@midscene/core 1.0.1-beta-20251208075922.0 → 1.0.1-beta-20251208112226.0

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 (83) hide show
  1. package/dist/es/agent/agent.mjs +22 -7
  2. package/dist/es/agent/agent.mjs.map +1 -1
  3. package/dist/es/agent/task-builder.mjs +30 -25
  4. package/dist/es/agent/task-builder.mjs.map +1 -1
  5. package/dist/es/agent/task-cache.mjs +1 -1
  6. package/dist/es/agent/task-cache.mjs.map +1 -1
  7. package/dist/es/agent/tasks.mjs +17 -7
  8. package/dist/es/agent/tasks.mjs.map +1 -1
  9. package/dist/es/agent/utils.mjs +1 -1
  10. package/dist/es/ai-model/conversation-history.mjs +19 -2
  11. package/dist/es/ai-model/conversation-history.mjs.map +1 -1
  12. package/dist/es/ai-model/llm-planning.mjs +3 -2
  13. package/dist/es/ai-model/llm-planning.mjs.map +1 -1
  14. package/dist/es/ai-model/prompt/llm-planning.mjs +20 -95
  15. package/dist/es/ai-model/prompt/llm-planning.mjs.map +1 -1
  16. package/dist/es/ai-model/service-caller/index.mjs +54 -39
  17. package/dist/es/ai-model/service-caller/index.mjs.map +1 -1
  18. package/dist/es/device/index.mjs +13 -7
  19. package/dist/es/device/index.mjs.map +1 -1
  20. package/dist/es/utils.mjs +2 -2
  21. package/dist/lib/agent/agent.js +24 -9
  22. package/dist/lib/agent/agent.js.map +1 -1
  23. package/dist/lib/agent/common.js +1 -1
  24. package/dist/lib/agent/execution-session.js +2 -2
  25. package/dist/lib/agent/index.js +2 -2
  26. package/dist/lib/agent/task-builder.js +32 -27
  27. package/dist/lib/agent/task-builder.js.map +1 -1
  28. package/dist/lib/agent/task-cache.js +3 -3
  29. package/dist/lib/agent/task-cache.js.map +1 -1
  30. package/dist/lib/agent/tasks.js +19 -9
  31. package/dist/lib/agent/tasks.js.map +1 -1
  32. package/dist/lib/agent/ui-utils.js +2 -2
  33. package/dist/lib/agent/utils.js +3 -3
  34. package/dist/lib/ai-model/conversation-history.js +21 -4
  35. package/dist/lib/ai-model/conversation-history.js.map +1 -1
  36. package/dist/lib/ai-model/index.js +2 -2
  37. package/dist/lib/ai-model/inspect.js +2 -2
  38. package/dist/lib/ai-model/llm-planning.js +5 -4
  39. package/dist/lib/ai-model/llm-planning.js.map +1 -1
  40. package/dist/lib/ai-model/prompt/assertion.js +2 -2
  41. package/dist/lib/ai-model/prompt/common.js +2 -2
  42. package/dist/lib/ai-model/prompt/describe.js +2 -2
  43. package/dist/lib/ai-model/prompt/extraction.js +2 -2
  44. package/dist/lib/ai-model/prompt/llm-locator.js +2 -2
  45. package/dist/lib/ai-model/prompt/llm-planning.js +22 -100
  46. package/dist/lib/ai-model/prompt/llm-planning.js.map +1 -1
  47. package/dist/lib/ai-model/prompt/llm-section-locator.js +2 -2
  48. package/dist/lib/ai-model/prompt/order-sensitive-judge.js +2 -2
  49. package/dist/lib/ai-model/prompt/playwright-generator.js +2 -2
  50. package/dist/lib/ai-model/prompt/ui-tars-locator.js +2 -2
  51. package/dist/lib/ai-model/prompt/ui-tars-planning.js +2 -2
  52. package/dist/lib/ai-model/prompt/util.js +2 -2
  53. package/dist/lib/ai-model/prompt/yaml-generator.js +2 -2
  54. package/dist/lib/ai-model/service-caller/index.js +55 -43
  55. package/dist/lib/ai-model/service-caller/index.js.map +1 -1
  56. package/dist/lib/ai-model/ui-tars-planning.js +2 -2
  57. package/dist/lib/common.js +2 -2
  58. package/dist/lib/device/device-options.js +1 -1
  59. package/dist/lib/device/index.js +15 -9
  60. package/dist/lib/device/index.js.map +1 -1
  61. package/dist/lib/image/index.js +2 -2
  62. package/dist/lib/index.js +2 -2
  63. package/dist/lib/report.js +2 -2
  64. package/dist/lib/service/index.js +2 -2
  65. package/dist/lib/service/utils.js +2 -2
  66. package/dist/lib/task-runner.js +2 -2
  67. package/dist/lib/tree.js +2 -2
  68. package/dist/lib/types.js +3 -3
  69. package/dist/lib/utils.js +4 -4
  70. package/dist/lib/yaml/builder.js +2 -2
  71. package/dist/lib/yaml/index.js +4 -4
  72. package/dist/lib/yaml/player.js +2 -2
  73. package/dist/lib/yaml/utils.js +2 -2
  74. package/dist/lib/yaml.js +1 -1
  75. package/dist/types/agent/agent.d.ts +1 -1
  76. package/dist/types/agent/task-builder.d.ts +4 -2
  77. package/dist/types/agent/tasks.d.ts +5 -2
  78. package/dist/types/ai-model/conversation-history.d.ts +6 -1
  79. package/dist/types/ai-model/llm-planning.d.ts +1 -0
  80. package/dist/types/ai-model/prompt/llm-planning.d.ts +0 -2
  81. package/dist/types/ai-model/service-caller/index.d.ts +1 -3
  82. package/dist/types/device/index.d.ts +10 -2
  83. package/package.json +5 -5
@@ -1 +1 @@
1
- {"version":3,"file":"agent/task-builder.mjs","sources":["../../../src/agent/task-builder.ts"],"sourcesContent":["import { findAllMidsceneLocatorField, parseActionParam } from '@/ai-model';\nimport type { AbstractInterface } from '@/device';\nimport type Service from '@/service';\nimport type {\n DetailedLocateParam,\n ElementCacheFeature,\n ExecutionTaskActionApply,\n ExecutionTaskApply,\n ExecutionTaskHitBy,\n ExecutionTaskPlanningLocateApply,\n LocateResultElement,\n LocateResultWithDump,\n PlanningAction,\n PlanningActionParamSleep,\n PlanningLocateParam,\n Rect,\n ServiceDump,\n} from '@/types';\nimport { ServiceError } from '@/types';\nimport { sleep } from '@/utils';\nimport type { IModelConfig } from '@midscene/shared/env';\nimport { generateElementByPosition } from '@midscene/shared/extractor';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type { TaskCache } from './task-cache';\nimport {\n ifPlanLocateParamIsBbox,\n matchElementFromCache,\n matchElementFromPlan,\n} from './utils';\n\nconst debug = getDebug('agent:task-builder');\n\nexport function locatePlanForLocate(param: string | DetailedLocateParam) {\n const locate = typeof param === 'string' ? { prompt: param } : param;\n const locatePlan: PlanningAction<PlanningLocateParam> = {\n type: 'Locate',\n param: locate,\n thought: '',\n };\n return locatePlan;\n}\n\ninterface TaskBuilderDeps {\n interfaceInstance: AbstractInterface;\n service: Service;\n taskCache?: TaskCache;\n}\n\ninterface BuildOptions {\n cacheable?: boolean;\n subTask?: boolean;\n}\n\ninterface PlanBuildContext {\n tasks: ExecutionTaskApply[];\n modelConfigForPlanning: IModelConfig;\n modelConfigForDefaultIntent: IModelConfig;\n cacheable?: boolean;\n subTask: boolean;\n}\n\nexport class TaskBuilder {\n private readonly interface: AbstractInterface;\n\n private readonly service: Service;\n\n private readonly taskCache?: TaskCache;\n\n constructor({ interfaceInstance, service, taskCache }: TaskBuilderDeps) {\n this.interface = interfaceInstance;\n this.service = service;\n this.taskCache = taskCache;\n }\n\n public async build(\n plans: PlanningAction[],\n modelConfigForPlanning: IModelConfig,\n modelConfigForDefaultIntent: IModelConfig,\n options?: BuildOptions,\n ): Promise<{ tasks: ExecutionTaskApply[] }> {\n const tasks: ExecutionTaskApply[] = [];\n const cacheable = options?.cacheable;\n\n const context: PlanBuildContext = {\n tasks,\n modelConfigForPlanning,\n modelConfigForDefaultIntent,\n cacheable,\n subTask: !!options?.subTask,\n };\n\n type PlanHandler = (plan: PlanningAction) => Promise<void> | void;\n\n const planHandlers = new Map<string, PlanHandler>([\n [\n 'Locate',\n (plan) =>\n this.handleLocatePlan(\n plan as PlanningAction<PlanningLocateParam>,\n context,\n ),\n ],\n ['Finished', (plan) => this.handleFinishedPlan(plan, context)],\n [\n 'Sleep',\n (plan) =>\n this.handleSleepPlan(\n plan as PlanningAction<PlanningActionParamSleep>,\n context,\n ),\n ],\n ]);\n\n const defaultHandler: PlanHandler = (plan) =>\n this.handleActionPlan(plan, context);\n\n for (const plan of plans) {\n const handler = planHandlers.get(plan.type) ?? defaultHandler;\n await handler(plan);\n }\n\n return {\n tasks,\n };\n }\n\n private handleFinishedPlan(\n plan: PlanningAction,\n context: PlanBuildContext,\n ): void {\n const taskActionFinished: ExecutionTaskActionApply<null> = {\n type: 'Action Space',\n subType: 'Finished',\n param: null,\n thought: plan.thought,\n subTask: context.subTask || undefined,\n executor: async () => {},\n };\n context.tasks.push(taskActionFinished);\n }\n\n private handleSleepPlan(\n plan: PlanningAction<PlanningActionParamSleep>,\n context: PlanBuildContext,\n ): void {\n const sleepTask = this.createSleepTask(plan.param, {\n thought: plan.thought,\n });\n if (context.subTask) {\n sleepTask.subTask = true;\n }\n context.tasks.push(sleepTask);\n }\n\n public createSleepTask(\n param: PlanningActionParamSleep,\n meta?: { thought?: string },\n ): ExecutionTaskActionApply<PlanningActionParamSleep> {\n return {\n type: 'Action Space',\n subType: 'Sleep',\n param,\n thought: meta?.thought,\n executor: async (taskParam) => {\n await sleep(taskParam?.timeMs || 3000);\n },\n };\n }\n\n private async handleLocatePlan(\n plan: PlanningAction<PlanningLocateParam>,\n context: PlanBuildContext,\n ): Promise<void> {\n const taskLocate = this.createLocateTask(plan, plan.param, context);\n context.tasks.push(taskLocate);\n }\n\n private async handleActionPlan(\n plan: PlanningAction,\n context: PlanBuildContext,\n ): Promise<void> {\n const planType = plan.type;\n const actionSpace = await this.interface.actionSpace();\n const action = actionSpace.find((item) => item.name === planType);\n const param = plan.param;\n\n if (!action) {\n throw new Error(`Action type '${planType}' not found`);\n }\n\n const locateFields = action\n ? findAllMidsceneLocatorField(action.paramSchema)\n : [];\n\n const requiredLocateFields = action\n ? findAllMidsceneLocatorField(action.paramSchema, true)\n : [];\n\n locateFields.forEach((field) => {\n if (param[field]) {\n if (ifPlanLocateParamIsBbox(param[field])) {\n debug(\n 'plan locate param is bbox, will match element from plan',\n param[field],\n );\n const elementFromPlan = matchElementFromPlan(param[field]);\n if (elementFromPlan) {\n param[field] = elementFromPlan;\n }\n } else {\n const locatePlan = locatePlanForLocate(param[field]);\n debug(\n 'will prepend locate param for field',\n `action.type=${planType}`,\n `param=${JSON.stringify(param[field])}`,\n `locatePlan=${JSON.stringify(locatePlan)}`,\n );\n const locateTask = this.createLocateTask(\n locatePlan,\n param[field],\n context,\n (result) => {\n param[field] = result;\n },\n );\n context.tasks.push(locateTask);\n }\n } else {\n assert(\n !requiredLocateFields.includes(field),\n `Required locate field '${field}' is not provided for action ${planType}`,\n );\n debug(`field '${field}' is not provided for action ${planType}`);\n }\n });\n\n const task: ExecutionTaskApply<\n 'Action Space',\n any,\n { success: boolean; action: string; param: any },\n void\n > = {\n type: 'Action Space',\n subType: planType,\n thought: plan.thought,\n param: plan.param,\n subTask: context.subTask || undefined,\n executor: async (param, taskContext) => {\n debug(\n 'executing action',\n planType,\n param,\n `taskContext.element.center: ${taskContext.element?.center}`,\n );\n\n const uiContext = taskContext.uiContext;\n assert(uiContext, 'uiContext is required for Action task');\n\n requiredLocateFields.forEach((field) => {\n assert(\n param[field],\n `field '${field}' is required for action ${planType} but not provided. Cannot execute action ${planType}.`,\n );\n });\n\n try {\n await Promise.all([\n (async () => {\n if (this.interface.beforeInvokeAction) {\n debug('will call \"beforeInvokeAction\" for interface');\n await this.interface.beforeInvokeAction(action.name, param);\n debug('called \"beforeInvokeAction\" for interface');\n }\n })(),\n sleep(200),\n ]);\n } catch (originalError: any) {\n const originalMessage =\n originalError?.message || String(originalError);\n throw new Error(\n `error in running beforeInvokeAction for ${action.name}: ${originalMessage}`,\n { cause: originalError },\n );\n }\n\n if (action.paramSchema) {\n try {\n param = parseActionParam(param, action.paramSchema);\n } catch (error: any) {\n throw new Error(\n `Invalid parameters for action ${action.name}: ${error.message}\\nParameters: ${JSON.stringify(param)}`,\n { cause: error },\n );\n }\n }\n\n debug('calling action', action.name);\n const actionFn = action.call.bind(this.interface);\n const actionResult = await actionFn(param, taskContext);\n debug('called action', action.name, 'result:', actionResult);\n\n const delayAfterRunner = action.delayAfterRunner ?? 300;\n if (delayAfterRunner > 0) {\n await sleep(delayAfterRunner);\n }\n\n try {\n if (this.interface.afterInvokeAction) {\n debug('will call \"afterInvokeAction\" for interface');\n await this.interface.afterInvokeAction(action.name, param);\n debug('called \"afterInvokeAction\" for interface');\n }\n } catch (originalError: any) {\n const originalMessage =\n originalError?.message || String(originalError);\n throw new Error(\n `error in running afterInvokeAction for ${action.name}: ${originalMessage}`,\n { cause: originalError },\n );\n }\n\n return {\n output: actionResult,\n };\n },\n };\n\n context.tasks.push(task);\n }\n\n private createLocateTask(\n plan: PlanningAction<PlanningLocateParam>,\n detailedLocateParam: DetailedLocateParam | string,\n context: PlanBuildContext,\n onResult?: (result: LocateResultElement) => void,\n ): ExecutionTaskPlanningLocateApply {\n const { cacheable, modelConfigForDefaultIntent } = context;\n\n let locateParam = detailedLocateParam;\n debug(\n 'modelConfigForDefaultIntent-locateParam',\n modelConfigForDefaultIntent,\n locateParam,\n );\n\n if (typeof locateParam === 'string') {\n locateParam = {\n prompt: locateParam,\n };\n }\n\n if (cacheable !== undefined) {\n locateParam = {\n ...locateParam,\n cacheable,\n };\n }\n\n const taskLocator: ExecutionTaskPlanningLocateApply = {\n type: 'Planning',\n subType: 'Locate',\n subTask: context.subTask || undefined,\n param: locateParam,\n thought: plan.thought,\n executor: async (param, taskContext) => {\n const { task } = taskContext;\n let { uiContext } = taskContext;\n\n assert(\n param?.prompt || param?.bbox,\n `No prompt or id or position or bbox to locate, param=${JSON.stringify(\n param,\n )}`,\n );\n\n if (!uiContext) {\n uiContext = await this.service.contextRetrieverFn();\n }\n\n assert(uiContext, 'uiContext is required for Service task');\n\n let locateDump: ServiceDump | undefined;\n let locateResult: LocateResultWithDump | undefined;\n\n const applyDump = (dump?: ServiceDump) => {\n if (!dump) {\n return;\n }\n locateDump = dump;\n task.log = {\n dump,\n };\n task.usage = dump.taskInfo?.usage;\n if (dump.taskInfo?.searchAreaUsage) {\n task.searchAreaUsage = dump.taskInfo.searchAreaUsage;\n }\n };\n\n // from xpath\n let rectFromXpath: Rect | undefined;\n if (param.xpath && this.interface.rectMatchesCacheFeature) {\n try {\n rectFromXpath = await this.interface.rectMatchesCacheFeature({\n xpaths: [param.xpath],\n });\n } catch (error) {\n // xpath locate failed, allow fallback to cache or AI locate\n rectFromXpath = undefined;\n }\n }\n const elementFromXpath = rectFromXpath\n ? generateElementByPosition(\n {\n x: rectFromXpath.left + rectFromXpath.width / 2,\n y: rectFromXpath.top + rectFromXpath.height / 2,\n },\n typeof param.prompt === 'string'\n ? param.prompt\n : param.prompt?.prompt || '',\n )\n : undefined;\n const userExpectedPathHitFlag = !!elementFromXpath;\n\n const cachePrompt = param.prompt;\n const locateCacheRecord = this.taskCache?.matchLocateCache(cachePrompt);\n const cacheEntry = locateCacheRecord?.cacheContent?.cache;\n\n const elementFromCache = userExpectedPathHitFlag\n ? null\n : await matchElementFromCache(\n {\n taskCache: this.taskCache,\n interfaceInstance: this.interface,\n },\n cacheEntry,\n cachePrompt,\n param.cacheable,\n );\n const cacheHitFlag = !!elementFromCache;\n\n const planHitFlag = false;\n\n let elementFromAiLocate: LocateResultElement | null | undefined;\n if (!userExpectedPathHitFlag && !cacheHitFlag && !planHitFlag) {\n try {\n locateResult = await this.service.locate(\n param,\n {\n context: uiContext,\n },\n modelConfigForDefaultIntent,\n );\n applyDump(locateResult.dump);\n elementFromAiLocate = locateResult.element;\n } catch (error) {\n if (error instanceof ServiceError) {\n applyDump(error.dump);\n }\n throw error;\n }\n }\n\n const element =\n elementFromXpath || elementFromCache || elementFromAiLocate;\n\n let currentCacheEntry: ElementCacheFeature | undefined;\n if (\n element &&\n this.taskCache &&\n !cacheHitFlag &&\n param?.cacheable !