@midscene/core 0.28.11-beta-20250922065131.0 → 0.28.12-beta-20250923080328.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (41) hide show
  1. package/dist/es/agent/agent.mjs +1 -1
  2. package/dist/es/agent/agent.mjs.map +1 -1
  3. package/dist/es/agent/tasks.mjs +37 -164
  4. package/dist/es/agent/tasks.mjs.map +1 -1
  5. package/dist/es/agent/ui-utils.mjs +27 -13
  6. package/dist/es/agent/ui-utils.mjs.map +1 -1
  7. package/dist/es/agent/utils.mjs +1 -1
  8. package/dist/es/ai-model/conversation-history.mjs +58 -0
  9. package/dist/es/ai-model/conversation-history.mjs.map +1 -0
  10. package/dist/es/ai-model/index.mjs +2 -1
  11. package/dist/es/ai-model/llm-planning.mjs +23 -3
  12. package/dist/es/ai-model/llm-planning.mjs.map +1 -1
  13. package/dist/es/ai-model/ui-tars-planning.mjs +26 -6
  14. package/dist/es/ai-model/ui-tars-planning.mjs.map +1 -1
  15. package/dist/es/utils.mjs +2 -2
  16. package/dist/es/yaml/player.mjs +0 -9
  17. package/dist/es/yaml/player.mjs.map +1 -1
  18. package/dist/lib/agent/agent.js +1 -1
  19. package/dist/lib/agent/agent.js.map +1 -1
  20. package/dist/lib/agent/tasks.js +36 -163
  21. package/dist/lib/agent/tasks.js.map +1 -1
  22. package/dist/lib/agent/ui-utils.js +27 -13
  23. package/dist/lib/agent/ui-utils.js.map +1 -1
  24. package/dist/lib/agent/utils.js +1 -1
  25. package/dist/lib/ai-model/conversation-history.js +92 -0
  26. package/dist/lib/ai-model/conversation-history.js.map +1 -0
  27. package/dist/lib/ai-model/index.js +6 -2
  28. package/dist/lib/ai-model/llm-planning.js +23 -3
  29. package/dist/lib/ai-model/llm-planning.js.map +1 -1
  30. package/dist/lib/ai-model/ui-tars-planning.js +26 -6
  31. package/dist/lib/ai-model/ui-tars-planning.js.map +1 -1
  32. package/dist/lib/utils.js +2 -2
  33. package/dist/lib/yaml/player.js +0 -9
  34. package/dist/lib/yaml/player.js.map +1 -1
  35. package/dist/types/agent/tasks.d.ts +3 -18
  36. package/dist/types/agent/ui-utils.d.ts +1 -1
  37. package/dist/types/ai-model/conversation-history.d.ts +18 -0
  38. package/dist/types/ai-model/index.d.ts +1 -0
  39. package/dist/types/ai-model/llm-planning.d.ts +2 -0
  40. package/dist/types/ai-model/ui-tars-planning.d.ts +6 -18
  41. package/package.json +3 -3
@@ -251,7 +251,7 @@ class Agent {
251
251
  const yaml = null == (_matchedCache_cacheContent1 = matchedCache.cacheContent) ? void 0 : _matchedCache_cacheContent1.yamlWorkflow;
252
252
  return this.runYaml(yaml);
253
253
  }
254
- const { output, executor } = await (isVlmUiTars ? this.taskExecutor.actionToGoal(taskPrompt, modelConfig) : this.taskExecutor.action(taskPrompt, modelConfig, this.opts.aiActionContext));
254
+ const { output, executor } = await this.taskExecutor.action(taskPrompt, modelConfig, this.opts.aiActionContext);
255
255
  if (this.taskCache && (null == output ? void 0 : output.yamlFlow) && false !== cacheable) {
256
256
  const yamlContent = {
257
257
  tasks: [
@@ -1 +1 @@
1
- {"version":3,"file":"agent/agent.mjs","sources":["webpack://@midscene/core/./src/agent/agent.ts"],"sourcesContent":["import {\n type AgentAssertOpt,\n type AgentDescribeElementAtPointResult,\n type AgentWaitForOpt,\n type DeviceAction,\n type ExecutionDump,\n type ExecutionRecorderItem,\n type ExecutionTask,\n type ExecutionTaskLog,\n type Executor,\n type GroupedActionDump,\n Insight,\n type InsightAction,\n type InsightExtractOption,\n type InsightExtractParam,\n type LocateOption,\n type LocateResultElement,\n type LocateValidatorResult,\n type LocatorValidatorOption,\n type MidsceneYamlScript,\n type OnTaskStartTip,\n type PlanningAction,\n type Rect,\n type ScrollParam,\n type TUserPrompt,\n type UIContext,\n} from '../index';\n\nimport yaml from 'js-yaml';\n\nimport {\n groupedActionDumpFileExt,\n reportHTMLContent,\n stringifyDumpData,\n writeLogFile,\n} from '@/utils';\nimport {\n ScriptPlayer,\n buildDetailedLocateParam,\n parseYamlScript,\n} from '../yaml/index';\n\nimport type { AbstractInterface } from '@/device';\nimport type { AgentOpt } from '@/types';\nimport {\n MIDSCENE_CACHE,\n ModelConfigManager,\n globalConfigManager,\n globalModelConfigManager,\n} from '@midscene/shared/env';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\n// import type { AndroidDeviceInputOpt } from '../device';\nimport { TaskCache } from './task-cache';\nimport { TaskExecutor, locatePlanForLocate } from './tasks';\nimport { locateParamStr, paramStr, taskTitleStr, typeStr } from './ui-utils';\nimport {\n commonContextParser,\n getReportFileName,\n parsePrompt,\n printReportMsg,\n} from './utils';\nimport { trimContextByViewport } from './utils';\n\nconst debug = getDebug('agent');\n\nconst distanceOfTwoPoints = (p1: [number, number], p2: [number, number]) => {\n const [x1, y1] = p1;\n const [x2, y2] = p2;\n return Math.round(Math.sqrt((x1 - x2) ** 2 + (y1 - y2) ** 2));\n};\n\nconst includedInRect = (point: [number, number], rect: Rect) => {\n const [x, y] = point;\n const { left, top, width, height } = rect;\n return x >= left && x <= left + width && y >= top && y <= top + height;\n};\n\nconst defaultInsightExtractOption: InsightExtractOption = {\n domIncluded: false,\n screenshotIncluded: true,\n};\n\nexport class Agent<\n InterfaceType extends AbstractInterface = AbstractInterface,\n> {\n interface: InterfaceType;\n\n insight: Insight;\n\n dump: GroupedActionDump;\n\n reportFile?: string | null;\n\n reportFileName?: string;\n\n taskExecutor: TaskExecutor;\n\n opts: AgentOpt;\n\n /**\n * If true, the agent will not perform any actions\n */\n dryMode = false;\n\n onTaskStartTip?: OnTaskStartTip;\n\n taskCache?: TaskCache;\n\n onDumpUpdate?: (dump: string) => void;\n\n destroyed = false;\n\n modelConfigManager: ModelConfigManager;\n\n /**\n * Frozen page context for consistent AI operations\n */\n private frozenUIContext?: UIContext;\n\n /**\n * Flag to track if VL model warning has been shown\n */\n private hasWarnedNonVLModel = false;\n\n // @deprecated use .interface instead\n get page() {\n return this.interface;\n }\n\n /**\n * Ensures VL model warning is shown once when needed\n */\n private ensureVLModelWarning() {\n if (\n !this.hasWarnedNonVLModel &&\n this.interface.interfaceType !== 'puppeteer' &&\n this.interface.interfaceType !== 'playwright' &&\n this.interface.interfaceType !== 'static' &&\n this.interface.interfaceType !== 'chrome-extension-proxy'\n ) {\n this.modelConfigManager.throwErrorIfNonVLModel();\n this.hasWarnedNonVLModel = true;\n }\n }\n\n constructor(interfaceInstance: InterfaceType, opts?: AgentOpt) {\n this.interface = interfaceInstance;\n this.opts = Object.assign(\n {\n generateReport: true,\n autoPrintReportMsg: true,\n groupName: 'Midscene Report',\n groupDescription: '',\n },\n opts || {},\n );\n\n if (opts?.modelConfig && typeof opts?.modelConfig !== 'function') {\n throw new Error(\n `opts.modelConfig must be one of function or undefined, but got ${typeof opts?.modelConfig}`,\n );\n }\n this.modelConfigManager = opts?.modelConfig\n ? new ModelConfigManager(opts.modelConfig)\n : globalModelConfigManager;\n\n this.onTaskStartTip = this.opts.onTaskStartTip;\n\n this.insight = new Insight(async (action: InsightAction) => {\n return this.getUIContext(action);\n });\n\n const useCache =\n typeof opts?.useCache === 'boolean'\n ? opts.useCache\n : globalConfigManager.getEnvConfigInBoolean(MIDSCENE_CACHE);\n\n if (opts?.cacheId) {\n this.taskCache = new TaskCache(\n opts.cacheId,\n useCache, // if we should use cache to match the element\n );\n }\n\n this.taskExecutor = new TaskExecutor(this.interface, this.insight, {\n taskCache: this.taskCache,\n onTaskStart: this.callbackOnTaskStartTip.bind(this),\n replanningCycleLimit: this.opts.replanningCycleLimit,\n });\n this.dump = this.resetDump();\n this.reportFileName =\n opts?.reportFileName ||\n getReportFileName(opts?.testId || this.interface.interfaceType || 'web');\n }\n\n async getActionSpace(): Promise<DeviceAction[]> {\n return this.interface.actionSpace();\n }\n\n async getUIContext(action?: InsightAction): Promise<UIContext> {\n // Check VL model configuration when UI context is first needed\n this.ensureVLModelWarning();\n\n // If page context is frozen, return the frozen context for all actions\n if (this.frozenUIContext) {\n debug('Using frozen page context for action:', action);\n return this.frozenUIContext;\n }\n\n if (this.interface.getContext) {\n debug('Using page.getContext for action:', action);\n return await this.interface.getContext();\n } else {\n debug('Using commonContextParser for action:', action);\n return await commonContextParser(this.interface, {\n uploadServerUrl: this.modelConfigManager.getUploadTestServerUrl(),\n });\n }\n }\n\n async _snapshotContext(): Promise<UIContext> {\n return await this.getUIContext('locate');\n }\n\n async setAIActionContext(prompt: string) {\n if (this.opts.aiActionContext) {\n console.warn(\n 'aiActionContext is already set, and it is called again, will override the previous setting',\n );\n }\n this.opts.aiActionContext = prompt;\n }\n\n resetDump() {\n this.dump = {\n groupName: this.opts.groupName!,\n groupDescription: this.opts.groupDescription,\n executions: [],\n modelBriefs: [],\n };\n\n return this.dump;\n }\n\n appendExecutionDump(execution: ExecutionDump) {\n // use trimContextByViewport to process execution\n const trimmedExecution = trimContextByViewport(execution);\n const currentDump = this.dump;\n currentDump.executions.push(trimmedExecution);\n }\n\n dumpDataString() {\n // update dump info\n this.dump.groupName = this.opts.groupName!;\n this.dump.groupDescription = this.opts.groupDescription;\n return stringifyDumpData(this.dump);\n }\n\n reportHTMLString() {\n return reportHTMLContent(this.dumpDataString());\n }\n\n writeOutActionDumps() {\n if (this.destroyed) {\n throw new Error(\n 'PageAgent has been destroyed. Cannot update report file.',\n );\n }\n const { generateReport, autoPrintReportMsg } = this.opts;\n this.reportFile = writeLogFile({\n fileName: this.reportFileName!,\n fileExt: groupedActionDumpFileExt,\n fileContent: this.dumpDataString(),\n type: 'dump',\n generateReport,\n });\n debug('writeOutActionDumps', this.reportFile);\n if (generateReport && autoPrintReportMsg && this.reportFile) {\n printReportMsg(this.reportFile);\n }\n }\n\n private async callbackOnTaskStartTip(task: ExecutionTask) {\n const param = paramStr(task);\n const tip = param ? `${typeStr(task)} - ${param}` : typeStr(task);\n\n if (this.onTaskStartTip) {\n await this.onTaskStartTip(tip);\n }\n }\n\n private async afterTaskRunning(executor: Executor, doNotThrowError = false) {\n this.appendExecutionDump(executor.dump());\n\n try {\n if (this.onDumpUpdate) {\n this.onDumpUpdate(this.dumpDataString());\n }\n } catch (error) {\n console.error('Error in onDumpUpdate', error);\n }\n\n this.writeOutActionDumps();\n\n if (executor.isInErrorState() && !doNotThrowError) {\n const errorTask = executor.latestErrorTask();\n throw new Error(`${errorTask?.errorMessage}\\n${errorTask?.errorStack}`, {\n cause: errorTask?.error,\n });\n }\n }\n\n async callActionInActionSpace<T = any>(\n type: string,\n opt?: T, // and all other action params\n ) {\n debug('callActionInActionSpace', type, ',', opt);\n\n const actionPlan: PlanningAction<T> = {\n type: type as any,\n param: (opt as any) || {},\n thought: '',\n };\n debug('actionPlan', actionPlan); // , ', in which the locateParam is', locateParam);\n\n const plans: PlanningAction[] = [actionPlan].filter(\n Boolean,\n ) as PlanningAction[];\n\n const title = taskTitleStr(\n type as any,\n locateParamStr((opt as any)?.locate || {}),\n );\n\n // assume all operation in action space is related to locating\n const modelConfig = this.modelConfigManager.getModelConfig('grounding');\n\n const { output, executor } = await this.taskExecutor.runPlans(\n title,\n plans,\n modelConfig,\n );\n await this.afterTaskRunning(executor);\n return output;\n }\n\n async aiTap(locatePrompt: TUserPrompt, opt?: LocateOption) {\n assert(locatePrompt, 'missing locate prompt for tap');\n\n const detailedLocateParam = buildDetailedLocateParam(locatePrompt, opt);\n\n return this.callActionInActionSpace('Tap', {\n locate: detailedLocateParam,\n });\n }\n\n async aiRightClick(locatePrompt: TUserPrompt, opt?: LocateOption) {\n assert(locatePrompt, 'missing locate prompt for right click');\n\n const detailedLocateParam = buildDetailedLocateParam(locatePrompt, opt);\n\n return this.callActionInActionSpace('RightClick', {\n locate: detailedLocateParam,\n });\n }\n\n async aiDoubleClick(locatePrompt: TUserPrompt, opt?: LocateOption) {\n assert(locatePrompt, 'missing locate prompt for double click');\n\n const detailedLocateParam = buildDetailedLocateParam(locatePrompt, opt);\n\n return this.callActionInActionSpace('DoubleClick', {\n locate: detailedLocateParam,\n });\n }\n\n async aiHover(locatePrompt: TUserPrompt, opt?: LocateOption) {\n assert(locatePrompt, 'missing locate prompt for hover');\n\n const detailedLocateParam = buildDetailedLocateParam(locatePrompt, opt);\n\n return this.callActionInActionSpace('Hover', {\n locate: detailedLocateParam,\n });\n }\n\n // New signature, always use locatePrompt as the first param\n async aiInput(\n locatePrompt: TUserPrompt,\n opt: LocateOption & { value: string } & { autoDismissKeyboard?: boolean },\n ): Promise<any>;\n\n // Legacy signature - deprecated\n /**\n * @deprecated Use aiInput(locatePrompt, opt) instead where opt contains the value\n */\n async aiInput(\n value: string,\n locatePrompt: TUserPrompt,\n opt?: LocateOption & { autoDismissKeyboard?: boolean }, // AndroidDeviceInputOpt &\n ): Promise<any>;\n\n // Implementation\n async aiInput(\n locatePromptOrValue: TUserPrompt | string,\n locatePromptOrOpt:\n | TUserPrompt\n | (LocateOption & { value: string } & { autoDismissKeyboard?: boolean }) // AndroidDeviceInputOpt &\n | undefined,\n optOrUndefined?: LocateOption, // AndroidDeviceInputOpt &\n ) {\n let value: string;\n let locatePrompt: TUserPrompt;\n let opt:\n | (LocateOption & { value: string } & { autoDismissKeyboard?: boolean }) // AndroidDeviceInputOpt &\n | undefined;\n\n // Check if using new signature (first param is locatePrompt, second has value)\n if (\n typeof locatePromptOrOpt === 'object' &&\n locatePromptOrOpt !