseElement = BaseElement, ContextType extends UIContext<ElementType> = UIContext<ElementType>> {
-    contextRetrieverFn: (action: InsightAction) => Promise<ContextType> | ContextType;
-    aiVendorFn: (...args: Array<any>) => Promise<any>;
-    onceDumpUpdatedFn?: DumpSubscriber;
-    taskInfo?: Omit<InsightTaskInfo, 'durationMs'>;
-    constructor(context: ContextType | ((action: InsightAction) => Promise<ContextType> | ContextType), opt?: InsightOptions);
-    locate(query: DetailedLocateParam, opt?: LocateOpts): Promise<LocateResult>;
-    extract<T>(dataDemand: InsightExtractParam, opt?: InsightExtractOption, multimodalPrompt?: TMultimodalPrompt): Promise<{
-        data: T;
-        thought?: string;
-        usage?: AIUsageInfo;
-    }>;
-    assert(assertion: TUserPrompt): Promise<InsightAssertionResponse>;
-    describe(target: Rect | [number, number], opt?: {
-        deepThink?: boolean;
-    }): Promise<Pick<AIDescribeElementResponse, 'description'>>;
-}
-
-export { AIDescribeElementResponse, AIUsageInfo, DetailedLocateParam, DumpSubscriber, ExecutionDump, ExecutionTask, ExecutionTaskApply, ExecutionTaskProgressOptions, Executor, Insight, InsightAction, InsightAssertionResponse, InsightExtractOption, InsightExtractParam, InsightOptions, InsightTaskInfo, LocateResult, TMultimodalPrompt, TUserPrompt, UIContext, Insight as default };

package/dist/es/index.js

@@ -1,570 +0,0 @@
-import {
-  getVersion
-} from "./chunk-O3KUKF2A.js";
-import {
-  AiAssert,
-  AiExtractElementInfo,
-  AiLocateElement,
-  AiLocateSection,
-  callAiFn,
-  callToGetJSONObject,
-  describeUserPage,
-  expandSearchArea,
-  plan
-} from "./chunk-DDYIQHOA.js";
-
-// src/ai-model/action-executor.ts
-import {
-  MIDSCENE_MODEL_NAME,
-  getAIConfig,
-  uiTarsModelVersion,
-  vlLocateMode
-} from "@midscene/shared/env";
-import { assert } from "@midscene/shared/utils";
-var Executor = class {
-  constructor(name, options) {
-    this.status = options?.tasks && options.tasks.length > 0 ? "pending" : "init";
-    this.name = name;
-    this.tasks = (options?.tasks || []).map(
-      (item) => this.markTaskAsPending(item)
-    );
-    this.onTaskStart = options?.onTaskStart;
-  }
-  markTaskAsPending(task) {
-    return {
-      status: "pending",
-      ...task
-    };
-  }
-  async append(task) {
-    assert(
-      this.status !== "error",
-      `executor is in error state, cannot append task
-error=${this.latestErrorTask()?.error}
-${this.latestErrorTask()?.errorStack}`
-    );
-    if (Array.isArray(task)) {
-      this.tasks.push(...task.map((item) => this.markTaskAsPending(item)));
-    } else {
-      this.tasks.push(this.markTaskAsPending(task));
-    }
-    if (this.status !== "running") {
-      this.status = "pending";
-    }
-  }
-  async flush() {
-    if (this.status === "init" && this.tasks.length > 0) {
-      console.warn(
-        "illegal state for executor, status is init but tasks are not empty"
-      );
-    }
-    assert(this.status !== "running", "executor is already running");
-    assert(this.status !== "completed", "executor is already completed");
-    assert(this.status !== "error", "executor is in error state");
-    const nextPendingIndex = this.tasks.findIndex(
-      (task) => task.status === "pending"
-    );
-    if (nextPendingIndex < 0) {
-      return;
-    }
-    this.status = "running";
-    let taskIndex = nextPendingIndex;
-    let successfullyCompleted = true;
-    let previousFindOutput;
-    while (taskIndex < this.tasks.length) {
-      const task = this.tasks[taskIndex];
-      assert(
-        task.status === "pending",
-        `task status should be pending, but got: ${task.status}`
-      );
-      task.timing = {
-        start: Date.now()
-      };
-      try {
-        task.status = "running";
-        try {
-          if (this.onTaskStart) {
-            await this.onTaskStart(task);
-          }
-        } catch (e) {