== false\n ) {\n if (this.interface.cacheFeatureForRect) {\n try {\n const feature = await this.interface.cacheFeatureForRect(\n element.rect,\n {\n targetDescription:\n typeof param.prompt === 'string'\n ? param.prompt\n : param.prompt?.prompt,\n modelConfig: modelConfigForDefaultIntent,\n },\n );\n if (feature && Object.keys(feature).length > 0) {\n debug(\n 'update cache, prompt: %s, cache: %o',\n cachePrompt,\n feature,\n );\n currentCacheEntry = feature;\n this.taskCache.updateOrAppendCacheRecord(\n {\n type: 'locate',\n prompt: cachePrompt,\n cache: feature,\n },\n locateCacheRecord,\n );\n } else {\n debug(\n 'no cache data returned, skip cache update, prompt: %s',\n cachePrompt,\n );\n }\n } catch (error) {\n debug('cacheFeatureForRect failed: %s', error);\n }\n } else {\n debug('cacheFeatureForRect is not supported, skip cache update');\n }\n }\n\n if (!element) {\n if (locateDump) {\n throw new ServiceError(\n `Element not found : ${param.prompt}`,\n locateDump,\n );\n }\n throw new Error(`Element not found: ${param.prompt}`);\n }\n\n let hitBy: ExecutionTaskHitBy | undefined;\n\n if (userExpectedPathHitFlag) {\n hitBy = {\n from: 'User expected path',\n context: {\n xpath: param.xpath,\n },\n };\n } else if (cacheHitFlag) {\n hitBy = {\n from: 'Cache',\n context: {\n cacheEntry,\n cacheToSave: currentCacheEntry,\n },\n };\n }\n\n onResult?.(element);\n\n return {\n output: {\n element,\n },\n hitBy,\n };\n },\n };\n\n return taskLocator;\n }\n}\n"],"names":["debug","getDebug","locatePlanForLocate","param","locate","locatePlan","TaskBuilder","plans","modelConfigForPlanning","modelConfigForDefaultIntent","options","tasks","cacheable","context","planHandlers","Map","plan","defaultHandler","handler","taskActionFinished","undefined","sleepTask","meta","taskParam","sleep","taskLocate","planType","actionSpace","action","item","Error","locateFields","findAllMidsceneLocatorField","requiredLocateFields","field","ifPlanLocateParamIsBbox","elementFromPlan","matchElementFromPlan","JSON","locateTask","result","assert","task","taskContext","uiContext","Promise","originalError","originalMessage","String","parseActionParam","error","actionFn","actionResult","delayAfterRunner","detailedLocateParam","onResult","locateParam","taskLocator","locateDump","locateResult","applyDump","dump","rectFromXpath","elementFromXpath","generateElementByPosition","userExpectedPathHitFlag","cachePrompt","locateCacheRecord","cacheEntry","elementFromCache","matchElementFromCache","cacheHitFlag","planHitFlag","elementFromAiLocate","ServiceError","element","currentCacheEntry","feature","Object","hitBy","interfaceInstance","service","taskCache"],"mappings":";;;;;;;;;;;;;;;;;AA+BA,MAAMA,QAAQC,SAAS;AAEhB,SAASC,oBAAoBC,KAAmC;IACrE,MAAMC,SAAS,AAAiB,YAAjB,OAAOD,QAAqB;QAAE,QAAQA;IAAM,IAAIA;IAC/D,MAAME,aAAkD;QACtD,MAAM;QACN,OAAOD;QACP,SAAS;IACX;IACA,OAAOC;AACT;AAqBO,MAAMC;IAaX,MAAa,MACXC,KAAuB,EACvBC,sBAAoC,EACpCC,2BAAyC,EACzCC,OAAsB,EACoB;QAC1C,MAAMC,QAA8B,EAAE;QACtC,MAAMC,YAAYF,SAAS;QAE3B,MAAMG,UAA4B;YAChCF;YACAH;YACAC;YACAG;YACA,SAAS,CAAC,CAACF,SAAS;QACtB;QAIA,MAAMI,eAAe,IAAIC,IAAyB;YAChD;gBACE;gBACA,CAACC,OACC,IAAI,CAAC,gBAAgB,CACnBA,MACAH;aAEL;YACD;gBAAC;gBAAY,CAACG,OAAS,IAAI,CAAC,kBAAkB,CAACA,MAAMH;aAAS;YAC9D;gBACE;gBACA,CAACG,OACC,IAAI,CAAC,eAAe,CAClBA,MACAH;aAEL;SACF;QAED,MAAMI,iBAA8B,CAACD,OACnC,IAAI,CAAC,gBAAgB,CAACA,MAAMH;QAE9B,KAAK,MAAMG,QAAQT,MAAO;YACxB,MAAMW,UAAUJ,aAAa,GAAG,CAACE,KAAK,IAAI,KAAKC;YAC/C,MAAMC,QAAQF;QAChB;QAEA,OAAO;YACLL;QACF;IACF;IAEQ,mBACNK,IAAoB,EACpBH,OAAyB,EACnB;QACN,MAAMM,qBAAqD;YACzD,MAAM;YACN,SAAS;YACT,OAAO;YACP,SAASH,KAAK,OAAO;YACrB,SAASH,QAAQ,OAAO,IAAIO;YAC5B,UAAU,WAAa;QACzB;QACAP,QAAQ,KAAK,CAAC,IAAI,CAACM;IACrB;IAEQ,gBACNH,IAA8C,EAC9CH,OAAyB,EACnB;QACN,MAAMQ,YAAY,IAAI,CAAC,eAAe,CAACL,KAAK,KAAK,EAAE;YACjD,SAASA,KAAK,OAAO;QACvB;QACA,IAAIH,QAAQ,OAAO,EACjBQ,UAAU,OAAO,GAAG;QAEtBR,QAAQ,KAAK,CAAC,IAAI,CAACQ;IACrB;IAEO,gBACLlB,KAA+B,EAC/BmB,IAA2B,EACyB;QACpD,OAAO;YACL,MAAM;YACN,SAAS;YACTnB;YACA,SAASmB,MAAM;YACf,UAAU,OAAOC;gBACf,MAAMC,MAAMD,WAAW,UAAU;YACnC;QACF;IACF;IAEA,MAAc,iBACZP,IAAyC,EACzCH,OAAyB,EACV;QACf,MAAMY,aAAa,IAAI,CAAC,gBAAgB,CAACT,MAAMA,KAAK,KAAK,EAAEH;QAC3DA,QAAQ,KAAK,CAAC,IAAI,CAACY;IACrB;IAEA,MAAc,iBACZT,IAAoB,EACpBH,OAAyB,EACV;QACf,MAAMa,WAAWV,KAAK,IAAI;QAC1B,MAAMW,cAAc,MAAM,IAAI,CAAC,SAAS,CAAC,WAAW;QACpD,MAAMC,SAASD,YAAY,IAAI,CAAC,CAACE,OAASA,KAAK,IAAI,KAAKH;QACxD,MAAMvB,QAAQa,KAAK,KAAK;QAExB,IAAI,CAACY,QACH,MAAM,IAAIE,MAAM,CAAC,aAAa,EAAEJ,SAAS,WAAW,CAAC;QAGvD,MAAMK,eAAeH,SACjBI,4BAA4BJ,OAAO,WAAW,IAC9C,EAAE;QAEN,MAAMK,uBAAuBL,SACzBI,4BAA4BJ,OAAO,WAAW,EAAE,QAChD,EAAE;QAENG,aAAa,OAAO,CAAC,CAACG;YACpB,IAAI/B,KAAK,CAAC+B,MAAM,EACd,IAAIC,wBAAwBhC,KAAK,CAAC+B,MAAM,GAAG;gBACzClC,MACE,2DACAG,KAAK,CAAC+B,MAAM;gBAEd,MAAME,kBAAkBC,qBAAqBlC,KAAK,CAAC+B,MAAM;gBACzD,IAAIE,iBACFjC,KAAK,CAAC+B,MAAM,GAAGE;YAEnB,OAAO;gBACL,MAAM/B,aAAaH,oBAAoBC,KAAK,CAAC+B,MAAM;gBACnDlC,MACE,uCACA,CAAC,YAAY,EAAE0B,UAAU,EACzB,CAAC,MAAM,EAAEY,KAAK,SAAS,CAACnC,KAAK,CAAC+B,MAAM,GAAG,EACvC,CAAC,WAAW,EAAEI,KAAK,SAAS,CAACjC,aAAa;gBAE5C,MAAMkC,aAAa,IAAI,CAAC,gBAAgB,CACtClC,YACAF,KAAK,CAAC+B,MAAM,EACZrB,SACA,CAAC2B;oBACCrC,KAAK,CAAC+B,MAAM,GAAGM;gBACjB;gBAEF3B,QAAQ,KAAK,CAAC,IAAI,CAAC0B;YACrB;iBACK;gBACLE,OACE,CAACR,qBAAqB,QAAQ,CAACC,QAC/B,CAAC,uBAAuB,EAAEA,MAAM,6BAA6B,EAAER,UAAU;gBAE3E1B,MAAM,CAAC,OAAO,EAAEkC,MAAM,6BAA6B,EAAER,UAAU;YACjE;QACF;QAEA,MAAMgB,OAKF;YACF,MAAM;YACN,SAAShB;YACT,SAASV,KAAK,OAAO;YACrB,OAAOA,KAAK,KAAK;YACjB,SAASH,QAAQ,OAAO,IAAIO;YAC5B,UAAU,OAAOjB,OAAOwC;gBACtB3C,MACE,oBACA0B,UACAvB,OACA,CAAC,4BAA4B,EAAEwC,YAAY,OAAO,EAAE,QAAQ;gBAG9D,MAAMC,YAAYD,YAAY,SAAS;gBACvCF,OAAOG,WAAW;gBAElBX,qBAAqB,OAAO,CAAC,CAACC;oBAC5BO,OACEtC,KAAK,CAAC+B,MAAM,EACZ,CAAC,OAAO,EAAEA,MAAM,yBAAyB,EAAER,SAAS,yCAAyC,EAAEA,SAAS,CAAC,CAAC;gBAE9G;gBAEA,IAAI;oBACF,MAAMmB,QAAQ,GAAG,CAAC;wBACf;4BACC,IAAI,IAAI,CAAC,SAAS,CAAC,kBAAkB,EAAE;gCACrC7C,MAAM;gCACN,MAAM,IAAI,CAAC,SAAS,CAAC,kBAAkB,CAAC4B,OAAO,IAAI,EAAEzB;gCACrDH,MAAM;4BACR;wBACF;wBACAwB,MAAM;qBACP;gBACH,EAAE,OAAOsB,eAAoB;oBAC3B,MAAMC,kBACJD,eAAe,WAAWE,OAAOF;oBACnC,MAAM,IAAIhB,MACR,CAAC,wCAAwC,EAAEF,OAAO,IAAI,CAAC,EAAE,EAAEmB,iBAAiB,EAC5E;wBAAE,OAAOD;oBAAc;gBAE3B;gBAEA,IAAIlB,OAAO,WAAW,EACpB,IAAI;oBACFzB,QAAQ8C,iBAAiB9C,OAAOyB,OAAO,WAAW;gBACpD,EAAE,OAAOsB,OAAY;oBACnB,MAAM,IAAIpB,MACR,CAAC,8BAA8B,EAAEF,OAAO,IAAI,CAAC,EAAE,EAAEsB,MAAM,OAAO,CAAC,cAAc,EAAEZ,KAAK,SAAS,CAACnC,QAAQ,EACtG;wBAAE,OAAO+C;oBAAM;gBAEnB;gBAGFlD,MAAM,kBAAkB4B,OAAO,IAAI;gBACnC,MAAMuB,WAAWvB,OAAO,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS;gBAChD,MAAMwB,eAAe,MAAMD,SAAShD,OAAOwC;gBAC3C3C,MAAM,iBAAiB4B,OAAO,IAAI,EAAE,WAAWwB;gBAE/C,MAAMC,mBAAmBzB,OAAO,gBAAgB,IAAI;gBACpD,IAAIyB,mBAAmB,GACrB,MAAM7B,MAAM6B;gBAGd,IAAI;oBACF,IAAI,IAAI,CAAC,SAAS,CAAC,iBAAiB,EAAE;wBACpCrD,MAAM;wBACN,MAAM,IAAI,CAAC,SAAS,CAAC,iBAAiB,CAAC4B,OAAO,IAAI,EAAEzB;wBACpDH,MAAM;oBACR;gBACF,EAAE,OAAO8C,eAAoB;oBAC3B,MAAMC,kBACJD,eAAe,WAAWE,OAAOF;oBACnC,MAAM,IAAIhB,MACR,CAAC,uCAAuC,EAAEF,OAAO,IAAI,CAAC,EAAE,EAAEmB,iBAAiB,EAC3E;wBAAE,OAAOD;oBAAc;gBAE3B;gBAEA,OAAO;oBACL,QAAQM;gBACV;YACF;QACF;QAEAvC,QAAQ,KAAK,CAAC,IAAI,CAAC6B;IACrB;IAEQ,iBACN1B,IAAyC,EACzCsC,mBAAiD,EACjDzC,OAAyB,EACzB0C,QAAgD,EACd;QAClC,MAAM,EAAE3C,SAAS,EAAEH,2BAA2B,EAAE,GAAGI;QAEnD,IAAI2C,cAAcF;QAClBtD,MACE,2CACAS,6BACA+C;QAGF,IAAI,AAAuB,YAAvB,OAAOA,aACTA,cAAc;YACZ,QAAQA;QACV;QAGF,IAAI5C,AAAcQ,WAAdR,WACF4C,cAAc;YACZ,GAAGA,WAAW;YACd5C;QACF;QAGF,MAAM6C,cAAgD;YACpD,MAAM;YACN,SAAS;YACT,SAAS5C,QAAQ,OAAO,IAAIO;YAC5B,OAAOoC;YACP,SAASxC,KAAK,OAAO;YACrB,UAAU,OAAOb,OAAOwC;gBACtB,MAAM,EAAED,IAAI,EAAE,GAAGC;gBACjB,IAAI,EAAEC,SAAS,EAAE,GAAGD;gBAEpBF,OACEtC,OAAO,UAAUA,OAAO,MACxB,CAAC,qDAAqD,EAAEmC,KAAK,SAAS,CACpEnC,QACC;gBAGL,IAAI,CAACyC,WACHA,YAAY,MAAM,IAAI,CAAC,OAAO,CAAC,kBAAkB;gBAGnDH,OAAOG,WAAW;gBAElB,IAAIc;gBACJ,IAAIC;gBAEJ,MAAMC,YAAY,CAACC;oBACjB,IAAI,CAACA,MACH;oBAEFH,aAAaG;oBACbnB,KAAK,GAAG,GAAG;wBACTmB;oBACF;oBACAnB,KAAK,KAAK,GAAGmB,KAAK,QAAQ,EAAE;oBAC5B,IAAIA,KAAK,QAAQ,EAAE,iBACjBnB,KAAK,eAAe,GAAGmB,KAAK,QAAQ,CAAC,eAAe;gBAExD;gBAGA,IAAIC;gBACJ,IAAI3D,MAAM,KAAK,IAAI,IAAI,CAAC,SAAS,CAAC,uBAAuB,EACvD,IAAI;oBACF2D,gBAAgB,MAAM,IAAI,CAAC,SAAS,CAAC,uBAAuB,CAAC;wBAC3D,QAAQ;4BAAC3D,MAAM,KAAK;yBAAC;oBACvB;gBACF,EAAE,OAAO+C,OAAO;oBAEdY,gBAAgB1C;gBAClB;gBAEF,MAAM2C,mBAAmBD,gBACrBE,0BACE;oBACE,GAAGF,cAAc,IAAI,GAAGA,cAAc,KAAK,GAAG;oBAC9C,GAAGA,cAAc,GAAG,GAAGA,cAAc,MAAM,GAAG;gBAChD,GACA,AAAwB,YAAxB,OAAO3D,MAAM,MAAM,GACfA,MAAM,MAAM,GACZA,MAAM,MAAM,EAAE,UAAU,MAE9BiB;gBACJ,MAAM6C,0BAA0B,CAAC,CAACF;gBAElC,MAAMG,cAAc/D,MAAM,MAAM;gBAChC,MAAMgE,oBAAoB,IAAI,CAAC,SAAS,EAAE,iBAAiBD;gBAC3D,MAAME,aAAaD,mBAAmB,cAAc;gBAEpD,MAAME,mBAAmBJ,0BACrB,OACA,MAAMK,sBACJ;oBACE,WAAW,IAAI,CAAC,SAAS;oBACzB,mBAAmB,IAAI,CAAC,SAAS;gBACnC,GACAF,YACAF,aACA/D,MAAM,SAAS;gBAErB,MAAMoE,eAAe,CAAC,CAACF;gBAEvB,MAAMG,cAAc;gBAEpB,IAAIC;gBACJ,IAAI,CAACR,2BAA2B,CAACM,gBAAgB,CAACC,aAChD,IAAI;oBACFb,eAAe,MAAM,IAAI,CAAC,OAAO,CAAC,MAAM,CACtCxD,OACA;wBACE,SAASyC;oBACX,GACAnC;oBAEFmD,UAAUD,aAAa,IAAI;oBAC3Bc,sBAAsBd,aAAa,OAAO;gBAC5C,EAAE,OAAOT,OAAO;oBACd,IAAIA,iBAAiBwB,cACnBd,UAAUV,MAAM,IAAI;oBAEtB,MAAMA;gBACR;gBAGF,MAAMyB,UACJZ,oBAAoBM,oBAAoBI;gBAE1C,IAAIG;gBACJ,IACED,WACA,IAAI,CAAC,SAAS,IACd,CAACJ,gBACDpE,OAAO,cAAc,OAErB,IAAI,IAAI,CAAC,SAAS,CAAC,mBAAmB,EACpC,IAAI;oBACF,MAAM0E,UAAU,MAAM,IAAI,CAAC,SAAS,CAAC,mBAAmB,CACtDF,QAAQ,IAAI,EACZ;wBACE,mBACE,AAAwB,YAAxB,OAAOxE,MAAM,MAAM,GACfA,MAAM,MAAM,GACZA,MAAM,MAAM,EAAE;wBACpB,aAAaM;oBACf;oBAEF,IAAIoE,WAAWC,OAAO,IAAI,CAACD,SAAS,MAAM,GAAG,GAAG;wBAC9C7E,MACE,uCACAkE,aACAW;wBAEFD,oBAAoBC;wBACpB,IAAI,CAAC,SAAS,CAAC,yBAAyB,CACtC;4BACE,MAAM;4BACN,QAAQX;4BACR,OAAOW;wBACT,GACAV;oBAEJ,OACEnE,MACE,yDACAkE;gBAGN,EAAE,OAAOhB,OAAO;oBACdlD,MAAM,kCAAkCkD;gBAC1C;qBAEAlD,MAAM;gBAIV,IAAI,CAAC2E,SAAS;oBACZ,IAAIjB,YACF,MAAM,IAAIgB,aACR,CAAC,oBAAoB,EAAEvE,MAAM,MAAM,EAAE,EACrCuD;oBAGJ,MAAM,IAAI5B,MAAM,CAAC,mBAAmB,EAAE3B,MAAM,MAAM,EAAE;gBACtD;gBAEA,IAAI4E;gBAEJ,IAAId,yBACFc,QAAQ;oBACN,MAAM;oBACN,SAAS;wBACP,OAAO5E,MAAM,KAAK;oBACpB;gBACF;qBACK,IAAIoE,cACTQ,QAAQ;oBACN,MAAM;oBACN,SAAS;wBACPX;wBACA,aAAaQ;oBACf;gBACF;gBAGFrB,WAAWoB;gBAEX,OAAO;oBACL,QAAQ;wBACNA;oBACF;oBACAI;gBACF;YACF;QACF;QAEA,OAAOtB;IACT;IAteA,YAAY,EAAEuB,iBAAiB,EAAEC,OAAO,EAAEC,SAAS,EAAmB,CAAE;QANxE,uBAAiB,aAAjB;QAEA,uBAAiB,WAAjB;QAEA,uBAAiB,aAAjB;QAGE,IAAI,CAAC,SAAS,GAAGF;QACjB,IAAI,CAAC,OAAO,GAAGC;QACf,IAAI,CAAC,SAAS,GAAGC;IACnB;AAmeF"}
1
+ {"version":3,"file":"agent/task-builder.mjs","sources":["../../../src/agent/task-builder.ts"],"sourcesContent":["import { findAllMidsceneLocatorField, parseActionParam } from '@/ai-model';\nimport type { AbstractInterface } from '@/device';\nimport type Service from '@/service';\nimport type {\n DetailedLocateParam,\n DeviceAction,\n ElementCacheFeature,\n ExecutionTaskActionApply,\n ExecutionTaskApply,\n ExecutionTaskHitBy,\n ExecutionTaskPlanningLocateApply,\n LocateResultElement,\n LocateResultWithDump,\n PlanningAction,\n PlanningActionParamSleep,\n PlanningLocateParam,\n Rect,\n ServiceDump,\n} from '@/types';\nimport { ServiceError } from '@/types';\nimport { sleep } from '@/utils';\nimport type { IModelConfig } from '@midscene/shared/env';\nimport { generateElementByPosition } from '@midscene/shared/extractor';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type { TaskCache } from './task-cache';\nimport {\n ifPlanLocateParamIsBbox,\n matchElementFromCache,\n matchElementFromPlan,\n} from './utils';\n\nconst debug = getDebug('agent:task-builder');\n\n/**\n * Check if a cache object is non-empty\n */\nfunction hasNonEmptyCache(cache: unknown): boolean {\n return (\n cache !== null &&\n cache !== undefined &&\n typeof cache === 'object' &&\n Object.keys(cache).length > 0\n );\n}\n\nexport function locatePlanForLocate(param: string | DetailedLocateParam) {\n const locate = typeof param === 'string' ? { prompt: param } : param;\n const locatePlan: PlanningAction<PlanningLocateParam> = {\n type: 'Locate',\n param: locate,\n thought: '',\n };\n return locatePlan;\n}\n\ninterface TaskBuilderDeps {\n interfaceInstance: AbstractInterface;\n service: Service;\n taskCache?: TaskCache;\n actionSpace: DeviceAction[];\n}\n\ninterface BuildOptions {\n cacheable?: boolean;\n subTask?: boolean;\n}\n\ninterface PlanBuildContext {\n tasks: ExecutionTaskApply[];\n modelConfigForPlanning: IModelConfig;\n modelConfigForDefaultIntent: IModelConfig;\n cacheable?: boolean;\n subTask: boolean;\n}\n\nexport class TaskBuilder {\n private readonly interface: AbstractInterface;\n\n private readonly service: Service;\n\n private readonly taskCache?: TaskCache;\n\n private readonly actionSpace: DeviceAction[];\n\n constructor({\n interfaceInstance,\n service,\n taskCache,\n actionSpace,\n }: TaskBuilderDeps) {\n this.interface = interfaceInstance;\n this.service = service;\n this.taskCache = taskCache;\n this.actionSpace = actionSpace;\n }\n\n public async build(\n plans: PlanningAction[],\n modelConfigForPlanning: IModelConfig,\n modelConfigForDefaultIntent: IModelConfig,\n options?: BuildOptions,\n ): Promise<{ tasks: ExecutionTaskApply[] }> {\n const tasks: ExecutionTaskApply[] = [];\n const cacheable = options?.cacheable;\n\n const context: PlanBuildContext = {\n tasks,\n modelConfigForPlanning,\n modelConfigForDefaultIntent,\n cacheable,\n subTask: !!options?.subTask,\n };\n\n type PlanHandler = (plan: PlanningAction) => Promise<void> | void;\n\n const planHandlers = new Map<string, PlanHandler>([\n [\n 'Locate',\n (plan) =>\n this.handleLocatePlan(\n plan as PlanningAction<PlanningLocateParam>,\n context,\n ),\n ],\n ['Finished', (plan) => this.handleFinishedPlan(plan, context)],\n [\n 'Sleep',\n (plan) =>\n this.handleSleepPlan(\n plan as PlanningAction<PlanningActionParamSleep>,\n context,\n ),\n ],\n ]);\n\n const defaultHandler: PlanHandler = (plan) =>\n this.handleActionPlan(plan, context);\n\n for (const plan of plans) {\n const handler = planHandlers.get(plan.type) ?? defaultHandler;\n await handler(plan);\n }\n\n return {\n tasks,\n };\n }\n\n private handleFinishedPlan(\n plan: PlanningAction,\n context: PlanBuildContext,\n ): void {\n const taskActionFinished: ExecutionTaskActionApply<null> = {\n type: 'Action Space',\n subType: 'Finished',\n param: null,\n thought: plan.thought,\n subTask: context.subTask || undefined,\n executor: async () => {},\n };\n context.tasks.push(taskActionFinished);\n }\n\n private handleSleepPlan(\n plan: PlanningAction<PlanningActionParamSleep>,\n context: PlanBuildContext,\n ): void {\n const sleepTask = this.createSleepTask(plan.param, {\n thought: plan.thought,\n });\n if (context.subTask) {\n sleepTask.subTask = true;\n }\n context.tasks.push(sleepTask);\n }\n\n public createSleepTask(\n param: PlanningActionParamSleep,\n meta?: { thought?: string },\n ): ExecutionTaskActionApply<PlanningActionParamSleep> {\n return {\n type: 'Action Space',\n subType: 'Sleep',\n param,\n thought: meta?.thought,\n executor: async (taskParam) => {\n await sleep(taskParam?.timeMs || 3000);\n },\n };\n }\n\n private async handleLocatePlan(\n plan: PlanningAction<PlanningLocateParam>,\n context: PlanBuildContext,\n ): Promise<void> {\n const taskLocate = this.createLocateTask(plan, plan.param, context);\n context.tasks.push(taskLocate);\n }\n\n private async handleActionPlan(\n plan: PlanningAction,\n context: PlanBuildContext,\n ): Promise<void> {\n const planType = plan.type;\n const actionSpace = this.actionSpace;\n const action = actionSpace.find((item) => item.name === planType);\n const param = plan.param;\n\n if (!action) {\n throw new Error(`Action type '${planType}' not found`);\n }\n\n const locateFields = action\n ? findAllMidsceneLocatorField(action.paramSchema)\n : [];\n\n const requiredLocateFields = action\n ? findAllMidsceneLocatorField(action.paramSchema, true)\n : [];\n\n locateFields.forEach((field) => {\n if (param[field]) {\n // Always use createLocateTask for all locate params (including bbox)\n // This ensures cache writing happens even when bbox is available\n const locatePlan = locatePlanForLocate(param[field]);\n debug(\n 'will prepend locate param for field',\n `action.type=${planType}`,\n `param=${JSON.stringify(param[field])}`,\n `locatePlan=${JSON.stringify(locatePlan)}`,\n `hasBbox=${ifPlanLocateParamIsBbox(param[field])}`,\n );\n const locateTask = this.createLocateTask(\n locatePlan,\n param[field],\n context,\n (result) => {\n param[field] = result;\n },\n );\n context.tasks.push(locateTask);\n } else {\n assert(\n !requiredLocateFields.includes(field),\n `Required locate field '${field}' is not provided for action ${planType}`,\n );\n debug(`field '${field}' is not provided for action ${planType}`);\n }\n });\n\n const task: ExecutionTaskApply<\n 'Action Space',\n any,\n { success: boolean; action: string; param: any },\n void\n > = {\n type: 'Action Space',\n subType: planType,\n thought: plan.thought,\n param: plan.param,\n subTask: context.subTask || undefined,\n executor: async (param, taskContext) => {\n debug(\n 'executing action',\n planType,\n param,\n `taskContext.element.center: ${taskContext.element?.center}`,\n );\n\n const uiContext = taskContext.uiContext;\n assert(uiContext, 'uiContext is required for Action task');\n\n requiredLocateFields.forEach((field) => {\n assert(\n param[field],\n `field '${field}' is required for action ${planType} but not provided. Cannot execute action ${planType}.