== null &&\n 'value' in locatePromptOrOpt\n ) {\n // New signature: aiInput(locatePrompt, opt)\n locatePrompt = locatePromptOrValue as TUserPrompt;\n const optWithValue = locatePromptOrOpt as LocateOption & {\n // AndroidDeviceInputOpt &\n value: string;\n autoDismissKeyboard?: boolean;\n };\n value = optWithValue.value;\n opt = optWithValue;\n } else {\n // Legacy signature: aiInput(value, locatePrompt, opt)\n value = locatePromptOrValue as string;\n locatePrompt = locatePromptOrOpt as TUserPrompt;\n opt = {\n ...optOrUndefined,\n value,\n };\n }\n\n assert(\n typeof value === 'string',\n 'input value must be a string, use empty string if you want to clear the input',\n );\n assert(locatePrompt, 'missing locate prompt for input');\n\n const detailedLocateParam = buildDetailedLocateParam(locatePrompt, opt);\n\n return this.callActionInActionSpace('Input', {\n ...(opt || {}),\n locate: detailedLocateParam,\n });\n }\n\n // New signature\n async aiKeyboardPress(\n locatePrompt: TUserPrompt,\n opt: LocateOption & { keyName: string },\n ): Promise<any>;\n\n // Legacy signature - deprecated\n /**\n * @deprecated Use aiKeyboardPress(locatePrompt, opt) instead where opt contains the keyName\n */\n async aiKeyboardPress(\n keyName: string,\n locatePrompt?: TUserPrompt,\n opt?: LocateOption,\n ): Promise<any>;\n\n // Implementation\n async aiKeyboardPress(\n locatePromptOrKeyName: TUserPrompt | string,\n locatePromptOrOpt:\n | TUserPrompt\n | (LocateOption & { keyName: string })\n | undefined,\n optOrUndefined?: LocateOption,\n ) {\n let keyName: string;\n let locatePrompt: TUserPrompt | undefined;\n let opt: (LocateOption & { keyName: string }) | undefined;\n\n // Check if using new signature (first param is locatePrompt, second has keyName)\n if (\n typeof locatePromptOrOpt === 'object' &&\n locatePromptOrOpt !== null &&\n 'keyName' in locatePromptOrOpt\n ) {\n // New signature: aiKeyboardPress(locatePrompt, opt)\n locatePrompt = locatePromptOrKeyName as TUserPrompt;\n opt = locatePromptOrOpt as LocateOption & {\n keyName: string;\n };\n } else {\n // Legacy signature: aiKeyboardPress(keyName, locatePrompt, opt)\n keyName = locatePromptOrKeyName as string;\n locatePrompt = locatePromptOrOpt as TUserPrompt | undefined;\n opt = {\n ...(optOrUndefined || {}),\n keyName,\n };\n }\n\n assert(opt?.keyName, 'missing keyName for keyboard press');\n\n const detailedLocateParam = locatePrompt\n ? buildDetailedLocateParam(locatePrompt, opt)\n : undefined;\n\n return this.callActionInActionSpace('KeyboardPress', {\n ...(opt || {}),\n locate: detailedLocateParam,\n });\n }\n\n // New signature\n async aiScroll(\n locatePrompt: TUserPrompt | undefined,\n opt: LocateOption & ScrollParam,\n ): Promise<any>;\n\n // Legacy signature - deprecated\n /**\n * @deprecated Use aiScroll(locatePrompt, opt) instead where opt contains the scroll parameters\n */\n async aiScroll(\n scrollParam: ScrollParam,\n locatePrompt?: TUserPrompt,\n opt?: LocateOption,\n ): Promise<any>;\n\n // Implementation\n async aiScroll(\n locatePromptOrScrollParam: TUserPrompt | ScrollParam | undefined,\n locatePromptOrOpt: TUserPrompt | (LocateOption & ScrollParam) | undefined,\n optOrUndefined?: LocateOption,\n ) {\n let scrollParam: ScrollParam;\n let locatePrompt: TUserPrompt | undefined;\n let opt: LocateOption | undefined;\n\n // Check if using new signature (first param is locatePrompt, second has scroll params)\n if (\n typeof locatePromptOrOpt === 'object' &&\n ('direction' in locatePromptOrOpt ||\n 'scrollType' in locatePromptOrOpt ||\n 'distance' in locatePromptOrOpt)\n ) {\n // New signature: aiScroll(locatePrompt, opt)\n locatePrompt = locatePromptOrScrollParam as TUserPrompt;\n opt = locatePromptOrOpt as LocateOption & ScrollParam;\n } else {\n // Legacy signature: aiScroll(scrollParam, locatePrompt, opt)\n scrollParam = locatePromptOrScrollParam as ScrollParam;\n locatePrompt = locatePromptOrOpt as TUserPrompt | undefined;\n opt = {\n ...(optOrUndefined || {}),\n ...(scrollParam || {}),\n };\n }\n\n const detailedLocateParam = buildDetailedLocateParam(\n locatePrompt || '',\n opt,\n );\n\n return this.callActionInActionSpace('Scroll', {\n ...(opt || {}),\n locate: detailedLocateParam,\n });\n }\n\n async aiAction(\n taskPrompt: string,\n opt?: {\n cacheable?: boolean;\n },\n ) {\n const modelConfig = this.modelConfigManager.getModelConfig('planning');\n\n const cacheable = opt?.cacheable;\n // if vlm-ui-tars, plan cache is not used\n const isVlmUiTars = modelConfig.vlMode === 'vlm-ui-tars';\n const matchedCache =\n isVlmUiTars || cacheable === false\n ? undefined\n : this.taskCache?.matchPlanCache(taskPrompt);\n if (matchedCache && this.taskCache?.isCacheResultUsed) {\n // log into report file\n const { executor } = await this.taskExecutor.loadYamlFlowAsPlanning(\n taskPrompt,\n matchedCache.cacheContent?.yamlWorkflow,\n );\n\n await this.afterTaskRunning(executor);\n\n debug('matched cache, will call .runYaml to run the action');\n const yaml = matchedCache.cacheContent?.yamlWorkflow;\n return this.runYaml(yaml);\n }\n\n const { output, executor } = await (isVlmUiTars\n ? this.taskExecutor.actionToGoal(taskPrompt, modelConfig)\n : this.taskExecutor.action(\n taskPrompt,\n modelConfig,\n this.opts.aiActionContext,\n ));\n\n // update cache\n if (this.taskCache && output?.yamlFlow && cacheable !== false) {\n const yamlContent: MidsceneYamlScript = {\n tasks: [\n {\n name: taskPrompt,\n flow: output.yamlFlow,\n },\n ],\n };\n const yamlFlowStr = yaml.dump(yamlContent);\n this.taskCache.updateOrAppendCacheRecord(\n {\n type: 'plan',\n prompt: taskPrompt,\n yamlWorkflow: yamlFlowStr,\n },\n matchedCache,\n );\n }\n\n await this.afterTaskRunning(executor);\n return output;\n }\n\n async aiQuery<ReturnType = any>(\n demand: InsightExtractParam,\n opt: InsightExtractOption = defaultInsightExtractOption,\n ): Promise<ReturnType> {\n const modelConfig = this.modelConfigManager.getModelConfig('VQA');\n const { output, executor } =\n await this.taskExecutor.createTypeQueryExecution(\n 'Query',\n demand,\n modelConfig,\n opt,\n );\n await this.afterTaskRunning(executor);\n return output as ReturnType;\n }\n\n async aiBoolean(\n prompt: TUserPrompt,\n opt: InsightExtractOption = defaultInsightExtractOption,\n ): Promise<boolean> {\n const modelConfig = this.modelConfigManager.getModelConfig('VQA');\n\n const { textPrompt, multimodalPrompt } = parsePrompt(prompt);\n const { output, executor } =\n await this.taskExecutor.createTypeQueryExecution(\n 'Boolean',\n textPrompt,\n modelConfig,\n opt,\n multimodalPrompt,\n );\n await this.afterTaskRunning(executor);\n return output as boolean;\n }\n\n async aiNumber(\n prompt: TUserPrompt,\n opt: InsightExtractOption = defaultInsightExtractOption,\n ): Promise<number> {\n const modelConfig = this.modelConfigManager.getModelConfig('VQA');\n\n const { textPrompt, multimodalPrompt } = parsePrompt(prompt);\n const { output, executor } =\n await this.taskExecutor.createTypeQueryExecution(\n 'Number',\n textPrompt,\n modelConfig,\n opt,\n multimodalPrompt,\n );\n await this.afterTaskRunning(executor);\n return output as number;\n }\n\n async aiString(\n prompt: TUserPrompt,\n opt: InsightExtractOption = defaultInsightExtractOption,\n ): Promise<string> {\n const modelConfig = this.modelConfigManager.getModelConfig('VQA');\n\n const { textPrompt, multimodalPrompt } = parsePrompt(prompt);\n const { output, executor } =\n await this.taskExecutor.createTypeQueryExecution(\n 'String',\n textPrompt,\n modelConfig,\n opt,\n multimodalPrompt,\n );\n await this.afterTaskRunning(executor);\n return output as string;\n }\n\n async aiAsk(\n prompt: TUserPrompt,\n opt: InsightExtractOption = defaultInsightExtractOption,\n ): Promise<string> {\n return this.aiString(prompt, opt);\n }\n\n async describeElementAtPoint(\n center: [number, number],\n opt?: {\n verifyPrompt?: boolean;\n retryLimit?: number;\n deepThink?: boolean;\n } & LocatorValidatorOption,\n ): Promise<AgentDescribeElementAtPointResult> {\n const { verifyPrompt = true, retryLimit = 3 } = opt || {};\n\n let success = false;\n let retryCount = 0;\n let resultPrompt = '';\n let deepThink = opt?.deepThink || false;\n let verifyResult: LocateValidatorResult | undefined;\n\n while (!success && retryCount < retryLimit) {\n if (retryCount >= 2) {\n deepThink = true;\n }\n debug(\n 'aiDescribe',\n center,\n 'verifyPrompt',\n verifyPrompt,\n 'retryCount',\n retryCount,\n 'deepThink',\n deepThink,\n );\n // use same intent as aiLocate\n const modelConfig = this.modelConfigManager.getModelConfig('grounding');\n\n const text = await this.insight.describe(center, modelConfig, {\n deepThink,\n });\n debug('aiDescribe text', text);\n assert(text.description, `failed to describe element at [${center}]`);\n resultPrompt = text.description;\n\n verifyResult = await this.verifyLocator(\n resultPrompt,\n deepThink ? { deepThink: true } : undefined,\n center,\n opt,\n );\n if (verifyResult.pass) {\n success = true;\n } else {\n retryCount++;\n }\n }\n\n return {\n prompt: resultPrompt,\n deepThink,\n verifyResult,\n };\n }\n\n async verifyLocator(\n prompt: string,\n locateOpt: LocateOption | undefined,\n expectCenter: [number, number],\n verifyLocateOption?: LocatorValidatorOption,\n ): Promise<LocateValidatorResult> {\n debug('verifyLocator', prompt, locateOpt, expectCenter, verifyLocateOption);\n\n const { center: verifyCenter, rect: verifyRect } = await this.aiLocate(\n prompt,\n locateOpt,\n );\n const distance = distanceOfTwoPoints(expectCenter, verifyCenter);\n const included = includedInRect(expectCenter, verifyRect);\n const pass =\n distance <= (verifyLocateOption?.centerDistanceThreshold || 20) ||\n included;\n const verifyResult = {\n pass,\n rect: verifyRect,\n center: verifyCenter,\n centerDistance: distance,\n };\n debug('aiDescribe verifyResult', verifyResult);\n return verifyResult;\n }\n\n async aiLocate(prompt: TUserPrompt, opt?: LocateOption) {\n const locateParam = buildDetailedLocateParam(prompt, opt);\n assert(locateParam, 'cannot get locate param for aiLocate');\n const locatePlan = locatePlanForLocate(locateParam);\n const plans = [locatePlan];\n const modelConfig = this.modelConfigManager.getModelConfig('grounding');\n\n const { executor, output } = await this.taskExecutor.runPlans(\n taskTitleStr('Locate', locateParamStr(locateParam)),\n plans,\n modelConfig,\n );\n await this.afterTaskRunning(executor);\n\n const { element } = output;\n\n return {\n rect: element?.rect,\n center: element?.center,\n scale: (await this.interface.size()).dpr,\n } as Pick<LocateResultElement, 'rect' | 'center'> & {\n scale: number;\n };\n }\n\n async aiAssert(\n assertion: TUserPrompt,\n msg?: string,\n opt?: AgentAssertOpt & InsightExtractOption,\n ) {\n const modelConfig = this.modelConfigManager.getModelConfig('VQA');\n\n const insightOpt: InsightExtractOption = {\n domIncluded: opt?.domIncluded ?? defaultInsightExtractOption.domIncluded,\n screenshotIncluded:\n opt?.screenshotIncluded ??\n defaultInsightExtractOption.screenshotIncluded,\n returnThought: opt?.returnThought ?? true,\n isWaitForAssert: opt?.isWaitForAssert,\n doNotThrowError: opt?.doNotThrowError,\n };\n\n const { output, executor, thought } = await this.taskExecutor.assert(\n assertion,\n modelConfig,\n insightOpt,\n );\n await this.afterTaskRunning(executor, true);\n\n const message = output\n ? undefined\n : `Assertion failed: ${msg || (typeof assertion === 'string' ? assertion : assertion.prompt)}\\nReason: ${\n thought || executor.latestErrorTask()?.error || '(no_reason)'\n }`;\n\n if (opt?.keepRawResponse) {\n return {\n pass: output,\n thought,\n message,\n };\n }\n\n if (!output) {\n throw new Error(message);\n }\n }\n\n async aiWaitFor(assertion: TUserPrompt, opt?: AgentWaitForOpt) {\n const modelConfig = this.modelConfigManager.getModelConfig('default');\n const { executor } = await this.taskExecutor.waitFor(\n assertion,\n {\n timeoutMs: opt?.timeoutMs || 15 * 1000,\n checkIntervalMs: opt?.checkIntervalMs || 3 * 1000,\n },\n modelConfig,\n );\n await this.afterTaskRunning(executor, true);\n\n if (executor.isInErrorState()) {\n const errorTask = executor.latestErrorTask();\n throw new Error(`${errorTask?.error}\\n${errorTask?.errorStack}`);\n }\n }\n\n async ai(taskPrompt: string, type = 'action') {\n if (type === 'action') {\n return this.aiAction(taskPrompt);\n }\n if (type === 'query') {\n return this.aiQuery(taskPrompt);\n }\n\n if (type === 'assert') {\n return this.aiAssert(taskPrompt);\n }\n\n if (type === 'tap') {\n return this.aiTap(taskPrompt);\n }\n\n if (type === 'rightClick') {\n return this.aiRightClick(taskPrompt);\n }\n\n if (type === 'doubleClick') {\n return this.aiDoubleClick(taskPrompt);\n }\n\n throw new Error(\n `Unknown type: ${type}, only support 'action', 'query', 'assert', 'tap', 'rightClick', 'doubleClick'`,\n );\n }\n\n async runYaml(yamlScriptContent: string): Promise<{\n result: Record<string, any>;\n }> {\n const script = parseYamlScript(yamlScriptContent, 'yaml');\n const player = new ScriptPlayer(script, async () => {\n return { agent: this, freeFn: [] };\n });\n await player.run();\n\n if (player.status === 'error') {\n const errors = player.taskStatusList\n .filter((task) => task.status === 'error')\n .map((task) => {\n return `task - ${task.name}: ${task.error?.message}`;\n })\n .join('\\n');\n throw new Error(`Error(s) occurred in running yaml script:\\n${errors}`);\n }\n\n return {\n result: player.result,\n };\n }\n\n async evaluateJavaScript(script: string) {\n assert(\n this.interface.evaluateJavaScript,\n 'evaluateJavaScript is not supported in current agent',\n );\n return this.interface.evaluateJavaScript(script);\n }\n\n async destroy() {\n await this.interface.destroy?.