-          console.error("error in onTaskStart", e);
-        }
-        assert(
-          ["Insight", "Action", "Planning"].indexOf(task.type) >= 0,
-          `unsupported task type: ${task.type}`
-        );
-        const { executor, param } = task;
-        assert(executor, `executor is required for task type: ${task.type}`);
-        let returnValue;
-        const executorContext = {
-          task,
-          element: previousFindOutput?.element
-        };
-        if (task.type === "Insight") {
-          assert(
-            task.subType === "Locate" || task.subType === "Query" || task.subType === "Assert" || task.subType === "Boolean" || task.subType === "Number" || task.subType === "String",
-            `unsupported insight subType: ${task.subType}`
-          );
-          returnValue = await task.executor(param, executorContext);
-          if (task.subType === "Locate") {
-            previousFindOutput = returnValue?.output;
-          }
-        } else if (task.type === "Action" || task.type === "Planning") {
-          returnValue = await task.executor(param, executorContext);
-        } else {
-          console.warn(
-            `unsupported task type: ${task.type}, will try to execute it directly`
-          );
-          returnValue = await task.executor(param, executorContext);
-        }
-        Object.assign(task, returnValue);
-        task.status = "finished";
-        task.timing.end = Date.now();
-        task.timing.cost = task.timing.end - task.timing.start;
-        taskIndex++;
-      } catch (e) {
-        successfullyCompleted = false;
-        task.error = e;
-        task.errorMessage = e?.message || (typeof e === "string" ? e : "error-without-message");
-        task.errorStack = e.stack;
-        task.status = "failed";
-        task.timing.end = Date.now();
-        task.timing.cost = task.timing.end - task.timing.start;
-        break;
-      }
-    }
-    for (let i = taskIndex + 1; i < this.tasks.length; i++) {
-      this.tasks[i].status = "cancelled";
-    }
-    if (successfullyCompleted) {
-      this.status = "completed";
-    } else {
-      this.status = "error";
-    }
-    if (this.tasks.length) {
-      const outputIndex = Math.min(taskIndex, this.tasks.length - 1);
-      const { thought, output } = this.tasks[outputIndex];
-      return {
-        thought,
-        output
-      };
-    }
-  }
-  isInErrorState() {
-    return this.status === "error";
-  }
-  latestErrorTask() {
-    if (this.status !== "error") {
-      return null;
-    }
-    const errorTaskIndex = this.tasks.findIndex(
-      (task) => task.status === "failed"
-    );
-    if (errorTaskIndex >= 0) {
-      return this.tasks[errorTaskIndex];
-    }
-    return null;
-  }
-  dump() {
-    let modelDescription = "";
-    if (vlLocateMode()) {
-      const uiTarsModelVer = uiTarsModelVersion();
-      if (uiTarsModelVer) {
-        modelDescription = `UI-TARS=${uiTarsModelVer}`;
-      } else {
-        modelDescription = `${vlLocateMode()} mode`;
-      }
-    }
-    const dumpData = {
-      sdkVersion: getVersion(),
-      model_name: getAIConfig(MIDSCENE_MODEL_NAME) || "",
-      model_description: modelDescription,
-      logTime: Date.now(),
-      name: this.name,
-      tasks: this.tasks
-    };
-    return dumpData;
-  }
-};
-
-// src/ai-model/prompt/describe.ts
-import { getPreferredLanguage } from "@midscene/shared/env";
-var elementDescriberInstruction = () => {
-  return `
-Describe the element in the red rectangle for precise identification. Use ${getPreferredLanguage()}.
-
-CRITICAL REQUIREMENTS:
-1. UNIQUENESS: The description must uniquely identify this element on the current page
-2. UNIVERSALITY: Use generic, reusable selectors that work across different contexts
-3. PRECISION: Be specific enough to distinguish from similar elements
-
-DESCRIPTION STRUCTURE:
-1. Element type (button, input, link, div, etc.)