`,\n );\n });\n\n try {\n await Promise.all([\n (async () => {\n if (this.interface.beforeInvokeAction) {\n debug('will call \"beforeInvokeAction\" for interface');\n await this.interface.beforeInvokeAction(action.name, param);\n debug('called \"beforeInvokeAction\" for interface');\n }\n })(),\n sleep(200),\n ]);\n } catch (originalError: any) {\n const originalMessage =\n originalError?.message || String(originalError);\n throw new Error(\n `error in running beforeInvokeAction for ${action.name}: ${originalMessage}`,\n { cause: originalError },\n );\n }\n\n if (action.paramSchema) {\n try {\n param = parseActionParam(param, action.paramSchema);\n } catch (error: any) {\n throw new Error(\n `Invalid parameters for action ${action.name}: ${error.message}\\nParameters: ${JSON.stringify(param)}`,\n { cause: error },\n );\n }\n }\n\n debug('calling action', action.name);\n const actionFn = action.call.bind(this.interface);\n const actionResult = await actionFn(param, taskContext);\n debug('called action', action.name, 'result:', actionResult);\n\n const delayAfterRunner = action.delayAfterRunner ?? 300;\n if (delayAfterRunner > 0) {\n await sleep(delayAfterRunner);\n }\n\n try {\n if (this.interface.afterInvokeAction) {\n debug('will call \"afterInvokeAction\" for interface');\n await this.interface.afterInvokeAction(action.name, param);\n debug('called \"afterInvokeAction\" for interface');\n }\n } catch (originalError: any) {\n const originalMessage =\n originalError?.message || String(originalError);\n throw new Error(\n `error in running afterInvokeAction for ${action.name}: ${originalMessage}`,\n { cause: originalError },\n );\n }\n\n return {\n output: actionResult,\n };\n },\n };\n\n context.tasks.push(task);\n }\n\n private createLocateTask(\n plan: PlanningAction<PlanningLocateParam>,\n detailedLocateParam: DetailedLocateParam | string,\n context: PlanBuildContext,\n onResult?: (result: LocateResultElement) => void,\n ): ExecutionTaskPlanningLocateApply {\n const { cacheable, modelConfigForDefaultIntent } = context;\n\n let locateParam = detailedLocateParam;\n\n if (typeof locateParam === 'string') {\n locateParam = {\n prompt: locateParam,\n };\n }\n\n if (cacheable !== undefined) {\n locateParam = {\n ...locateParam,\n cacheable,\n };\n }\n\n const taskLocator: ExecutionTaskPlanningLocateApply = {\n type: 'Planning',\n subType: 'Locate',\n subTask: context.subTask || undefined,\n param: locateParam,\n thought: plan.thought,\n executor: async (param, taskContext) => {\n const { task } = taskContext;\n let { uiContext } = taskContext;\n\n assert(\n param?.prompt || param?.bbox,\n `No prompt or id or position or bbox to locate, param=${JSON.stringify(\n param,\n )}`,\n );\n\n if (!uiContext) {\n uiContext = await this.service.contextRetrieverFn();\n }\n\n assert(uiContext, 'uiContext is required for Service task');\n\n let locateDump: ServiceDump | undefined;\n let locateResult: LocateResultWithDump | undefined;\n\n const applyDump = (dump?: ServiceDump) => {\n if (!dump) {\n return;\n }\n locateDump = dump;\n task.log = {\n dump,\n };\n task.usage = dump.taskInfo?.usage;\n if (dump.taskInfo?.searchAreaUsage) {\n task.searchAreaUsage = dump.taskInfo.searchAreaUsage;\n }\n };\n\n // from bbox (plan hit)\n const elementFromBbox = ifPlanLocateParamIsBbox(param)\n ? matchElementFromPlan(param)\n : undefined;\n const isPlanHit = !!elementFromBbox;\n\n // from xpath\n let rectFromXpath: Rect | undefined;\n if (\n !isPlanHit &&\n param.xpath &&\n this.interface.rectMatchesCacheFeature\n ) {\n try {\n rectFromXpath = await this.interface.rectMatchesCacheFeature({\n xpaths: [param.xpath],\n });\n } catch {\n // xpath locate failed, allow fallback to cache or AI locate\n }\n }\n const elementFromXpath = rectFromXpath\n ? generateElementByPosition(\n {\n x: rectFromXpath.left + rectFromXpath.width / 2,\n y: rectFromXpath.top + rectFromXpath.height / 2,\n },\n typeof param.prompt === 'string'\n ? param.prompt\n : param.prompt?.prompt || '',\n )\n : undefined;\n const isXpathHit = !!elementFromXpath;\n\n const cachePrompt = param.prompt;\n const locateCacheRecord = this.taskCache?.matchLocateCache(cachePrompt);\n const cacheEntry = locateCacheRecord?.cacheContent?.cache;\n\n const elementFromCache =\n isPlanHit || isXpathHit\n ? null\n : await matchElementFromCache(\n {\n taskCache: this.taskCache,\n interfaceInstance: this.interface,\n },\n cacheEntry,\n cachePrompt,\n param.cacheable,\n );\n const isCacheHit = !!elementFromCache;\n\n let elementFromAiLocate: LocateResultElement | null | undefined;\n if (!isXpathHit && !isCacheHit && !isPlanHit) {\n try {\n locateResult = await this.service.locate(\n param,\n {\n context: uiContext,\n },\n modelConfigForDefaultIntent,\n );\n applyDump(locateResult.dump);\n elementFromAiLocate = locateResult.element;\n } catch (error) {\n if (error instanceof ServiceError) {\n applyDump(error.dump);\n }\n throw error;\n }\n }\n\n const element =\n elementFromBbox ||\n elementFromXpath ||\n elementFromCache ||\n elementFromAiLocate;\n\n // Check if locate cache already exists (for planHitFlag case)\n const locateCacheAlreadyExists = hasNonEmptyCache(\n locateCacheRecord?.cacheContent?.cache,\n );\n\n let currentCacheEntry: ElementCacheFeature | undefined;\n // Write cache if:\n // 1. element found\n // 2. taskCache enabled\n // 3. not a cache hit (otherwise we'd be writing what we just read)\n // 4. not already cached (for bbox/plan hit case, avoid redundant writes)\n // 5. cacheable is not explicitly false\n if (\n element &&\n this.taskCache &&\n !isCacheHit &&\n !locateCacheAlreadyExists &&\n param?.cacheable !== false\n ) {\n if (this.interface.cacheFeatureForRect) {\n try {\n const feature = await this.interface.cacheFeatureForRect(\n element.rect,\n {\n targetDescription:\n typeof param.prompt === 'string'\n ? param.prompt\n : param.prompt?.prompt,\n modelConfig: modelConfigForDefaultIntent,\n },\n );\n if (hasNonEmptyCache(feature)) {\n debug(\n 'update cache, prompt: %s, cache: %o',\n cachePrompt,\n feature,\n );\n currentCacheEntry = feature;\n this.taskCache.updateOrAppendCacheRecord(\n {\n type: 'locate',\n prompt: cachePrompt,\n cache: feature,\n },\n locateCacheRecord,\n );\n } else {\n debug(\n 'no cache data returned, skip cache update, prompt: %s',\n cachePrompt,\n );\n }\n } catch (error) {\n debug('cacheFeatureForRect failed: %s', error);\n }\n } else {\n debug('cacheFeatureForRect is not supported, skip cache update');\n }\n }\n\n if (!element) {\n if (locateDump) {\n throw new ServiceError(\n `Element not found : ${param.prompt}`,\n locateDump,\n );\n }\n throw new Error(`Element not found: ${param.prompt}`);\n }\n\n let hitBy: ExecutionTaskHitBy | undefined;\n\n if (isPlanHit) {\n hitBy = {\n from: 'Plan',\n context: {\n bbox: param.bbox,\n },\n };\n } else if (isXpathHit) {\n hitBy = {\n from: 'User expected path',\n context: {\n xpath: param.xpath,\n },\n };\n } else if (isCacheHit) {\n hitBy = {\n from: 'Cache',\n context: {\n cacheEntry,\n cacheToSave: currentCacheEntry,\n },\n };\n }\n\n onResult?.(element);\n\n return {\n output: {\n element,\n },\n hitBy,\n };\n },\n };\n\n return taskLocator;\n }\n}\n"],"names":["debug","getDebug","hasNonEmptyCache","cache","Object","locatePlanForLocate","param","locate","locatePlan","TaskBuilder","plans","modelConfigForPlanning","modelConfigForDefaultIntent","options","tasks","cacheable","context","planHandlers","Map","plan","defaultHandler","handler","taskActionFinished","undefined","sleepTask","meta","taskParam","sleep","taskLocate","planType","actionSpace","action","item","Error","locateFields","findAllMidsceneLocatorField","requiredLocateFields","field","JSON","ifPlanLocateParamIsBbox","locateTask","result","assert","task","taskContext","uiContext","Promise","originalError","originalMessage","String","parseActionParam","error","actionFn","actionResult","delayAfterRunner","detailedLocateParam","onResult","locateParam","taskLocator","locateDump","locateResult","applyDump","dump","elementFromBbox","matchElementFromPlan","isPlanHit","rectFromXpath","elementFromXpath","generateElementByPosition","isXpathHit","cachePrompt","locateCacheRecord","cacheEntry","elementFromCache","matchElementFromCache","isCacheHit","elementFromAiLocate","ServiceError","element","locateCacheAlreadyExists","currentCacheEntry","feature","hitBy","interfaceInstance","service","taskCache"],"mappings":";;;;;;;;;;;;;;;;;AAgCA,MAAMA,QAAQC,SAAS;AAKvB,SAASC,iBAAiBC,KAAc;IACtC,OACEA,QAAAA,SAEA,AAAiB,YAAjB,OAAOA,SACPC,OAAO,IAAI,CAACD,OAAO,MAAM,GAAG;AAEhC;AAEO,SAASE,oBAAoBC,KAAmC;IACrE,MAAMC,SAAS,AAAiB,YAAjB,OAAOD,QAAqB;QAAE,QAAQA;IAAM,IAAIA;IAC/D,MAAME,aAAkD;QACtD,MAAM;QACN,OAAOD;QACP,SAAS;IACX;IACA,OAAOC;AACT;AAsBO,MAAMC;IAqBX,MAAa,MACXC,KAAuB,EACvBC,sBAAoC,EACpCC,2BAAyC,EACzCC,OAAsB,EACoB;QAC1C,MAAMC,QAA8B,EAAE;QACtC,MAAMC,YAAYF,SAAS;QAE3B,MAAMG,UAA4B;YAChCF;YACAH;YACAC;YACAG;YACA,SAAS,CAAC,CAACF,SAAS;QACtB;QAIA,MAAMI,eAAe,IAAIC,IAAyB;YAChD;gBACE;gBACA,CAACC,OACC,IAAI,CAAC,gBAAgB,CACnBA,MACAH;aAEL;YACD;gBAAC;gBAAY,CAACG,OAAS,IAAI,CAAC,kBAAkB,CAACA,MAAMH;aAAS;YAC9D;gBACE;gBACA,CAACG,OACC,IAAI,CAAC,eAAe,CAClBA,MACAH;aAEL;SACF;QAED,MAAMI,iBAA8B,CAACD,OACnC,IAAI,CAAC,gBAAgB,CAACA,MAAMH;QAE9B,KAAK,MAAMG,QAAQT,MAAO;YACxB,MAAMW,UAAUJ,aAAa,GAAG,CAACE,KAAK,IAAI,KAAKC;YAC/C,MAAMC,QAAQF;QAChB;QAEA,OAAO;YACLL;QACF;IACF;IAEQ,mBACNK,IAAoB,EACpBH,OAAyB,EACnB;QACN,MAAMM,qBAAqD;YACzD,MAAM;YACN,SAAS;YACT,OAAO;YACP,SAASH,KAAK,OAAO;YACrB,SAASH,QAAQ,OAAO,IAAIO;YAC5B,UAAU,WAAa;QACzB;QACAP,QAAQ,KAAK,CAAC,IAAI,CAACM;IACrB;IAEQ,gBACNH,IAA8C,EAC9CH,OAAyB,EACnB;QACN,MAAMQ,YAAY,IAAI,CAAC,eAAe,CAACL,KAAK,KAAK,EAAE;YACjD,SAASA,KAAK,OAAO;QACvB;QACA,IAAIH,QAAQ,OAAO,EACjBQ,UAAU,OAAO,GAAG;QAEtBR,QAAQ,KAAK,CAAC,IAAI,CAACQ;IACrB;IAEO,gBACLlB,KAA+B,EAC/BmB,IAA2B,EACyB;QACpD,OAAO;YACL,MAAM;YACN,SAAS;YACTnB;YACA,SAASmB,MAAM;YACf,UAAU,OAAOC;gBACf,MAAMC,MAAMD,WAAW,UAAU;YACnC;QACF;IACF;IAEA,MAAc,iBACZP,IAAyC,EACzCH,OAAyB,EACV;QACf,MAAMY,aAAa,IAAI,CAAC,gBAAgB,CAACT,MAAMA,KAAK,KAAK,EAAEH;QAC3DA,QAAQ,KAAK,CAAC,IAAI,CAACY;IACrB;IAEA,MAAc,iBACZT,IAAoB,EACpBH,OAAyB,EACV;QACf,MAAMa,WAAWV,KAAK,IAAI;QAC1B,MAAMW,cAAc,IAAI,CAAC,WAAW;QACpC,MAAMC,SAASD,YAAY,IAAI,CAAC,CAACE,OAASA,KAAK,IAAI,KAAKH;QACxD,MAAMvB,QAAQa,KAAK,KAAK;QAExB,IAAI,CAACY,QACH,MAAM,IAAIE,MAAM,CAAC,aAAa,EAAEJ,SAAS,WAAW,CAAC;QAGvD,MAAMK,eAAeH,SACjBI,4BAA4BJ,OAAO,WAAW,IAC9C,EAAE;QAEN,MAAMK,uBAAuBL,SACzBI,4BAA4BJ,OAAO,WAAW,EAAE,QAChD,EAAE;QAENG,aAAa,OAAO,CAAC,CAACG;YACpB,IAAI/B,KAAK,CAAC+B,MAAM,EAAE;gBAGhB,MAAM7B,aAAaH,oBAAoBC,KAAK,CAAC+B,MAAM;gBACnDrC,MACE,uCACA,CAAC,YAAY,EAAE6B,UAAU,EACzB,CAAC,MAAM,EAAES,KAAK,SAAS,CAAChC,KAAK,CAAC+B,MAAM,GAAG,EACvC,CAAC,WAAW,EAAEC,KAAK,SAAS,CAAC9B,aAAa,EAC1C,CAAC,QAAQ,EAAE+B,wBAAwBjC,KAAK,CAAC+B,MAAM,GAAG;gBAEpD,MAAMG,aAAa,IAAI,CAAC,gBAAgB,CACtChC,YACAF,KAAK,CAAC+B,MAAM,EACZrB,SACA,CAACyB;oBACCnC,KAAK,CAAC+B,MAAM,GAAGI;gBACjB;gBAEFzB,QAAQ,KAAK,CAAC,IAAI,CAACwB;YACrB,OAAO;gBACLE,OACE,CAACN,qBAAqB,QAAQ,CAACC,QAC/B,CAAC,uBAAuB,EAAEA,MAAM,6BAA6B,EAAER,UAAU;gBAE3E7B,MAAM,CAAC,OAAO,EAAEqC,MAAM,6BAA6B,EAAER,UAAU;YACjE;QACF;QAEA,MAAMc,OAKF;YACF,MAAM;YACN,SAASd;YACT,SAASV,KAAK,OAAO;YACrB,OAAOA,KAAK,KAAK;YACjB,SAASH,QAAQ,OAAO,IAAIO;YAC5B,UAAU,OAAOjB,OAAOsC;gBACtB5C,MACE,oBACA6B,UACAvB,OACA,CAAC,4BAA4B,EAAEsC,YAAY,OAAO,EAAE,QAAQ;gBAG9D,MAAMC,YAAYD,YAAY,SAAS;gBACvCF,OAAOG,WAAW;gBAElBT,qBAAqB,OAAO,CAAC,CAACC;oBAC5BK,OACEpC,KAAK,CAAC+B,MAAM,EACZ,CAAC,OAAO,EAAEA,MAAM,yBAAyB,EAAER,SAAS,yCAAyC,EAAEA,SAAS,CAAC,CAAC;gBAE9G;gBAEA,IAAI;oBACF,MAAMiB,QAAQ,GAAG,CAAC;wBACf;4BACC,IAAI,IAAI,CAAC,SAAS,CAAC,kBAAkB,EAAE;gCACrC9C,MAAM;gCACN,MAAM,IAAI,CAAC,SAAS,CAAC,kBAAkB,CAAC+B,OAAO,IAAI,EAAEzB;gCACrDN,MAAM;4BACR;wBACF;wBACA2B,MAAM;qBACP;gBACH,EAAE,OAAOoB,eAAoB;oBAC3B,MAAMC,kBACJD,eAAe,WAAWE,OAAOF;oBACnC,MAAM,IAAId,MACR,CAAC,wCAAwC,EAAEF,OAAO,IAAI,CAAC,EAAE,EAAEiB,iBAAiB,EAC5E;wBAAE,OAAOD;oBAAc;gBAE3B;gBAEA,IAAIhB,OAAO,WAAW,EACpB,IAAI;oBACFzB,QAAQ4C,iBAAiB5C,OAAOyB,OAAO,WAAW;gBACpD,EAAE,OAAOoB,OAAY;oBACnB,MAAM,IAAIlB,MACR,CAAC,8BAA8B,EAAEF,OAAO,IAAI,CAAC,EAAE,EAAEoB,MAAM,OAAO,CAAC,cAAc,EAAEb,KAAK,SAAS,CAAChC,QAAQ,EACtG;wBAAE,OAAO6C;oBAAM;gBAEnB;gBAGFnD,MAAM,kBAAkB+B,OAAO,IAAI;gBACnC,MAAMqB,WAAWrB,OAAO,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS;gBAChD,MAAMsB,eAAe,MAAMD,SAAS9C,OAAOsC;gBAC3C5C,MAAM,iBAAiB+B,OAAO,IAAI,EAAE,WAAWsB;gBAE/C,MAAMC,mBAAmBvB,OAAO,gBAAgB,IAAI;gBACpD,IAAIuB,mBAAmB,GACrB,MAAM3B,MAAM2B;gBAGd,IAAI;oBACF,IAAI,IAAI,CAAC,SAAS,CAAC,iBAAiB,EAAE;wBACpCtD,MAAM;wBACN,MAAM,IAAI,CAAC,SAAS,CAAC,iBAAiB,CAAC+B,OAAO,IAAI,EAAEzB;wBACpDN,MAAM;oBACR;gBACF,EAAE,OAAO+C,eAAoB;oBAC3B,MAAMC,kBACJD,eAAe,WAAWE,OAAOF;oBACnC,MAAM,IAAId,MACR,CAAC,uCAAuC,EAAEF,OAAO,IAAI,CAAC,EAAE,EAAEiB,iBAAiB,EAC3E;wBAAE,OAAOD;oBAAc;gBAE3B;gBAEA,OAAO;oBACL,QAAQM;gBACV;YACF;QACF;QAEArC,QAAQ,KAAK,CAAC,IAAI,CAAC2B;IACrB;IAEQ,iBACNxB,IAAyC,EACzCoC,mBAAiD,EACjDvC,OAAyB,EACzBwC,QAAgD,EACd;QAClC,MAAM,EAAEzC,SAAS,EAAEH,2BAA2B,EAAE,GAAGI;QAEnD,IAAIyC,cAAcF;QAElB,IAAI,AAAuB,YAAvB,OAAOE,aACTA,cAAc;YACZ,QAAQA;QACV;QAGF,IAAI1C,AAAcQ,WAAdR,WACF0C,cAAc;YACZ,GAAGA,WAAW;YACd1C;QACF;QAGF,MAAM2C,cAAgD;YACpD,MAAM;YACN,SAAS;YACT,SAAS1C,QAAQ,OAAO,IAAIO;YAC5B,OAAOkC;YACP,SAAStC,KAAK,OAAO;YACrB,UAAU,OAAOb,OAAOsC;gBACtB,MAAM,EAAED,IAAI,EAAE,GAAGC;gBACjB,IAAI,EAAEC,SAAS,EAAE,GAAGD;gBAEpBF,OACEpC,OAAO,UAAUA,OAAO,MACxB,CAAC,qDAAqD,EAAEgC,KAAK,SAAS,CACpEhC,QACC;gBAGL,IAAI,CAACuC,WACHA,YAAY,MAAM,IAAI,CAAC,OAAO,CAAC,kBAAkB;gBAGnDH,OAAOG,WAAW;gBAElB,IAAIc;gBACJ,IAAIC;gBAEJ,MAAMC,YAAY,CAACC;oBACjB,IAAI,CAACA,MACH;oBAEFH,aAAaG;oBACbnB,KAAK,GAAG,GAAG;wBACTmB;oBACF;oBACAnB,KAAK,KAAK,GAAGmB,KAAK,QAAQ,EAAE;oBAC5B,IAAIA,KAAK,QAAQ,EAAE,iBACjBnB,KAAK,eAAe,GAAGmB,KAAK,QAAQ,CAAC,eAAe;gBAExD;gBAGA,MAAMC,kBAAkBxB,wBAAwBjC,SAC5C0D,qBAAqB1D,SACrBiB;gBACJ,MAAM0C,YAAY,CAAC,CAACF;gBAGpB,IAAIG;gBACJ,IACE,CAACD,aACD3D,MAAM,KAAK,IACX,IAAI,CAAC,SAAS,CAAC,uBAAuB,EAEtC,IAAI;oBACF4D,gBAAgB,MAAM,IAAI,CAAC,SAAS,CAAC,uBAAuB,CAAC;wBAC3D,QAAQ;4BAAC5D,MAAM,KAAK;yBAAC;oBACvB;gBACF,EAAE,OAAM,CAER;gBAEF,MAAM6D,mBAAmBD,gBACrBE,0BACE;oBACE,GAAGF,cAAc,IAAI,GAAGA,cAAc,KAAK,GAAG;oBAC9C,GAAGA,cAAc,GAAG,GAAGA,cAAc,MAAM,GAAG;gBAChD,GACA,AAAwB,YAAxB,OAAO5D,MAAM,MAAM,GACfA,MAAM,MAAM,GACZA,MAAM,MAAM,EAAE,UAAU,MAE9BiB;gBACJ,MAAM8C,aAAa,CAAC,CAACF;gBAErB,MAAMG,cAAchE,MAAM,MAAM;gBAChC,MAAMiE,oBAAoB,IAAI,CAAC,SAAS,EAAE,iBAAiBD;gBAC3D,MAAME,aAAaD,mBAAmB,cAAc;gBAEpD,MAAME,mBACJR,aAAaI,aACT,OACA,MAAMK,sBACJ;oBACE,WAAW,IAAI,CAAC,SAAS;oBACzB,mBAAmB,IAAI,CAAC,SAAS;gBACnC,GACAF,YACAF,aACAhE,MAAM,SAAS;gBAEvB,MAAMqE,aAAa,CAAC,CAACF;gBAErB,IAAIG;gBACJ,IAAI,CAACP,cAAc,CAACM,cAAc,CAACV,WACjC,IAAI;oBACFL,eAAe,MAAM,IAAI,CAAC,OAAO,CAAC,MAAM,CACtCtD,OACA;wBACE,SAASuC;oBACX,GACAjC;oBAEFiD,UAAUD,aAAa,IAAI;oBAC3BgB,sBAAsBhB,aAAa,OAAO;gBAC5C,EAAE,OAAOT,OAAO;oBACd,IAAIA,iBAAiB0B,cACnBhB,UAAUV,MAAM,IAAI;oBAEtB,MAAMA;gBACR;gBAGF,MAAM2B,UACJf,mBACAI,oBACAM,oBACAG;gBAGF,MAAMG,2BAA2B7E,iBAC/BqE,mBAAmB,cAAc;gBAGnC,IAAIS;gBAOJ,IACEF,WACA,IAAI,CAAC,SAAS,IACd,CAACH,cACD,CAACI,4BACDzE,OAAO,cAAc,OAErB,IAAI,IAAI,CAAC,SAAS,CAAC,mBAAmB,EACpC,IAAI;oBACF,MAAM2E,UAAU,MAAM,IAAI,CAAC,SAAS,CAAC,mBAAmB,CACtDH,QAAQ,IAAI,EACZ;wBACE,mBACE,AAAwB,YAAxB,OAAOxE,MAAM,MAAM,GACfA,MAAM,MAAM,GACZA,MAAM,MAAM,EAAE;wBACpB,aAAaM;oBACf;oBAEF,IAAIV,iBAAiB+E,UAAU;wBAC7BjF,MACE,uCACAsE,aACAW;wBAEFD,oBAAoBC;wBACpB,IAAI,CAAC,SAAS,CAAC,yBAAyB,CACtC;4BACE,MAAM;4BACN,QAAQX;4BACR,OAAOW;wBACT,GACAV;oBAEJ,OACEvE,MACE,yDACAsE;gBAGN,EAAE,OAAOnB,OAAO;oBACdnD,MAAM,kCAAkCmD;gBAC1C;qBAEAnD,MAAM;gBAIV,IAAI,CAAC8E,SAAS;oBACZ,IAAInB,YACF,MAAM,IAAIkB,aACR,CAAC,oBAAoB,EAAEvE,MAAM,MAAM,EAAE,EACrCqD;oBAGJ,MAAM,IAAI1B,MAAM,CAAC,mBAAmB,EAAE3B,MAAM,MAAM,EAAE;gBACtD;gBAEA,IAAI4E;gBAEJ,IAAIjB,WACFiB,QAAQ;oBACN,MAAM;oBACN,SAAS;wBACP,MAAM5E,MAAM,IAAI;oBAClB;gBACF;qBACK,IAAI+D,YACTa,QAAQ;oBACN,MAAM;oBACN,SAAS;wBACP,OAAO5E,MAAM,KAAK;oBACpB;gBACF;qBACK,IAAIqE,YACTO,QAAQ;oBACN,MAAM;oBACN,SAAS;wBACPV;wBACA,aAAaQ;oBACf;gBACF;gBAGFxB,WAAWsB;gBAEX,OAAO;oBACL,QAAQ;wBACNA;oBACF;oBACAI;gBACF;YACF;QACF;QAEA,OAAOxB;IACT;IA7fA,YAAY,EACVyB,iBAAiB,EACjBC,OAAO,EACPC,SAAS,EACTvD,WAAW,EACK,CAAE;QAbpB,uBAAiB,aAAjB;QAEA,uBAAiB,WAAjB;QAEA,uBAAiB,aAAjB;QAEA,uBAAiB,eAAjB;QAQE,IAAI,CAAC,SAAS,GAAGqD;QACjB,IAAI,CAAC,OAAO,GAAGC;QACf,IAAI,CAAC,SAAS,GAAGC;QACjB,IAAI,CAAC,WAAW,GAAGvD;IACrB;AAofF"}
@@ -26,9 +26,9 @@ const cacheFileExt = '.cache.yaml';
26
26
  class TaskCache {
27
27
  matchCache(prompt, type) {
28
28
  if (!this.isCacheResultUsed) return;
29
+ const promptStr = 'string' == typeof prompt ? prompt : JSON.stringify(prompt);