();\n this.resetDump(); // reset dump to release memory\n this.destroyed = true;\n }\n\n async logScreenshot(\n title?: string,\n opt?: {\n content: string;\n },\n ) {\n // 1. screenshot\n const base64 = await this.interface.screenshotBase64();\n const now = Date.now();\n // 2. build recorder\n const recorder: ExecutionRecorderItem[] = [\n {\n type: 'screenshot',\n ts: now,\n screenshot: base64,\n },\n ];\n // 3. build ExecutionTaskLog\n const task: ExecutionTaskLog = {\n type: 'Log',\n subType: 'Screenshot',\n status: 'finished',\n recorder,\n timing: {\n start: now,\n end: now,\n cost: 0,\n },\n param: {\n content: opt?.content || '',\n },\n executor: async () => {},\n };\n // 4. build ExecutionDump\n const executionDump: ExecutionDump = {\n sdkVersion: '',\n logTime: now,\n name: `Log - ${title || 'untitled'}`,\n description: opt?.content || '',\n tasks: [task],\n };\n // 5. append to execution dump\n this.appendExecutionDump(executionDump);\n\n try {\n this.onDumpUpdate?.(this.dumpDataString());\n } catch (error) {\n console.error('Failed to update dump', error);\n }\n\n this.writeOutActionDumps();\n }\n\n _unstableLogContent() {\n const { groupName, groupDescription, executions } = this.dump;\n const newExecutions = Array.isArray(executions)\n ? executions.map((execution: any) => {\n const { tasks, ...restExecution } = execution;\n let newTasks = tasks;\n if (Array.isArray(tasks)) {\n newTasks = tasks.map((task: any) => {\n // only remove uiContext and log from task\n const { uiContext, log, ...restTask } = task;\n return restTask;\n });\n }\n return { ...restExecution, ...(newTasks ? { tasks: newTasks } : {}) };\n })\n : [];\n return {\n groupName,\n groupDescription,\n executions: newExecutions,\n };\n }\n\n /**\n * Freezes the current page context to be reused in subsequent AI operations\n * This avoids recalculating page context for each operation\n */\n async freezePageContext(): Promise<void> {\n debug('Freezing page context');\n const context = await this._snapshotContext();\n // Mark the context as frozen\n context._isFrozen = true;\n this.frozenUIContext = context;\n debug('Page context frozen successfully');\n }\n\n /**\n * Unfreezes the page context, allowing AI operations to calculate context dynamically\n */\n async unfreezePageContext(): Promise<void> {\n debug('Unfreezing page context');\n this.frozenUIContext = undefined;\n debug('Page context unfrozen successfully');\n }\n}\n\nexport const createAgent = (\n interfaceInstance: AbstractInterface,\n opts?: AgentOpt,\n) => {\n return new Agent(interfaceInstance, opts);\n};\n"],"names":["debug","getDebug","distanceOfTwoPoints","p1","p2","x1","y1","x2","y2","Math","includedInRect","point","rect","x","y","left","top","width","height","defaultInsightExtractOption","Agent","action","commonContextParser","prompt","console","execution","trimmedExecution","trimContextByViewport","currentDump","stringifyDumpData","reportHTMLContent","Error","generateReport","autoPrintReportMsg","writeLogFile","groupedActionDumpFileExt","printReportMsg","task","param","paramStr","tip","typeStr","executor","doNotThrowError","error","errorTask","type","opt","actionPlan","plans","Boolean","title","taskTitleStr","locateParamStr","modelConfig","output","locatePrompt","assert","detailedLocateParam","buildDetailedLocateParam","locatePromptOrValue","locatePromptOrOpt","optOrUndefined","value","optWithValue","locatePromptOrKeyName","keyName","undefined","locatePromptOrScrollParam","scrollParam","taskPrompt","_this_taskCache","_this_taskCache1","cacheable","isVlmUiTars","matchedCache","_matchedCache_cacheContent","_matchedCache_cacheContent1","yaml","yamlContent","yamlFlowStr","demand","textPrompt","multimodalPrompt","parsePrompt","center","verifyPrompt","retryLimit","success","retryCount","resultPrompt","deepThink","verifyResult","text","locateOpt","expectCenter","verifyLocateOption","verifyCenter","verifyRect","distance","included","pass","locateParam","locatePlan","locatePlanForLocate","element","assertion","msg","_executor_latestErrorTask","insightOpt","thought","message","yamlScriptContent","script","parseYamlScript","player","ScriptPlayer","errors","_task_error","_this_interface","base64","now","Date","recorder","executionDump","_this","groupName","groupDescription","executions","newExecutions","Array","tasks","restExecution","newTasks","uiContext","log","restTask","context","interfaceInstance","opts","Object","ModelConfigManager","globalModelConfigManager","Insight","useCache","globalConfigManager","MIDSCENE_CACHE","TaskCache","TaskExecutor","getReportFileName","createAgent"],"mappings":";;;;;;;;;;;;;;;;;;;;;AAgEA,MAAMA,QAAQC,SAAS;AAEvB,MAAMC,sBAAsB,CAACC,IAAsBC;IACjD,MAAM,CAACC,IAAIC,GAAG,GAAGH;IACjB,MAAM,CAACI,IAAIC,GAAG,GAAGJ;IACjB,OAAOK,KAAK,KAAK,CAACA,KAAK,IAAI,CAAEJ,AAAAA,CAAAA,KAAKE,EAAC,KAAM,IAAKD,AAAAA,CAAAA,KAAKE,EAAC,KAAM;AAC5D;AAEA,MAAME,iBAAiB,CAACC,OAAyBC;IAC/C,MAAM,CAACC,GAAGC,EAAE,GAAGH;IACf,MAAM,EAAEI,IAAI,EAAEC,GAAG,EAAEC,KAAK,EAAEC,MAAM,EAAE,GAAGN;IACrC,OAAOC,KAAKE,QAAQF,KAAKE,OAAOE,SAASH,KAAKE,OAAOF,KAAKE,MAAME;AAClE;AAEA,MAAMC,8BAAoD;IACxD,aAAa;IACb,oBAAoB;AACtB;AAEO,MAAMC;IA2CX,IAAI,OAAO;QACT,OAAO,IAAI,CAAC,SAAS;IACvB;IAKQ,uBAAuB;QAC7B,IACE,CAAC,IAAI,CAAC,mBAAmB,IACzB,AAAiC,gBAAjC,IAAI,CAAC,SAAS,CAAC,aAAa,IAC5B,AAAiC,iBAAjC,IAAI,CAAC,SAAS,CAAC,aAAa,IAC5B,AAAiC,aAAjC,IAAI,CAAC,SAAS,CAAC,aAAa,IAC5B,AAAiC,6BAAjC,IAAI,CAAC,SAAS,CAAC,aAAa,EAC5B;YACA,IAAI,CAAC,kBAAkB,CAAC,sBAAsB;YAC9C,IAAI,CAAC,mBAAmB,GAAG;QAC7B;IACF;IAoDA,MAAM,iBAA0C;QAC9C,OAAO,IAAI,CAAC,SAAS,CAAC,WAAW;IACnC;IAEA,MAAM,aAAaC,MAAsB,EAAsB;QAE7D,IAAI,CAAC,oBAAoB;QAGzB,IAAI,IAAI,CAAC,eAAe,EAAE;YACxBrB,MAAM,yCAAyCqB;YAC/C,OAAO,IAAI,CAAC,eAAe;QAC7B;QAEA,IAAI,IAAI,CAAC,SAAS,CAAC,UAAU,EAAE;YAC7BrB,MAAM,qCAAqCqB;YAC3C,OAAO,MAAM,IAAI,CAAC,SAAS,CAAC,UAAU;QACxC;QACErB,MAAM,yCAAyCqB;QAC/C,OAAO,MAAMC,oBAAoB,IAAI,CAAC,SAAS,EAAE;YAC/C,iBAAiB,IAAI,CAAC,kBAAkB,CAAC,sBAAsB;QACjE;IAEJ;IAEA,MAAM,mBAAuC;QAC3C,OAAO,MAAM,IAAI,CAAC,YAAY,CAAC;IACjC;IAEA,MAAM,mBAAmBC,MAAc,EAAE;QACvC,IAAI,IAAI,CAAC,IAAI,CAAC,eAAe,EAC3BC,QAAQ,IAAI,CACV;QAGJ,IAAI,CAAC,IAAI,CAAC,eAAe,GAAGD;IAC9B;IAEA,YAAY;QACV,IAAI,CAAC,IAAI,GAAG;YACV,WAAW,IAAI,CAAC,IAAI,CAAC,SAAS;YAC9B,kBAAkB,IAAI,CAAC,IAAI,CAAC,gBAAgB;YAC5C,YAAY,EAAE;YACd,aAAa,EAAE;QACjB;QAEA,OAAO,IAAI,CAAC,IAAI;IAClB;IAEA,oBAAoBE,SAAwB,EAAE;QAE5C,MAAMC,mBAAmBC,sBAAsBF;QAC/C,MAAMG,cAAc,IAAI,CAAC,IAAI;QAC7BA,YAAY,UAAU,CAAC,IAAI,CAACF;IAC9B;IAEA,iBAAiB;QAEf,IAAI,CAAC,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS;QACzC,IAAI,CAAC,IAAI,CAAC,gBAAgB,GAAG,IAAI,CAAC,IAAI,CAAC,gBAAgB;QACvD,OAAOG,kBAAkB,IAAI,CAAC,IAAI;IACpC;IAEA,mBAAmB;QACjB,OAAOC,kBAAkB,IAAI,CAAC,cAAc;IAC9C;IAEA,sBAAsB;QACpB,IAAI,IAAI,CAAC,SAAS,EAChB,MAAM,IAAIC,MACR;QAGJ,MAAM,EAAEC,cAAc,EAAEC,kBAAkB,EAAE,GAAG,IAAI,CAAC,IAAI;QACxD,IAAI,CAAC,UAAU,GAAGC,aAAa;YAC7B,UAAU,IAAI,CAAC,cAAc;YAC7B,SAASC;YACT,aAAa,IAAI,CAAC,cAAc;YAChC,MAAM;YACNH;QACF;QACAhC,MAAM,uBAAuB,IAAI,CAAC,UAAU;QAC5C,IAAIgC,kBAAkBC,sBAAsB,IAAI,CAAC,UAAU,EACzDG,eAAe,IAAI,CAAC,UAAU;IAElC;IAEA,MAAc,uBAAuBC,IAAmB,EAAE;QACxD,MAAMC,QAAQC,SAASF;QACvB,MAAMG,MAAMF,QAAQ,GAAGG,QAAQJ,MAAM,GAAG,EAAEC,OAAO,GAAGG,QAAQJ;QAE5D,IAAI,IAAI,CAAC,cAAc,EACrB,MAAM,IAAI,CAAC,cAAc,CAACG;IAE9B;IAEA,MAAc,iBAAiBE,QAAkB,EAAEC,kBAAkB,KAAK,EAAE;QAC1E,IAAI,CAAC,mBAAmB,CAACD,SAAS,IAAI;QAEtC,IAAI;YACF,IAAI,IAAI,CAAC,YAAY,EACnB,IAAI,CAAC,YAAY,CAAC,IAAI,CAAC,cAAc;QAEzC,EAAE,OAAOE,OAAO;YACdpB,QAAQ,KAAK,CAAC,yBAAyBoB;QACzC;QAEA,IAAI,CAAC,mBAAmB;QAExB,IAAIF,SAAS,cAAc,MAAM,CAACC,iBAAiB;YACjD,MAAME,YAAYH,SAAS,eAAe;YAC1C,MAAM,IAAIX,MAAM,GAAGc,QAAAA,YAAAA,KAAAA,IAAAA,UAAW,YAAY,CAAC,EAAE,EAAEA,QAAAA,YAAAA,KAAAA,IAAAA,UAAW,UAAU,EAAE,EAAE;gBACtE,OAAOA,QAAAA,YAAAA,KAAAA,IAAAA,UAAW,KAAK;YACzB;QACF;IACF;IAEA,MAAM,wBACJC,IAAY,EACZC,GAAO,EACP;QACA/C,MAAM,2BAA2B8C,MAAM,KAAKC;QAE5C,MAAMC,aAAgC;YACpC,MAAMF;YACN,OAAQC,OAAe,CAAC;YACxB,SAAS;QACX;QACA/C,MAAM,cAAcgD;QAEpB,MAAMC,QAA0B;YAACD;SAAW,CAAC,MAAM,CACjDE;QAGF,MAAMC,QAAQC,aACZN,MACAO,eAAe,AAACN,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAa,MAAM,AAAD,KAAK,CAAC;QAI1C,MAAMO,cAAc,IAAI,CAAC,kBAAkB,CAAC,cAAc,CAAC;QAE3D,MAAM,EAAEC,MAAM,EAAEb,QAAQ,EAAE,GAAG,MAAM,IAAI,CAAC,YAAY,CAAC,QAAQ,CAC3DS,OACAF,OACAK;QAEF,MAAM,IAAI,CAAC,gBAAgB,CAACZ;QAC5B,OAAOa;IACT;IAEA,MAAM,MAAMC,YAAyB,EAAET,GAAkB,EAAE;QACzDU,OAAOD,cAAc;QAErB,MAAME,sBAAsBC,yBAAyBH,cAAcT;QAEnE,OAAO,IAAI,CAAC,uBAAuB,CAAC,OAAO;YACzC,QAAQW;QACV;IACF;IAEA,MAAM,aAAaF,YAAyB,EAAET,GAAkB,EAAE;QAChEU,OAAOD,cAAc;QAErB,MAAME,sBAAsBC,yBAAyBH,cAAcT;QAEnE,OAAO,IAAI,CAAC,uBAAuB,CAAC,cAAc;YAChD,QAAQW;QACV;IACF;IAEA,MAAM,cAAcF,YAAyB,EAAET,GAAkB,EAAE;QACjEU,OAAOD,cAAc;QAErB,MAAME,sBAAsBC,yBAAyBH,cAAcT;QAEnE,OAAO,IAAI,CAAC,uBAAuB,CAAC,eAAe;YACjD,QAAQW;QACV;IACF;IAEA,MAAM,QAAQF,YAAyB,EAAET,GAAkB,EAAE;QAC3DU,OAAOD,cAAc;QAErB,MAAME,sBAAsBC,yBAAyBH,cAAcT;QAEnE,OAAO,IAAI,CAAC,uBAAuB,CAAC,SAAS;YAC3C,QAAQW;QACV;IACF;IAmBA,MAAM,QACJE,mBAAyC,EACzCC,iBAGa,EACbC,cAA6B,EAC7B;QACA,IAAIC;QACJ,IAAIP;QACJ,IAAIT;QAKJ,IACE,AAA6B,YAA7B,OAAOc,qBACPA,AAAsB,SAAtBA,qBACA,WAAWA,mBACX;YAEAL,eAAeI;YACf,MAAMI,eAAeH;YAKrBE,QAAQC,aAAa,KAAK;YAC1BjB,MAAMiB;QACR,OAAO;YAELD,QAAQH;YACRJ,eAAeK;YACfd,MAAM;gBACJ,GAAGe,cAAc;gBACjBC;YACF;QACF;QAEAN,OACE,AAAiB,YAAjB,OAAOM,OACP;QAEFN,OAAOD,cAAc;QAErB,MAAME,sBAAsBC,yBAAyBH,cAAcT;QAEnE,OAAO,IAAI,CAAC,uBAAuB,CAAC,SAAS;YAC3C,GAAIA,OAAO,CAAC,CAAC;YACb,QAAQW;QACV;IACF;IAmBA,MAAM,gBACJO,qBAA2C,EAC3CJ,iBAGa,EACbC,cAA6B,EAC7B;QACA,IAAII;QACJ,IAAIV;QACJ,IAAIT;QAGJ,IACE,AAA6B,YAA7B,OAAOc,qBACPA,AAAsB,SAAtBA,qBACA,aAAaA,mBACb;YAEAL,eAAeS;YACflB,MAAMc;QAGR,OAAO;YAELK,UAAUD;YACVT,eAAeK;YACfd,MAAM;gBACJ,GAAIe,kBAAkB,CAAC,CAAC;gBACxBI;YACF;QACF;QAEAT,OAAOV,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,OAAO,EAAE;QAErB,MAAMW,sBAAsBF,eACxBG,yBAAyBH,cAAcT,OACvCoB;QAEJ,OAAO,IAAI,CAAC,uBAAuB,CAAC,iBAAiB;YACnD,GAAIpB,OAAO,CAAC,CAAC;YACb,QAAQW;QACV;IACF;IAmBA,MAAM,SACJU,yBAAgE,EAChEP,iBAAyE,EACzEC,cAA6B,EAC7B;QACA,IAAIO;QACJ,IAAIb;QACJ,IAAIT;QAGJ,IACE,AAA6B,YAA7B,OAAOc,qBACN,gBAAeA,qBACd,gBAAgBA,qBAChB,cAAcA,iBAAgB,GAChC;YAEAL,eAAeY;YACfrB,MAAMc;QACR,OAAO;YAELQ,cAAcD;YACdZ,eAAeK;YACfd,MAAM;gBACJ,GAAIe,kBAAkB,CAAC,CAAC;gBACxB,GAAIO,eAAe,CAAC,CAAC;YACvB;QACF;QAEA,MAAMX,sBAAsBC,yBAC1BH,gBAAgB,IAChBT;QAGF,OAAO,IAAI,CAAC,uBAAuB,CAAC,UAAU;YAC5C,GAAIA,OAAO,CAAC,CAAC;YACb,QAAQW;QACV;IACF;IAEA,MAAM,SACJY,UAAkB,EAClBvB,GAEC,EACD;YASMwB,iBACcC;QATpB,MAAMlB,cAAc,IAAI,CAAC,kBAAkB,CAAC,cAAc,CAAC;QAE3D,MAAMmB,YAAY1B,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,SAAS;QAEhC,MAAM2B,cAAcpB,AAAuB,kBAAvBA,YAAY,MAAM;QACtC,MAAMqB,eACJD,eAAeD,AAAc,UAAdA,YACXN,SAAAA,QACAI,CAAAA,kBAAAA,IAAI,CAAC,SAAS,AAAD,IAAbA,KAAAA,IAAAA,gBAAgB,cAAc,CAACD;QACrC,IAAIK,gBAAAA,SAAgBH,CAAAA,mBAAAA,IAAI,CAAC,SAAS,AAAD,IAAbA,KAAAA,IAAAA,iBAAgB,iBAAiB,AAAD,GAAG;gBAInDI,4BAMWC;YARb,MAAM,EAAEnC,QAAQ,EAAE,GAAG,MAAM,IAAI,CAAC,YAAY,CAAC,sBAAsB,CACjE4B,YAAAA,QACAM,CAAAA,6BAAAA,aAAa,YAAY,AAAD,IAAxBA,KAAAA,IAAAA,2BAA2B,YAAY;YAGzC,MAAM,IAAI,CAAC,gBAAgB,CAAClC;YAE5B1C,MAAM;YACN,MAAM8E,OAAO,QAAAD,CAAAA,8BAAAA,aAAa,YAAY,AAAD,IAAxBA,KAAAA,IAAAA,4BAA2B,YAAY;YACpD,OAAO,IAAI,CAAC,OAAO,CAACC;QACtB;QAEA,MAAM,EAAEvB,MAAM,EAAEb,QAAQ,EAAE,GAAG,MAAOgC,CAAAA,cAChC,IAAI,CAAC,YAAY,CAAC,YAAY,CAACJ,YAAYhB,eAC3C,IAAI,CAAC,YAAY,CAAC,MAAM,CACtBgB,YACAhB,aACA,IAAI,CAAC,IAAI,CAAC,eAAe;QAI/B,IAAI,IAAI,CAAC,SAAS,IAAIC,CAAAA,QAAAA,SAAAA,KAAAA,IAAAA,OAAQ,QAAQ,AAAD,KAAKkB,AAAc,UAAdA,WAAqB;YAC7D,MAAMM,cAAkC;gBACtC,OAAO;oBACL;wBACE,MAAMT;wBACN,MAAMf,OAAO,QAAQ;oBACvB;iBACD;YACH;YACA,MAAMyB,cAAcF,QAAAA,IAAS,CAACC;YAC9B,IAAI,CAAC,SAAS,CAAC,yBAAyB,CACtC;gBACE,MAAM;gBACN,QAAQT;gBACR,cAAcU;YAChB,GACAL;QAEJ;QAEA,MAAM,IAAI,CAAC,gBAAgB,CAACjC;QAC5B,OAAOa;IACT;IAEA,MAAM,QACJ0B,MAA2B,EAC3BlC,MAA4B5B,2BAA2B,EAClC;QACrB,MAAMmC,cAAc,IAAI,CAAC,kBAAkB,CAAC,cAAc,CAAC;QAC3D,MAAM,EAAEC,MAAM,EAAEb,QAAQ,EAAE,GACxB,MAAM,IAAI,CAAC,YAAY,CAAC,wBAAwB,CAC9C,SACAuC,QACA3B,aACAP;QAEJ,MAAM,IAAI,CAAC,gBAAgB,CAACL;QAC5B,OAAOa;IACT;IAEA,MAAM,UACJhC,MAAmB,EACnBwB,MAA4B5B,2BAA2B,EACrC;QAClB,MAAMmC,cAAc,IAAI,CAAC,kBAAkB,CAAC,cAAc,CAAC;QAE3D,MAAM,EAAE4B,UAAU,EAAEC,gBAAgB,EAAE,GAAGC,YAAY7D;QACrD,MAAM,EAAEgC,MAAM,EAAEb,QAAQ,EAAE,GACxB,MAAM,IAAI,CAAC,YAAY,CAAC,wBAAwB,CAC9C,WACAwC,YACA5B,aACAP,KACAoC;QAEJ,MAAM,IAAI,CAAC,gBAAgB,CAACzC;QAC5B,OAAOa;IACT;IAEA,MAAM,SACJhC,MAAmB,EACnBwB,MAA4B5B,2BAA2B,EACtC;QACjB,MAAMmC,cAAc,IAAI,CAAC,kBAAkB,CAAC,cAAc,CAAC;QAE3D,MAAM,EAAE4B,UAAU,EAAEC,gBAAgB,EAAE,GAAGC,YAAY7D;QACrD,MAAM,EAAEgC,MAAM,EAAEb,QAAQ,EAAE,GACxB,MAAM,IAAI,CAAC,YAAY,CAAC,wBAAwB,