-2. Primary identifier (in order of preference):
-   - Unique text content: "with text 'Login'"
-   - Unique attribute: "with aria-label 'Search'"
-   - Unique class/ID: "with class 'primary-button'"
-   - Unique position: "in header navigation"
-3. Secondary identifiers (if needed for uniqueness):
-   - Visual features: "blue background", "with icon"
-   - Relative position: "below search bar", "in sidebar"
-   - Parent context: "in login form", "in main menu"
-
-GUIDELINES:
-- Keep description under 25 words
-- Prioritize semantic identifiers over visual ones
-- Use consistent terminology across similar elements
-- Avoid page-specific or temporary content
-- Don't mention the red rectangle or selection box
-- Focus on stable, reusable characteristics
-
-EXAMPLES:
-- "Login button with text 'Sign In'"
-- "Search input with placeholder 'Enter keywords'"
-- "Navigation link with text 'Home' in header"
-- "Submit button in contact form"
-- "Menu icon with aria-label 'Open menu'"
-
-Return JSON:
-{
-  "description": "unique element identifier",
-  "error"?: "error message if any"
-}`;
-};
-
-// src/insight/index.ts
-import {
-  MIDSCENE_FORCE_DEEP_THINK,
-  MIDSCENE_USE_QWEN_VL,
-  getAIConfigInBoolean,
-  vlLocateMode as vlLocateMode2
-} from "@midscene/shared/env";
-import { compositeElementInfoImg, cropByRect } from "@midscene/shared/img";
-import { getDebug } from "@midscene/shared/logger";
-import { assert as assert2 } from "@midscene/shared/utils";
-
-// src/insight/utils.ts
-import { MIDSCENE_MODEL_NAME as MIDSCENE_MODEL_NAME2, getAIConfig as getAIConfig2 } from "@midscene/shared/env";
-import { uuid } from "@midscene/shared/utils";
-function emitInsightDump(data, dumpSubscriber) {
-  const baseData = {
-    sdkVersion: getVersion(),
-    logTime: Date.now(),
-    model_name: getAIConfig2(MIDSCENE_MODEL_NAME2) || ""
-  };
-  const finalData = {
-    logId: uuid(),
-    ...baseData,
-    ...data
-  };
-  dumpSubscriber?.(finalData);
-}
-
-// src/insight/index.ts
-var debug = getDebug("ai:insight");
-var Insight = class {
-  constructor(context, opt) {
-    this.aiVendorFn = callAiFn;
-    assert2(context, "context is required for Insight");
-    if (typeof context === "function") {
-      this.contextRetrieverFn = context;
-    } else {
-      this.contextRetrieverFn = () => Promise.resolve(context);
-    }
-    if (typeof opt?.aiVendorFn !== "undefined") {
-      this.aiVendorFn = opt.aiVendorFn;
-    }
-    if (typeof opt?.taskInfo !== "undefined") {
-      this.taskInfo = opt.taskInfo;
-    }
-  }
-  async locate(query, opt) {
-    const { callAI } = opt || {};
-    const queryPrompt = typeof query === "string" ? query : query.prompt;
-    assert2(queryPrompt, "query is required for locate");
-    const dumpSubscriber = this.onceDumpUpdatedFn;
-    this.onceDumpUpdatedFn = void 0;
-    assert2(typeof query === "object", "query should be an object for locate");
-    const globalDeepThinkSwitch = getAIConfigInBoolean(
-      MIDSCENE_FORCE_DEEP_THINK
-    );
-    if (globalDeepThinkSwitch) {
-      debug("globalDeepThinkSwitch", globalDeepThinkSwitch);
-    }
-    let searchAreaPrompt;
-    if (query.deepThink || globalDeepThinkSwitch) {
-      searchAreaPrompt = query.prompt;
-    }
-    if (searchAreaPrompt && !vlLocateMode2()) {
-      console.warn(
-        'The "deepThink" feature is not supported with multimodal LLM. Please config VL model for Midscene. https://midscenejs.com/choose-a-model'
-      );
-      searchAreaPrompt = void 0;
-    }
-    const context = opt?.context || await this.contextRetrieverFn("locate");
-    let searchArea = void 0;
-    let searchAreaRawResponse = void 0;
-    let searchAreaUsage = void 0;