29
30
  for(let i = 0; i < this.cacheOriginalLength; i++){
30
31
  const item = this.cache.caches[i];
31
- const promptStr = 'string' == typeof prompt ? prompt : JSON.stringify(prompt);
32
32
  const key = `${type}:${promptStr}:${i}`;
33
33
  if (item.type === type && isDeepStrictEqual(item.prompt, prompt) && !this.matchedCacheIndices.has(key)) {
34
34
  if ('locate' === item.type) {
@@ -1 +1 @@
1
- {"version":3,"file":"agent/task-cache.mjs","sources":["../../../src/agent/task-cache.ts"],"sourcesContent":["import assert from 'node:assert';\nimport { existsSync, mkdirSync, readFileSync, writeFileSync } from 'node:fs';\nimport { dirname, join } from 'node:path';\nimport { isDeepStrictEqual } from 'node:util';\nimport type { TUserPrompt } from '@/ai-model';\nimport type { ElementCacheFeature } from '@/types';\nimport { getMidsceneRunSubDir } from '@midscene/shared/common';\nimport {\n MIDSCENE_CACHE_MAX_FILENAME_LENGTH,\n globalConfigManager,\n} from '@midscene/shared/env';\nimport { getDebug } from '@midscene/shared/logger';\nimport { ifInBrowser, ifInWorker } from '@midscene/shared/utils';\nimport { generateHashId } from '@midscene/shared/utils';\nimport { replaceIllegalPathCharsAndSpace } from '@midscene/shared/utils';\nimport yaml from 'js-yaml';\nimport semver from 'semver';\nimport { getMidsceneVersion } from './utils';\n\nconst DEFAULT_CACHE_MAX_FILENAME_LENGTH = 200;\n\nexport const debug = getDebug('cache');\n\nexport interface PlanningCache {\n type: 'plan';\n prompt: string;\n yamlWorkflow: string;\n}\n\nexport interface LocateCache {\n type: 'locate';\n prompt: TUserPrompt;\n cache?: ElementCacheFeature;\n /** @deprecated kept for backward compatibility */\n xpaths?: string[];\n}\n\nexport interface MatchCacheResult<T extends PlanningCache | LocateCache> {\n cacheContent: T;\n updateFn: (cb: (cache: T) => void) => void;\n}\n\nexport type CacheFileContent = {\n midsceneVersion: string;\n cacheId: string;\n caches: Array<PlanningCache | LocateCache>;\n};\n\nconst lowestSupportedMidsceneVersion = '0.16.10';\nexport const cacheFileExt = '.cache.yaml';\n\nexport class TaskCache {\n cacheId: string;\n\n cacheFilePath?: string;\n\n cache: CacheFileContent;\n\n isCacheResultUsed: boolean; // a flag to indicate if the cache result should be used\n cacheOriginalLength: number;\n\n readOnlyMode: boolean; // a flag to indicate if the cache is in read-only mode\n\n writeOnlyMode: boolean; // a flag to indicate if the cache is in write-only mode\n\n private matchedCacheIndices: Set<string> = new Set(); // Track matched records\n\n constructor(\n cacheId: string,\n isCacheResultUsed: boolean,\n cacheFilePath?: string,\n options: { readOnly?: boolean; writeOnly?: boolean } = {},\n ) {\n assert(cacheId, 'cacheId is required');\n let safeCacheId = replaceIllegalPathCharsAndSpace(cacheId);\n const cacheMaxFilenameLength =\n globalConfigManager.getEnvConfigInNumber(\n MIDSCENE_CACHE_MAX_FILENAME_LENGTH,\n ) || DEFAULT_CACHE_MAX_FILENAME_LENGTH;\n if (Buffer.byteLength(safeCacheId, 'utf8') > cacheMaxFilenameLength) {\n const prefix = safeCacheId.slice(0, 32);\n const hash = generateHashId(undefined, safeCacheId);\n safeCacheId = `${prefix}-${hash}`;\n }\n this.cacheId = safeCacheId;\n\n this.cacheFilePath =\n ifInBrowser || ifInWorker\n ? undefined\n : cacheFilePath ||\n join(getMidsceneRunSubDir('cache'), `${this.cacheId}${cacheFileExt}`);\n const readOnlyMode = Boolean(options?.readOnly);\n const writeOnlyMode = Boolean(options?.writeOnly);\n\n if (readOnlyMode && writeOnlyMode) {\n throw new Error('TaskCache cannot be both read-only and write-only');\n }\n\n this.isCacheResultUsed = writeOnlyMode ? false : isCacheResultUsed;\n this.readOnlyMode = readOnlyMode;\n this.writeOnlyMode = writeOnlyMode;\n\n let cacheContent;\n if (this.cacheFilePath && !this.writeOnlyMode) {\n cacheContent = this.loadCacheFromFile();\n }\n if (!cacheContent) {\n cacheContent = {\n midsceneVersion: getMidsceneVersion(),\n cacheId: this.cacheId,\n caches: [],\n };\n }\n this.cache = cacheContent;\n this.cacheOriginalLength = this.isCacheResultUsed\n ? this.cache.caches.length\n : 0;\n }\n\n matchCache(\n prompt: TUserPrompt,\n type: 'plan' | 'locate',\n ): MatchCacheResult<PlanningCache | LocateCache> | undefined {\n if (!this.isCacheResultUsed) {\n return undefined;\n }\n // Find the first unused matching cache\n for (let i = 0; i < this.cacheOriginalLength; i++) {\n const item = this.cache.caches[i];\n const promptStr =\n typeof prompt === 'string' ? prompt : JSON.stringify(prompt);\n const key = `${type}:${promptStr}:${i}`;\n if (\n item.type === type &&\n isDeepStrictEqual(item.prompt, prompt) &&\n !this.matchedCacheIndices.has(key)\n ) {\n if (item.type === 'locate') {\n const locateItem = item as LocateCache;\n if (!locateItem.cache && Array.isArray(locateItem.xpaths)) {\n locateItem.cache = { xpaths: locateItem.xpaths };\n }\n if ('xpaths' in locateItem) {\n locateItem.xpaths = undefined;\n }\n }\n this.matchedCacheIndices.add(key);\n debug(\n 'cache found and marked as used, type: %s, prompt: %s, index: %d',\n type,\n prompt,\n i,\n );\n return {\n cacheContent: item,\n updateFn: (cb: (cache: PlanningCache | LocateCache) => void) => {\n debug(\n 'will call updateFn to update cache, type: %s, prompt: %s, index: %d',\n type,\n prompt,\n i,\n );\n cb(item);\n\n if (this.readOnlyMode) {\n debug(\n 'read-only mode, cache updated in memory but not flushed to file',\n );\n return;\n }\n\n debug(\n 'cache updated, will flush to file, type: %s, prompt: %s, index: %d',\n type,\n prompt,\n i,\n );\n this.flushCacheToFile();\n },\n };\n }\n }\n debug('no unused cache found, type: %s, prompt: %s', type, prompt);\n return undefined;\n }\n\n matchPlanCache(prompt: string): MatchCacheResult<PlanningCache> | undefined {\n return this.matchCache(prompt, 'plan') as\n | MatchCacheResult<PlanningCache>\n | undefined;\n }\n\n matchLocateCache(\n prompt: TUserPrompt,\n ): MatchCacheResult<LocateCache> | undefined {\n return this.matchCache(prompt, 'locate') as\n | MatchCacheResult<LocateCache>\n | undefined;\n }\n\n appendCache(cache: PlanningCache | LocateCache) {\n debug('will append cache', cache);\n this.cache.caches.push(cache);\n\n if (this.readOnlyMode) {\n debug('read-only mode, cache appended to memory but not flushed to file');\n return;\n }\n\n this.flushCacheToFile();\n }\n\n loadCacheFromFile() {\n const cacheFile = this.cacheFilePath;\n assert(cacheFile, 'cache file path is required');\n\n if (!existsSync(cacheFile)) {\n debug('no cache file found, path: %s', cacheFile);\n return undefined;\n }\n\n // detect old cache file\n const jsonTypeCacheFile = cacheFile.replace(cacheFileExt, '.json');\n if (existsSync(jsonTypeCacheFile) && this.isCacheResultUsed) {\n console.warn(\n `An outdated cache file from an earlier version of Midscene has been detected. Since version 0.17, we have implemented an improved caching strategy. Please delete the old file located at: ${jsonTypeCacheFile}.`,\n );\n return undefined;\n }\n\n try {\n const data = readFileSync(cacheFile, 'utf8');\n const jsonData = yaml.load(data) as CacheFileContent;\n\n const version = getMidsceneVersion();\n if (!version) {\n debug('no midscene version info, will not read cache from file');\n return undefined;\n }\n\n if (\n semver.lt(jsonData.midsceneVersion, lowestSupportedMidsceneVersion) &&\n !jsonData.midsceneVersion.includes('beta') // for internal test\n ) {\n console.warn(\n `You are using an old version of Midscene cache file, and we cannot match any info from it. Starting from Midscene v0.17, we changed our strategy to use xpath for cache info, providing better performance.\\nPlease delete the existing cache and rebuild it. Sorry for the inconvenience.\\ncache file: ${cacheFile}`,\n );\n return undefined;\n }\n\n debug(\n 'cache loaded from file, path: %s, cache version: %s, record length: %s',\n cacheFile,\n jsonData.midsceneVersion,\n jsonData.caches.length,\n );\n jsonData.midsceneVersion = getMidsceneVersion(); // update the version\n return jsonData;\n } catch (err) {\n debug(\n 'cache file exists but load failed, path: %s, error: %s',\n cacheFile,\n err,\n );\n return undefined;\n }\n }\n\n flushCacheToFile(options?: { cleanUnused?: boolean }) {\n const version = getMidsceneVersion();\n if (!version) {\n debug('no midscene version info, will not write cache to file');\n return;\n }\n\n if (!this.cacheFilePath) {\n debug('no cache file path, will not write cache to file');\n return;\n }\n\n // Clean unused caches if requested\n if (options?.cleanUnused) {\n // Skip cleaning in write-only mode or when cache is not used\n if (this.isCacheResultUsed) {\n const originalLength = this.cache.caches.length;\n\n // Collect indices of used caches\n const usedIndices = new Set<number>();\n for (const key of this.matchedCacheIndices) {\n // key format: \"type:prompt:index\"\n const parts = key.split(':');\n const index = Number.parseInt(parts[parts.length - 1], 10);\n if (!Number.isNaN(index)) {\n usedIndices.add(index);\n }\n }\n\n // Filter: keep used caches and newly added caches\n this.cache.caches = this.cache.caches.filter((_, index) => {\n const isUsed = usedIndices.has(index);\n const isNew = index >= this.cacheOriginalLength;\n return isUsed || isNew;\n });\n\n const removedCount = originalLength - this.cache.caches.length;\n if (removedCount > 0) {\n debug('cleaned %d unused cache record(s)', removedCount);\n } else {\n debug('no unused cache to clean');\n }\n } else {\n debug('skip cleaning: cache is not used for reading');\n }\n }\n\n try {\n const dir = dirname(this.cacheFilePath);\n if (!existsSync(dir)) {\n mkdirSync(dir, { recursive: true });\n debug('created cache directory: %s', dir);\n }\n\n // Sort caches to ensure plan entries come before locate entries for better readability\n // Create a sorted copy for writing to disk while keeping in-memory order unchanged\n const sortedCaches = [...this.cache.caches].sort((a, b) => {\n if (a.type === 'plan' && b.type === 'locate') return -1;\n if (a.type === 'locate' && b.type === 'plan') return 1;\n return 0;\n });\n\n const cacheToWrite = {\n ...this.cache,\n caches: sortedCaches,\n };\n\n const yamlData = yaml.dump(cacheToWrite);\n writeFileSync(this.cacheFilePath, yamlData);\n debug('cache flushed to file: %s', this.cacheFilePath);\n } catch (err) {\n debug(\n 'write cache to file failed, path: %s, error: %s',\n this.cacheFilePath,\n err,\n );\n }\n }\n\n updateOrAppendCacheRecord(\n newRecord: PlanningCache | LocateCache,\n cachedRecord?: MatchCacheResult<PlanningCache | LocateCache>,\n ) {\n if (cachedRecord) {\n // update existing record\n if (newRecord.type === 'plan') {\n cachedRecord.updateFn((cache) => {\n (cache as PlanningCache).yamlWorkflow = newRecord.yamlWorkflow;\n });\n } else {\n cachedRecord.updateFn((cache) => {\n const locateCache = cache as LocateCache;\n locateCache.cache = newRecord.cache;\n if ('xpaths' in locateCache) {\n locateCache.xpaths = undefined;\n }\n });\n }\n } else {\n this.appendCache(newRecord);\n }\n }\n}\n"],"names":["DEFAULT_CACHE_MAX_FILENAME_LENGTH","debug","getDebug","lowestSupportedMidsceneVersion","cacheFileExt","TaskCache","prompt","type","i","item","promptStr","JSON","key","isDeepStrictEqual","locateItem","Array","undefined","cb","cache","cacheFile","assert","existsSync","jsonTypeCacheFile","console","data","readFileSync","jsonData","yaml","version","getMidsceneVersion","semver","err","options","originalLength","usedIndices","Set","parts","index","Number","_","isUsed","isNew","removedCount","dir","dirname","mkdirSync","sortedCaches","a","b","cacheToWrite","yamlData","writeFileSync","newRecord","cachedRecord","locateCache","cacheId","isCacheResultUsed","cacheFilePath","safeCacheId","replaceIllegalPathCharsAndSpace","cacheMaxFilenameLength","globalConfigManager","MIDSCENE_CACHE_MAX_FILENAME_LENGTH","Buffer","prefix","hash","generateHashId","ifInBrowser","ifInWorker","join","getMidsceneRunSubDir","readOnlyMode","Boolean","writeOnlyMode","Error","cacheContent"],"mappings":";;;;;;;;;;;;;;;;;;;;;AAmBA,MAAMA,oCAAoC;AAEnC,MAAMC,QAAQC,SAAS;AA2B9B,MAAMC,iCAAiC;AAChC,MAAMC,eAAe;AAErB,MAAMC;IAoEX,WACEC,MAAmB,EACnBC,IAAuB,EACoC;QAC3D,IAAI,CAAC,IAAI,CAAC,iBAAiB,EACzB;QAGF,IAAK,IAAIC,IAAI,GAAGA,IAAI,IAAI,CAAC,mBAAmB,EAAEA,IAAK;YACjD,MAAMC,OAAO,IAAI,CAAC,KAAK,CAAC,MAAM,CAACD,EAAE;YACjC,MAAME,YACJ,AAAkB,YAAlB,OAAOJ,SAAsBA,SAASK,KAAK,SAAS,CAACL;YACvD,MAAMM,MAAM,GAAGL,KAAK,CAAC,EAAEG,UAAU,CAAC,EAAEF,GAAG;YACvC,IACEC,KAAK,IAAI,KAAKF,QACdM,kBAAkBJ,KAAK,MAAM,EAAEH,WAC/B,CAAC,IAAI,CAAC,mBAAmB,CAAC,GAAG,CAACM,MAC9B;gBACA,IAAIH,AAAc,aAAdA,KAAK,IAAI,EAAe;oBAC1B,MAAMK,aAAaL;oBACnB,IAAI,CAACK,WAAW,KAAK,IAAIC,MAAM,OAAO,CAACD,WAAW,MAAM,GACtDA,WAAW,KAAK,GAAG;wBAAE,QAAQA,WAAW,MAAM;oBAAC;oBAEjD,IAAI,YAAYA,YACdA,WAAW,MAAM,GAAGE;gBAExB;gBACA,IAAI,CAAC,mBAAmB,CAAC,GAAG,CAACJ;gBAC7BX,MACE,mEACAM,MACAD,QACAE;gBAEF,OAAO;oBACL,cAAcC;oBACd,UAAU,CAACQ;wBACThB,MACE,uEACAM,MACAD,QACAE;wBAEFS,GAAGR;wBAEH,IAAI,IAAI,CAAC,YAAY,EAAE,YACrBR,MACE;wBAKJA,MACE,sEACAM,MACAD,QACAE;wBAEF,IAAI,CAAC,gBAAgB;oBACvB;gBACF;YACF;QACF;QACAP,MAAM,+CAA+CM,MAAMD;IAE7D;IAEA,eAAeA,MAAc,EAA+C;QAC1E,OAAO,IAAI,CAAC,UAAU,CAACA,QAAQ;IAGjC;IAEA,iBACEA,MAAmB,EACwB;QAC3C,OAAO,IAAI,CAAC,UAAU,CAACA,QAAQ;IAGjC;IAEA,YAAYY,KAAkC,EAAE;QAC9CjB,MAAM,qBAAqBiB;QAC3B,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,CAACA;QAEvB,IAAI,IAAI,CAAC,YAAY,EAAE,YACrBjB,MAAM;QAIR,IAAI,CAAC,gBAAgB;IACvB;IAEA,oBAAoB;QAClB,MAAMkB,YAAY,IAAI,CAAC,aAAa;QACpCC,YAAOD,WAAW;QAElB,IAAI,CAACE,WAAWF,YAAY,YAC1BlB,MAAM,iCAAiCkB;QAKzC,MAAMG,oBAAoBH,UAAU,OAAO,CAACf,cAAc;QAC1D,IAAIiB,WAAWC,sBAAsB,IAAI,CAAC,iBAAiB,EAAE,YAC3DC,QAAQ,IAAI,CACV,CAAC,2LAA2L,EAAED,kBAAkB,CAAC,CAAC;QAKtN,IAAI;YACF,MAAME,OAAOC,aAAaN,WAAW;YACrC,MAAMO,WAAWC,QAAAA,IAAS,CAACH;YAE3B,MAAMI,UAAUC;YAChB,IAAI,CAACD,SAAS,YACZ3B,MAAM;YAIR,IACE6B,OAAO,EAAE,CAACJ,SAAS,eAAe,EAAEvB,mCACpC,CAACuB,SAAS,eAAe,CAAC,QAAQ,CAAC,SACnC,YACAH,QAAQ,IAAI,CACV,CAAC,wSAAwS,EAAEJ,WAAW;YAK1TlB,MACE,0EACAkB,WACAO,SAAS,eAAe,EACxBA,SAAS,MAAM,CAAC,MAAM;YAExBA,SAAS,eAAe,GAAGG;YAC3B,OAAOH;QACT,EAAE,OAAOK,KAAK;YACZ9B,MACE,0DACAkB,WACAY;YAEF;QACF;IACF;IAEA,iBAAiBC,OAAmC,EAAE;QACpD,MAAMJ,UAAUC;QAChB,IAAI,CAACD,SAAS,YACZ3B,MAAM;QAIR,IAAI,CAAC,IAAI,CAAC,aAAa,EAAE,YACvBA,MAAM;QAKR,IAAI+B,SAAS,aAEX,IAAI,IAAI,CAAC,iBAAiB,EAAE;YAC1B,MAAMC,iBAAiB,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM;YAG/C,MAAMC,cAAc,IAAIC;YACxB,KAAK,MAAMvB,OAAO,IAAI,CAAC,mBAAmB,CAAE;gBAE1C,MAAMwB,QAAQxB,IAAI,KAAK,CAAC;gBACxB,MAAMyB,QAAQC,OAAO,QAAQ,CAACF,KAAK,CAACA,MAAM,MAAM,GAAG,EAAE,EAAE;gBACvD,IAAI,CAACE,OAAO,KAAK,CAACD,QAChBH,YAAY,GAAG,CAACG;YAEpB;YAGA,IAAI,CAAC,KAAK,CAAC,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,CAACE,GAAGF;gBAC/C,MAAMG,SAASN,YAAY,GAAG,CAACG;gBAC/B,MAAMI,QAAQJ,SAAS,IAAI,CAAC,mBAAmB;gBAC/C,OAAOG,UAAUC;YACnB;YAEA,MAAMC,eAAeT,iBAAiB,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM;YAC1DS,eAAe,IACjBzC,MAAM,qCAAqCyC,gBAE3CzC,MAAM;QAEV,OACEA,MAAM;QAIV,IAAI;YACF,MAAM0C,MAAMC,QAAQ,IAAI,CAAC,aAAa;YACtC,IAAI,CAACvB,WAAWsB,MAAM;gBACpBE,UAAUF,KAAK;oBAAE,WAAW;gBAAK;gBACjC1C,MAAM,+BAA+B0C;YACvC;YAIA,MAAMG,eAAe;mBAAI,IAAI,CAAC,KAAK,CAAC,MAAM;aAAC,CAAC,IAAI,CAAC,CAACC,GAAGC;gBACnD,IAAID,AAAW,WAAXA,EAAE,IAAI,IAAeC,AAAW,aAAXA,EAAE,IAAI,EAAe,OAAO;gBACrD,IAAID,AAAW,aAAXA,EAAE,IAAI,IAAiBC,AAAW,WAAXA,EAAE,IAAI,EAAa,OAAO;gBACrD,OAAO;YACT;YAEA,MAAMC,eAAe;gBACnB,GAAG,IAAI,CAAC,KAAK;gBACb,QAAQH;YACV;YAEA,MAAMI,WAAWvB,QAAAA,IAAS,CAACsB;YAC3BE,cAAc,IAAI,CAAC,aAAa,EAAED;YAClCjD,MAAM,6BAA6B,IAAI,CAAC,aAAa;QACvD,EAAE,OAAO8B,KAAK;YACZ9B,MACE,mDACA,IAAI,CAAC,aAAa,EAClB8B;QAEJ;IACF;IAEA,0BACEqB,SAAsC,EACtCC,YAA4D,EAC5D;QACA,IAAIA,cAEF,IAAID,AAAmB,WAAnBA,UAAU,IAAI,EAChBC,aAAa,QAAQ,CAAC,CAACnC;YACpBA,MAAwB,YAAY,GAAGkC,UAAU,YAAY;QAChE;aAEAC,aAAa,QAAQ,CAAC,CAACnC;YACrB,MAAMoC,cAAcpC;YACpBoC,YAAY,KAAK,GAAGF,UAAU,KAAK;YACnC,IAAI,YAAYE,aACdA,YAAY,MAAM,GAAGtC;QAEzB;aAGF,IAAI,CAAC,WAAW,CAACoC;IAErB;IA9SA,YACEG,OAAe,EACfC,iBAA0B,EAC1BC,aAAsB,EACtBzB,UAAuD,CAAC,CAAC,CACzD;QApBF;QAEA;QAEA;QAEA;QACA;QAEA;QAEA;QAEA,uBAAQ,uBAAmC,IAAIG;QAQ7Cf,YAAOmC,SAAS;QAChB,IAAIG,cAAcC,gCAAgCJ;QAClD,MAAMK,yBACJC,oBAAoB,oBAAoB,CACtCC,uCACG9D;QACP,IAAI+D,OAAO,UAAU,CAACL,aAAa,UAAUE,wBAAwB;YACnE,MAAMI,SAASN,YAAY,KAAK,CAAC,GAAG;YACpC,MAAMO,OAAOC,eAAelD,QAAW0C;YACvCA,cAAc,GAAGM,OAAO,CAAC,EAAEC,MAAM;QACnC;QACA,IAAI,CAAC,OAAO,GAAGP;QAEf,IAAI,CAAC,aAAa,GAChBS,eAAeC,aACXpD,SACAyC,iBACAY,KAAKC,qBAAqB,UAAU,GAAG,IAAI,CAAC,OAAO,GAAGlE,cAAc;QAC1E,MAAMmE,eAAeC,QAAQxC,SAAS;QACtC,MAAMyC,gBAAgBD,QAAQxC,SAAS;QAEvC,IAAIuC,gBAAgBE,eAClB,MAAM,IAAIC,MAAM;QAGlB,IAAI,CAAC,iBAAiB,GAAGD,gBAAgB,QAAQjB;QACjD,IAAI,CAAC,YAAY,GAAGe;QACpB,IAAI,CAAC,aAAa,GAAGE;QAErB,IAAIE;QACJ,IAAI,IAAI,CAAC,aAAa,IAAI,CAAC,IAAI,CAAC,aAAa,EAC3CA,eAAe,IAAI,CAAC,iBAAiB;QAEvC,IAAI,CAACA,cACHA,eAAe;YACb,iBAAiB9C;YACjB,SAAS,IAAI,CAAC,OAAO;YACrB,QAAQ,EAAE;QACZ;QAEF,IAAI,CAAC,KAAK,GAAG8C;QACb,IAAI,CAAC,mBAAmB,GAAG,IAAI,CAAC,iBAAiB,GAC7C,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,GACxB;IACN;AA6PF"}