CAC9C,UACAwC,YACA5B,aACAP,KACAoC;QAEJ,MAAM,IAAI,CAAC,gBAAgB,CAACzC;QAC5B,OAAOa;IACT;IAEA,MAAM,SACJhC,MAAmB,EACnBwB,MAA4B5B,2BAA2B,EACtC;QACjB,MAAMmC,cAAc,IAAI,CAAC,kBAAkB,CAAC,cAAc,CAAC;QAE3D,MAAM,EAAE4B,UAAU,EAAEC,gBAAgB,EAAE,GAAGC,YAAY7D;QACrD,MAAM,EAAEgC,MAAM,EAAEb,QAAQ,EAAE,GACxB,MAAM,IAAI,CAAC,YAAY,CAAC,wBAAwB,CAC9C,UACAwC,YACA5B,aACAP,KACAoC;QAEJ,MAAM,IAAI,CAAC,gBAAgB,CAACzC;QAC5B,OAAOa;IACT;IAEA,MAAM,MACJhC,MAAmB,EACnBwB,MAA4B5B,2BAA2B,EACtC;QACjB,OAAO,IAAI,CAAC,QAAQ,CAACI,QAAQwB;IAC/B;IAEA,MAAM,uBACJsC,MAAwB,EACxBtC,GAI0B,EACkB;QAC5C,MAAM,EAAEuC,eAAe,IAAI,EAAEC,aAAa,CAAC,EAAE,GAAGxC,OAAO,CAAC;QAExD,IAAIyC,UAAU;QACd,IAAIC,aAAa;QACjB,IAAIC,eAAe;QACnB,IAAIC,YAAY5C,AAAAA,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,SAAS,AAAD,KAAK;QAClC,IAAI6C;QAEJ,MAAO,CAACJ,WAAWC,aAAaF,WAAY;YAC1C,IAAIE,cAAc,GAChBE,YAAY;YAEd3F,MACE,cACAqF,QACA,gBACAC,cACA,cACAG,YACA,aACAE;YAGF,MAAMrC,cAAc,IAAI,CAAC,kBAAkB,CAAC,cAAc,CAAC;YAE3D,MAAMuC,OAAO,MAAM,IAAI,CAAC,OAAO,CAAC,QAAQ,CAACR,QAAQ/B,aAAa;gBAC5DqC;YACF;YACA3F,MAAM,mBAAmB6F;YACzBpC,OAAOoC,KAAK,WAAW,EAAE,CAAC,+BAA+B,EAAER,OAAO,CAAC,CAAC;YACpEK,eAAeG,KAAK,WAAW;YAE/BD,eAAe,MAAM,IAAI,CAAC,aAAa,CACrCF,cACAC,YAAY;gBAAE,WAAW;YAAK,IAAIxB,QAClCkB,QACAtC;YAEF,IAAI6C,aAAa,IAAI,EACnBJ,UAAU;iBAEVC;QAEJ;QAEA,OAAO;YACL,QAAQC;YACRC;YACAC;QACF;IACF;IAEA,MAAM,cACJrE,MAAc,EACduE,SAAmC,EACnCC,YAA8B,EAC9BC,kBAA2C,EACX;QAChChG,MAAM,iBAAiBuB,QAAQuE,WAAWC,cAAcC;QAExD,MAAM,EAAE,QAAQC,YAAY,EAAE,MAAMC,UAAU,EAAE,GAAG,MAAM,IAAI,CAAC,QAAQ,CACpE3E,QACAuE;QAEF,MAAMK,WAAWjG,oBAAoB6F,cAAcE;QACnD,MAAMG,WAAW1F,eAAeqF,cAAcG;QAC9C,MAAMG,OACJF,YAAaH,CAAAA,CAAAA,QAAAA,qBAAAA,KAAAA,IAAAA,mBAAoB,uBAAuB,AAAD,KAAK,EAAC,KAC7DI;QACF,MAAMR,eAAe;YACnBS;YACA,MAAMH;YACN,QAAQD;YACR,gBAAgBE;QAClB;QACAnG,MAAM,2BAA2B4F;QACjC,OAAOA;IACT;IAEA,MAAM,SAASrE,MAAmB,EAAEwB,GAAkB,EAAE;QACtD,MAAMuD,cAAc3C,yBAAyBpC,QAAQwB;QACrDU,OAAO6C,aAAa;QACpB,MAAMC,aAAaC,oBAAoBF;QACvC,MAAMrD,QAAQ;YAACsD;SAAW;QAC1B,MAAMjD,cAAc,IAAI,CAAC,kBAAkB,CAAC,cAAc,CAAC;QAE3D,MAAM,EAAEZ,QAAQ,EAAEa,MAAM,EAAE,GAAG,MAAM,IAAI,CAAC,YAAY,CAAC,QAAQ,CAC3DH,aAAa,UAAUC,eAAeiD,eACtCrD,OACAK;QAEF,MAAM,IAAI,CAAC,gBAAgB,CAACZ;QAE5B,MAAM,EAAE+D,OAAO,EAAE,GAAGlD;QAEpB,OAAO;YACL,MAAMkD,QAAAA,UAAAA,KAAAA,IAAAA,QAAS,IAAI;YACnB,QAAQA,QAAAA,UAAAA,KAAAA,IAAAA,QAAS,MAAM;YACvB,OAAQ,OAAM,IAAI,CAAC,SAAS,CAAC,IAAI,EAAC,EAAG,GAAG;QAC1C;IAGF;IAEA,MAAM,SACJC,SAAsB,EACtBC,GAAY,EACZ5D,GAA2C,EAC3C;YAuBiB6D;QAtBjB,MAAMtD,cAAc,IAAI,CAAC,kBAAkB,CAAC,cAAc,CAAC;QAE3D,MAAMuD,aAAmC;YACvC,aAAa9D,AAAAA,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,WAAW,AAAD,KAAK5B,4BAA4B,WAAW;YACxE,oBACE4B,AAAAA,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,kBAAkB,AAAD,KACtB5B,4BAA4B,kBAAkB;YAChD,eAAe4B,AAAAA,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,aAAa,AAAD,KAAK;YACrC,iBAAiBA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,eAAe;YACrC,iBAAiBA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,eAAe;QACvC;QAEA,MAAM,EAAEQ,MAAM,EAAEb,QAAQ,EAAEoE,OAAO,EAAE,GAAG,MAAM,IAAI,CAAC,YAAY,CAAC,MAAM,CAClEJ,WACApD,aACAuD;QAEF,MAAM,IAAI,CAAC,gBAAgB,CAACnE,UAAU;QAEtC,MAAMqE,UAAUxD,SACZY,SACA,CAAC,kBAAkB,EAAEwC,OAAQ,CAAqB,YAArB,OAAOD,YAAyBA,YAAYA,UAAU,MAAK,EAAG,UAAU,EACnGI,WAAAA,SAAWF,CAAAA,4BAAAA,SAAS,eAAe,EAAC,IAAzBA,KAAAA,IAAAA,0BAA4B,KAAK,AAAD,KAAK,eAChD;QAEN,IAAI7D,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,eAAe,EACtB,OAAO;YACL,MAAMQ;YACNuD;YACAC;QACF;QAGF,IAAI,CAACxD,QACH,MAAM,IAAIxB,MAAMgF;IAEpB;IAEA,MAAM,UAAUL,SAAsB,EAAE3D,GAAqB,EAAE;QAC7D,MAAMO,cAAc,IAAI,CAAC,kBAAkB,CAAC,cAAc,CAAC;QAC3D,MAAM,EAAEZ,QAAQ,EAAE,GAAG,MAAM,IAAI,CAAC,YAAY,CAAC,OAAO,CAClDgE,WACA;YACE,WAAW3D,AAAAA,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,SAAS,AAAD,KAAK;YAC7B,iBAAiBA,AAAAA,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,eAAe,AAAD,KAAK;QAC3C,GACAO;QAEF,MAAM,IAAI,CAAC,gBAAgB,CAACZ,UAAU;QAEtC,IAAIA,SAAS,cAAc,IAAI;YAC7B,MAAMG,YAAYH,SAAS,eAAe;YAC1C,MAAM,IAAIX,MAAM,GAAGc,QAAAA,YAAAA,KAAAA,IAAAA,UAAW,KAAK,CAAC,EAAE,EAAEA,QAAAA,YAAAA,KAAAA,IAAAA,UAAW,UAAU,EAAE;QACjE;IACF;IAEA,MAAM,GAAGyB,UAAkB,EAAExB,OAAO,QAAQ,EAAE;QAC5C,IAAIA,AAAS,aAATA,MACF,OAAO,IAAI,CAAC,QAAQ,CAACwB;QAEvB,IAAIxB,AAAS,YAATA,MACF,OAAO,IAAI,CAAC,OAAO,CAACwB;QAGtB,IAAIxB,AAAS,aAATA,MACF,OAAO,IAAI,CAAC,QAAQ,CAACwB;QAGvB,IAAIxB,AAAS,UAATA,MACF,OAAO,IAAI,CAAC,KAAK,CAACwB;QAGpB,IAAIxB,AAAS,iBAATA,MACF,OAAO,IAAI,CAAC,YAAY,CAACwB;QAG3B,IAAIxB,AAAS,kBAATA,MACF,OAAO,IAAI,CAAC,aAAa,CAACwB;QAG5B,MAAM,IAAIvC,MACR,CAAC,cAAc,EAAEe,KAAK,8EAA8E,CAAC;IAEzG;IAEA,MAAM,QAAQkE,iBAAyB,EAEpC;QACD,MAAMC,SAASC,gBAAgBF,mBAAmB;QAClD,MAAMG,SAAS,IAAIC,aAAaH,QAAQ,UAC/B;gBAAE,OAAO,IAAI;gBAAE,QAAQ,EAAE;YAAC;QAEnC,MAAME,OAAO,GAAG;QAEhB,IAAIA,AAAkB,YAAlBA,OAAO,MAAM,EAAc;YAC7B,MAAME,SAASF,OAAO,cAAc,CACjC,MAAM,CAAC,CAAC9E,OAASA,AAAgB,YAAhBA,KAAK,MAAM,EAC5B,GAAG,CAAC,CAACA;oBAC2BiF;gBAA/B,OAAO,CAAC,OAAO,EAAEjF,KAAK,IAAI,CAAC,EAAE,EAAE,QAAAiF,CAAAA,cAAAA,KAAK,KAAK,AAAD,IAATA,KAAAA,IAAAA,YAAY,OAAO,EAAE;YACtD,GACC,IAAI,CAAC;YACR,MAAM,IAAIvF,MAAM,CAAC,2CAA2C,EAAEsF,QAAQ;QACxE;QAEA,OAAO;YACL,QAAQF,OAAO,MAAM;QACvB;IACF;IAEA,MAAM,mBAAmBF,MAAc,EAAE;QACvCxD,OACE,IAAI,CAAC,SAAS,CAAC,kBAAkB,EACjC;QAEF,OAAO,IAAI,CAAC,SAAS,CAAC,kBAAkB,CAACwD;IAC3C;IAEA,MAAM,UAAU;YACRM,yBAAAA;QAAN,eAAMA,CAAAA,0BAAAA,AAAAA,CAAAA,kBAAAA,IAAI,CAAC,SAAS,AAAD,EAAE,OAAO,AAAD,IAArBA,KAAAA,IAAAA,wBAAAA,IAAAA,CAAAA,gBAAAA;QACN,IAAI,CAAC,SAAS;QACd,IAAI,CAAC,SAAS,GAAG;IACnB;IAEA,MAAM,cACJpE,KAAc,EACdJ,GAEC,EACD;QAEA,MAAMyE,SAAS,MAAM,IAAI,CAAC,SAAS,CAAC,gBAAgB;QACpD,MAAMC,MAAMC,KAAK,GAAG;QAEpB,MAAMC,WAAoC;YACxC;gBACE,MAAM;gBACN,IAAIF;gBACJ,YAAYD;YACd;SACD;QAED,MAAMnF,OAAyB;YAC7B,MAAM;YACN,SAAS;YACT,QAAQ;YACRsF;YACA,QAAQ;gBACN,OAAOF;gBACP,KAAKA;gBACL,MAAM;YACR;YACA,OAAO;gBACL,SAAS1E,AAAAA,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,OAAO,AAAD,KAAK;YAC3B;YACA,UAAU,WAAa;QACzB;QAEA,MAAM6E,gBAA+B;YACnC,YAAY;YACZ,SAASH;YACT,MAAM,CAAC,MAAM,EAAEtE,SAAS,YAAY;YACpC,aAAaJ,AAAAA,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,OAAO,AAAD,KAAK;YAC7B,OAAO;gBAACV;aAAK;QACf;QAEA,IAAI,CAAC,mBAAmB,CAACuF;QAEzB,IAAI;gBACFC,oBAAAA;oBAAAA,CAAAA,qBAAAA,AAAAA,CAAAA,QAAAA,IAAI,AAAD,EAAE,YAAY,AAAD,KAAhBA,mBAAAA,IAAAA,CAAAA,OAAoB,IAAI,CAAC,cAAc;QACzC,EAAE,OAAOjF,OAAO;YACdpB,QAAQ,KAAK,CAAC,yBAAyBoB;QACzC;QAEA,IAAI,CAAC,mBAAmB;IAC1B;IAEA,sBAAsB;QACpB,MAAM,EAAEkF,SAAS,EAAEC,gBAAgB,EAAEC,UAAU,EAAE,GAAG,IAAI,CAAC,IAAI;QAC7D,MAAMC,gBAAgBC,MAAM,OAAO,CAACF,cAChCA,WAAW,GAAG,CAAC,CAACvG;YACd,MAAM,EAAE0G,KAAK,EAAE,GAAGC,eAAe,GAAG3G;YACpC,IAAI4G,WAAWF;YACf,IAAID,MAAM,OAAO,CAACC,QAChBE,WAAWF,MAAM,GAAG,CAAC,CAAC9F;gBAEpB,MAAM,EAAEiG,SAAS,EAAEC,GAAG,EAAE,GAAGC,UAAU,GAAGnG;gBACxC,OAAOmG;YACT;YAEF,OAAO;gBAAE,GAAGJ,aAAa;gBAAE,GAAIC,WAAW;oBAAE,OAAOA;gBAAS,IAAI,CAAC,CAAC;YAAE;QACtE,KACA,EAAE;QACN,OAAO;YACLP;YACAC;YACA,YAAYE;QACd;IACF;IAMA,MAAM,oBAAmC;QACvCjI,MAAM;QACN,MAAMyI,UAAU,MAAM,IAAI,CAAC,gBAAgB;QAE3CA,QAAQ,SAAS,GAAG;QACpB,IAAI,CAAC,eAAe,GAAGA;QACvBzI,MAAM;IACR;IAKA,MAAM,sBAAqC;QACzCA,MAAM;QACN,IAAI,CAAC,eAAe,GAAGmE;QACvBnE,MAAM;IACR;IA54BA,YAAY0I,iBAAgC,EAAEC,IAAe,CAAE;QA5D/D;QAEA;QAEA;QAEA;QAEA;QAEA;QAEA;QAKA,kCAAU;QAEV;QAEA;QAEA;QAEA,oCAAY;QAEZ;QAKA,uBAAQ,mBAAR;QAKA,uBAAQ,uBAAsB;QAwB5B,IAAI,CAAC,SAAS,GAAGD;QACjB,IAAI,CAAC,IAAI,GAAGE,OAAO,MAAM,CACvB;YACE,gBAAgB;YAChB,oBAAoB;YACpB,WAAW;YACX,kBAAkB;QACpB,GACAD,QAAQ,CAAC;QAGX,IAAIA,AAAAA,CAAAA,QAAAA,OAAAA,KAAAA,IAAAA,KAAM,WAAW,AAAD,KAAK,AAA6B,cAA7B,OAAOA,CAAAA,QAAAA,OAAAA,KAAAA,IAAAA,KAAM,WAAW,AAAD,GAC9C,MAAM,IAAI5G,MACR,CAAC,+DAA+D,EAAE,OAAO4G,CAAAA,QAAAA,OAAAA,KAAAA,IAAAA,KAAM,WAAW,AAAD,GAAG;QAGhG,IAAI,CAAC,kBAAkB,GAAGA,AAAAA,CAAAA,QAAAA,OAAAA,KAAAA,IAAAA,KAAM,WAAW,AAAD,IACtC,IAAIE,mBAAmBF,KAAK,WAAW,IACvCG;QAEJ,IAAI,CAAC,cAAc,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc;QAE9C,IAAI,CAAC,OAAO,GAAG,IAAIC,QAAQ,OAAO1H,SACzB,IAAI,CAAC,YAAY,CAACA;QAG3B,MAAM2H,WACJ,AAA0B,aAA1B,OAAOL,CAAAA,QAAAA,OAAAA,KAAAA,IAAAA,KAAM,QAAQ,AAAD,IAChBA,KAAK,QAAQ,GACbM,oBAAoB,qBAAqB,CAACC;QAEhD,IAAIP,QAAAA,OAAAA,KAAAA,IAAAA,KAAM,OAAO,EACf,IAAI,CAAC,SAAS,GAAG,IAAIQ,UACnBR,KAAK,OAAO,EACZK;QAIJ,IAAI,CAAC,YAAY,GAAG,IAAII,aAAa,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,OAAO,EAAE;YACjE,WAAW,IAAI,CAAC,SAAS;YACzB,aAAa,IAAI,CAAC,sBAAsB,CAAC,IAAI,CAAC,IAAI;YAClD,sBAAsB,IAAI,CAAC,IAAI,CAAC,oBAAoB;QACtD;QACA,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,SAAS;QAC1B,IAAI,CAAC,cAAc,GACjBT,AAAAA,CAAAA,QAAAA,OAAAA,KAAAA,IAAAA,KAAM,cAAc,AAAD,KACnBU,kBAAkBV,AAAAA,CAAAA,QAAAA,OAAAA,KAAAA,IAAAA,KAAM,MAAM,AAAD,KAAK,IAAI,CAAC,SAAS,CAAC,aAAa,IAAI;IACtE;AA61BF;AAEO,MAAMW,cAAc,CACzBZ,mBACAC,OAEO,IAAIvH,MAAMsH,mBAAmBC"}
1
+ {"version":3,"file":"agent/agent.mjs","sources":["webpack://@midscene/core/./src/agent/agent.ts"],"sourcesContent":["import {\n type AgentAssertOpt,\n type AgentDescribeElementAtPointResult,\n type AgentWaitForOpt,\n type DeviceAction,\n type ExecutionDump,\n type ExecutionRecorderItem,\n type ExecutionTask,\n type ExecutionTaskLog,\n type Executor,\n type GroupedActionDump,\n Insight,\n type InsightAction,\n type InsightExtractOption,\n type InsightExtractParam,\n type LocateOption,\n type LocateResultElement,\n type LocateValidatorResult,\n type LocatorValidatorOption,\n type MidsceneYamlScript,\n type OnTaskStartTip,\n type PlanningAction,\n type Rect,\n type ScrollParam,\n type TUserPrompt,\n type UIContext,\n} from '../index';\n\nimport yaml from 'js-yaml';\n\nimport {\n groupedActionDumpFileExt,\n reportHTMLContent,\n stringifyDumpData,\n writeLogFile,\n} from '@/utils';\nimport {\n ScriptPlayer,\n buildDetailedLocateParam,\n parseYamlScript,\n} from '../yaml/index';\n\nimport type { AbstractInterface } from '@/device';\nimport type { AgentOpt } from '@/types';\nimport {\n MIDSCENE_CACHE,\n ModelConfigManager,\n globalConfigManager,\n globalModelConfigManager,\n} from '@midscene/shared/env';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\n// import type { AndroidDeviceInputOpt } from '../device';\nimport { TaskCache } from './task-cache';\nimport { TaskExecutor, locatePlanForLocate } from './tasks';\nimport { locateParamStr, paramStr, taskTitleStr, typeStr } from './ui-utils';\nimport {\n commonContextParser,\n getReportFileName,\n parsePrompt,\n printReportMsg,\n} from './utils';\nimport { trimContextByViewport } from './utils';\n\nconst debug = getDebug('agent');\n\nconst distanceOfTwoPoints = (p1: [number, number], p2: [number, number]) => {\n const [x1, y1] = p1;\n const [x2, y2] = p2;\n return Math.round(Math.sqrt((x1 - x2) ** 2 + (y1 - y2) ** 2));\n};\n\nconst includedInRect = (point: [number, number], rect: Rect) => {\n const [x, y] = point;\n const { left, top, width, height } = rect;\n return x >= left && x <= left + width && y >= top && y <= top + height;\n};\n\nconst defaultInsightExtractOption: InsightExtractOption = {\n domIncluded: false,\n screenshotIncluded: true,\n};\n\nexport class Agent<\n InterfaceType extends AbstractInterface = AbstractInterface,\n> {\n interface: InterfaceType;\n\n insight: Insight;\n\n dump: GroupedActionDump;\n\n reportFile?: string | null;\n\n reportFileName?: string;\n\n taskExecutor: TaskExecutor;\n\n opts: AgentOpt;\n\n /**\n * If true, the agent will not perform any actions\n */\n dryMode = false;\n\n onTaskStartTip?: OnTaskStartTip;\n\n taskCache?: TaskCache;\n\n onDumpUpdate?: (dump: string) => void;\n\n destroyed = false;\n\n modelConfigManager: ModelConfigManager;\n\n /**\n * Frozen page context for consistent AI operations\n */\n private frozenUIContext?: UIContext;\n\n /**\n * Flag to track if VL model warning has been shown\n */\n private hasWarnedNonVLModel = false;\n\n // @deprecated use .interface instead\n get page() {\n return this.interface;\n }\n\n /**\n * Ensures VL model warning is shown once when needed\n */\n private ensureVLModelWarning() {\n if (\n !this.hasWarnedNonVLModel &&\n this.interface.interfaceType !== 'puppeteer' &&\n this.interface.interfaceType !== 'playwright' &&\n this.interface.interfaceType !== 'static' &&\n this.interface.interfaceType !== 'chrome-extension-proxy'\n ) {\n this.modelConfigManager.throwErrorIfNonVLModel();\n this.hasWarnedNonVLModel = true;\n }\n }\n\n constructor(interfaceInstance: InterfaceType, opts?: AgentOpt) {\n this.interface = interfaceInstance;\n this.opts = Object.assign(\n {\n generateReport: true,\n autoPrintReportMsg: true,\n groupName: 'Midscene Report',\n groupDescription: '',\n },\n opts || {},\n );\n\n if (opts?.modelConfig && typeof opts?.modelConfig !