-    let searchAreaResponse = void 0;
-    if (searchAreaPrompt) {
-      searchAreaResponse = await AiLocateSection({
-        context,
-        sectionDescription: searchAreaPrompt
-      });
-      assert2(
-        searchAreaResponse.rect,
-        `cannot find search area for "${searchAreaPrompt}"${searchAreaResponse.error ? `: ${searchAreaResponse.error}` : ""}`
-      );
-      searchAreaRawResponse = searchAreaResponse.rawResponse;
-      searchAreaUsage = searchAreaResponse.usage;
-      searchArea = searchAreaResponse.rect;
-    }
-    const startTime = Date.now();
-    const {
-      parseResult,
-      rect,
-      elementById,
-      rawResponse,
-      usage,
-      isOrderSensitive
-    } = await AiLocateElement({
-      callAI: callAI || this.aiVendorFn,
-      context,
-      targetElementDescription: queryPrompt,
-      searchConfig: searchAreaResponse
-    });
-    const timeCost = Date.now() - startTime;
-    const taskInfo = {
-      ...this.taskInfo ? this.taskInfo : {},
-      durationMs: timeCost,
-      rawResponse: JSON.stringify(rawResponse),
-      formatResponse: JSON.stringify(parseResult),
-      usage,
-      searchArea,
-      searchAreaRawResponse,
-      searchAreaUsage
-    };
-    let errorLog;
-    if (parseResult.errors?.length) {
-      errorLog = `AI model failed to locate: 
-${parseResult.errors.join("\n")}`;
-    }
-    const dumpData = {
-      type: "locate",
-      userQuery: {
-        element: queryPrompt
-      },
-      matchedElement: [],
-      matchedRect: rect,
-      data: null,
-      taskInfo,
-      deepThink: !!searchArea,
-      error: errorLog
-    };
-    const elements = [];
-    (parseResult.elements || []).forEach((item) => {
-      if ("id" in item) {
-        const element = elementById(item?.id);
-        if (!element) {
-          console.warn(
-            `locate: cannot find element id=${item.id}. Maybe an unstable response from AI model`
-          );
-          return;
-        }
-        elements.push(element);
-      }
-    });
-    emitInsightDump(
-      {
-        ...dumpData,
-        matchedElement: elements
-      },
-      dumpSubscriber
-    );
-    if (errorLog) {
-      throw new Error(errorLog);
-    }
-    assert2(
-      elements.length <= 1,
-      `locate: multiple elements found, length = ${elements.length}`
-    );
-    if (elements.length === 1) {
-      return {
-        element: {
-          id: elements[0].id,
-          indexId: elements[0].indexId,
-          center: elements[0].center,
-          rect: elements[0].rect,
-          xpaths: elements[0].xpaths || [],
-          attributes: elements[0].attributes,
-          isOrderSensitive
-        },
-        rect
-      };
-    }
-    return {
-      element: null,
-      rect
-    };
-  }
-  async extract(dataDemand, opt, multimodalPrompt) {
-    assert2(
-      typeof dataDemand === "object" || typeof dataDemand === "string",
-      `dataDemand should be object or string, but get ${typeof dataDemand}`
-    );
-    const dumpSubscriber = this.onceDumpUpdatedFn;
-    this.onceDumpUpdatedFn = void 0;
-    const context = await this.contextRetrieverFn("extract");
-    const startTime = Date.now();
-    const { parseResult, usage } = await AiExtractElementInfo({
-      context,
-      dataQuery: dataDemand,
-      multimodalPrompt,
-      extractOption: opt
-    });
-    const timeCost = Date.now() - startTime;
-    const taskInfo = {
-      ...this.taskInfo ? this.taskInfo : {},
-      durationMs: timeCost,
-      rawResponse: JSON.stringify(parseResult)
-    };
-    let errorLog;
-    if (parseResult.errors?.length) {
-      errorLog = `AI response error: 
-${parseResult.errors.join("\n")}`;