1
+ {"version":3,"file":"agent/task-cache.mjs","sources":["../../../src/agent/task-cache.ts"],"sourcesContent":["import assert from 'node:assert';\nimport { existsSync, mkdirSync, readFileSync, writeFileSync } from 'node:fs';\nimport { dirname, join } from 'node:path';\nimport { isDeepStrictEqual } from 'node:util';\nimport type { TUserPrompt } from '@/ai-model';\nimport type { ElementCacheFeature } from '@/types';\nimport { getMidsceneRunSubDir } from '@midscene/shared/common';\nimport {\n MIDSCENE_CACHE_MAX_FILENAME_LENGTH,\n globalConfigManager,\n} from '@midscene/shared/env';\nimport { getDebug } from '@midscene/shared/logger';\nimport { ifInBrowser, ifInWorker } from '@midscene/shared/utils';\nimport { generateHashId } from '@midscene/shared/utils';\nimport { replaceIllegalPathCharsAndSpace } from '@midscene/shared/utils';\nimport yaml from 'js-yaml';\nimport semver from 'semver';\nimport { getMidsceneVersion } from './utils';\n\nconst DEFAULT_CACHE_MAX_FILENAME_LENGTH = 200;\n\nexport const debug = getDebug('cache');\n\nexport interface PlanningCache {\n type: 'plan';\n prompt: string;\n yamlWorkflow: string;\n}\n\nexport interface LocateCache {\n type: 'locate';\n prompt: TUserPrompt;\n cache?: ElementCacheFeature;\n /** @deprecated kept for backward compatibility */\n xpaths?: string[];\n}\n\nexport interface MatchCacheResult<T extends PlanningCache | LocateCache> {\n cacheContent: T;\n updateFn: (cb: (cache: T) => void) => void;\n}\n\nexport type CacheFileContent = {\n midsceneVersion: string;\n cacheId: string;\n caches: Array<PlanningCache | LocateCache>;\n};\n\nconst lowestSupportedMidsceneVersion = '0.16.10';\nexport const cacheFileExt = '.cache.yaml';\n\nexport class TaskCache {\n cacheId: string;\n\n cacheFilePath?: string;\n\n cache: CacheFileContent;\n\n isCacheResultUsed: boolean; // a flag to indicate if the cache result should be used\n cacheOriginalLength: number;\n\n readOnlyMode: boolean; // a flag to indicate if the cache is in read-only mode\n\n writeOnlyMode: boolean; // a flag to indicate if the cache is in write-only mode\n\n private matchedCacheIndices: Set<string> = new Set(); // Track matched records\n\n constructor(\n cacheId: string,\n isCacheResultUsed: boolean,\n cacheFilePath?: string,\n options: { readOnly?: boolean; writeOnly?: boolean } = {},\n ) {\n assert(cacheId, 'cacheId is required');\n let safeCacheId = replaceIllegalPathCharsAndSpace(cacheId);\n const cacheMaxFilenameLength =\n globalConfigManager.getEnvConfigInNumber(\n MIDSCENE_CACHE_MAX_FILENAME_LENGTH,\n ) || DEFAULT_CACHE_MAX_FILENAME_LENGTH;\n if (Buffer.byteLength(safeCacheId, 'utf8') > cacheMaxFilenameLength) {\n const prefix = safeCacheId.slice(0, 32);\n const hash = generateHashId(undefined, safeCacheId);\n safeCacheId = `${prefix}-${hash}`;\n }\n this.cacheId = safeCacheId;\n\n this.cacheFilePath =\n ifInBrowser || ifInWorker\n ? undefined\n : cacheFilePath ||\n join(getMidsceneRunSubDir('cache'), `${this.cacheId}${cacheFileExt}`);\n const readOnlyMode = Boolean(options?.readOnly);\n const writeOnlyMode = Boolean(options?.writeOnly);\n\n if (readOnlyMode && writeOnlyMode) {\n throw new Error('TaskCache cannot be both read-only and write-only');\n }\n\n this.isCacheResultUsed = writeOnlyMode ? false : isCacheResultUsed;\n this.readOnlyMode = readOnlyMode;\n this.writeOnlyMode = writeOnlyMode;\n\n let cacheContent;\n if (this.cacheFilePath && !this.writeOnlyMode) {\n cacheContent = this.loadCacheFromFile();\n }\n if (!cacheContent) {\n cacheContent = {\n midsceneVersion: getMidsceneVersion(),\n cacheId: this.cacheId,\n caches: [],\n };\n }\n this.cache = cacheContent;\n this.cacheOriginalLength = this.isCacheResultUsed\n ? this.cache.caches.length\n : 0;\n }\n\n matchCache(\n prompt: TUserPrompt,\n type: 'plan' | 'locate',\n ): MatchCacheResult<PlanningCache | LocateCache> | undefined {\n if (!this.isCacheResultUsed) {\n return undefined;\n }\n // Find the first unused matching cache\n const promptStr =\n typeof prompt === 'string' ? prompt : JSON.stringify(prompt);\n for (let i = 0; i < this.cacheOriginalLength; i++) {\n const item = this.cache.caches[i];\n const key = `${type}:${promptStr}:${i}`;\n if (\n item.type === type &&\n isDeepStrictEqual(item.prompt, prompt) &&\n !this.matchedCacheIndices.has(key)\n ) {\n if (item.type === 'locate') {\n const locateItem = item as LocateCache;\n if (!locateItem.cache && Array.isArray(locateItem.xpaths)) {\n locateItem.cache = { xpaths: locateItem.xpaths };\n }\n if ('xpaths' in locateItem) {\n locateItem.xpaths = undefined;\n }\n }\n this.matchedCacheIndices.add(key);\n debug(\n 'cache found and marked as used, type: %s, prompt: %s, index: %d',\n type,\n prompt,\n i,\n );\n return {\n cacheContent: item,\n updateFn: (cb: (cache: PlanningCache | LocateCache) => void) => {\n debug(\n 'will call updateFn to update cache, type: %s, prompt: %s, index: %d',\n type,\n prompt,\n i,\n );\n cb(item);\n\n if (this.readOnlyMode) {\n debug(\n 'read-only mode, cache updated in memory but not flushed to file',\n );\n return;\n }\n\n debug(\n 'cache updated, will flush to file, type: %s, prompt: %s, index: %d',\n type,\n prompt,\n i,\n );\n this.flushCacheToFile();\n },\n };\n }\n }\n debug('no unused cache found, type: %s, prompt: %s', type, prompt);\n return undefined;\n }\n\n matchPlanCache(prompt: string): MatchCacheResult<PlanningCache> | undefined {\n return this.matchCache(prompt, 'plan') as\n | MatchCacheResult<PlanningCache>\n | undefined;\n }\n\n matchLocateCache(\n prompt: TUserPrompt,\n ): MatchCacheResult<LocateCache> | undefined {\n return this.matchCache(prompt, 'locate') as\n | MatchCacheResult<LocateCache>\n | undefined;\n }\n\n appendCache(cache: PlanningCache | LocateCache) {\n debug('will append cache', cache);\n this.cache.caches.push(cache);\n\n if (this.readOnlyMode) {\n debug('read-only mode, cache appended to memory but not flushed to file');\n return;\n }\n\n this.flushCacheToFile();\n }\n\n loadCacheFromFile() {\n const cacheFile = this.cacheFilePath;\n assert(cacheFile, 'cache file path is required');\n\n if (!existsSync(cacheFile)) {\n debug('no cache file found, path: %s', cacheFile);\n return undefined;\n }\n\n // detect old cache file\n const jsonTypeCacheFile = cacheFile.replace(cacheFileExt, '.json');\n if (existsSync(jsonTypeCacheFile) && this.isCacheResultUsed) {\n console.warn(\n `An outdated cache file from an earlier version of Midscene has been detected. Since version 0.17, we have implemented an improved caching strategy. Please delete the old file located at: ${jsonTypeCacheFile}.`,\n );\n return undefined;\n }\n\n try {\n const data = readFileSync(cacheFile, 'utf8');\n const jsonData = yaml.load(data) as CacheFileContent;\n\n const version = getMidsceneVersion();\n if (!version) {\n debug('no midscene version info, will not read cache from file');\n return undefined;\n }\n\n if (\n semver.lt(jsonData.midsceneVersion, lowestSupportedMidsceneVersion) &&\n !jsonData.midsceneVersion.includes('beta') // for internal test\n ) {\n console.warn(\n `You are using an old version of Midscene cache file, and we cannot match any info from it. Starting from Midscene v0.17, we changed our strategy to use xpath for cache info, providing better performance.\\nPlease delete the existing cache and rebuild it. Sorry for the inconvenience.\\ncache file: ${cacheFile}`,\n );\n return undefined;\n }\n\n debug(\n 'cache loaded from file, path: %s, cache version: %s, record length: %s',\n cacheFile,\n jsonData.midsceneVersion,\n jsonData.caches.length,\n );\n jsonData.midsceneVersion = getMidsceneVersion(); // update the version\n return jsonData;\n } catch (err) {\n debug(\n 'cache file exists but load failed, path: %s, error: %s',\n cacheFile,\n err,\n );\n return undefined;\n }\n }\n\n flushCacheToFile(options?: { cleanUnused?: boolean }) {\n const version = getMidsceneVersion();\n if (!version) {\n debug('no midscene version info, will not write cache to file');\n return;\n }\n\n if (!this.cacheFilePath) {\n debug('no cache file path, will not write cache to file');\n return;\n }\n\n // Clean unused caches if requested\n if (options?.cleanUnused) {\n // Skip cleaning in write-only mode or when cache is not used\n if (this.isCacheResultUsed) {\n const originalLength = this.cache.caches.length;\n\n // Collect indices of used caches\n const usedIndices = new Set<number>();\n for (const key of this.matchedCacheIndices) {\n // key format: \"type:prompt:index\"\n const parts = key.split(':');\n const index = Number.parseInt(parts[parts.length - 1], 10);\n if (!Number.isNaN(index)) {\n usedIndices.add(index);\n }\n }\n\n // Filter: keep used caches and newly added caches\n this.cache.caches = this.cache.caches.filter((_, index) => {\n const isUsed = usedIndices.has(index);\n const isNew = index >= this.cacheOriginalLength;\n return isUsed || isNew;\n });\n\n const removedCount = originalLength - this.cache.caches.length;\n if (removedCount > 0) {\n debug('cleaned %d unused cache record(s)', removedCount);\n } else {\n debug('no unused cache to clean');\n }\n } else {\n debug('skip cleaning: cache is not used for reading');\n }\n }\n\n try {\n const dir = dirname(this.cacheFilePath);\n if (!existsSync(dir)) {\n mkdirSync(dir, { recursive: true });\n debug('created cache directory: %s', dir);\n }\n\n // Sort caches to ensure plan entries come before locate entries for better readability\n // Create a sorted copy for writing to disk while keeping in-memory order unchanged\n const sortedCaches = [...this.cache.caches].sort((a, b) => {\n if (a.type === 'plan' && b.type === 'locate') return -1;\n if (a.type === 'locate' && b.type === 'plan') return 1;\n return 0;\n });\n\n const cacheToWrite = {\n ...this.cache,\n caches: sortedCaches,\n };\n\n const yamlData = yaml.dump(cacheToWrite);\n writeFileSync(this.cacheFilePath, yamlData);\n debug('cache flushed to file: %s', this.cacheFilePath);\n } catch (err) {\n debug(\n 'write cache to file failed, path: %s, error: %s',\n this.cacheFilePath,\n err,\n );\n }\n }\n\n updateOrAppendCacheRecord(\n newRecord: PlanningCache | LocateCache,\n cachedRecord?: MatchCacheResult<PlanningCache | LocateCache>,\n ) {\n if (cachedRecord) {\n // update existing record\n if (newRecord.type === 'plan') {\n cachedRecord.updateFn((cache) => {\n (cache as PlanningCache).yamlWorkflow = newRecord.yamlWorkflow;\n });\n } else {\n cachedRecord.updateFn((cache) => {\n const locateCache = cache as LocateCache;\n locateCache.cache = newRecord.cache;\n if ('xpaths' in locateCache) {\n locateCache.xpaths = undefined;\n }\n });\n }\n } else {\n this.appendCache(newRecord);\n }\n }\n}\n"],"names":["DEFAULT_CACHE_MAX_FILENAME_LENGTH","debug","getDebug","lowestSupportedMidsceneVersion","cacheFileExt","TaskCache","prompt","type","promptStr","JSON","i","item","key","isDeepStrictEqual","locateItem","Array","undefined","cb","cache","cacheFile","assert","existsSync","jsonTypeCacheFile","console","data","readFileSync","jsonData","yaml","version","getMidsceneVersion","semver","err","options","originalLength","usedIndices","Set","parts","index","Number","_","isUsed","isNew","removedCount","dir","dirname","mkdirSync","sortedCaches","a","b","cacheToWrite","yamlData","writeFileSync","newRecord","cachedRecord","locateCache","cacheId","isCacheResultUsed","cacheFilePath","safeCacheId","replaceIllegalPathCharsAndSpace","cacheMaxFilenameLength","globalConfigManager","MIDSCENE_CACHE_MAX_FILENAME_LENGTH","Buffer","prefix","hash","generateHashId","ifInBrowser","ifInWorker","join","getMidsceneRunSubDir","readOnlyMode","Boolean","writeOnlyMode","Error","cacheContent"],"mappings":";;;;;;;;;;;;;;;;;;;;;AAmBA,MAAMA,oCAAoC;AAEnC,MAAMC,QAAQC,SAAS;AA2B9B,MAAMC,iCAAiC;AAChC,MAAMC,eAAe;AAErB,MAAMC;IAoEX,WACEC,MAAmB,EACnBC,IAAuB,EACoC;QAC3D,IAAI,CAAC,IAAI,CAAC,iBAAiB,EACzB;QAGF,MAAMC,YACJ,AAAkB,YAAlB,OAAOF,SAAsBA,SAASG,KAAK,SAAS,CAACH;QACvD,IAAK,IAAII,IAAI,GAAGA,IAAI,IAAI,CAAC,mBAAmB,EAAEA,IAAK;YACjD,MAAMC,OAAO,IAAI,CAAC,KAAK,CAAC,MAAM,CAACD,EAAE;YACjC,MAAME,MAAM,GAAGL,KAAK,CAAC,EAAEC,UAAU,CAAC,EAAEE,GAAG;YACvC,IACEC,KAAK,IAAI,KAAKJ,QACdM,kBAAkBF,KAAK,MAAM,EAAEL,WAC/B,CAAC,IAAI,CAAC,mBAAmB,CAAC,GAAG,CAACM,MAC9B;gBACA,IAAID,AAAc,aAAdA,KAAK,IAAI,EAAe;oBAC1B,MAAMG,aAAaH;oBACnB,IAAI,CAACG,WAAW,KAAK,IAAIC,MAAM,OAAO,CAACD,WAAW,MAAM,GACtDA,WAAW,KAAK,GAAG;wBAAE,QAAQA,WAAW,MAAM;oBAAC;oBAEjD,IAAI,YAAYA,YACdA,WAAW,MAAM,GAAGE;gBAExB;gBACA,IAAI,CAAC,mBAAmB,CAAC,GAAG,CAACJ;gBAC7BX,MACE,mEACAM,MACAD,QACAI;gBAEF,OAAO;oBACL,cAAcC;oBACd,UAAU,CAACM;wBACThB,MACE,uEACAM,MACAD,QACAI;wBAEFO,GAAGN;wBAEH,IAAI,IAAI,CAAC,YAAY,EAAE,YACrBV,MACE;wBAKJA,MACE,sEACAM,MACAD,QACAI;wBAEF,IAAI,CAAC,gBAAgB;oBACvB;gBACF;YACF;QACF;QACAT,MAAM,+CAA+CM,MAAMD;IAE7D;IAEA,eAAeA,MAAc,EAA+C;QAC1E,OAAO,IAAI,CAAC,UAAU,CAACA,QAAQ;IAGjC;IAEA,iBACEA,MAAmB,EACwB;QAC3C,OAAO,IAAI,CAAC,UAAU,CAACA,QAAQ;IAGjC;IAEA,YAAYY,KAAkC,EAAE;QAC9CjB,MAAM,qBAAqBiB;QAC3B,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,CAACA;QAEvB,IAAI,IAAI,CAAC,YAAY,EAAE,YACrBjB,MAAM;QAIR,IAAI,CAAC,gBAAgB;IACvB;IAEA,oBAAoB;QAClB,MAAMkB,YAAY,IAAI,CAAC,aAAa;QACpCC,YAAOD,WAAW;QAElB,IAAI,CAACE,WAAWF,YAAY,YAC1BlB,MAAM,iCAAiCkB;QAKzC,MAAMG,oBAAoBH,UAAU,OAAO,CAACf,cAAc;QAC1D,IAAIiB,WAAWC,sBAAsB,IAAI,CAAC,iBAAiB,EAAE,YAC3DC,QAAQ,IAAI,CACV,CAAC,2LAA2L,EAAED,kBAAkB,CAAC,CAAC;QAKtN,IAAI;YACF,MAAME,OAAOC,aAAaN,WAAW;YACrC,MAAMO,WAAWC,QAAAA,IAAS,CAACH;YAE3B,MAAMI,UAAUC;YAChB,IAAI,CAACD,SAAS,YACZ3B,MAAM;YAIR,IACE6B,OAAO,EAAE,CAACJ,SAAS,eAAe,EAAEvB,mCACpC,CAACuB,SAAS,eAAe,CAAC,QAAQ,CAAC,SACnC,YACAH,QAAQ,IAAI,CACV,CAAC,wSAAwS,EAAEJ,WAAW;YAK1TlB,MACE,0EACAkB,WACAO,SAAS,eAAe,EACxBA,SAAS,MAAM,CAAC,MAAM;YAExBA,SAAS,eAAe,GAAGG;YAC3B,OAAOH;QACT,EAAE,OAAOK,KAAK;YACZ9B,MACE,0DACAkB,WACAY;YAEF;QACF;IACF;IAEA,iBAAiBC,OAAmC,EAAE;QACpD,MAAMJ,UAAUC;QAChB,IAAI,CAACD,SAAS,YACZ3B,MAAM;QAIR,IAAI,CAAC,IAAI,CAAC,aAAa,EAAE,YACvBA,MAAM;QAKR,IAAI+B,SAAS,aAEX,IAAI,IAAI,CAAC,iBAAiB,EAAE;YAC1B,MAAMC,iBAAiB,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM;YAG/C,MAAMC,cAAc,IAAIC;YACxB,KAAK,MAAMvB,OAAO,IAAI,CAAC,mBAAmB,CAAE;gBAE1C,MAAMwB,QAAQxB,IAAI,KAAK,CAAC;gBACxB,MAAMyB,QAAQC,OAAO,QAAQ,CAACF,KAAK,CAACA,MAAM,MAAM,GAAG,EAAE,EAAE;gBACvD,IAAI,CAACE,OAAO,KAAK,CAACD,QAChBH,YAAY,GAAG,CAACG;YAEpB;YAGA,IAAI,CAAC,KAAK,CAAC,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,CAACE,GAAGF;gBAC/C,MAAMG,SAASN,YAAY,GAAG,CAACG;gBAC/B,MAAMI,QAAQJ,SAAS,IAAI,CAAC,mBAAmB;gBAC/C,OAAOG,UAAUC;YACnB;YAEA,MAAMC,eAAeT,iBAAiB,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM;YAC1DS,eAAe,IACjBzC,MAAM,qCAAqCyC,gBAE3CzC,MAAM;QAEV,OACEA,MAAM;QAIV,IAAI;YACF,MAAM0C,MAAMC,QAAQ,IAAI,CAAC,aAAa;YACtC,IAAI,CAACvB,WAAWsB,MAAM;gBACpBE,UAAUF,KAAK;oBAAE,WAAW;gBAAK;gBACjC1C,MAAM,+BAA+B0C;YACvC;YAIA,MAAMG,eAAe;mBAAI,IAAI,CAAC,KAAK,CAAC,MAAM;aAAC,CAAC,IAAI,CAAC,CAACC,GAAGC;gBACnD,IAAID,AAAW,WAAXA,EAAE,IAAI,IAAeC,AAAW,aAAXA,EAAE,IAAI,EAAe,OAAO;gBACrD,IAAID,AAAW,aAAXA,EAAE,IAAI,IAAiBC,AAAW,WAAXA,EAAE,IAAI,EAAa,OAAO;gBACrD,OAAO;YACT;YAEA,MAAMC,eAAe;gBACnB,GAAG,IAAI,CAAC,KAAK;gBACb,QAAQH;YACV;YAEA,MAAMI,WAAWvB,QAAAA,IAAS,CAACsB;YAC3BE,cAAc,IAAI,CAAC,aAAa,EAAED;YAClCjD,MAAM,6BAA6B,IAAI,CAAC,aAAa;QACvD,EAAE,OAAO8B,KAAK;YACZ9B,MACE,mDACA,IAAI,CAAC,aAAa,EAClB8B;QAEJ;IACF;IAEA,0BACEqB,SAAsC,EACtCC,YAA4D,EAC5D;QACA,IAAIA,cAEF,IAAID,AAAmB,WAAnBA,UAAU,IAAI,EAChBC,aAAa,QAAQ,CAAC,CAACnC;YACpBA,MAAwB,YAAY,GAAGkC,UAAU,YAAY;QAChE;aAEAC,aAAa,QAAQ,CAAC,CAACnC;YACrB,MAAMoC,cAAcpC;YACpBoC,YAAY,KAAK,GAAGF,UAAU,KAAK;YACnC,IAAI,YAAYE,aACdA,YAAY,MAAM,GAAGtC;QAEzB;aAGF,IAAI,CAAC,WAAW,CAACoC;IAErB;IA9SA,YACEG,OAAe,EACfC,iBAA0B,EAC1BC,aAAsB,EACtBzB,UAAuD,CAAC,CAAC,CACzD;QApBF;QAEA;QAEA;QAEA;QACA;QAEA;QAEA;QAEA,uBAAQ,uBAAmC,IAAIG;QAQ7Cf,YAAOmC,SAAS;QAChB,IAAIG,cAAcC,gCAAgCJ;QAClD,MAAMK,yBACJC,oBAAoB,oBAAoB,CACtCC,uCACG9D;QACP,IAAI+D,OAAO,UAAU,CAACL,aAAa,UAAUE,wBAAwB;YACnE,MAAMI,SAASN,YAAY,KAAK,CAAC,GAAG;YACpC,MAAMO,OAAOC,eAAelD,QAAW0C;YACvCA,cAAc,GAAGM,OAAO,CAAC,EAAEC,MAAM;QACnC;QACA,IAAI,CAAC,OAAO,GAAGP;QAEf,IAAI,CAAC,aAAa,GAChBS,eAAeC,aACXpD,SACAyC,iBACAY,KAAKC,qBAAqB,UAAU,GAAG,IAAI,CAAC,OAAO,GAAGlE,cAAc;QAC1E,MAAMmE,eAAeC,QAAQxC,SAAS;QACtC,MAAMyC,gBAAgBD,QAAQxC,SAAS;QAEvC,IAAIuC,gBAAgBE,eAClB,MAAM,IAAIC,MAAM;QAGlB,IAAI,CAAC,iBAAiB,GAAGD,gBAAgB,QAAQjB;QACjD,IAAI,CAAC,YAAY,GAAGe;QACpB,IAAI,CAAC,aAAa,GAAGE;QAErB,IAAIE;QACJ,IAAI,IAAI,CAAC,aAAa,IAAI,CAAC,IAAI,CAAC,aAAa,EAC3CA,eAAe,IAAI,CAAC,iBAAiB;QAEvC,IAAI,CAACA,cACHA,eAAe;YACb,iBAAiB9C;YACjB,SAAS,IAAI,CAAC,OAAO;YACrB,QAAQ,EAAE;QACZ;QAEF,IAAI,CAAC,KAAK,GAAG8C;QACb,IAAI,CAAC,mBAAmB,GAAG,IAAI,CAAC,iBAAiB,GAC7C,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,GACxB;IACN;AA6PF"}