== 'function') {\n throw new Error(\n `opts.modelConfig must be one of function or undefined, but got ${typeof opts?.modelConfig}`,\n );\n }\n this.modelConfigManager = opts?.modelConfig\n ? new ModelConfigManager(opts.modelConfig)\n : globalModelConfigManager;\n\n this.onTaskStartTip = this.opts.onTaskStartTip;\n\n this.insight = new Insight(async (action: InsightAction) => {\n return this.getUIContext(action);\n });\n\n const useCache =\n typeof opts?.useCache === 'boolean'\n ? opts.useCache\n : globalConfigManager.getEnvConfigInBoolean(MIDSCENE_CACHE);\n\n if (opts?.cacheId) {\n this.taskCache = new TaskCache(\n opts.cacheId,\n useCache, // if we should use cache to match the element\n );\n }\n\n this.taskExecutor = new TaskExecutor(this.interface, this.insight, {\n taskCache: this.taskCache,\n onTaskStart: this.callbackOnTaskStartTip.bind(this),\n replanningCycleLimit: this.opts.replanningCycleLimit,\n });\n this.dump = this.resetDump();\n this.reportFileName =\n opts?.reportFileName ||\n getReportFileName(opts?.testId || this.interface.interfaceType || 'web');\n }\n\n async getActionSpace(): Promise<DeviceAction[]> {\n return this.interface.actionSpace();\n }\n\n async getUIContext(action?: InsightAction): Promise<UIContext> {\n // Check VL model configuration when UI context is first needed\n this.ensureVLModelWarning();\n\n // If page context is frozen, return the frozen context for all actions\n if (this.frozenUIContext) {\n debug('Using frozen page context for action:', action);\n return this.frozenUIContext;\n }\n\n if (this.interface.getContext) {\n debug('Using page.getContext for action:', action);\n return await this.interface.getContext();\n } else {\n debug('Using commonContextParser for action:', action);\n return await commonContextParser(this.interface, {\n uploadServerUrl: this.modelConfigManager.getUploadTestServerUrl(),\n });\n }\n }\n\n async _snapshotContext(): Promise<UIContext> {\n return await this.getUIContext('locate');\n }\n\n async setAIActionContext(prompt: string) {\n if (this.opts.aiActionContext) {\n console.warn(\n 'aiActionContext is already set, and it is called again, will override the previous setting',\n );\n }\n this.opts.aiActionContext = prompt;\n }\n\n resetDump() {\n this.dump = {\n groupName: this.opts.groupName!,\n groupDescription: this.opts.groupDescription,\n executions: [],\n modelBriefs: [],\n };\n\n return this.dump;\n }\n\n appendExecutionDump(execution: ExecutionDump) {\n // use trimContextByViewport to process execution\n const trimmedExecution = trimContextByViewport(execution);\n const currentDump = this.dump;\n currentDump.executions.push(trimmedExecution);\n }\n\n dumpDataString() {\n // update dump info\n this.dump.groupName = this.opts.groupName!;\n this.dump.groupDescription = this.opts.groupDescription;\n return stringifyDumpData(this.dump);\n }\n\n reportHTMLString() {\n return reportHTMLContent(this.dumpDataString());\n }\n\n writeOutActionDumps() {\n if (this.destroyed) {\n throw new Error(\n 'PageAgent has been destroyed. Cannot update report file.',\n );\n }\n const { generateReport, autoPrintReportMsg } = this.opts;\n this.reportFile = writeLogFile({\n fileName: this.reportFileName!,\n fileExt: groupedActionDumpFileExt,\n fileContent: this.dumpDataString(),\n type: 'dump',\n generateReport,\n });\n debug('writeOutActionDumps', this.reportFile);\n if (generateReport && autoPrintReportMsg && this.reportFile) {\n printReportMsg(this.reportFile);\n }\n }\n\n private async callbackOnTaskStartTip(task: ExecutionTask) {\n const param = paramStr(task);\n const tip = param ? `${typeStr(task)} - ${param}` : typeStr(task);\n\n if (this.onTaskStartTip) {\n await this.onTaskStartTip(tip);\n }\n }\n\n private async afterTaskRunning(executor: Executor, doNotThrowError = false) {\n this.appendExecutionDump(executor.dump());\n\n try {\n if (this.onDumpUpdate) {\n this.onDumpUpdate(this.dumpDataString());\n }\n } catch (error) {\n console.error('Error in onDumpUpdate', error);\n }\n\n this.writeOutActionDumps();\n\n if (executor.isInErrorState() && !doNotThrowError) {\n const errorTask = executor.latestErrorTask();\n throw new Error(`${errorTask?.errorMessage}\\n${errorTask?.errorStack}`, {\n cause: errorTask?.error,\n });\n }\n }\n\n async callActionInActionSpace<T = any>(\n type: string,\n opt?: T, // and all other action params\n ) {\n debug('callActionInActionSpace', type, ',', opt);\n\n const actionPlan: PlanningAction<T> = {\n type: type as any,\n param: (opt as any) || {},\n thought: '',\n };\n debug('actionPlan', actionPlan); // , ', in which the locateParam is', locateParam);\n\n const plans: PlanningAction[] = [actionPlan].filter(\n Boolean,\n ) as PlanningAction[];\n\n const title = taskTitleStr(\n type as any,\n locateParamStr((opt as any)?.locate || {}),\n );\n\n // assume all operation in action space is related to locating\n const modelConfig = this.modelConfigManager.getModelConfig('grounding');\n\n const { output, executor } = await this.taskExecutor.runPlans(\n title,\n plans,\n modelConfig,\n );\n await this.afterTaskRunning(executor);\n return output;\n }\n\n async aiTap(locatePrompt: TUserPrompt, opt?: LocateOption) {\n assert(locatePrompt, 'missing locate prompt for tap');\n\n const detailedLocateParam = buildDetailedLocateParam(locatePrompt, opt);\n\n return this.callActionInActionSpace('Tap', {\n locate: detailedLocateParam,\n });\n }\n\n async aiRightClick(locatePrompt: TUserPrompt, opt?: LocateOption) {\n assert(locatePrompt, 'missing locate prompt for right click');\n\n const detailedLocateParam = buildDetailedLocateParam(locatePrompt, opt);\n\n return this.callActionInActionSpace('RightClick', {\n locate: detailedLocateParam,\n });\n }\n\n async aiDoubleClick(locatePrompt: TUserPrompt, opt?: LocateOption) {\n assert(locatePrompt, 'missing locate prompt for double click');\n\n const detailedLocateParam = buildDetailedLocateParam(locatePrompt, opt);\n\n return this.callActionInActionSpace('DoubleClick', {\n locate: detailedLocateParam,\n });\n }\n\n async aiHover(locatePrompt: TUserPrompt, opt?: LocateOption) {\n assert(locatePrompt, 'missing locate prompt for hover');\n\n const detailedLocateParam = buildDetailedLocateParam(locatePrompt, opt);\n\n return this.callActionInActionSpace('Hover', {\n locate: detailedLocateParam,\n });\n }\n\n // New signature, always use locatePrompt as the first param\n async aiInput(\n locatePrompt: TUserPrompt,\n opt: LocateOption & { value: string } & { autoDismissKeyboard?: boolean },\n ): Promise<any>;\n\n // Legacy signature - deprecated\n /**\n * @deprecated Use aiInput(locatePrompt, opt) instead where opt contains the value\n */\n async aiInput(\n value: string,\n locatePrompt: TUserPrompt,\n opt?: LocateOption & { autoDismissKeyboard?: boolean }, // AndroidDeviceInputOpt &\n ): Promise<any>;\n\n // Implementation\n async aiInput(\n locatePromptOrValue: TUserPrompt | string,\n locatePromptOrOpt:\n | TUserPrompt\n | (LocateOption & { value: string } & { autoDismissKeyboard?: boolean }) // AndroidDeviceInputOpt &\n | undefined,\n optOrUndefined?: LocateOption, // AndroidDeviceInputOpt &\n ) {\n let value: string;\n let locatePrompt: TUserPrompt;\n let opt:\n | (LocateOption & { value: string } & { autoDismissKeyboard?: boolean }) // AndroidDeviceInputOpt &\n | undefined;\n\n // Check if using new signature (first param is locatePrompt, second has value)\n if (\n typeof locatePromptOrOpt === 'object' &&\n locatePromptOrOpt !== null &&\n 'value' in locatePromptOrOpt\n ) {\n // New signature: aiInput(locatePrompt, opt)\n locatePrompt = locatePromptOrValue as TUserPrompt;\n const optWithValue = locatePromptOrOpt as LocateOption & {\n // AndroidDeviceInputOpt &\n value: string;\n autoDismissKeyboard?: boolean;\n };\n value = optWithValue.value;\n opt = optWithValue;\n } else {\n // Legacy signature: aiInput(value, locatePrompt, opt)\n value = locatePromptOrValue as string;\n locatePrompt = locatePromptOrOpt as TUserPrompt;\n opt = {\n ...optOrUndefined,\n value,\n };\n }\n\n assert(\n typeof value === 'string',\n 'input value must be a string, use empty string if you want to clear the input',\n );\n assert(locatePrompt, 'missing locate prompt for input');\n\n const detailedLocateParam = buildDetailedLocateParam(locatePrompt, opt);\n\n return this.callActionInActionSpace('Input', {\n ...(opt || {}),\n locate: detailedLocateParam,\n });\n }\n\n // New signature\n async aiKeyboardPress(\n locatePrompt: TUserPrompt,\n opt: LocateOption & { keyName: string },\n ): Promise<any>;\n\n // Legacy signature - deprecated\n /**\n * @deprecated Use aiKeyboardPress(locatePrompt, opt) instead where opt contains the keyName\n */\n async aiKeyboardPress(\n keyName: string,\n locatePrompt?: TUserPrompt,\n opt?: LocateOption,\n ): Promise<any>;\n\n // Implementation\n async aiKeyboardPress(\n locatePromptOrKeyName: TUserPrompt | string,\n locatePromptOrOpt:\n | TUserPrompt\n | (LocateOption & { keyName: string })\n | undefined,\n optOrUndefined?: LocateOption,\n ) {\n let keyName: string;\n let locatePrompt: TUserPrompt | undefined;\n let opt: (LocateOption & { keyName: string }) | undefined;\n\n // Check if using new signature (first param is locatePrompt, second has keyName)\n if (\n typeof locatePromptOrOpt === 'object' &&\n locatePromptOrOpt !== null &&\n 'keyName' in locatePromptOrOpt\n ) {\n // New signature: aiKeyboardPress(locatePrompt, opt)\n locatePrompt = locatePromptOrKeyName as TUserPrompt;\n opt = locatePromptOrOpt as LocateOption & {\n keyName: string;\n };\n } else {\n // Legacy signature: aiKeyboardPress(keyName, locatePrompt, opt)\n keyName = locatePromptOrKeyName as string;\n locatePrompt = locatePromptOrOpt as TUserPrompt | undefined;\n opt = {\n ...(optOrUndefined || {}),\n keyName,\n };\n }\n\n assert(opt?.keyName, 'missing keyName for keyboard press');\n\n const detailedLocateParam = locatePrompt\n ? buildDetailedLocateParam(locatePrompt, opt)\n : undefined;\n\n return this.callActionInActionSpace('KeyboardPress', {\n ...(opt || {}),\n locate: detailedLocateParam,\n });\n }\n\n // New signature\n async aiScroll(\n locatePrompt: TUserPrompt | undefined,\n opt: LocateOption & ScrollParam,\n ): Promise<any>;\n\n // Legacy signature - deprecated\n /**\n * @deprecated Use aiScroll(locatePrompt, opt) instead where opt contains the scroll parameters\n */\n async aiScroll(\n scrollParam: ScrollParam,\n locatePrompt?: TUserPrompt,\n opt?: LocateOption,\n ): Promise<any>;\n\n // Implementation\n async aiScroll(\n locatePromptOrScrollParam: TUserPrompt | ScrollParam | undefined,\n locatePromptOrOpt: TUserPrompt | (LocateOption & ScrollParam) | undefined,\n optOrUndefined?: LocateOption,\n ) {\n let scrollParam: ScrollParam;\n let locatePrompt: TUserPrompt | undefined;\n let opt: LocateOption | undefined;\n\n // Check if using new signature (first param is locatePrompt, second has scroll params)\n if (\n typeof locatePromptOrOpt === 'object' &&\n ('direction' in locatePromptOrOpt ||\n 'scrollType' in locatePromptOrOpt ||\n 'distance' in locatePromptOrOpt)\n ) {\n // New signature: aiScroll(locatePrompt, opt)\n locatePrompt = locatePromptOrScrollParam as TUserPrompt;\n opt = locatePromptOrOpt as LocateOption & ScrollParam;\n } else {\n // Legacy signature: aiScroll(scrollParam, locatePrompt, opt)\n scrollParam = locatePromptOrScrollParam as ScrollParam;\n locatePrompt = locatePromptOrOpt as TUserPrompt | undefined;\n opt = {\n ...(optOrUndefined || {}),\n ...(scrollParam || {}),\n };\n }\n\n const detailedLocateParam = buildDetailedLocateParam(\n locatePrompt || '',\n opt,\n );\n\n return this.callActionInActionSpace('Scroll', {\n ...(opt || {}),\n locate: detailedLocateParam,\n });\n }\n\n async aiAction(\n taskPrompt: string,\n opt?: {\n cacheable?: boolean;\n },\n ) {\n const modelConfig = this.modelConfigManager.getModelConfig('planning');\n\n const cacheable = opt?.cacheable;\n // if vlm-ui-tars, plan cache is not used\n const isVlmUiTars = modelConfig.vlMode === 'vlm-ui-tars';\n const matchedCache =\n isVlmUiTars || cacheable === false\n ? undefined\n : this.taskCache?.matchPlanCache(taskPrompt);\n if (matchedCache && this.taskCache?.isCacheResultUsed) {\n // log into report file\n const { executor } = await this.taskExecutor.loadYamlFlowAsPlanning(\n taskPrompt,\n matchedCache.cacheContent?.yamlWorkflow,\n );\n\n await this.afterTaskRunning(executor);\n\n debug('matched cache, will call .runYaml to run the action');\n const yaml = matchedCache.cacheContent?.yamlWorkflow;\n return this.runYaml(yaml);\n }\n\n const { output, executor } = await this.taskExecutor.action(\n taskPrompt,\n modelConfig,\n this.opts.aiActionContext,\n );\n\n // update cache\n if (this.taskCache && output?.yamlFlow && cacheable !