-    }
-    const dumpData = {
-      type: "extract",
-      userQuery: {
-        dataDemand
-      },
-      matchedElement: [],
-      data: null,
-      taskInfo,
-      error: errorLog
-    };
-    const { data, thought } = parseResult || {};
-    emitInsightDump(
-      {
-        ...dumpData,
-        data
-      },
-      dumpSubscriber
-    );
-    if (errorLog && !data) {
-      throw new Error(errorLog);
-    }
-    return {
-      data,
-      thought,
-      usage
-    };
-  }
-  async assert(assertion) {
-    const dumpSubscriber = this.onceDumpUpdatedFn;
-    this.onceDumpUpdatedFn = void 0;
-    const context = await this.contextRetrieverFn("assert");
-    const startTime = Date.now();
-    const assertResult = await AiAssert({
-      assertion,
-      context
-    });
-    const timeCost = Date.now() - startTime;
-    const taskInfo = {
-      ...this.taskInfo ? this.taskInfo : {},
-      durationMs: timeCost,
-      rawResponse: JSON.stringify(assertResult.content)
-    };
-    const { thought, pass } = assertResult.content;
-    const dumpData = {
-      type: "assert",
-      userQuery: {
-        assertion
-      },
-      matchedElement: [],
-      data: null,
-      taskInfo,
-      assertionPass: pass,
-      assertionThought: thought,
-      error: pass ? void 0 : thought
-    };
-    emitInsightDump(dumpData, dumpSubscriber);
-    return {
-      pass,
-      thought,
-      usage: assertResult.usage
-    };
-  }
-  async describe(target, opt) {
-    assert2(target, "target is required for insight.describe");
-    const context = await this.contextRetrieverFn("describe");
-    const { screenshotBase64, size } = context;
-    assert2(screenshotBase64, "screenshot is required for insight.describe");
-    const systemPrompt = elementDescriberInstruction();
-    const defaultRectSize = 30;
-    const targetRect = Array.isArray(target) ? {
-      left: Math.floor(target[0] - defaultRectSize / 2),
-      top: Math.floor(target[1] - defaultRectSize / 2),
-      width: defaultRectSize,
-      height: defaultRectSize
-    } : target;
-    let imagePayload = await compositeElementInfoImg({
-      inputImgBase64: screenshotBase64,
-      size,
-      elementsPositionInfo: [
-        {
-          rect: targetRect
-        }
-      ],
-      borderThickness: 3
-    });
-    if (opt?.deepThink) {
-      const searchArea = expandSearchArea(targetRect, context.size);
-      debug("describe: set searchArea", searchArea);
-      imagePayload = await cropByRect(
-        imagePayload,
-        searchArea,
-        getAIConfigInBoolean(MIDSCENE_USE_QWEN_VL)
-      );
-    }
-    const msgs = [
-      { role: "system", content: systemPrompt },
-      {
-        role: "user",
-        content: [
-          {
-            type: "image_url",
-            image_url: {
-              url: imagePayload,
-              detail: "high"
-            }
-          }
-        ]
-      }
-    ];
-    const callAIFn = this.aiVendorFn || callToGetJSONObject;
-    const res = await callAIFn(msgs, 4 /* DESCRIBE_ELEMENT */);
-    const { content } = res;
-    assert2(!content.error, `describe failed: ${content.error}`);
-    assert2(content.description, "failed to describe the element");
-    return content;
-  }
-};
-
-// src/index.ts
-import { getAIConfig as getAIConfig3, MIDSCENE_MODEL_NAME as MIDSCENE_MODEL_NAME3 } from "@midscene/shared/env";
-var src_default = Insight;
-export {
-  AiAssert,
-  AiLocateElement,
-  Executor,
-  Insight,
-  MIDSCENE_MODEL_NAME3 as MIDSCENE_MODEL_NAME,
-  src_default as default,
-  describeUserPage,
-  getAIConfig3 as getAIConfig,
-  getVersion,
-  plan
-};
-
-//# sourceMappingURL=index.js.map