@@ -31,6 +31,9 @@ class TaskExecutor {
31
31
  onTaskUpdate: this.hooks?.onTaskUpdate
32
32
  });
33
33
  }
34
+ getActionSpace() {
35
+ return this.providedActionSpace;
36
+ }
34
37
  async convertPlanToExecutable(plans, modelConfigForPlanning, modelConfigForDefaultIntent, options) {
35
38
  return this.taskBuilder.build(plans, modelConfigForPlanning, modelConfigForDefaultIntent, options);
36
39
  }
@@ -81,7 +84,7 @@ class TaskExecutor {
81
84
  runner
82
85
  };
83
86
  }
84
- async action(userPrompt, modelConfigForPlanning, modelConfigForDefaultIntent, includeBboxInPlanning, backgroundKnowledge, cacheable, replanningCycleLimitOverride) {
87
+ async action(userPrompt, modelConfigForPlanning, modelConfigForDefaultIntent, includeBboxInPlanning, backgroundKnowledge, cacheable, replanningCycleLimitOverride, imagesIncludeCount) {
85
88
  this.conversationHistory.reset();
86
89
  const session = this.createExecutionSession(taskTitleStr('Action', userPrompt));
87
90
  const runner = session.getRunner();
@@ -96,7 +99,8 @@ class TaskExecutor {
96
99
  subType: 'Plan',
97
100
  param: {
98
101
  userInstruction: userPrompt,
99
- aiActionContext: backgroundKnowledge
102
+ aiActionContext: backgroundKnowledge,
103
+ imagesIncludeCount
100
104
  },
101
105
  executor: async (param, executorContext)=>{
102
106
  const startTime = Date.now();
@@ -104,8 +108,7 @@ class TaskExecutor {
104
108
  assert(uiContext, 'uiContext is required for Planning task');
105
109
  const { vlMode } = modelConfigForPlanning;
106
110
  const uiTarsModelVersion = 'vlm-ui-tars' === vlMode ? modelConfigForPlanning.uiTarsModelVersion : void 0;
107
- assert(this.interface.actionSpace, 'actionSpace for device is not implemented');
108
- const actionSpace = await this.interface.actionSpace();
111
+ const actionSpace = this.getActionSpace();
109
112
  debug('actionSpace for this interface is:', actionSpace.map((action)=>action.name).join(', '));
110
113
  assert(Array.isArray(actionSpace), 'actionSpace must be an array');
111
114
  if (0 === actionSpace.length) console.warn(`ActionSpace for ${this.interface.interfaceType} is empty. This may lead to unexpected behavior.`);
@@ -116,7 +119,8 @@ class TaskExecutor {
116
119
  actionSpace,
117
120
  modelConfig: modelConfigForPlanning,
118
121
  conversationHistory: this.conversationHistory,
119
- includeBbox: includeBboxInPlanning
122
+ includeBbox: includeBboxInPlanning,
123
+ imagesIncludeCount
120
124
  });
121
125
  debug('planResult', JSON.stringify(planResult, null, 2));
122
126
  const { actions, log, more_actions_needed_by_instruction, error, usage, rawResponse, sleep } = planResult;
@@ -163,15 +167,18 @@ class TaskExecutor {
163
167
  return session.appendErrorPlan(`Error converting plans to executable tasks: ${error}, plans: ${JSON.stringify(plans)}`);
164
168
  }
165
169
  if (this.conversationHistory.pendingFeedbackMessage) console.warn('unconsumed pending feedback message detected, this may lead to unexpected planning result:', this.conversationHistory.pendingFeedbackMessage);
170
+ let errorFlag = false;
166
171
  try {
167
172
  await session.appendAndRun(executables.tasks);
168
173
  } catch (error) {
174
+ errorFlag = true;
169
175
  errorCountInOnePlanningLoop++;
170
176
  this.conversationHistory.pendingFeedbackMessage = `Error executing running tasks: ${error?.message || String(error)}`;
171
177
  debug('error when executing running tasks, but continue to run if it is not too many errors:', error instanceof Error ? error.message : String(error), 'current error count in one planning loop:', errorCountInOnePlanningLoop);
172
178
  }
173
179
  if (errorCountInOnePlanningLoop > maxErrorCountAllowedInOnePlanningLoop) return session.appendErrorPlan('Too many errors in one planning loop');
174
- if (!planResult?.more_actions_needed_by_instruction) break;
180
+ if (!planResult?.more_actions_needed_by_instruction) if (errorFlag) debug('more_actions_needed_by_instruction is false, but there are errors in one planning loop, continue to run');
181
+ else break;
175
182
  ++replanCount;
176
183
  if (replanCount > replanningCycleLimit) {
177
184
  const errorMsg = `Replanned ${replanningCycleLimit} times, exceeding the limit. Please configure a larger value for replanningCycleLimit (or use MIDSCENE_REPLANNING_CYCLE_LIMIT) to handle more complex tasks.`;
@@ -322,6 +329,7 @@ class TaskExecutor {
322
329
  _define_property(this, "interface", void 0);
323
330
  _define_property(this, "service", void 0);
324
331
  _define_property(this, "taskCache", void 0);
332
+ _define_property(this, "providedActionSpace", void 0);
325
333
  _define_property(this, "taskBuilder", void 0);
326
334
  _define_property(this, "conversationHistory", void 0);
327
335
  _define_property(this, "onTaskStartCallback", void 0);
@@ -334,10 +342,12 @@ class TaskExecutor {
334
342
  this.replanningCycleLimit = opts.replanningCycleLimit;
335
343
  this.hooks = opts.hooks;
336
344
  this.conversationHistory = new ConversationHistory();
345
+ this.providedActionSpace = opts.actionSpace;
337
346
  this.taskBuilder = new TaskBuilder({
338
347
  interfaceInstance,
339
348
  service,
340
- taskCache: opts.taskCache
349
+ taskCache: opts.taskCache,
350
+ actionSpace: this.getActionSpace()
341
351
  });
342
352
  }
343
353
  }
@@ -1 +1 @@
1
- {"version":3,"file":"agent/tasks.mjs","sources":["../../../src/agent/tasks.ts"],"sourcesContent":["import { ConversationHistory, plan, uiTarsPlanning } from '@/ai-model';\nimport type { TMultimodalPrompt, TUserPrompt } from '@/common';\nimport type { AbstractInterface } from '@/device';\nimport type Service from '@/service';\nimport type { TaskRunner } from '@/task-runner';\nimport { TaskExecutionError } from '@/task-runner';\nimport type {\n ExecutionTaskApply,\n ExecutionTaskInsightQueryApply,\n ExecutionTaskPlanningApply,\n ExecutionTaskProgressOptions,\n InterfaceType,\n MidsceneYamlFlowItem,\n PlanningAIResponse,\n PlanningAction,\n PlanningActionParamSleep,\n PlanningActionParamWaitFor,\n ServiceDump,\n ServiceExtractOption,\n ServiceExtractParam,\n} from '@/types';\nimport { ServiceError } from '@/types';\nimport type { IModelConfig } from '@midscene/shared/env';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport { ExecutionSession } from './execution-session';\nimport { TaskBuilder } from './task-builder';\nimport type { TaskCache } from './task-cache';\nexport { locatePlanForLocate } from './task-builder';\nimport { descriptionOfTree } from '@midscene/shared/extractor';\nimport { taskTitleStr } from './ui-utils';\nimport { parsePrompt } from './utils';\n\ninterface ExecutionResult<OutputType = any> {\n output: OutputType;\n thought?: string;\n runner: TaskRunner;\n}\n\ninterface TaskExecutorHooks {\n onTaskUpdate?: (\n runner: TaskRunner,\n error?: TaskExecutionError,\n ) => Promise<void> | void;\n}\n\nconst debug = getDebug('device-task-executor');\nconst maxErrorCountAllowedInOnePlanningLoop = 5;\n\nexport { TaskExecutionError };\n\nexport class TaskExecutor {\n interface: AbstractInterface;\n\n service: Service;\n\n taskCache?: TaskCache;\n\n private readonly taskBuilder: TaskBuilder;\n\n private conversationHistory: ConversationHistory;\n\n onTaskStartCallback?: ExecutionTaskProgressOptions['onTaskStart'];\n\n private readonly hooks?: TaskExecutorHooks;\n\n replanningCycleLimit?: number;\n\n // @deprecated use .interface instead\n get page() {\n return this.interface;\n }\n\n constructor(\n interfaceInstance: AbstractInterface,\n service: Service,\n opts: {\n taskCache?: TaskCache;\n onTaskStart?: ExecutionTaskProgressOptions['onTaskStart'];\n replanningCycleLimit?: number;\n hooks?: TaskExecutorHooks;\n },\n ) {\n this.interface = interfaceInstance;\n this.service = service;\n this.taskCache = opts.taskCache;\n this.onTaskStartCallback = opts?.onTaskStart;\n this.replanningCycleLimit = opts.replanningCycleLimit;\n this.hooks = opts.hooks;\n this.conversationHistory = new ConversationHistory();\n this.taskBuilder = new TaskBuilder({\n interfaceInstance,\n service,\n taskCache: opts.taskCache,\n });\n }\n\n private createExecutionSession(\n title: string,\n options?: { tasks?: ExecutionTaskApply[] },\n ) {\n return new ExecutionSession(\n title,\n () => Promise.resolve(this.service.contextRetrieverFn()),\n {\n onTaskStart: this.onTaskStartCallback,\n tasks: options?.tasks,\n onTaskUpdate: this.hooks?.onTaskUpdate,\n },\n );\n }\n\n public async convertPlanToExecutable(\n plans: PlanningAction[],\n modelConfigForPlanning: IModelConfig,\n modelConfigForDefaultIntent: IModelConfig,\n options?: {\n cacheable?: boolean;\n subTask?: boolean;\n },\n ) {\n return this.taskBuilder.build(\n plans,\n modelConfigForPlanning,\n modelConfigForDefaultIntent,\n options,\n );\n }\n\n async loadYamlFlowAsPlanning(userInstruction: string, yamlString: string) {\n const session = this.createExecutionSession(\n taskTitleStr('Action', userInstruction),\n );\n\n const task: ExecutionTaskPlanningApply = {\n type: 'Planning',\n subType: 'LoadYaml',\n param: {\n userInstruction,\n },\n executor: async (param, executorContext) => {\n const { uiContext } = executorContext;\n assert(uiContext, 'uiContext is required for Planning task');\n return {\n output: {\n actions: [],\n more_actions_needed_by_instruction: false,\n log: '',\n yamlString,\n },\n cache: {\n hit: true,\n },\n hitBy: {\n from: 'Cache',\n context: {\n yamlString,\n },\n },\n };\n },\n };\n const runner = session.getRunner();\n await session.appendAndRun(task);\n\n return {\n runner,\n };\n }\n\n async runPlans(\n title: string,\n plans: PlanningAction[],\n modelConfigForPlanning: IModelConfig,\n modelConfigForDefaultIntent: IModelConfig,\n ): Promise<ExecutionResult> {\n const session = this.createExecutionSession(title);\n const { tasks } = await this.convertPlanToExecutable(\n plans,\n modelConfigForPlanning,\n modelConfigForDefaultIntent,\n );\n const runner = session.getRunner();\n const result = await session.appendAndRun(tasks);\n const { output } = result ?? {};\n return {\n output,\n runner,\n };\n }\n\n async action(\n userPrompt: string,\n modelConfigForPlanning: IModelConfig,\n modelConfigForDefaultIntent: IModelConfig,\n includeBboxInPlanning: boolean,\n backgroundKnowledge?: string,\n cacheable?: boolean,\n replanningCycleLimitOverride?: number,\n ): Promise<\n ExecutionResult<\n | {\n yamlFlow?: MidsceneYamlFlowItem[]; // for cache use\n }\n | undefined\n >\n > {\n this.conversationHistory.reset();\n\n const session = this.createExecutionSession(\n taskTitleStr('Action', userPrompt),\n );\n const runner = session.getRunner();\n\n let replanCount = 0;\n const yamlFlow: MidsceneYamlFlowItem[] = [];\n const replanningCycleLimit =\n replanningCycleLimitOverride ?? this.replanningCycleLimit;\n assert(\n replanningCycleLimit !== undefined,\n 'replanningCycleLimit is required for TaskExecutor.action',\n );\n\n let errorCountInOnePlanningLoop = 0; // count the number of errors in one planning loop\n\n // Main planning loop - unified plan/replan logic\n while (true) {\n const result = await session.appendAndRun(\n {\n type: 'Planning',\n subType: 'Plan',\n param: {\n userInstruction: userPrompt,\n aiActionContext: backgroundKnowledge,\n },\n executor: async (param, executorContext) => {\n const startTime = Date.now();\n const { uiContext } = executorContext;\n assert(uiContext, 'uiContext is required for Planning task');\n const { vlMode } = modelConfigForPlanning;\n const uiTarsModelVersion =\n vlMode === 'vlm-ui-tars'\n ? modelConfigForPlanning.uiTarsModelVersion\n : undefined;\n\n assert(\n this.interface.actionSpace,\n 'actionSpace for device is not implemented',\n );\n const actionSpace = await this.interface.actionSpace();\n debug(\n 'actionSpace for this interface is:',\n actionSpace.map((action) => action.name).join(', '),\n );\n assert(Array.isArray(actionSpace), 'actionSpace must be an array');\n if (actionSpace.length === 0) {\n console.warn(\n `ActionSpace for ${this.interface.interfaceType} is empty. This may lead to unexpected behavior.`,\n );\n }\n\n const planResult = await (uiTarsModelVersion\n ? uiTarsPlanning\n : plan)(param.userInstruction, {\n context: uiContext,\n actionContext: param.aiActionContext,\n interfaceType: this.interface.interfaceType as InterfaceType,\n actionSpace,\n modelConfig: modelConfigForPlanning,\n conversationHistory: this.conversationHistory,\n includeBbox: includeBboxInPlanning,\n });\n debug('planResult', JSON.stringify(planResult, null, 2));\n\n const {\n actions,\n log,\n more_actions_needed_by_instruction,\n error,\n usage,\n rawResponse,\n sleep,\n } = planResult;\n\n executorContext.task.log = {\n ...(executorContext.task.log || {}),\n rawResponse,\n };\n executorContext.task.usage = usage;\n executorContext.task.output = {\n actions: actions || [],\n more_actions_needed_by_instruction,\n log,\n yamlFlow: planResult.yamlFlow,\n };\n executorContext.uiContext = uiContext;\n\n const finalActions = [...(actions || [])];\n\n if (sleep) {\n const timeNow = Date.now();\n const timeRemaining = sleep - (timeNow - startTime);\n if (timeRemaining > 0) {\n finalActions.push(this.sleepPlan(timeRemaining));\n }\n }\n\n assert(!error, `Failed to continue: ${error}\\n${log || ''}`);\n\n return {\n cache: {\n hit: false,\n },\n } as any;\n },\n },\n {\n allowWhenError: true,\n },\n );\n\n const planResult = result?.output as PlanningAIResponse | undefined;\n\n // Execute planned actions\n const plans = planResult?.actions || [];\n yamlFlow.push(...(planResult?.yamlFlow || []));\n\n let executables: Awaited<ReturnType<typeof this.convertPlanToExecutable>>;\n try {\n executables = await this.convertPlanToExecutable(\n plans,\n modelConfigForPlanning,\n modelConfigForDefaultIntent,\n {\n cacheable,\n subTask: true,\n },\n );\n } catch (error) {\n return session.appendErrorPlan(\n `Error converting plans to executable tasks: ${error}, plans: ${JSON.stringify(\n plans,\n )}`,\n );\n }\n if (this.conversationHistory.pendingFeedbackMessage) {\n console.warn(\n 'unconsumed pending feedback message detected, this may lead to unexpected planning result:',\n this.conversationHistory.pendingFeedbackMessage,\n );\n }\n try {\n await session.appendAndRun(executables.tasks);\n } catch (error: any) {\n errorCountInOnePlanningLoop++;\n this.conversationHistory.pendingFeedbackMessage = `Error executing running tasks: ${error?.message || String(error)}`;\n debug(\n 'error when executing running tasks, but continue to run if it is not too many errors:',\n error instanceof Error ? error.message : String(error),\n 'current error count in one planning loop:',\n errorCountInOnePlanningLoop,\n );\n }\n\n if (errorCountInOnePlanningLoop > maxErrorCountAllowedInOnePlanningLoop) {\n return session.appendErrorPlan('Too many errors in one planning loop');\n }\n\n // Check if task is complete\n if (!planResult?.more_actions_needed_by_instruction) {\n break;\n }\n\n // Increment replan count for next iteration\n ++replanCount;\n\n if (replanCount > replanningCycleLimit) {\n const errorMsg = `Replanned ${replanningCycleLimit} times, exceeding the limit. Please configure a larger value for replanningCycleLimit (or use MIDSCENE_REPLANNING_CYCLE_LIMIT) to handle more complex tasks.`;\n return session.appendErrorPlan(errorMsg);\n }\n\n if (!this.conversationHistory.pendingFeedbackMessage) {\n this.conversationHistory.pendingFeedbackMessage =\n 'I have finished the action previously planned.';\n }\n }\n\n const finalResult = {\n output: {\n yamlFlow,\n },\n runner,\n };\n return finalResult;\n }\n\n private createTypeQueryTask(\n type: 'Query' | 'Boolean' | 'Number' | 'String' | 'Assert' | 'WaitFor',\n demand: ServiceExtractParam,\n modelConfig: IModelConfig,\n opt?: ServiceExtractOption,\n multimodalPrompt?: TMultimodalPrompt,\n ) {\n const queryTask: ExecutionTaskInsightQueryApply = {\n type: 'Insight',\n subType: type,\n param: {\n dataDemand: multimodalPrompt\n ? ({\n demand,\n multimodalPrompt,\n } as never)\n : demand, // for user param presentation in report right sidebar\n },\n executor: async (param, taskContext) => {\n const { task } = taskContext;\n let queryDump: ServiceDump | undefined;\n const applyDump = (dump: ServiceDump) => {\n queryDump = dump;\n task.log = {\n dump,\n };\n };\n\n // Get context for query operations\n const uiContext = taskContext.uiContext;\n assert(uiContext, 'uiContext is required for Query task');\n\n const ifTypeRestricted = type !