== false) {\n const yamlContent: MidsceneYamlScript = {\n tasks: [\n {\n name: taskPrompt,\n flow: output.yamlFlow,\n },\n ],\n };\n const yamlFlowStr = yaml.dump(yamlContent);\n this.taskCache.updateOrAppendCacheRecord(\n {\n type: 'plan',\n prompt: taskPrompt,\n yamlWorkflow: yamlFlowStr,\n },\n matchedCache,\n );\n }\n\n await this.afterTaskRunning(executor);\n return output;\n }\n\n async aiQuery<ReturnType = any>(\n demand: InsightExtractParam,\n opt: InsightExtractOption = defaultInsightExtractOption,\n ): Promise<ReturnType> {\n const modelConfig = this.modelConfigManager.getModelConfig('VQA');\n const { output, executor } =\n await this.taskExecutor.createTypeQueryExecution(\n 'Query',\n demand,\n modelConfig,\n opt,\n );\n await this.afterTaskRunning(executor);\n return output as ReturnType;\n }\n\n async aiBoolean(\n prompt: TUserPrompt,\n opt: InsightExtractOption = defaultInsightExtractOption,\n ): Promise<boolean> {\n const modelConfig = this.modelConfigManager.getModelConfig('VQA');\n\n const { textPrompt, multimodalPrompt } = parsePrompt(prompt);\n const { output, executor } =\n await this.taskExecutor.createTypeQueryExecution(\n 'Boolean',\n textPrompt,\n modelConfig,\n opt,\n multimodalPrompt,\n );\n await this.afterTaskRunning(executor);\n return output as boolean;\n }\n\n async aiNumber(\n prompt: TUserPrompt,\n opt: InsightExtractOption = defaultInsightExtractOption,\n ): Promise<number> {\n const modelConfig = this.modelConfigManager.getModelConfig('VQA');\n\n const { textPrompt, multimodalPrompt } = parsePrompt(prompt);\n const { output, executor } =\n await this.taskExecutor.createTypeQueryExecution(\n 'Number',\n textPrompt,\n modelConfig,\n opt,\n multimodalPrompt,\n );\n await this.afterTaskRunning(executor);\n return output as number;\n }\n\n async aiString(\n prompt: TUserPrompt,\n opt: InsightExtractOption = defaultInsightExtractOption,\n ): Promise<string> {\n const modelConfig = this.modelConfigManager.getModelConfig('VQA');\n\n const { textPrompt, multimodalPrompt } = parsePrompt(prompt);\n const { output, executor } =\n await this.taskExecutor.createTypeQueryExecution(\n 'String',\n textPrompt,\n modelConfig,\n opt,\n multimodalPrompt,\n );\n await this.afterTaskRunning(executor);\n return output as string;\n }\n\n async aiAsk(\n prompt: TUserPrompt,\n opt: InsightExtractOption = defaultInsightExtractOption,\n ): Promise<string> {\n return this.aiString(prompt, opt);\n }\n\n async describeElementAtPoint(\n center: [number, number],\n opt?: {\n verifyPrompt?: boolean;\n retryLimit?: number;\n deepThink?: boolean;\n } & LocatorValidatorOption,\n ): Promise<AgentDescribeElementAtPointResult> {\n const { verifyPrompt = true, retryLimit = 3 } = opt || {};\n\n let success = false;\n let retryCount = 0;\n let resultPrompt = '';\n let deepThink = opt?.deepThink || false;\n let verifyResult: LocateValidatorResult | undefined;\n\n while (!success && retryCount < retryLimit) {\n if (retryCount >= 2) {\n deepThink = true;\n }\n debug(\n 'aiDescribe',\n center,\n 'verifyPrompt',\n verifyPrompt,\n 'retryCount',\n retryCount,\n 'deepThink',\n deepThink,\n );\n // use same intent as aiLocate\n const modelConfig = this.modelConfigManager.getModelConfig('grounding');\n\n const text = await this.insight.describe(center, modelConfig, {\n deepThink,\n });\n debug('aiDescribe text', text);\n assert(text.description, `failed to describe element at [${center}]`);\n resultPrompt = text.description;\n\n verifyResult = await this.verifyLocator(\n resultPrompt,\n deepThink ? { deepThink: true } : undefined,\n center,\n opt,\n );\n if (verifyResult.pass) {\n success = true;\n } else {\n retryCount++;\n }\n }\n\n return {\n prompt: resultPrompt,\n deepThink,\n verifyResult,\n };\n }\n\n async verifyLocator(\n prompt: string,\n locateOpt: LocateOption | undefined,\n expectCenter: [number, number],\n verifyLocateOption?: LocatorValidatorOption,\n ): Promise<LocateValidatorResult> {\n debug('verifyLocator', prompt, locateOpt, expectCenter, verifyLocateOption);\n\n const { center: verifyCenter, rect: verifyRect } = await this.aiLocate(\n prompt,\n locateOpt,\n );\n const distance = distanceOfTwoPoints(expectCenter, verifyCenter);\n const included = includedInRect(expectCenter, verifyRect);\n const pass =\n distance <= (verifyLocateOption?.centerDistanceThreshold || 20) ||\n included;\n const verifyResult = {\n pass,\n rect: verifyRect,\n center: verifyCenter,\n centerDistance: distance,\n };\n debug('aiDescribe verifyResult', verifyResult);\n return verifyResult;\n }\n\n async aiLocate(prompt: TUserPrompt, opt?: LocateOption) {\n const locateParam = buildDetailedLocateParam(prompt, opt);\n assert(locateParam, 'cannot get locate param for aiLocate');\n const locatePlan = locatePlanForLocate(locateParam);\n const plans = [locatePlan];\n const modelConfig = this.modelConfigManager.getModelConfig('grounding');\n\n const { executor, output } = await this.taskExecutor.runPlans(\n taskTitleStr('Locate', locateParamStr(locateParam)),\n plans,\n modelConfig,\n );\n await this.afterTaskRunning(executor);\n\n const { element } = output;\n\n return {\n rect: element?.rect,\n center: element?.center,\n scale: (await this.interface.size()).dpr,\n } as Pick<LocateResultElement, 'rect' | 'center'> & {\n scale: number;\n };\n }\n\n async aiAssert(\n assertion: TUserPrompt,\n msg?: string,\n opt?: AgentAssertOpt & InsightExtractOption,\n ) {\n const modelConfig = this.modelConfigManager.getModelConfig('VQA');\n\n const insightOpt: InsightExtractOption = {\n domIncluded: opt?.domIncluded ?? defaultInsightExtractOption.domIncluded,\n screenshotIncluded:\n opt?.screenshotIncluded ??\n defaultInsightExtractOption.screenshotIncluded,\n returnThought: opt?.returnThought ?? true,\n isWaitForAssert: opt?.isWaitForAssert,\n doNotThrowError: opt?.doNotThrowError,\n };\n\n const { output, executor, thought } = await this.taskExecutor.assert(\n assertion,\n modelConfig,\n insightOpt,\n );\n await this.afterTaskRunning(executor, true);\n\n const message = output\n ? undefined\n : `Assertion failed: ${msg || (typeof assertion === 'string' ? assertion : assertion.prompt)}\\nReason: ${\n thought || executor.latestErrorTask()?.error || '(no_reason)'\n }`;\n\n if (opt?.keepRawResponse) {\n return {\n pass: output,\n thought,\n message,\n };\n }\n\n if (!output) {\n throw new Error(message);\n }\n }\n\n async aiWaitFor(assertion: TUserPrompt, opt?: AgentWaitForOpt) {\n const modelConfig = this.modelConfigManager.getModelConfig('default');\n const { executor } = await this.taskExecutor.waitFor(\n assertion,\n {\n timeoutMs: opt?.timeoutMs || 15 * 1000,\n checkIntervalMs: opt?.checkIntervalMs || 3 * 1000,\n },\n modelConfig,\n );\n await this.afterTaskRunning(executor, true);\n\n if (executor.isInErrorState()) {\n const errorTask = executor.latestErrorTask();\n throw new Error(`${errorTask?.error}\\n${errorTask?.errorStack}`);\n }\n }\n\n async ai(taskPrompt: string, type = 'action') {\n if (type === 'action') {\n return this.aiAction(taskPrompt);\n }\n if (type === 'query') {\n return this.aiQuery(taskPrompt);\n }\n\n if (type === 'assert') {\n return this.aiAssert(taskPrompt);\n }\n\n if (type === 'tap') {\n return this.aiTap(taskPrompt);\n }\n\n if (type === 'rightClick') {\n return this.aiRightClick(taskPrompt);\n }\n\n if (type === 'doubleClick') {\n return this.aiDoubleClick(taskPrompt);\n }\n\n throw new Error(\n `Unknown type: ${type}, only support 'action', 'query', 'assert', 'tap', 'rightClick', 'doubleClick'`,\n );\n }\n\n async runYaml(yamlScriptContent: string): Promise<{\n result: Record<string, any>;\n }> {\n const script = parseYamlScript(yamlScriptContent, 'yaml');\n const player = new ScriptPlayer(script, async () => {\n return { agent: this, freeFn: [] };\n });\n await player.run();\n\n if (player.status === 'error') {\n const errors = player.taskStatusList\n .filter((task) => task.status === 'error')\n .map((task) => {\n return `task - ${task.name}: ${task.error?.message}`;\n })\n .join('\\n');\n throw new Error(`Error(s) occurred in running yaml script:\\n${errors}`);\n }\n\n return {\n result: player.result,\n };\n }\n\n async evaluateJavaScript(script: string) {\n assert(\n this.interface.evaluateJavaScript,\n 'evaluateJavaScript is not supported in current agent',\n );\n return this.interface.evaluateJavaScript(script);\n }\n\n async destroy() {\n await this.interface.destroy?.();\n this.resetDump(); // reset dump to release memory\n this.destroyed = true;\n }\n\n async logScreenshot(\n title?: string,\n opt?: {\n content: string;\n },\n ) {\n // 1. screenshot\n const base64 = await this.interface.screenshotBase64();\n const now = Date.now();\n // 2. build recorder\n const recorder: ExecutionRecorderItem[] = [\n {\n type: 'screenshot',\n ts: now,\n screenshot: base64,\n },\n ];\n // 3. build ExecutionTaskLog\n const task: ExecutionTaskLog = {\n type: 'Log',\n subType: 'Screenshot',\n status: 'finished',\n recorder,\n timing: {\n start: now,\n end: now,\n cost: 0,\n },\n param: {\n content: opt?.content || '',\n },\n executor: async () => {},\n };\n // 4. build ExecutionDump\n const executionDump: ExecutionDump = {\n sdkVersion: '',\n logTime: now,\n name: `Log - ${title || 'untitled'}`,\n description: opt?.content || '',\n tasks: [task],\n };\n // 5. append to execution dump\n this.appendExecutionDump(executionDump);\n\n try {\n this.onDumpUpdate?.(this.dumpDataString());\n } catch (error) {\n console.error('Failed to update dump', error);\n }\n\n this.writeOutActionDumps();\n }\n\n _unstableLogContent() {\n const { groupName, groupDescription, executions } = this.dump;\n const newExecutions = Array.isArray(executions)\n ? executions.map((execution: any) => {\n const { tasks, ...restExecution } = execution;\n let newTasks = tasks;\n if (Array.isArray(tasks)) {\n newTasks = tasks.map((task: any) => {\n // only remove uiContext and log from task\n const { uiContext, log, ...restTask } = task;\n return restTask;\n });\n }\n return { ...restExecution, ...(newTasks ? { tasks: newTasks } : {}) };\n })\n : [];\n return {\n groupName,\n groupDescription,\n executions: newExecutions,\n };\n }\n\n /**\n * Freezes the current page context to be reused in subsequent AI operations\n * This avoids recalculating page context for each operation\n */\n async freezePageContext(): Promise<void> {\n debug('Freezing page context');\n const context = await this._snapshotContext();\n // Mark the context as frozen\n context._isFrozen = true;\n this.frozenUIContext = context;\n debug('Page context frozen successfully');\n }\n\n /**\n * Unfreezes the page context, allowing AI operations to calculate context dynamically\n */\n async unfreezePageContext(): Promise<void> {\n debug('Unfreezing page context');\n this.frozenUIContext = undefined;\n debug('Page context unfrozen successfully');\n }\n}\n\nexport const createAgent = (\n interfaceInstance: AbstractInterface,\n opts?: AgentOpt,\n) => {\n return new Agent(interfaceInstance, opts);\n};\n"],"names":["debug","getDebug","distanceOfTwoPoints","p1","p2","x1","y1","x2","y2","Math","includedInRect","point","rect","x","y","left","top","width","height","defaultInsightExtractOption","Agent","action","commonContextParser","prompt","console","execution","trimmedExecution","trimContextByViewport","currentDump","stringifyDumpData","reportHTMLContent","Error","generateReport","autoPrintReportMsg","writeLogFile","groupedActionDumpFileExt","printReportMsg","task","param","paramStr","tip","typeStr","executor","doNotThrowError","error","errorTask","type","opt","actionPlan","plans","Boolean","title","taskTitleStr","locateParamStr","modelConfig","output","locatePrompt","assert","detailedLocateParam","buildDetailedLocateParam","locatePromptOrValue","locatePromptOrOpt","optOrUndefined","value","optWithValue","locatePromptOrKeyName","keyName","undefined","locatePromptOrScrollParam","scrollParam","taskPrompt","_this_taskCache","_this_taskCache1","cacheable","isVlmUiTars","matchedCache","_matchedCache_cacheContent","_matchedCache_cacheContent1","yaml","yamlContent","yamlFlowStr","demand","textPrompt","multimodalPrompt","parsePrompt","center","verifyPrompt","retryLimit","success","retryCount","resultPrompt","deepThink","verifyResult","text","locateOpt","expectCenter","verifyLocateOption","verifyCenter","verifyRect","distance","included","pass","locateParam","locatePlan","locatePlanForLocate","element","assertion","msg","_executor_latestErrorTask","insightOpt","thought","message","yamlScriptContent","script","parseYamlScript","player","ScriptPlayer","errors","_task_error","_this_interface","base64","now","Date","recorder","executionDump","_this","groupName","groupDescription","executions","newExecutions","Array","tasks","restExecution","newTasks","uiContext","log","restTask","context","interfaceInstance","opts","Object","ModelConfigManager","globalModelConfigManager","Insight","useCache","globalConfigManager","MIDSCENE_CACHE","TaskCache","TaskExecutor","getReportFileName","createAgent"],"mappings":";;;;;;;;;;;;;;;;;;;;;AAgEA,MAAMA,QAAQC,SAAS;AAEvB,MAAMC,sBAAsB,CAACC,IAAsBC;IACjD,MAAM,CAACC,IAAIC,GAAG,GAAGH;IACjB,MAAM,CAACI,IAAIC,GAAG,GAAGJ;IACjB,OAAOK,KAAK,KAAK,CAACA,KAAK,IAAI,CAAEJ,AAAAA,CAAAA,KAAKE,EAAC,KAAM,IAAKD,AAAAA,CAAAA,KAAKE,EAAC,KAAM;AAC5D;AAEA,MAAME,iBAAiB,CAACC,OAAyBC;IAC/C,MAAM,CAACC,GAAGC,EAAE,GAAGH;IACf,MAAM,EAAEI,IAAI,EAAEC,GAAG,EAAEC,KAAK,EAAEC,MAAM,EAAE,GAAGN;IACrC,OAAOC,KAAKE,QAAQF,KAAKE,OAAOE,SAASH,KAAKE,OAAOF,KAAKE,MAAME;AAClE;AAEA,MAAMC,8BAAoD;IACxD,aAAa;IACb,oBAAoB;AACtB;AAEO,MAAMC;IA2CX,IAAI,OAAO;QACT,OAAO,IAAI,CAAC,SAAS;IACvB;IAKQ,uBAAuB;QAC7B,IACE,CAAC,IAAI,CAAC,mBAAmB,IACzB,AAAiC,gBAAjC,IAAI,CAAC,SAAS,CAAC,aAAa,IAC5B,AAAiC,iBAAjC,IAAI,CAAC,SAAS,CAAC,aAAa,IAC5B,AAAiC,aAAjC,IAAI,CAAC,SAAS,CAAC,aAAa,IAC5B,AAAiC,6BAAjC,IAAI,CAAC,SAAS,CAAC,aAAa,EAC5B;YACA,IAAI,CAAC,kBAAkB,CAAC,sBAAsB;YAC9C,IAAI,CAAC,mBAAmB,GAAG;QAC7B;IACF;IAoDA,MAAM,iBAA0C;QAC9C,OAAO,IAAI,CAAC,SAAS,CAAC,WAAW;IACnC;IAEA,MAAM,aAAaC,MAAsB,EAAsB;QAE7D,IAAI,CAAC,oBAAoB;QAGzB,IAAI,IAAI,CAAC,eAAe,EAAE;YACxBrB,MAAM,yCAAyCqB;YAC/C,OAAO,IAAI,CAAC,eAAe;QAC7B;QAEA,IAAI,IAAI,CAAC,SAAS,CAAC,UAAU,EAAE;YAC7BrB,MAAM,qCAAqCqB;YAC3C,OAAO,MAAM,IAAI,CAAC,SAAS,CAAC,UAAU;QACxC;QACErB,MAAM,yCAAyCqB;QAC/C,OAAO,MAAMC,oBAAoB,IAAI,CAAC,SAAS,EAAE;YAC/C,iBAAiB,IAAI,CAAC,kBAAkB,CAAC,sBAAsB;QACjE;IAEJ;IAEA,MAAM,mBAAuC;QAC3C,OAAO,MAAM,IAAI,CAAC,YAAY,CAAC;IACjC;IAEA,MAAM,mBAAmBC,MAAc,EAAE;QACvC,IAAI,IAAI,CAAC,IAAI,CAAC,eAAe,EAC3BC,QAAQ,IAAI,CACV;QAGJ,IAAI,CAAC,IAAI,CAAC,eAAe,GAAGD;IAC9B;IAEA,YAAY;QACV,IAAI,CAAC,IAAI,GAAG;YACV,WAAW,IAAI,CAAC,IAAI,CAAC,SAAS;YAC9B,kBAAkB,IAAI,CAAC,IAAI,CAAC,gBAAgB;YAC5C,YAAY,EAAE;YACd,aAAa,EAAE;QACjB;QAEA,OAAO,IAAI,CAAC,IAAI;IAClB;IAEA,oBAAoBE,SAAwB,EAAE;QAE5C,MAAMC,mBAAmBC,sBAAsBF;QAC/C,MAAMG,cAAc,IAAI,CAAC,IAAI;QAC7BA,YAAY,UAAU,CAAC,IAAI,CAACF;IAC9B;IAEA,iBAAiB;QAEf,IAAI,CAAC,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS;QACzC,IAAI,CAAC,IAAI,CAAC,gBAAgB,GAAG,IAAI,CAAC,IAAI,CAAC,gBAAgB;QACvD,OAAOG,kBAAkB,IAAI,CAAC,IAAI;IACpC;IAEA,mBAAmB;QACjB,OAAOC,kBAAkB,IAAI,CAAC,cAAc;IAC9C;IAEA,sBAAsB;QACpB,IAAI,IAAI,CAAC,SAAS,EAChB,MAAM,IAAIC,MACR;QAGJ,MAAM,EAAEC,cAAc,EAAEC,kBAAkB,EAAE,GAAG,IAAI,CAAC,IAAI;QACxD,IAAI,CAAC,UAAU,GAAGC,aAAa;YAC7B,UAAU,IAAI,CAAC,cAAc;YAC7B,SAASC;YACT,aAAa,IAAI,CAAC,cAAc;YAChC,MAAM;YACNH;QACF;QACAhC,MAAM,uBAAuB,IAAI,CAAC,UAAU;QAC5C,IAAIgC,kBAAkBC,sBAAsB,IAAI,CAAC,UAAU,EACzDG,eAAe,IAAI,CAAC,UAAU;IAElC;IAEA,MAAc,uBAAuBC,IAAmB,EAAE;QACxD,MAAMC,QAAQC,SAASF;QACvB,MAAMG,MAAMF,QAAQ,GAAGG,QAAQJ,MAAM,GAAG,EAAEC,OAAO,GAAGG,QAAQJ;QAE5D,IAAI,IAAI,CAAC,cAAc,EACrB,MAAM,IAAI,CAAC,cAAc,CAACG;IAE9B;IAEA,MAAc,iBAAiBE,QAAkB,EAAEC,kBAAkB,KAAK,EAAE;QAC1E,IAAI,CAAC,mBAAmB,CAACD,SAAS,IAAI;QAEtC,IAAI;YACF,IAAI,IAAI,CAAC,YAAY,EACnB,IAAI,CAAC,YAAY,CAAC,IAAI,CAAC,cAAc;QAEzC,EAAE,OAAOE,OAAO;YACdpB,QAAQ,KAAK,CAAC,yBAAyBoB;QACzC;QAEA,IAAI,CAAC,mBAAmB;QAExB,IAAIF,SAAS,cAAc,MAAM,CAACC,iBAAiB;YACjD,MAAME,YAAYH,SAAS,eAAe;YAC1C,MAAM,IAAIX,MAAM,GAAGc,QAAAA,YAAAA,KAAAA,IAAAA,UAAW,YAAY,CAAC,EAAE,EAAEA,QAAAA,YAAAA,KAAAA,IAAAA,UAAW,UAAU,EAAE,EAAE;gBACtE,OAAOA,QAAAA,YAAAA,KAAAA,IAAAA,UAAW,KAAK;YACzB;QACF;IACF;IAEA,MAAM,wBACJC,IAAY,EACZC,GAAO,EACP;QACA/C,MAAM,2BAA2B8C,MAAM,KAAKC;QAE5C,MAAMC,aAAgC;YACpC,MAAMF;YACN,OAAQC,OAAe,CAAC;YACxB,SAAS;QACX;QACA/C,MAAM,cAAcgD;QAEpB,MAAMC,QAA0B;YAACD;SAAW,CAAC,MAAM,CACjDE;QAGF,MAAMC,QAAQC,aACZN,MACAO,eAAe,AAACN,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAa,MAAM,AAAD,KAAK,CAAC;QAI1C,MAAMO,cAAc,IAAI,CAAC,kBAAkB,CAAC,cAAc,CAAC;QAE3D,MAAM,EAAEC,MAAM,EAAEb,QAAQ,EAAE,GAAG,MAAM,IAAI,CAAC,YAAY,CAAC,QAAQ,CAC3DS,OACAF,OACAK;QAEF,MAAM,IAAI,CAAC,gBAAgB,CAACZ;QAC5B,OAAOa;IACT;IAEA,MAAM,MAAMC,YAAyB,EAAET,GAAkB,EAAE;QACzDU,OAAOD,cAAc;QAErB,MAAME,sBAAsBC,yBAAyBH,cAAcT;QAEnE,OAAO,IAAI,CAAC,uBAAuB,CAAC,OAAO;YACzC,QAAQW;QACV;IACF;IAEA,MAAM,aAAaF,YAAyB,EAAET,GAAkB,EAAE;QAChEU,OAAOD,cAAc;QAErB,MAAME,sBAAsBC,yBAAyBH,cAAcT;QAEnE,OAAO,IAAI,CAAC,uBAAuB,CAAC,cAAc;YAChD,QAAQW;QACV;IACF;IAEA,MAAM,cAAcF,YAAyB,EAAET,GAAkB,EAAE;QACjEU,OAAOD,cAAc;QAErB,MAAME,sBAAsBC,yBAAyBH,cAAcT;QAEnE,OAAO,IAAI,CAAC,uBAAuB,CAAC,eAAe;YACjD,QAAQW;QACV;IACF;IAEA,MAAM,QAAQF,YAAyB,EAAET,GAAkB,EAAE;QAC3DU,OAAOD,cAAc;QAErB,MAAME,sBAAsBC,yBAAyBH,cAAcT;QAEnE,OAAO,IAAI,CAAC,uBAAuB,CAAC,SAAS;YAC3C,QAAQW;QACV;IACF;IAmBA,MAAM,QACJE,mBAAyC,EACzCC,iBAGa,EACbC,cAA6B,EAC7B;QACA,IAAIC;QACJ,IAAIP;QACJ,IAAIT;QAKJ,IACE,AAA6B,YAA7B,OAAOc,qBACPA,AAAsB,SAAtBA,qBACA,WAAWA,mBACX;YAEAL,eAAeI;YACf,MAAMI,eAAeH;YAKrBE,QAAQC,aAAa,KAAK;YAC1BjB,MAAMiB;QACR,OAAO;YAELD,QAAQH;YACRJ,eAAeK;YACfd,MAAM;gBACJ,GAAGe,cAAc;gBACjBC;YACF;QACF;QAEAN,OACE,AAAiB,YAAjB,OAAOM,OACP;QAEFN,OAAOD,cAAc;QAErB,MAAME,sBAAsBC,yBAAyBH,cAAcT;QAEnE,OAAO,IAAI,CAAC,uBAAuB,CAAC,SAAS;YAC3C,GAAIA,OAAO,CAAC,CAAC;YACb,QAAQW;QACV;IACF;IAmBA,MAAM,gBACJO,qBAA2C,EAC3CJ,iBAGa,EACbC,cAA6B,EAC7B;QACA,IAAII;QACJ,IAAIV;QACJ,IAAIT;QAGJ,IACE,AAA6B,YAA7B,OAAOc,qBACPA,AAAsB,SAAtBA,qBACA,aAAaA,mBACb;YAEAL,eAAeS;YACflB,MAAMc;QAGR,OAAO;YAELK,UAAUD;YACVT,eAAeK;YACfd,MAAM;gBACJ,GAAIe,kBAAkB,CAAC,CAAC;gBACxBI;YACF;QACF;QAEAT,OAAOV,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,OAAO,EAAE;QAErB,MAAMW,sBAAsBF,eACxBG,yBAAyBH,cAAcT,OACvCoB;QAEJ,OAAO,IAAI,CAAC,uBAAuB,CAAC,iBAAiB;YACnD,GAAIpB,OAAO,CAAC,CAAC;YACb,QAAQW;QACV;IACF;IAmBA,MAAM,SACJU,yBAAgE,EAChEP,iBAAyE,EACzEC,cAA6B,EAC7B;QACA,IAAIO;QACJ,IAAIb;QACJ,IAAIT;QAGJ,IACE,AAA6B,YAA7B,OAAOc,qBACN,gBAAeA,qBACd,gBAAgBA,qBAChB,cAAcA,iBAAgB,GAChC;YAEAL,eAAeY;YACfrB,MAAMc;QACR,OAAO;YAELQ,cAAcD;YACdZ,eAAeK;YACfd,MAAM;gBACJ,GAAIe,kBAAkB,CAAC,CAAC;gBACxB,GAAIO,eAAe,CAAC,CAAC;YACvB;QACF;QAEA,MAAMX,sBAAsBC,yBAC1BH,gBAAgB,IAChBT;QAGF,OAAO,IAAI,CAAC,uBAAuB,CAAC,UAAU;YAC5C,GAAIA,OAAO,CAAC,CAAC;YACb,QAAQW;QACV;IACF;IAEA,MAAM,SACJY,UAAkB,EAClBvB,GAEC,EACD;YASMwB,iBACcC;QATpB,MAAMlB,cAAc,IAAI,CAAC,kBAAkB,CAAC,cAAc,CAAC;QAE3D,MAAMmB,YAAY1B,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,SAAS;QAEhC,MAAM2B,cAAcpB,AAAuB,kBAAvBA,YAAY,MAAM;QACtC,MAAMqB,eACJD,eAAeD,AAAc,UAAdA,YACXN,SAAAA,QACAI,CAAAA,kBAAAA,IAAI,CAAC,SAAS,AAAD,IAAbA,KAAAA,IAAAA,gBAAgB,cAAc,CAACD;QACrC,IAAIK,gBAAAA,SAAgBH,CAAAA,mBAAAA,IAAI,CAAC,SAAS,AAAD,IAAbA,KAAAA,IAAAA,iBAAgB,iBAAiB,AAAD,GAAG;gBAInDI,4BAMWC;YARb,MAAM,EAAEnC,QAAQ,EAAE,GAAG,MAAM,IAAI,CAAC,YAAY,CAAC,sBAAsB,CACjE4B,YAAAA,QACAM,CAAAA,6BAAAA,aAAa,YAAY,AAAD,IAAxBA,KAAAA,IAAAA,2BAA2B,YAAY;YAGzC,MAAM,IAAI,CAAC,gBAAgB,CAAClC;YAE5B1C,MAAM;YACN,MAAM8E,OAAO,QAAAD,CAAAA,8BAAAA,aAAa,YAAY,AAAD,IAAxBA,KAAAA,IAAAA,4BAA2B,YAAY;YACpD,OAAO,IAAI,CAAC,OAAO,CAACC;QACtB;QAEA,MAAM,EAAEvB,MAAM,EAAEb,QAAQ,EAAE,GAAG,MAAM,IAAI,CAAC,YAAY,CAAC,MAAM,CACzD4B,YACAhB,aACA,IAAI,CAAC,IAAI,CAAC,eAAe;QAI3B,IAAI,IAAI,CAAC,SAAS,IAAIC,CAAAA,QAAAA,SAAAA,KAAAA,IAAAA,OAAQ,QAAQ,AAAD,KAAKkB,AAAc,UAAdA,WAAqB;YAC7D,MAAMM,cAAkC;gBACtC,OAAO;oBACL;wBACE,MAAMT;wBACN,MAAMf,OAAO,QAAQ;oBACvB;iBACD;YACH;YACA,MAAMyB,cAAcF,QAAAA,IAAS,CAACC;YAC9B,IAAI,CAAC,SAAS,CAAC,yBAAyB,CACtC;gBACE,MAAM;gBACN,QAAQT;gBACR,cAAcU;YAChB,GACAL;QAEJ;QAEA,MAAM,IAAI,CAAC,gBAAgB,CAACjC;QAC5B,OAAOa;IACT;IAEA,MAAM,QACJ0B,MAA2B,EAC3BlC,MAA4B5B,2BAA2B,EAClC;QACrB,MAAMmC,cAAc,IAAI,CAAC,kBAAkB,CAAC,cAAc,CAAC;QAC3D,MAAM,EAAEC,MAAM,EAAEb,QAAQ,EAAE,GACxB,MAAM,IAAI,CAAC,YAAY,CAAC,wBAAwB,CAC9C,SACAuC,QACA3B,aACAP;QAEJ,MAAM,IAAI,CAAC,gBAAgB,CAACL;QAC5B,OAAOa;IACT;IAEA,MAAM,UACJhC,MAAmB,EACnBwB,MAA4B5B,2BAA2B,EACrC;QAClB,MAAMmC,cAAc,IAAI,CAAC,kBAAkB,CAAC,cAAc,CAAC;QAE3D,MAAM,EAAE4B,UAAU,EAAEC,gBAAgB,EAAE,GAAGC,YAAY7D;QACrD,MAAM,EAAEgC,MAAM,EAAEb,QAAQ,EAAE,GACxB,MAAM,IAAI,CAAC,YAAY,CAAC,wBAAwB,CAC9C,WACAwC,YACA5B,aACAP,KACAoC;QAEJ,MAAM,IAAI,CAAC,gBAAgB,CAACzC;QAC5B,OAAOa;IACT;IAEA,MAAM,SACJhC,MAAmB,EACnBwB,MAA4B5B,2BAA2B,EACtC;QACjB,MAAMmC,cAAc,IAAI,CAAC,kBAAkB,CAAC,cAAc,CAAC;QAE3D,MAAM,EAAE4B,UAAU,EAAEC,gBAAgB,EAAE,GAAGC,YAAY7D;QACrD,MAAM,EAAEgC,MAAM,EAAEb,QAAQ,EAAE,GACxB,MAAM,IAAI,CAAC,YAAY,CAAC,wBAAwB,CAC9C,UACAwC,YACA5B,aACAP,KACAoC;QAEJ,MAAM,IAAI,CAAC,gBAAgB,CAACzC;QAC5B,OAAOa;IACT;IAEA,MAAM,SACJhC,MAAmB,EACnBwB,MAA4B5B,2BAA2B,EACtC;QACjB,MAAMmC,cAAc,IAAI,CAAC,kBAAkB,CAAC,cAAc,CAAC;QAE3D,MAAM,EAAE4B,UAAU,EAAEC,gBAAgB,EAAE,GAAGC,YAAY7D;QACrD,MAAM,EAAEgC,MAAM,EAAEb,QAAQ,EAAE,GACxB,MAAM,IAAI,CAAC,YAAY,CAAC,wBAAwB,CAC9C,UACAwC,YACA5B,aACAP,KACAoC;QAEJ,MAAM,IAAI,CAAC,gBAAgB,CAACzC;QAC5B,OAAOa;IACT;IAEA,MAAM,MACJhC,MAAmB,EACnBwB,MAA4B5B,2BAA2B,EACtC;QACjB,OAAO,IAAI,CAAC,QAAQ,CAACI,QAAQwB;IAC/B;IAEA,MAAM,uBACJsC,MAAwB,EACxBtC,GAI0B,EACkB;QAC5C,MAAM,EAAEuC,eAAe,IAAI,EAAEC,aAAa,CAAC,EAAE,GAAGxC,OAAO,CAAC;QAExD,IAAIyC,UAAU;QACd,IAAIC,aAAa;QACjB,IAAIC,eAAe;QACnB,IAAIC,YAAY5C,AAAAA,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,SAAS,AAAD,KAAK;QAClC,IAAI6C;QAEJ,MAAO,CAACJ,WAAWC,aAAaF,WAAY;YAC1C,IAAIE,cAAc,GAChBE,YAAY;YAEd3F,MACE,cACAqF,QACA,gBACAC,cACA,cACAG,YACA,aACAE;YAGF,MAAMrC,cAAc,IAAI,CAAC,kBAAkB,CAAC,cAAc,CAAC;YAE3D,MAAMuC,OAAO,MAAM,IAAI,CAAC,OAAO,CAAC,QAAQ,CAACR,QAAQ/B,aAAa;gBAC5DqC;YACF;YACA3F,MAAM,mBAAmB6F;YACzBpC,OAAOoC,KAAK,WAAW,EAAE,CAAC,+BAA+B,EAAER,OAAO,CAAC,CAAC;YACpEK,eAAeG,KAAK,WAAW;YAE/BD,eAAe,MAAM,IAAI,CAAC,aAAa,CACrCF,cACAC,YAAY;gBAAE,WAAW;YAAK,IAAIxB,QAClCkB,QACAtC;YAEF,IAAI6C,aAAa,IAAI,EACnBJ,UAAU;iBAEVC;QAEJ;QAEA,OAAO;YACL,QAAQC;YACRC;YACAC;QACF;IACF;IAEA,MAAM,cACJrE,MAAc,EACduE,SAAmC,EACnCC,YAA8B,EAC9BC,kBAA2C,EACX;QAChChG,MAAM,iBAAiBuB,QAAQuE,WAAWC,cAAcC;QAExD,MAAM,EAAE,QAAQC,YAAY,EAAE,MAAMC,UAAU,EAAE,GAAG,MAAM,IAAI,CAAC,QAAQ,CACpE3E,QACAuE;QAEF,MAAMK,WAAWjG,oBAAoB6F,cAAcE;QACnD,MAAMG,WAAW1F,eAAeqF,cAAcG;QAC9C,MAAMG,OACJF,YAAaH,CAAAA,CAAAA,QAAAA,qBAAAA,KAAAA,IAAAA,mBAAoB,uBAAuB,AAAD,KAAK,EAAC,KAC7DI;QACF,MAAMR,eAAe;YACnBS;YACA,MAAMH;YACN,QAAQD;YACR,gBAAgBE;QAClB;QACAnG,MAAM,2BAA2B4F;QACjC,OAAOA;IACT;IAEA,MAAM,SAASrE,MAAmB,EAAEwB,GAAkB,EAAE;QACtD,MAAMuD,cAAc3C,yBAAyBpC,QAAQwB;QACrDU,OAAO6C,aAAa;QACpB,MAAMC,aAAaC,oBAAoBF;QACvC,MAAMrD,QAAQ;YAACsD;SAAW;QAC1B,MAAMjD,cAAc,IAAI,CAAC,kBAAkB,CAAC,cAAc,CAAC;QAE3D,MAAM,EAAEZ,QAAQ,EAAEa,MAAM,EAAE,GAAG,MAAM,IAAI,CAAC,YAAY,CAAC,QAAQ,CAC3DH,aAAa,UAAUC,eAAeiD,eACtCrD,OACAK;QAEF,MAAM,IAAI,CAAC,gBAAgB,CAACZ;QAE5B,MAAM,EAAE+D,OAAO,EAAE,GAAGlD;QAEpB,OAAO;YACL,MAAMkD,QAAAA,UAAAA,KAAAA,IAAAA,QAAS,IAAI;YACnB,QAAQA,QAAAA,UAAAA,KAAAA,IAAAA,QAAS,MAAM;YACvB,OAAQ,OAAM,IAAI,CAAC,SAAS,CAAC,IAAI,EAAC,EAAG,GAAG;QAC1C;IAGF;IAEA,MAAM,SACJC,SAAsB,EACtBC,GAAY,EACZ5D,GAA2C,EAC3C;YAuBiB6D;QAtBjB,MAAMtD,cAAc,IAAI,CAAC,kBAAkB,CAAC,cAAc,CAAC;QAE3D,MAAMuD,aAAmC;YACvC,aAAa9D,AAAAA,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,WAAW,AAAD,KAAK5B,4BAA4B,WAAW;YACxE,oBACE4B,AAAAA,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,kBAAkB,AAAD,KACtB5B,4BAA4B,kBAAkB;YAChD,eAAe4B,AAAAA,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,aAAa,AAAD,KAAK;YACrC,iBAAiBA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,eAAe;YACrC,iBAAiBA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,eAAe;QACvC;QAEA,MAAM,EAAEQ,MAAM,EAAEb,QAAQ,EAAEoE,OAAO,EAAE,GAAG,MAAM,IAAI,CAAC,YAAY,CAAC,MAAM,CAClEJ,WACApD,aACAuD;QAEF,MAAM,IAAI,CAAC,gBAAgB,CAACnE,UAAU;QAEtC,MAAMqE,UAAUxD,SACZY,SACA,CAAC,kBAAkB,EAAEwC,OAAQ,CAAqB,YAArB,OAAOD,YAAyBA,YAAYA,UAAU,MAAK,EAAG,UAAU,EACnGI,WAAAA,SAAWF,CAAAA,4BAAAA,SAAS,eAAe,EAAC,IAAzBA,KAAAA,IAAAA,0BAA4B,KAAK,AAAD,KAAK,eAChD;QAEN,IAAI7D,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,eAAe,EACtB,OAAO;YACL,MAAMQ;YACNuD;YACAC;QACF;QAGF,IAAI,CAACxD,QACH,MAAM,IAAIxB,MAAMgF;IAEpB;IAEA,MAAM,UAAUL,SAAsB,EAAE3D,GAAqB,EAAE;QAC7D,MAAMO,cAAc,IAAI,CAAC,kBAAkB,CAAC,cAAc,CAAC;QAC3D,MAAM,EAAEZ,QAAQ,EAAE,GAAG,MAAM,IAAI,CAAC,YAAY,CAAC,OAAO,CAClDgE,WACA;YACE,WAAW3D,AAAAA,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,SAAS,AAAD,KAAK;YAC7B,iBAAiBA,AAAAA,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,eAAe,AAAD,KAAK;QAC3C,GACAO;QAEF,MAAM,IAAI,CAAC,gBAAgB,CAACZ,UAAU;QAEtC,IAAIA,SAAS,cAAc,IAAI;YAC7B,MAAMG,YAAYH,SAAS,eAAe;YAC1C,MAAM,IAAIX,MAAM,GAAGc,QAAAA,YAAAA,KAAAA,IAAAA,UAAW,KAAK,CAAC,EAAE,EAAEA,QAAAA,YAAAA,KAAAA,IAAAA,UAAW,UAAU,EAAE;QACjE;IACF;IAEA,MAAM,GAAGyB,UAAkB,EAAExB,OAAO,QAAQ,EAAE;QAC5C,IAAIA,AAAS,aAATA,MACF,OAAO,IAAI,CAAC,QAAQ,CAACwB;QAEvB,IAAIxB,AAAS,YAATA,MACF,OAAO,IAAI,CAAC,OAAO,CAACwB;QAGtB,IAAIxB,AAAS,aAATA,MACF,OAAO,IAAI,CAAC,QAAQ,CAACwB;QAGvB,IAAIxB,AAAS,UAATA,MACF,OAAO,IAAI,CAAC,KAAK,CAACwB;QAGpB,IAAIxB,AAAS,iBAATA,MACF,OAAO,IAAI,CAAC,YAAY,CAACwB;QAG3B,IAAIxB,AAAS,kBAATA,MACF,OAAO,IAAI,CAAC,aAAa,CAACwB;QAG5B,MAAM,IAAIvC,MACR,CAAC,cAAc,EAAEe,KAAK,8EAA8E,CAAC;IAEzG;IAEA,MAAM,QAAQkE,iBAAyB,EAEpC;QACD,MAAMC,SAASC,gBAAgBF,mBAAmB;QAClD,MAAMG,SAAS,IAAIC,aAAaH,QAAQ,UAC/B;gBAAE,OAAO,IAAI;gBAAE,QAAQ,EAAE;YAAC;QAEnC,MAAME,OAAO,GAAG;QAEhB,IAAIA,AAAkB,YAAlBA,OAAO,MAAM,EAAc;YAC7B,MAAME,SAASF,OAAO,cAAc,CACjC,MAAM,CAAC,CAAC9E,OAASA,AAAgB,YAAhBA,KAAK,MAAM,EAC5B,GAAG,CAAC,CAACA;oBAC2BiF;gBAA/B,OAAO,CAAC,OAAO,EAAEjF,KAAK,IAAI,CAAC,EAAE,EAAE,QAAAiF,CAAAA,cAAAA,KAAK,KAAK,AAAD,IAATA,KAAAA,IAAAA,YAAY,OAAO,EAAE;YACtD,GACC,IAAI,CAAC;YACR,MAAM,IAAIvF,MAAM,CAAC,2CAA2C,EAAEsF,QAAQ;QACxE;QAEA,OAAO;YACL,QAAQF,OAAO,MAAM;QACvB;IACF;IAEA,MAAM,mBAAmBF,MAAc,EAAE;QACvCxD,OACE,IAAI,CAAC,SAAS,CAAC,kBAAkB,EACjC;QAEF,OAAO,IAAI,CAAC,SAAS,CAAC,kBAAkB,CAACwD;IAC3C;IAEA,MAAM,UAAU;YACRM,yBAAAA;QAAN,eAAMA,CAAAA,0BAAAA,AAAAA,CAAAA,kBAAAA,IAAI,CAAC,SAAS,AAAD,EAAE,OAAO,AAAD,IAArBA,KAAAA,IAAAA,wBAAAA,IAAAA,CAAAA,gBAAAA;QACN,IAAI,CAAC,SAAS;QACd,IAAI,CAAC,SAAS,GAAG;IACnB;IAEA,MAAM,cACJpE,KAAc,EACdJ,GAEC,EACD;QAEA,MAAMyE,SAAS,MAAM,IAAI,CAAC,SAAS,CAAC,gBAAgB;QACpD,MAAMC,MAAMC,KAAK,GAAG;QAEpB,MAAMC,WAAoC;YACxC;gBACE,MAAM;gBACN,IAAIF;gBACJ,YAAYD;YACd;SACD;QAED,MAAMnF,OAAyB;YAC7B,MAAM;YACN,SAAS;YACT,QAAQ;YACRsF;YACA,QAAQ;gBACN,OAAOF;gBACP,KAAKA;gBACL,MAAM;YACR;YACA,OAAO;gBACL,SAAS1E,AAAAA,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,OAAO,AAAD,KAAK;YAC3B;YACA,UAAU,WAAa;QACzB;QAEA,MAAM6E,gBAA+B;YACnC,YAAY;YACZ,SAASH;YACT,MAAM,CAAC,MAAM,EAAEtE,SAAS,YAAY;YACpC,aAAaJ,AAAAA,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,OAAO,AAAD,KAAK;YAC7B,OAAO;gBAACV;aAAK;QACf;QAEA,IAAI,CAAC,mBAAmB,CAACuF;QAEzB,IAAI;gBACFC,oBAAAA;oBAAAA,CAAAA,qBAAAA,AAAAA,CAAAA,QAAAA,IAAI,AAAD,EAAE,YAAY,AAAD,KAAhBA,mBAAAA,IAAAA,CAAAA,OAAoB,IAAI,CAAC,cAAc;QACzC,EAAE,OAAOjF,OAAO;YACdpB,QAAQ,KAAK,CAAC,yBAAyBoB;QACzC;QAEA,IAAI,CAAC,mBAAmB;IAC1B;IAEA,sBAAsB;QACpB,MAAM,EAAEkF,SAAS,EAAEC,gBAAgB,EAAEC,UAAU,EAAE,GAAG,IAAI,CAAC,IAAI;QAC7D,MAAMC,gBAAgBC,MAAM,OAAO,CAACF,cAChCA,WAAW,GAAG,CAAC,CAACvG;YACd,MAAM,EAAE0G,KAAK,EAAE,GAAGC,eAAe,GAAG3G;YACpC,IAAI4G,WAAWF;YACf,IAAID,MAAM,OAAO,CAACC,QAChBE,WAAWF,MAAM,GAAG,CAAC,CAAC9F;gBAEpB,MAAM,EAAEiG,SAAS,EAAEC,GAAG,EAAE,GAAGC,UAAU,GAAGnG;gBACxC,OAAOmG;YACT;YAEF,OAAO;gBAAE,GAAGJ,aAAa;gBAAE,GAAIC,WAAW;oBAAE,OAAOA;gBAAS,IAAI,CAAC,CAAC;YAAE;QACtE,KACA,EAAE;QACN,OAAO;YACLP;YACAC;YACA,YAAYE;QACd;IACF;IAMA,MAAM,oBAAmC;QACvCjI,MAAM;QACN,MAAMyI,UAAU,MAAM,IAAI,CAAC,gBAAgB;QAE3CA,QAAQ,SAAS,GAAG;QACpB,IAAI,CAAC,eAAe,GAAGA;QACvBzI,MAAM;IACR;IAKA,MAAM,sBAAqC;QACzCA,MAAM;QACN,IAAI,CAAC,eAAe,GAAGmE;QACvBnE,MAAM;IACR;IA14BA,YAAY0I,iBAAgC,EAAEC,IAAe,CAAE;QA5D/D;QAEA;QAEA;QAEA;QAEA;QAEA;QAEA;QAKA,kCAAU;QAEV;QAEA;QAEA;QAEA,oCAAY;QAEZ;QAKA,uBAAQ,mBAAR;QAKA,uBAAQ,uBAAsB;QAwB5B,IAAI,CAAC,SAAS,GAAGD;QACjB,IAAI,CAAC,IAAI,GAAGE,OAAO,MAAM,CACvB;YACE,gBAAgB;YAChB,oBAAoB;YACpB,WAAW;YACX,kBAAkB;QACpB,GACAD,QAAQ,CAAC;QAGX,IAAIA,AAAAA,CAAAA,QAAAA,OAAAA,KAAAA,IAAAA,KAAM,WAAW,AAAD,KAAK,AAA6B,cAA7B,OAAOA,CAAAA,QAAAA,OAAAA,KAAAA,IAAAA,KAAM,WAAW,AAAD,GAC9C,MAAM,IAAI5G,MACR,CAAC,+DAA+D,EAAE,OAAO4G,CAAAA,QAAAA,OAAAA,KAAAA,IAAAA,KAAM,WAAW,AAAD,GAAG;QAGhG,IAAI,CAAC,kBAAkB,GAAGA,AAAAA,CAAAA,QAAAA,OAAAA,KAAAA,IAAAA,KAAM,WAAW,AAAD,IACtC,IAAIE,mBAAmBF,KAAK,WAAW,IACvCG;QAEJ,IAAI,CAAC,cAAc,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc;QAE9C,IAAI,CAAC,OAAO,GAAG,IAAIC,QAAQ,OAAO1H,SACzB,IAAI,CAAC,YAAY,CAACA;QAG3B,MAAM2H,WACJ,AAA0B,aAA1B,OAAOL,CAAAA,QAAAA,OAAAA,KAAAA,IAAAA,KAAM,QAAQ,AAAD,IAChBA,KAAK,QAAQ,GACbM,oBAAoB,qBAAqB,CAACC;QAEhD,IAAIP,QAAAA,OAAAA,KAAAA,IAAAA,KAAM,OAAO,EACf,IAAI,CAAC,SAAS,GAAG,IAAIQ,UACnBR,KAAK,OAAO,EACZK;QAIJ,IAAI,CAAC,YAAY,GAAG,IAAII,aAAa,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,OAAO,EAAE;YACjE,WAAW,IAAI,CAAC,SAAS;YACzB,aAAa,IAAI,CAAC,sBAAsB,CAAC,IAAI,CAAC,IAAI;YAClD,sBAAsB,IAAI,CAAC,IAAI,CAAC,oBAAoB;QACtD;QACA,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,SAAS;QAC1B,IAAI,CAAC,cAAc,GACjBT,AAAAA,CAAAA,QAAAA,OAAAA,KAAAA,IAAAA,KAAM,cAAc,AAAD,KACnBU,kBAAkBV,AAAAA,CAAAA,QAAAA,OAAAA,KAAAA,IAAAA,KAAM,MAAM,AAAD,KAAK,IAAI,CAAC,SAAS,CAAC,aAAa,IAAI;IACtE;AA21BF;AAEO,MAAMW,cAAc,CACzBZ,mBACAC,OAEO,IAAIvH,MAAMsH,mBAAmBC"}