== 'Query';\n let demandInput = demand;\n let keyOfResult = 'result';\n if (ifTypeRestricted && (type === 'Assert' || type === 'WaitFor')) {\n keyOfResult = 'StatementIsTruthy';\n const booleanPrompt =\n type === 'Assert'\n ? `Boolean, whether the following statement is true: ${demand}`\n : `Boolean, the user wants to do some 'wait for' operation, please check whether the following statement is true: ${demand}`;\n demandInput = {\n [keyOfResult]: booleanPrompt,\n };\n } else if (ifTypeRestricted) {\n demandInput = {\n [keyOfResult]: `${type}, ${demand}`,\n };\n }\n\n let extractResult;\n\n let extraPageDescription = '';\n if (opt?.domIncluded && this.interface.getElementsNodeTree) {\n debug('appending tree info for page');\n const tree = await this.interface.getElementsNodeTree();\n extraPageDescription = await descriptionOfTree(\n tree,\n 200,\n false,\n opt?.domIncluded === 'visible-only',\n );\n }\n\n try {\n extractResult = await this.service.extract<any>(\n demandInput,\n modelConfig,\n opt,\n extraPageDescription,\n multimodalPrompt,\n );\n } catch (error) {\n if (error instanceof ServiceError) {\n applyDump(error.dump);\n }\n throw error;\n }\n\n const { data, usage, thought, dump } = extractResult;\n applyDump(dump);\n\n let outputResult = data;\n if (ifTypeRestricted) {\n // If AI returned a plain string instead of structured format, use it directly\n if (typeof data === 'string') {\n outputResult = data;\n } else if (type === 'WaitFor') {\n if (data === null || data === undefined) {\n outputResult = false;\n } else {\n outputResult = (data as any)[keyOfResult];\n }\n } else if (data === null || data === undefined) {\n outputResult = null;\n } else {\n assert(\n data?.[keyOfResult] !== undefined,\n 'No result in query data',\n );\n outputResult = (data as any)[keyOfResult];\n }\n }\n\n if (type === 'Assert' && !outputResult) {\n task.usage = usage;\n task.thought = thought;\n throw new Error(`Assertion failed: ${thought}`);\n }\n\n return {\n output: outputResult,\n log: queryDump,\n usage,\n thought,\n };\n },\n };\n\n return queryTask;\n }\n async createTypeQueryExecution<T>(\n type: 'Query' | 'Boolean' | 'Number' | 'String' | 'Assert',\n demand: ServiceExtractParam,\n modelConfig: IModelConfig,\n opt?: ServiceExtractOption,\n multimodalPrompt?: TMultimodalPrompt,\n ): Promise<ExecutionResult<T>> {\n const session = this.createExecutionSession(\n taskTitleStr(\n type,\n typeof demand === 'string' ? demand : JSON.stringify(demand),\n ),\n );\n\n const queryTask = await this.createTypeQueryTask(\n type,\n demand,\n modelConfig,\n opt,\n multimodalPrompt,\n );\n\n const runner = session.getRunner();\n const result = await session.appendAndRun(queryTask);\n\n if (!result) {\n throw new Error(\n 'result of taskExecutor.flush() is undefined in function createTypeQueryTask',\n );\n }\n\n const { output, thought } = result;\n\n return {\n output,\n thought,\n runner,\n };\n }\n\n private sleepPlan(timeMs: number): PlanningAction<PlanningActionParamSleep> {\n return {\n type: 'Sleep',\n param: {\n timeMs,\n },\n };\n }\n\n async taskForSleep(timeMs: number, _modelConfig: IModelConfig) {\n return this.taskBuilder.createSleepTask({\n timeMs,\n });\n }\n\n async waitFor(\n assertion: TUserPrompt,\n opt: PlanningActionParamWaitFor,\n modelConfig: IModelConfig,\n ): Promise<ExecutionResult<void>> {\n const { textPrompt, multimodalPrompt } = parsePrompt(assertion);\n\n const description = `waitFor: ${textPrompt}`;\n const session = this.createExecutionSession(\n taskTitleStr('WaitFor', description),\n );\n const runner = session.getRunner();\n const { timeoutMs, checkIntervalMs } = opt;\n\n assert(assertion, 'No assertion for waitFor');\n assert(timeoutMs, 'No timeoutMs for waitFor');\n assert(checkIntervalMs, 'No checkIntervalMs for waitFor');\n\n assert(\n checkIntervalMs <= timeoutMs,\n `wrong config for waitFor: checkIntervalMs must be less than timeoutMs, config: {checkIntervalMs: ${checkIntervalMs}, timeoutMs: ${timeoutMs}}`,\n );\n\n const overallStartTime = Date.now();\n let lastCheckStart = overallStartTime;\n let errorThought = '';\n // Continue checking as long as the previous iteration began within the timeout window.\n while (lastCheckStart - overallStartTime <= timeoutMs) {\n const currentCheckStart = Date.now();\n lastCheckStart = currentCheckStart;\n const queryTask = await this.createTypeQueryTask(\n 'WaitFor',\n textPrompt,\n modelConfig,\n undefined,\n multimodalPrompt,\n );\n\n const result = (await session.appendAndRun(queryTask)) as\n | {\n output: boolean;\n thought?: string;\n }\n | undefined;\n\n if (result?.output) {\n return {\n output: undefined,\n runner,\n };\n }\n\n errorThought =\n result?.thought ||\n (!result && `No result from assertion: ${textPrompt}`) ||\n `unknown error when waiting for assertion: ${textPrompt}`;\n const now = Date.now();\n if (now - currentCheckStart < checkIntervalMs) {\n const timeRemaining = checkIntervalMs - (now - currentCheckStart);\n const sleepTask = this.taskBuilder.createSleepTask({\n timeMs: timeRemaining,\n });\n await session.append(sleepTask);\n }\n }\n\n return session.appendErrorPlan(`waitFor timeout: ${errorThought}`);\n }\n}\n"],"names":["debug","getDebug","maxErrorCountAllowedInOnePlanningLoop","TaskExecutor","title","options","ExecutionSession","Promise","plans","modelConfigForPlanning","modelConfigForDefaultIntent","userInstruction","yamlString","session","taskTitleStr","task","param","executorContext","uiContext","assert","runner","tasks","result","output","userPrompt","includeBboxInPlanning","backgroundKnowledge","cacheable","replanningCycleLimitOverride","replanCount","yamlFlow","replanningCycleLimit","undefined","errorCountInOnePlanningLoop","startTime","Date","vlMode","uiTarsModelVersion","actionSpace","action","Array","console","planResult","uiTarsPlanning","plan","JSON","actions","log","more_actions_needed_by_instruction","error","usage","rawResponse","sleep","finalActions","timeNow","timeRemaining","executables","String","Error","errorMsg","finalResult","type","demand","modelConfig","opt","multimodalPrompt","queryTask","taskContext","queryDump","applyDump","dump","ifTypeRestricted","demandInput","keyOfResult","booleanPrompt","extractResult","extraPageDescription","tree","descriptionOfTree","ServiceError","data","thought","outputResult","timeMs","_modelConfig","assertion","textPrompt","parsePrompt","description","timeoutMs","checkIntervalMs","overallStartTime","lastCheckStart","errorThought","currentCheckStart","now","sleepTask","interfaceInstance","service","opts","ConversationHistory","TaskBuilder"],"mappings":";;;;;;;;;;;;;;;;;;;;AA8CA,MAAMA,QAAQC,SAAS;AACvB,MAAMC,wCAAwC;AAIvC,MAAMC;IAkBX,IAAI,OAAO;QACT,OAAO,IAAI,CAAC,SAAS;IACvB;IA0BQ,uBACNC,KAAa,EACbC,OAA0C,EAC1C;QACA,OAAO,IAAIC,iBACTF,OACA,IAAMG,QAAQ,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,kBAAkB,KACrD;YACE,aAAa,IAAI,CAAC,mBAAmB;YACrC,OAAOF,SAAS;YAChB,cAAc,IAAI,CAAC,KAAK,EAAE;QAC5B;IAEJ;IAEA,MAAa,wBACXG,KAAuB,EACvBC,sBAAoC,EACpCC,2BAAyC,EACzCL,OAGC,EACD;QACA,OAAO,IAAI,CAAC,WAAW,CAAC,KAAK,CAC3BG,OACAC,wBACAC,6BACAL;IAEJ;IAEA,MAAM,uBAAuBM,eAAuB,EAAEC,UAAkB,EAAE;QACxE,MAAMC,UAAU,IAAI,CAAC,sBAAsB,CACzCC,aAAa,UAAUH;QAGzB,MAAMI,OAAmC;YACvC,MAAM;YACN,SAAS;YACT,OAAO;gBACLJ;YACF;YACA,UAAU,OAAOK,OAAOC;gBACtB,MAAM,EAAEC,SAAS,EAAE,GAAGD;gBACtBE,OAAOD,WAAW;gBAClB,OAAO;oBACL,QAAQ;wBACN,SAAS,EAAE;wBACX,oCAAoC;wBACpC,KAAK;wBACLN;oBACF;oBACA,OAAO;wBACL,KAAK;oBACP;oBACA,OAAO;wBACL,MAAM;wBACN,SAAS;4BACPA;wBACF;oBACF;gBACF;YACF;QACF;QACA,MAAMQ,SAASP,QAAQ,SAAS;QAChC,MAAMA,QAAQ,YAAY,CAACE;QAE3B,OAAO;YACLK;QACF;IACF;IAEA,MAAM,SACJhB,KAAa,EACbI,KAAuB,EACvBC,sBAAoC,EACpCC,2BAAyC,EACf;QAC1B,MAAMG,UAAU,IAAI,CAAC,sBAAsB,CAACT;QAC5C,MAAM,EAAEiB,KAAK,EAAE,GAAG,MAAM,IAAI,CAAC,uBAAuB,CAClDb,OACAC,wBACAC;QAEF,MAAMU,SAASP,QAAQ,SAAS;QAChC,MAAMS,SAAS,MAAMT,QAAQ,YAAY,CAACQ;QAC1C,MAAM,EAAEE,MAAM,EAAE,GAAGD,UAAU,CAAC;QAC9B,OAAO;YACLC;YACAH;QACF;IACF;IAEA,MAAM,OACJI,UAAkB,EAClBf,sBAAoC,EACpCC,2BAAyC,EACzCe,qBAA8B,EAC9BC,mBAA4B,EAC5BC,SAAmB,EACnBC,4BAAqC,EAQrC;QACA,IAAI,CAAC,mBAAmB,CAAC,KAAK;QAE9B,MAAMf,UAAU,IAAI,CAAC,sBAAsB,CACzCC,aAAa,UAAUU;QAEzB,MAAMJ,SAASP,QAAQ,SAAS;QAEhC,IAAIgB,cAAc;QAClB,MAAMC,WAAmC,EAAE;QAC3C,MAAMC,uBACJH,gCAAgC,IAAI,CAAC,oBAAoB;QAC3DT,OACEY,AAAyBC,WAAzBD,sBACA;QAGF,IAAIE,8BAA8B;QAGlC,MAAO,KAAM;YACX,MAAMX,SAAS,MAAMT,QAAQ,YAAY,CACvC;gBACE,MAAM;gBACN,SAAS;gBACT,OAAO;oBACL,iBAAiBW;oBACjB,iBAAiBE;gBACnB;gBACA,UAAU,OAAOV,OAAOC;oBACtB,MAAMiB,YAAYC,KAAK,GAAG;oBAC1B,MAAM,EAAEjB,SAAS,EAAE,GAAGD;oBACtBE,OAAOD,WAAW;oBAClB,MAAM,EAAEkB,MAAM,EAAE,GAAG3B;oBACnB,MAAM4B,qBACJD,AAAW,kBAAXA,SACI3B,uBAAuB,kBAAkB,GACzCuB;oBAENb,OACE,IAAI,CAAC,SAAS,CAAC,WAAW,EAC1B;oBAEF,MAAMmB,cAAc,MAAM,IAAI,CAAC,SAAS,CAAC,WAAW;oBACpDtC,MACE,sCACAsC,YAAY,GAAG,CAAC,CAACC,SAAWA,OAAO,IAAI,EAAE,IAAI,CAAC;oBAEhDpB,OAAOqB,MAAM,OAAO,CAACF,cAAc;oBACnC,IAAIA,AAAuB,MAAvBA,YAAY,MAAM,EACpBG,QAAQ,IAAI,CACV,CAAC,gBAAgB,EAAE,IAAI,CAAC,SAAS,CAAC,aAAa,CAAC,gDAAgD,CAAC;oBAIrG,MAAMC,aAAa,MAAOL,AAAAA,CAAAA,qBACtBM,iBACAC,IAAG,EAAG5B,MAAM,eAAe,EAAE;wBAC/B,SAASE;wBACT,eAAeF,MAAM,eAAe;wBACpC,eAAe,IAAI,CAAC,SAAS,CAAC,aAAa;wBAC3CsB;wBACA,aAAa7B;wBACb,qBAAqB,IAAI,CAAC,mBAAmB;wBAC7C,aAAagB;oBACf;oBACAzB,MAAM,cAAc6C,KAAK,SAAS,CAACH,YAAY,MAAM;oBAErD,MAAM,EACJI,OAAO,EACPC,GAAG,EACHC,kCAAkC,EAClCC,KAAK,EACLC,KAAK,EACLC,WAAW,EACXC,KAAK,EACN,GAAGV;oBAEJzB,gBAAgB,IAAI,CAAC,GAAG,GAAG;wBACzB,GAAIA,gBAAgB,IAAI,CAAC,GAAG,IAAI,CAAC,CAAC;wBAClCkC;oBACF;oBACAlC,gBAAgB,IAAI,CAAC,KAAK,GAAGiC;oBAC7BjC,gBAAgB,IAAI,CAAC,MAAM,GAAG;wBAC5B,SAAS6B,WAAW,EAAE;wBACtBE;wBACAD;wBACA,UAAUL,WAAW,QAAQ;oBAC/B;oBACAzB,gBAAgB,SAAS,GAAGC;oBAE5B,MAAMmC,eAAe;2BAAKP,WAAW,EAAE;qBAAE;oBAEzC,IAAIM,OAAO;wBACT,MAAME,UAAUnB,KAAK,GAAG;wBACxB,MAAMoB,gBAAgBH,QAASE,CAAAA,UAAUpB,SAAQ;wBACjD,IAAIqB,gBAAgB,GAClBF,aAAa,IAAI,CAAC,IAAI,CAAC,SAAS,CAACE;oBAErC;oBAEApC,OAAO,CAAC8B,OAAO,CAAC,oBAAoB,EAAEA,MAAM,EAAE,EAAEF,OAAO,IAAI;oBAE3D,OAAO;wBACL,OAAO;4BACL,KAAK;wBACP;oBACF;gBACF;YACF,GACA;gBACE,gBAAgB;YAClB;YAGF,MAAML,aAAapB,QAAQ;YAG3B,MAAMd,QAAQkC,YAAY,WAAW,EAAE;YACvCZ,SAAS,IAAI,IAAKY,YAAY,YAAY,EAAE;YAE5C,IAAIc;YACJ,IAAI;gBACFA,cAAc,MAAM,IAAI,CAAC,uBAAuB,CAC9ChD,OACAC,wBACAC,6BACA;oBACEiB;oBACA,SAAS;gBACX;YAEJ,EAAE,OAAOsB,OAAO;gBACd,OAAOpC,QAAQ,eAAe,CAC5B,CAAC,4CAA4C,EAAEoC,MAAM,SAAS,EAAEJ,KAAK,SAAS,CAC5ErC,QACC;YAEP;YACA,IAAI,IAAI,CAAC,mBAAmB,CAAC,sBAAsB,EACjDiC,QAAQ,IAAI,CACV,8FACA,IAAI,CAAC,mBAAmB,CAAC,sBAAsB;YAGnD,IAAI;gBACF,MAAM5B,QAAQ,YAAY,CAAC2C,YAAY,KAAK;YAC9C,EAAE,OAAOP,OAAY;gBACnBhB;gBACA,IAAI,CAAC,mBAAmB,CAAC,sBAAsB,GAAG,CAAC,+BAA+B,EAAEgB,OAAO,WAAWQ,OAAOR,QAAQ;gBACrHjD,MACE,yFACAiD,iBAAiBS,QAAQT,MAAM,OAAO,GAAGQ,OAAOR,QAChD,6CACAhB;YAEJ;YAEA,IAAIA,8BAA8B/B,uCAChC,OAAOW,QAAQ,eAAe,CAAC;YAIjC,IAAI,CAAC6B,YAAY,oCACf;YAIF,EAAEb;YAEF,IAAIA,cAAcE,sBAAsB;gBACtC,MAAM4B,WAAW,CAAC,UAAU,EAAE5B,qBAAqB,4JAA4J,CAAC;gBAChN,OAAOlB,QAAQ,eAAe,CAAC8C;YACjC;YAEA,IAAI,CAAC,IAAI,CAAC,mBAAmB,CAAC,sBAAsB,EAClD,IAAI,CAAC,mBAAmB,CAAC,sBAAsB,GAC7C;QAEN;QAEA,MAAMC,cAAc;YAClB,QAAQ;gBACN9B;YACF;YACAV;QACF;QACA,OAAOwC;IACT;IAEQ,oBACNC,IAAsE,EACtEC,MAA2B,EAC3BC,WAAyB,EACzBC,GAA0B,EAC1BC,gBAAoC,EACpC;QACA,MAAMC,YAA4C;YAChD,MAAM;YACN,SAASL;YACT,OAAO;gBACL,YAAYI,mBACP;oBACCH;oBACAG;gBACF,IACAH;YACN;YACA,UAAU,OAAO9C,OAAOmD;gBACtB,MAAM,EAAEpD,IAAI,EAAE,GAAGoD;gBACjB,IAAIC;gBACJ,MAAMC,YAAY,CAACC;oBACjBF,YAAYE;oBACZvD,KAAK,GAAG,GAAG;wBACTuD;oBACF;gBACF;gBAGA,MAAMpD,YAAYiD,YAAY,SAAS;gBACvChD,OAAOD,WAAW;gBAElB,MAAMqD,mBAAmBV,AAAS,YAATA;gBACzB,IAAIW,cAAcV;gBAClB,IAAIW,cAAc;gBAClB,IAAIF,oBAAqBV,CAAAA,AAAS,aAATA,QAAqBA,AAAS,cAATA,IAAiB,GAAI;oBACjEY,cAAc;oBACd,MAAMC,gBACJb,AAAS,aAATA,OACI,CAAC,kDAAkD,EAAEC,QAAQ,GAC7D,CAAC,+GAA+G,EAAEA,QAAQ;oBAChIU,cAAc;wBACZ,CAACC,YAAY,EAAEC;oBACjB;gBACF,OAAO,IAAIH,kBACTC,cAAc;oBACZ,CAACC,YAAY,EAAE,GAAGZ,KAAK,EAAE,EAAEC,QAAQ;gBACrC;gBAGF,IAAIa;gBAEJ,IAAIC,uBAAuB;gBAC3B,IAAIZ,KAAK,eAAe,IAAI,CAAC,SAAS,CAAC,mBAAmB,EAAE;oBAC1DhE,MAAM;oBACN,MAAM6E,OAAO,MAAM,IAAI,CAAC,SAAS,CAAC,mBAAmB;oBACrDD,uBAAuB,MAAME,kBAC3BD,MACA,KACA,OACAb,KAAK,gBAAgB;gBAEzB;gBAEA,IAAI;oBACFW,gBAAgB,MAAM,IAAI,CAAC,OAAO,CAAC,OAAO,CACxCH,aACAT,aACAC,KACAY,sBACAX;gBAEJ,EAAE,OAAOhB,OAAO;oBACd,IAAIA,iBAAiB8B,cACnBV,UAAUpB,MAAM,IAAI;oBAEtB,MAAMA;gBACR;gBAEA,MAAM,EAAE+B,IAAI,EAAE9B,KAAK,EAAE+B,OAAO,EAAEX,IAAI,EAAE,GAAGK;gBACvCN,UAAUC;gBAEV,IAAIY,eAAeF;gBACnB,IAAIT,kBAEF,IAAI,AAAgB,YAAhB,OAAOS,MACTE,eAAeF;qBACV,IAAInB,AAAS,cAATA,MAEPqB,eADEF,QAAAA,OACa,QAECA,IAAY,CAACP,YAAY;qBAEtC,IAAIO,QAAAA,MACTE,eAAe;qBACV;oBACL/D,OACE6D,MAAM,CAACP,YAAY,KAAKzC,QACxB;oBAEFkD,eAAgBF,IAAY,CAACP,YAAY;gBAC3C;gBAGF,IAAIZ,AAAS,aAATA,QAAqB,CAACqB,cAAc;oBACtCnE,KAAK,KAAK,GAAGmC;oBACbnC,KAAK,OAAO,GAAGkE;oBACf,MAAM,IAAIvB,MAAM,CAAC,kBAAkB,EAAEuB,SAAS;gBAChD;gBAEA,OAAO;oBACL,QAAQC;oBACR,KAAKd;oBACLlB;oBACA+B;gBACF;YACF;QACF;QAEA,OAAOf;IACT;IACA,MAAM,yBACJL,IAA0D,EAC1DC,MAA2B,EAC3BC,WAAyB,EACzBC,GAA0B,EAC1BC,gBAAoC,EACP;QAC7B,MAAMpD,UAAU,IAAI,CAAC,sBAAsB,CACzCC,aACE+C,MACA,AAAkB,YAAlB,OAAOC,SAAsBA,SAASjB,KAAK,SAAS,CAACiB;QAIzD,MAAMI,YAAY,MAAM,IAAI,CAAC,mBAAmB,CAC9CL,MACAC,QACAC,aACAC,KACAC;QAGF,MAAM7C,SAASP,QAAQ,SAAS;QAChC,MAAMS,SAAS,MAAMT,QAAQ,YAAY,CAACqD;QAE1C,IAAI,CAAC5C,QACH,MAAM,IAAIoC,MACR;QAIJ,MAAM,EAAEnC,MAAM,EAAE0D,OAAO,EAAE,GAAG3D;QAE5B,OAAO;YACLC;YACA0D;YACA7D;QACF;IACF;IAEQ,UAAU+D,MAAc,EAA4C;QAC1E,OAAO;YACL,MAAM;YACN,OAAO;gBACLA;YACF;QACF;IACF;IAEA,MAAM,aAAaA,MAAc,EAAEC,YAA0B,EAAE;QAC7D,OAAO,IAAI,CAAC,WAAW,CAAC,eAAe,CAAC;YACtCD;QACF;IACF;IAEA,MAAM,QACJE,SAAsB,EACtBrB,GAA+B,EAC/BD,WAAyB,EACO;QAChC,MAAM,EAAEuB,UAAU,EAAErB,gBAAgB,EAAE,GAAGsB,YAAYF;QAErD,MAAMG,cAAc,CAAC,SAAS,EAAEF,YAAY;QAC5C,MAAMzE,UAAU,IAAI,CAAC,sBAAsB,CACzCC,aAAa,WAAW0E;QAE1B,MAAMpE,SAASP,QAAQ,SAAS;QAChC,MAAM,EAAE4E,SAAS,EAAEC,eAAe,EAAE,GAAG1B;QAEvC7C,OAAOkE,WAAW;QAClBlE,OAAOsE,WAAW;QAClBtE,OAAOuE,iBAAiB;QAExBvE,OACEuE,mBAAmBD,WACnB,CAAC,iGAAiG,EAAEC,gBAAgB,aAAa,EAAED,UAAU,CAAC,CAAC;QAGjJ,MAAME,mBAAmBxD,KAAK,GAAG;QACjC,IAAIyD,iBAAiBD;QACrB,IAAIE,eAAe;QAEnB,MAAOD,iBAAiBD,oBAAoBF,UAAW;YACrD,MAAMK,oBAAoB3D,KAAK,GAAG;YAClCyD,iBAAiBE;YACjB,MAAM5B,YAAY,MAAM,IAAI,CAAC,mBAAmB,CAC9C,WACAoB,YACAvB,aACA/B,QACAiC;YAGF,MAAM3C,SAAU,MAAMT,QAAQ,YAAY,CAACqD;YAO3C,IAAI5C,QAAQ,QACV,OAAO;gBACL,QAAQU;gBACRZ;YACF;YAGFyE,eACEvE,QAAQ,WACP,CAACA,UAAU,CAAC,0BAA0B,EAAEgE,YAAY,IACrD,CAAC,0CAA0C,EAAEA,YAAY;YAC3D,MAAMS,MAAM5D,KAAK,GAAG;YACpB,IAAI4D,MAAMD,oBAAoBJ,iBAAiB;gBAC7C,MAAMnC,gBAAgBmC,kBAAmBK,CAAAA,MAAMD,iBAAgB;gBAC/D,MAAME,YAAY,IAAI,CAAC,WAAW,CAAC,eAAe,CAAC;oBACjD,QAAQzC;gBACV;gBACA,MAAM1C,QAAQ,MAAM,CAACmF;YACvB;QACF;QAEA,OAAOnF,QAAQ,eAAe,CAAC,CAAC,iBAAiB,EAAEgF,cAAc;IACnE;IAtjBA,YACEI,iBAAoC,EACpCC,OAAgB,EAChBC,IAKC,CACD;QA9BF;QAEA;QAEA;QAEA,uBAAiB,eAAjB;QAEA,uBAAQ,uBAAR;QAEA;QAEA,uBAAiB,SAAjB;QAEA;QAiBE,IAAI,CAAC,SAAS,GAAGF;QACjB,IAAI,CAAC,OAAO,GAAGC;QACf,IAAI,CAAC,SAAS,GAAGC,KAAK,SAAS;QAC/B,IAAI,CAAC,mBAAmB,GAAGA,MAAM;QACjC,IAAI,CAAC,oBAAoB,GAAGA,KAAK,oBAAoB;QACrD,IAAI,CAAC,KAAK,GAAGA,KAAK,KAAK;QACvB,IAAI,CAAC,mBAAmB,GAAG,IAAIC;QAC/B,IAAI,CAAC,WAAW,GAAG,IAAIC,YAAY;YACjCJ;YACAC;YACA,WAAWC,KAAK,SAAS;QAC3B;IACF;AAiiBF"}
1