@@ -1,4 +1,4 @@
1
- import { elementByPositionWithElementInfo, findAllMidsceneLocatorField, resizeImageForUiTars, vlmPlanning } from "../ai-model/index.mjs";
1
+ import { ConversationHistory, elementByPositionWithElementInfo, findAllMidsceneLocatorField, vlmPlanning } from "../ai-model/index.mjs";
2
2
  import { Executor, plan as external_index_mjs_plan } from "../index.mjs";
3
3
  import { sleep as external_utils_mjs_sleep } from "../utils.mjs";
4
4
  import { NodeType } from "@midscene/shared/constants";
@@ -19,6 +19,7 @@ function _define_property(obj, key, value) {
19
19
  }
20
20
  const debug = getDebug('device-task-executor');
21
21
  const defaultReplanningCycleLimit = 10;
22
+ const defaultVlmUiTarsReplanningCycleLimit = 40;
22
23
  function locatePlanForLocate(param) {
23
24
  const locate = 'string' == typeof param ? {
24
25
  prompt: param
@@ -389,31 +390,32 @@ class TaskExecutor {
389
390
  executor: taskExecutor
390
391
  };
391
392
  }
392
- planningTaskFromPrompt(userInstruction, opts) {
393
- const { log, actionContext, modelConfig } = opts;
393
+ createPlanningTask(userInstruction, actionContext, modelConfig) {
394
394
  const task = {
395
395
  type: 'Planning',
396
396
  subType: 'Plan',
397
397
  locate: null,
398
398
  param: {
399
- userInstruction,
400
- log
399
+ userInstruction
401
400
  },
402
401
  executor: async (param, executorContext)=>{
403
402
  const startTime = Date.now();
404
403
  const { uiContext } = await this.setupPlanningContext(executorContext);
404
+ const { vlMode } = modelConfig;
405
+ const uiTarsModelVersion = 'vlm-ui-tars' === vlMode ? modelConfig.uiTarsModelVersion : void 0;
405
406
  assert(this.interface.actionSpace, 'actionSpace for device is not implemented');
406
407
  const actionSpace = await this.interface.actionSpace();
407
408
  debug('actionSpace for this interface is:', actionSpace.map((action)=>action.name).join(', '));
408
409
  assert(Array.isArray(actionSpace), 'actionSpace must be an array');
409
410
  if (0 === actionSpace.length) console.warn(`ActionSpace for ${this.interface.interfaceType} is empty. This may lead to unexpected behavior.`);
410
- const planResult = await external_index_mjs_plan(param.userInstruction, {
411
+ const planResult = await (uiTarsModelVersion ? vlmPlanning : external_index_mjs_plan)(param.userInstruction, {
411
412
  context: uiContext,
412
413
  log: param.log,
413
414
  actionContext,
414
415
  interfaceType: this.interface.interfaceType,
415
416
  actionSpace,
416
- modelConfig
417
+ modelConfig,
418
+ conversationHistory: this.conversationHistory
417
419
  });
418
420
  const { actions, log, more_actions_needed_by_instruction, error, usage, rawResponse, sleep } = planResult;
419
421
  executorContext.task.log = {
@@ -450,64 +452,6 @@ class TaskExecutor {
450
452
  };
451
453
  return task;
452
454
  }
453
- planningTaskToGoal(userInstruction, opts) {
454
- const task = {
455
- type: 'Planning',
456
- subType: 'Plan',
457
- locate: null,
458
- param: {
459
- userInstruction
460
- },
461
- executor: async (param, executorContext)=>{
462
- var _actions_;
463
- const { uiContext } = await this.setupPlanningContext(executorContext);
464
- const { modelConfig } = opts;
465
- const { uiTarsModelVersion } = modelConfig;
466
- const imagePayload = await resizeImageForUiTars(uiContext.screenshotBase64, uiContext.size, uiTarsModelVersion);
467
- this.appendConversationHistory({
468
- role: 'user',
469
- content: [
470
- {
471
- type: 'image_url',
472
- image_url: {
473
- url: imagePayload
474
- }
475
- }
476
- ]
477
- });
478
- const planResult = await vlmPlanning({
479
- userInstruction: param.userInstruction,
480
- conversationHistory: this.conversationHistory,
481
- size: uiContext.size,
482
- modelConfig
483
- });
484
- const { actions, action_summary, usage } = planResult;
485
- executorContext.task.log = {
486
- ...executorContext.task.log || {},
487
- rawResponse: planResult.rawResponse
488
- };
489
- executorContext.task.usage = usage;
490
- this.appendConversationHistory({
491
- role: 'assistant',
492
- content: action_summary
493
- });
494
- return {
495
- output: {
496
- actions,
497
- thought: null == (_actions_ = actions[0]) ? void 0 : _actions_.thought,
498
- actionType: actions[0].type,
499
- more_actions_needed_by_instruction: true,
500
- log: '',
501
- yamlFlow: planResult.yamlFlow
502
- },
503
- cache: {
504
- hit: false
505
- }
506
- };
507
- }
508
- };
509
- return task;
510
- }
511
455
  async runPlans(title, plans, modelConfig) {
512
456
  const taskExecutor = new Executor(title, {
513
457
  onTaskStart: this.onTaskStartCallback
@@ -521,24 +465,23 @@ class TaskExecutor {
521
465
  executor: taskExecutor
522
466
  };
523
467
  }
468
+ getReplanningCycleLimit(isVlmUiTars) {
469
+ return this.replanningCycleLimit || globalConfigManager.getEnvConfigInNumber(MIDSCENE_REPLANNING_CYCLE_LIMIT) || (isVlmUiTars ? defaultVlmUiTarsReplanningCycleLimit : defaultReplanningCycleLimit);
470
+ }
524
471
  async action(userPrompt, modelConfig, actionContext) {
472
+ this.conversationHistory.reset();
525
473
  const taskExecutor = new Executor(taskTitleStr('Action', userPrompt), {
526
474
  onTaskStart: this.onTaskStartCallback
527
475
  });
528
- let planningTask = this.planningTaskFromPrompt(userPrompt, {
529
- log: void 0,
530
- actionContext,
531
- modelConfig
532
- });
533
476
  let replanCount = 0;
534
- const logList = [];
535
477
  const yamlFlow = [];
536
- const replanningCycleLimit = this.replanningCycleLimit || globalConfigManager.getEnvConfigInNumber(MIDSCENE_REPLANNING_CYCLE_LIMIT) || defaultReplanningCycleLimit;
537
- while(planningTask){
478
+ const replanningCycleLimit = this.getReplanningCycleLimit('vlm-ui-tars' === modelConfig.vlMode);
479
+ while(true){
538
480
  if (replanCount > replanningCycleLimit) {
539
481
  const errorMsg = 'Replanning too many times, please split the task into multiple steps';
540
482
  return this.appendErrorPlan(taskExecutor, errorMsg, modelConfig);
541
483
  }
484
+ const planningTask = this.createPlanningTask(userPrompt, actionContext, modelConfig);
542
485
  await taskExecutor.append(planningTask);
543
486
  const result = await taskExecutor.flush();
544
487
  const planResult = null == result ? void 0 : result.output;
@@ -560,16 +503,7 @@ class TaskExecutor {
560
503
  output: void 0,
561
504
  executor: taskExecutor
562
505
  };
563
- if (null == planResult ? void 0 : planResult.log) logList.push(planResult.log);
564
- if (!planResult.more_actions_needed_by_instruction) {
565
- planningTask = null;
566
- break;
567
- }
568
- planningTask = this.planningTaskFromPrompt(userPrompt, {
569
- log: logList.length > 0 ? `- ${logList.join('\n- ')}` : void 0,
570
- actionContext,
571
- modelConfig
572
- });
506
+ if (!planResult.more_actions_needed_by_instruction) break;
573
507
  replanCount++;
574
508
  }
575
509
  return {
@@ -579,52 +513,6 @@ class TaskExecutor {
579
513
  executor: taskExecutor
580
514
  };
581
515
  }
582
- async actionToGoal(userPrompt, modelConfig) {
583
- const taskExecutor = new Executor(taskTitleStr('Action', userPrompt), {
584
- onTaskStart: this.onTaskStartCallback
585
- });
586
- this.conversationHistory = [];
587
- const isCompleted = false;
588
- let currentActionCount = 0;
589
- const maxActionNumber = 40;
590
- const yamlFlow = [];
591
- while(!isCompleted && currentActionCount < maxActionNumber){
592
- currentActionCount++;
593
- debug('actionToGoal, currentActionCount:', currentActionCount, 'userPrompt:', userPrompt);
594
- const planningTask = this.planningTaskToGoal(userPrompt, {
595
- modelConfig
596
- });
597
- await taskExecutor.append(planningTask);
598
- const result = await taskExecutor.flush();
599
- if (taskExecutor.isInErrorState()) return {
600
- output: void 0,
601
- executor: taskExecutor
602
- };
603
- if (!result) throw new Error('result of taskExecutor.flush() is undefined in function actionToGoal');
604
- const { output } = result;
605
- const plans = output.actions;
606
- yamlFlow.push(...output.yamlFlow || []);
607
- let executables;
608
- try {
609
- executables = await this.convertPlanToExecutable(plans, modelConfig);
610
- taskExecutor.append(executables.tasks);
611
- } catch (error) {
612
- return this.appendErrorPlan(taskExecutor, `Error converting plans to executable tasks: ${error}, plans: ${JSON.stringify(plans)}`, modelConfig);
613
- }
614
- await taskExecutor.flush();
615
- if (taskExecutor.isInErrorState()) return {
616
- output: void 0,
617
- executor: taskExecutor
618
- };
619
- if ('Finished' === plans[0].type) break;
620
- }
621
- return {
622
- output: {
623
- yamlFlow
624
- },
625
- executor: taskExecutor
626
- };
627
- }
628
516
  createTypeQueryTask(type, demand, modelConfig, opt, multimodalPrompt) {
629
517
  const queryTask = {
630
518
  type: 'Insight',
@@ -657,21 +545,18 @@ class TaskExecutor {
657
545
  ];
658
546
  const ifTypeRestricted = 'Query' !== type;
659
547
  let demandInput = demand;
660
- let keyOfResult = 'result';
661
- if (ifTypeRestricted && 'Assert' === type) {
662
- keyOfResult = 'StatementIsTruthy';
548
+ if (ifTypeRestricted) {
549
+ const returnType = 'Assert' === type ? 'Boolean' : type;
663
550
  demandInput = {
664
- [keyOfResult]: `Boolean, ${demand}`
551
+ result: `${returnType}, ${demand}`
665
552
  };
666
- } else if (ifTypeRestricted) demandInput = {
667
- [keyOfResult]: `${type}, ${demand}`
668
- };
553
+ }
669
554
  const { data, usage, thought } = await this.insight.extract(demandInput, modelConfig, opt, multimodalPrompt);
670
555
  let outputResult = data;
671
556
  if (ifTypeRestricted) if ('string' == typeof data) outputResult = data;
672
557
  else {
673
- assert((null == data ? void 0 : data[keyOfResult]) !== void 0, 'No result in query data');
674
- outputResult = data[keyOfResult];
558
+ assert((null == data ? void 0 : data.result) !== void 0, 'No result in query data');
559
+ outputResult = data.result;
675
560
  }
676
561
  return {
677
562
  output: outputResult,
@@ -705,16 +590,6 @@ class TaskExecutor {
705
590
  const { textPrompt, multimodalPrompt } = parsePrompt(assertion);
706
591
  return await this.createTypeQueryExecution('Assert', textPrompt, modelConfig, opt, multimodalPrompt);
707
592
  }
708
- appendConversationHistory(conversationHistory) {
709
- if ('user' === conversationHistory.role) {
710
- const userImgItems = this.conversationHistory.filter((item)=>'user' === item.role);
711
- if (userImgItems.length >= 4 && 'user' === conversationHistory.role) {
712
- const firstUserImgIndex = this.conversationHistory.findIndex((item)=>'user' === item.role);
713
- if (firstUserImgIndex >= 0) this.conversationHistory.splice(firstUserImgIndex, 1);
714
- }
715
- }
716
- this.conversationHistory.push(conversationHistory);
717
- }
718
593
  async appendErrorPlan(taskExecutor, errorMsg, modelConfig) {
719
594
  const errorPlan = {
720
595
  type: 'Error',
@@ -733,19 +608,6 @@ class TaskExecutor {
733
608
  executor: taskExecutor
734
609
  };
735
610
  }
736
- async taskForSleep(timeMs, modelConfig) {
737
- const sleepPlan = {
738
- type: 'Sleep',
739
- param: {
740
- timeMs
741
- },
742
- locate: null
743
- };
744
- const { tasks: sleepTasks } = await this.convertPlanToExecutable([
745
- sleepPlan
746
- ], modelConfig);
747
- return this.prependExecutorWithScreenshot(sleepTasks[0]);
748
- }
749
611
  async waitFor(assertion, opt, modelConfig) {
750
612
  const { textPrompt, multimodalPrompt } = parsePrompt(assertion);
751
613
  const description = `waitFor: ${textPrompt}`;
@@ -778,8 +640,18 @@ class TaskExecutor {
778
640
  const now = Date.now();
779
641
  if (now - startTime < checkIntervalMs) {
780
642
  const timeRemaining = checkIntervalMs - (now - startTime);
781
- const sleepTask = await this.taskForSleep(timeRemaining, modelConfig);
782
- await taskExecutor.append(sleepTask);
643
+ const sleepPlan = {
644
+ type: 'Sleep',
645
+ param: {
646
+ timeMs: timeRemaining
647
+ },
648
+ locate: null
649
+ };
650
+ const { tasks: sleepTasks } = await this.convertPlanToExecutable([
651
+ sleepPlan
652
+ ], modelConfig);
653
+ await taskExecutor.append(this.prependExecutorWithScreenshot(sleepTasks[0]));
654
+ await taskExecutor.flush();
783
655
  }
784
656
  }
785
657
  return this.appendErrorPlan(taskExecutor, `waitFor timeout: ${errorThought}`, modelConfig);
@@ -788,7 +660,7 @@ class TaskExecutor {
788
660
  _define_property(this, "interface", void 0);
789
661
  _define_property(this, "insight", void 0);
790
662
  _define_property(this, "taskCache", void 0);
791
- _define_property(this, "conversationHistory", []);
663
+ _define_property(this, "conversationHistory", void 0);
792
664
  _define_property(this, "onTaskStartCallback", void 0);
793
665
  _define_property(this, "replanningCycleLimit", void 0);
794
666
  this.interface = interfaceInstance;
@@ -796,6 +668,7 @@ class TaskExecutor {
796
668
  this.taskCache = opts.taskCache;
797
669
  this.onTaskStartCallback = null == opts ? void 0 : opts.onTaskStart;
798
670
  this.replanningCycleLimit = opts.replanningCycleLimit;
671
+ this.conversationHistory = new ConversationHistory();
799
672
  }
800
673
  }
801
674
  export { TaskExecutor, locatePlanForLocate };