+ {"version":3,"file":"agent/tasks.mjs","sources":["../../../src/agent/tasks.ts"],"sourcesContent":["import { ConversationHistory, plan, uiTarsPlanning } from '@/ai-model';\nimport type { TMultimodalPrompt, TUserPrompt } from '@/common';\nimport type { AbstractInterface } from '@/device';\nimport type Service from '@/service';\nimport type { TaskRunner } from '@/task-runner';\nimport { TaskExecutionError } from '@/task-runner';\nimport type {\n ExecutionTaskApply,\n ExecutionTaskInsightQueryApply,\n ExecutionTaskPlanningApply,\n ExecutionTaskProgressOptions,\n InterfaceType,\n MidsceneYamlFlowItem,\n PlanningAIResponse,\n PlanningAction,\n PlanningActionParamSleep,\n PlanningActionParamWaitFor,\n ServiceDump,\n ServiceExtractOption,\n ServiceExtractParam,\n DeviceAction,\n} from '@/types';\nimport { ServiceError } from '@/types';\nimport type { IModelConfig } from '@midscene/shared/env';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport { ExecutionSession } from './execution-session';\nimport { TaskBuilder } from './task-builder';\nimport type { TaskCache } from './task-cache';\nexport { locatePlanForLocate } from './task-builder';\nimport { descriptionOfTree } from '@midscene/shared/extractor';\nimport { taskTitleStr } from './ui-utils';\nimport { parsePrompt } from './utils';\n\ninterface ExecutionResult<OutputType = any> {\n output: OutputType;\n thought?: string;\n runner: TaskRunner;\n}\n\ninterface TaskExecutorHooks {\n onTaskUpdate?: (\n runner: TaskRunner,\n error?: TaskExecutionError,\n ) => Promise<void> | void;\n}\n\nconst debug = getDebug('device-task-executor');\nconst maxErrorCountAllowedInOnePlanningLoop = 5;\n\nexport { TaskExecutionError };\n\nexport class TaskExecutor {\n interface: AbstractInterface;\n\n service: Service;\n\n taskCache?: TaskCache;\n\n private readonly providedActionSpace: DeviceAction[];\n\n private readonly taskBuilder: TaskBuilder;\n\n private conversationHistory: ConversationHistory;\n\n onTaskStartCallback?: ExecutionTaskProgressOptions['onTaskStart'];\n\n private readonly hooks?: TaskExecutorHooks;\n\n replanningCycleLimit?: number;\n\n // @deprecated use .interface instead\n get page() {\n return this.interface;\n }\n\n constructor(\n interfaceInstance: AbstractInterface,\n service: Service,\n opts: {\n taskCache?: TaskCache;\n onTaskStart?: ExecutionTaskProgressOptions['onTaskStart'];\n replanningCycleLimit?: number;\n hooks?: TaskExecutorHooks;\n actionSpace: DeviceAction[];\n },\n ) {\n this.interface = interfaceInstance;\n this.service = service;\n this.taskCache = opts.taskCache;\n this.onTaskStartCallback = opts?.onTaskStart;\n this.replanningCycleLimit = opts.replanningCycleLimit;\n this.hooks = opts.hooks;\n this.conversationHistory = new ConversationHistory();\n this.providedActionSpace = opts.actionSpace;\n this.taskBuilder = new TaskBuilder({\n interfaceInstance,\n service,\n taskCache: opts.taskCache,\n actionSpace: this.getActionSpace(),\n });\n }\n\n private createExecutionSession(\n title: string,\n options?: { tasks?: ExecutionTaskApply[] },\n ) {\n return new ExecutionSession(\n title,\n () => Promise.resolve(this.service.contextRetrieverFn()),\n {\n onTaskStart: this.onTaskStartCallback,\n tasks: options?.tasks,\n onTaskUpdate: this.hooks?.onTaskUpdate,\n },\n );\n }\n\n private getActionSpace(): DeviceAction[] {\n return this.providedActionSpace;\n }\n\n public async convertPlanToExecutable(\n plans: PlanningAction[],\n modelConfigForPlanning: IModelConfig,\n modelConfigForDefaultIntent: IModelConfig,\n options?: {\n cacheable?: boolean;\n subTask?: boolean;\n },\n ) {\n return this.taskBuilder.build(\n plans,\n modelConfigForPlanning,\n modelConfigForDefaultIntent,\n options,\n );\n }\n\n async loadYamlFlowAsPlanning(userInstruction: string, yamlString: string) {\n const session = this.createExecutionSession(\n taskTitleStr('Action', userInstruction),\n );\n\n const task: ExecutionTaskPlanningApply = {\n type: 'Planning',\n subType: 'LoadYaml',\n param: {\n userInstruction,\n },\n executor: async (param, executorContext) => {\n const { uiContext } = executorContext;\n assert(uiContext, 'uiContext is required for Planning task');\n return {\n output: {\n actions: [],\n more_actions_needed_by_instruction: false,\n log: '',\n yamlString,\n },\n cache: {\n hit: true,\n },\n hitBy: {\n from: 'Cache',\n context: {\n yamlString,\n },\n },\n };\n },\n };\n const runner = session.getRunner();\n await session.appendAndRun(task);\n\n return {\n runner,\n };\n }\n\n async runPlans(\n title: string,\n plans: PlanningAction[],\n modelConfigForPlanning: IModelConfig,\n modelConfigForDefaultIntent: IModelConfig,\n ): Promise<ExecutionResult> {\n const session = this.createExecutionSession(title);\n const { tasks } = await this.convertPlanToExecutable(\n plans,\n modelConfigForPlanning,\n modelConfigForDefaultIntent,\n );\n const runner = session.getRunner();\n const result = await session.appendAndRun(tasks);\n const { output } = result ?? {};\n return {\n output,\n runner,\n };\n }\n\n async action(\n userPrompt: string,\n modelConfigForPlanning: IModelConfig,\n modelConfigForDefaultIntent: IModelConfig,\n includeBboxInPlanning: boolean,\n backgroundKnowledge?: string,\n cacheable?: boolean,\n replanningCycleLimitOverride?: number,\n imagesIncludeCount?: number,\n ): Promise<\n ExecutionResult<\n | {\n yamlFlow?: MidsceneYamlFlowItem[]; // for cache use\n }\n | undefined\n >\n > {\n this.conversationHistory.reset();\n\n const session = this.createExecutionSession(\n taskTitleStr('Action', userPrompt),\n );\n const runner = session.getRunner();\n\n let replanCount = 0;\n const yamlFlow: MidsceneYamlFlowItem[] = [];\n const replanningCycleLimit =\n replanningCycleLimitOverride ?? this.replanningCycleLimit;\n assert(\n replanningCycleLimit !== undefined,\n 'replanningCycleLimit is required for TaskExecutor.action',\n );\n\n let errorCountInOnePlanningLoop = 0; // count the number of errors in one planning loop\n\n // Main planning loop - unified plan/replan logic\n while (true) {\n const result = await session.appendAndRun(\n {\n type: 'Planning',\n subType: 'Plan',\n param: {\n userInstruction: userPrompt,\n aiActionContext: backgroundKnowledge,\n imagesIncludeCount,\n },\n executor: async (param, executorContext) => {\n const startTime = Date.now();\n const { uiContext } = executorContext;\n assert(uiContext, 'uiContext is required for Planning task');\n const { vlMode } = modelConfigForPlanning;\n const uiTarsModelVersion =\n vlMode === 'vlm-ui-tars'\n ? modelConfigForPlanning.uiTarsModelVersion\n : undefined;\n\n const actionSpace = this.getActionSpace();\n debug(\n 'actionSpace for this interface is:',\n actionSpace.map((action) => action.name).join(', '),\n );\n assert(Array.isArray(actionSpace), 'actionSpace must be an array');\n if (actionSpace.length === 0) {\n console.warn(\n `ActionSpace for ${this.interface.interfaceType} is empty. This may lead to unexpected behavior.`,\n );\n }\n\n const planResult = await (uiTarsModelVersion\n ? uiTarsPlanning\n : plan)(param.userInstruction, {\n context: uiContext,\n actionContext: param.aiActionContext,\n interfaceType: this.interface.interfaceType as InterfaceType,\n actionSpace,\n modelConfig: modelConfigForPlanning,\n conversationHistory: this.conversationHistory,\n includeBbox: includeBboxInPlanning,\n imagesIncludeCount,\n });\n debug('planResult', JSON.stringify(planResult, null, 2));\n\n const {\n actions,\n log,\n more_actions_needed_by_instruction,\n error,\n usage,\n rawResponse,\n sleep,\n } = planResult;\n\n executorContext.task.log = {\n ...(executorContext.task.log || {}),\n rawResponse,\n };\n executorContext.task.usage = usage;\n executorContext.task.output = {\n actions: actions || [],\n more_actions_needed_by_instruction,\n log,\n yamlFlow: planResult.yamlFlow,\n };\n executorContext.uiContext = uiContext;\n\n const finalActions = [...(actions || [])];\n\n if (sleep) {\n const timeNow = Date.now();\n const timeRemaining = sleep - (timeNow - startTime);\n if (timeRemaining > 0) {\n finalActions.push(this.sleepPlan(timeRemaining));\n }\n }\n\n assert(!error, `Failed to continue: ${error}\\n${log || ''}`);\n\n return {\n cache: {\n hit: false,\n },\n } as any;\n },\n },\n {\n allowWhenError: true,\n },\n );\n\n const planResult = result?.output as PlanningAIResponse | undefined;\n\n // Execute planned actions\n const plans = planResult?.actions || [];\n yamlFlow.push(...(planResult?.yamlFlow || []));\n\n let executables: Awaited<ReturnType<typeof this.convertPlanToExecutable>>;\n try {\n executables = await this.convertPlanToExecutable(\n plans,\n modelConfigForPlanning,\n modelConfigForDefaultIntent,\n {\n cacheable,\n subTask: true,\n },\n );\n } catch (error) {\n return session.appendErrorPlan(\n `Error converting plans to executable tasks: ${error}, plans: ${JSON.stringify(\n plans,\n )}`,\n );\n }\n if (this.conversationHistory.pendingFeedbackMessage) {\n console.warn(\n 'unconsumed pending feedback message detected, this may lead to unexpected planning result:',\n this.conversationHistory.pendingFeedbackMessage,\n );\n }\n let errorFlag = false;\n try {\n await session.appendAndRun(executables.tasks);\n } catch (error: any) {\n errorFlag = true;\n errorCountInOnePlanningLoop++;\n this.conversationHistory.pendingFeedbackMessage = `Error executing running tasks: ${error?.message || String(error)}`;\n debug(\n 'error when executing running tasks, but continue to run if it is not too many errors:',\n error instanceof Error ? error.message : String(error),\n 'current error count in one planning loop:',\n errorCountInOnePlanningLoop,\n );\n }\n\n if (errorCountInOnePlanningLoop > maxErrorCountAllowedInOnePlanningLoop) {\n return session.appendErrorPlan('Too many errors in one planning loop');\n }\n\n // Check if task is complete\n if (!planResult?.more_actions_needed_by_instruction) {\n if (errorFlag) {\n debug(\n 'more_actions_needed_by_instruction is false, but there are errors in one planning loop, continue to run',\n );\n } else {\n break;\n }\n }\n\n // Increment replan count for next iteration\n ++replanCount;\n\n if (replanCount > replanningCycleLimit) {\n const errorMsg = `Replanned ${replanningCycleLimit} times, exceeding the limit. Please configure a larger value for replanningCycleLimit (or use MIDSCENE_REPLANNING_CYCLE_LIMIT) to handle more complex tasks.`;\n return session.appendErrorPlan(errorMsg);\n }\n\n if (!this.conversationHistory.pendingFeedbackMessage) {\n this.conversationHistory.pendingFeedbackMessage =\n 'I have finished the action previously planned.';\n }\n }\n\n const finalResult = {\n output: {\n yamlFlow,\n },\n runner,\n };\n return finalResult;\n }\n\n private createTypeQueryTask(\n type: 'Query' | 'Boolean' | 'Number' | 'String' | 'Assert' | 'WaitFor',\n demand: ServiceExtractParam,\n modelConfig: IModelConfig,\n opt?: ServiceExtractOption,\n multimodalPrompt?: TMultimodalPrompt,\n ) {\n const queryTask: ExecutionTaskInsightQueryApply = {\n type: 'Insight',\n subType: type,\n param: {\n dataDemand: multimodalPrompt\n ? ({\n demand,\n multimodalPrompt,\n } as never)\n : demand, // for user param presentation in report right sidebar\n },\n executor: async (param, taskContext) => {\n const { task } = taskContext;\n let queryDump: ServiceDump | undefined;\n const applyDump = (dump: ServiceDump) => {\n queryDump = dump;\n task.log = {\n dump,\n };\n };\n\n // Get context for query operations\n const uiContext = taskContext.uiContext;\n assert(uiContext, 'uiContext is required for Query task');\n\n const ifTypeRestricted = type !== 'Query';\n let demandInput = demand;\n let keyOfResult = 'result';\n if (ifTypeRestricted && (type === 'Assert' || type === 'WaitFor')) {\n keyOfResult = 'StatementIsTruthy';\n const booleanPrompt =\n type === 'Assert'\n ? `Boolean, whether the following statement is true: ${demand}`\n : `Boolean, the user wants to do some 'wait for' operation, please check whether the following statement is true: ${demand}`;\n demandInput = {\n [keyOfResult]: booleanPrompt,\n };\n } else if (ifTypeRestricted) {\n demandInput = {\n [keyOfResult]: `${type}, ${demand}`,\n };\n }\n\n let extractResult;\n\n let extraPageDescription = '';\n if (opt?.domIncluded && this.interface.getElementsNodeTree) {\n debug('appending tree info for page');\n const tree = await this.interface.getElementsNodeTree();\n extraPageDescription = await descriptionOfTree(\n tree,\n 200,\n false,\n opt?.domIncluded === 'visible-only',\n );\n }\n\n try {\n extractResult = await this.service.extract<any>(\n demandInput,\n modelConfig,\n opt,\n extraPageDescription,\n multimodalPrompt,\n );\n } catch (error) {\n if (error instanceof ServiceError) {\n applyDump(error.dump);\n }\n throw error;\n }\n\n const { data, usage, thought, dump } = extractResult;\n applyDump(dump);\n\n let outputResult = data;\n if (ifTypeRestricted) {\n // If AI returned a plain string instead of structured format, use it directly\n if (typeof data === 'string') {\n outputResult = data;\n } else if (type === 'WaitFor') {\n if (data === null || data === undefined) {\n outputResult = false;\n } else {\n outputResult = (data as any)[keyOfResult];\n }\n } else if (data === null || data === undefined) {\n outputResult = null;\n } else {\n assert(\n data?.[keyOfResult] !== undefined,\n 'No result in query data',\n );\n outputResult = (data as any)[keyOfResult];\n }\n }\n\n if (type === 'Assert' && !outputResult) {\n task.usage = usage;\n task.thought = thought;\n throw new Error(`Assertion failed: ${thought}`);\n }\n\n return {\n output: outputResult,\n log: queryDump,\n usage,\n thought,\n };\n },\n };\n\n return queryTask;\n }\n async createTypeQueryExecution<T>(\n type: 'Query' | 'Boolean' | 'Number' | 'String' | 'Assert',\n demand: ServiceExtractParam,\n modelConfig: IModelConfig,\n opt?: ServiceExtractOption,\n multimodalPrompt?: TMultimodalPrompt,\n ): Promise<ExecutionResult<T>> {\n const session = this.createExecutionSession(\n taskTitleStr(\n type,\n typeof demand === 'string' ? demand : JSON.stringify(demand),\n ),\n );\n\n const queryTask = await this.createTypeQueryTask(\n type,\n demand,\n modelConfig,\n opt,\n multimodalPrompt,\n );\n\n const runner = session.getRunner();\n const result = await session.appendAndRun(queryTask);\n\n if (!result) {\n throw new Error(\n 'result of taskExecutor.flush() is undefined in function createTypeQueryTask',\n );\n }\n\n const { output, thought } = result;\n\n return {\n output,\n thought,\n runner,\n };\n }\n\n private sleepPlan(timeMs: number): PlanningAction<PlanningActionParamSleep> {\n return {\n type: 'Sleep',\n param: {\n timeMs,\n },\n };\n }\n\n async taskForSleep(timeMs: number, _modelConfig: IModelConfig) {\n return this.taskBuilder.createSleepTask({\n timeMs,\n });\n }\n\n async waitFor(\n assertion: TUserPrompt,\n opt: PlanningActionParamWaitFor,\n modelConfig: IModelConfig,\n ): Promise<ExecutionResult<void>> {\n const { textPrompt, multimodalPrompt } = parsePrompt(assertion);\n\n const description = `waitFor: ${textPrompt}`;\n const session = this.createExecutionSession(\n taskTitleStr('WaitFor', description),\n );\n const runner = session.getRunner();\n const { timeoutMs, checkIntervalMs } = opt;\n\n assert(assertion, 'No assertion for waitFor');\n assert(timeoutMs, 'No timeoutMs for waitFor');\n assert(checkIntervalMs, 'No checkIntervalMs for waitFor');\n\n assert(\n checkIntervalMs <= timeoutMs,\n `wrong config for waitFor: checkIntervalMs must be less than timeoutMs, config: {checkIntervalMs: ${checkIntervalMs}, timeoutMs: ${timeoutMs}}`,\n );\n\n const overallStartTime = Date.now();\n let lastCheckStart = overallStartTime;\n let errorThought = '';\n // Continue checking as long as the previous iteration began within the timeout window.\n while (lastCheckStart - overallStartTime <= timeoutMs) {\n const currentCheckStart = Date.now();\n lastCheckStart = currentCheckStart;\n const queryTask = await this.createTypeQueryTask(\n 'WaitFor',\n textPrompt,\n modelConfig,\n undefined,\n multimodalPrompt,\n );\n\n const result = (await session.appendAndRun(queryTask)) as\n | {\n output: boolean;\n thought?: string;\n }\n | undefined;\n\n if (result?.output) {\n return {\n output: undefined,\n runner,\n };\n }\n\n errorThought =\n result?.thought ||\n (!result && `No result from assertion: ${textPrompt}`) ||\n `unknown error when waiting for assertion: ${textPrompt}`;\n const now = Date.now();\n if (now - currentCheckStart < checkIntervalMs) {\n const timeRemaining = checkIntervalMs - (now - currentCheckStart);\n const sleepTask = this.taskBuilder.createSleepTask({\n timeMs: timeRemaining,\n });\n await session.append(sleepTask);\n }\n }\n\n return session.appendErrorPlan(`waitFor timeout: ${errorThought}`);\n }\n}\n"],"names":["debug","getDebug","maxErrorCountAllowedInOnePlanningLoop","TaskExecutor","title","options","ExecutionSession","Promise","plans","modelConfigForPlanning","modelConfigForDefaultIntent","userInstruction","yamlString","session","taskTitleStr","task","param","executorContext","uiContext","assert","runner","tasks","result","output","userPrompt","includeBboxInPlanning","backgroundKnowledge","cacheable","replanningCycleLimitOverride","imagesIncludeCount","replanCount","yamlFlow","replanningCycleLimit","undefined","errorCountInOnePlanningLoop","startTime","Date","vlMode","uiTarsModelVersion","actionSpace","action","Array","console","planResult","uiTarsPlanning","plan","JSON","actions","log","more_actions_needed_by_instruction","error","usage","rawResponse","sleep","finalActions","timeNow","timeRemaining","executables","errorFlag","String","Error","errorMsg","finalResult","type","demand","modelConfig","opt","multimodalPrompt","queryTask","taskContext","queryDump","applyDump","dump","ifTypeRestricted","demandInput","keyOfResult","booleanPrompt","extractResult","extraPageDescription","tree","descriptionOfTree","ServiceError","data","thought","outputResult","timeMs","_modelConfig","assertion","textPrompt","parsePrompt","description","timeoutMs","checkIntervalMs","overallStartTime","lastCheckStart","errorThought","currentCheckStart","now","sleepTask","interfaceInstance","service","opts","ConversationHistory","TaskBuilder"],"mappings":";;;;;;;;;;;;;;;;;;;;AA+CA,MAAMA,QAAQC,SAAS;AACvB,MAAMC,wCAAwC;AAIvC,MAAMC;IAoBX,IAAI,OAAO;QACT,OAAO,IAAI,CAAC,SAAS;IACvB;IA6BQ,uBACNC,KAAa,EACbC,OAA0C,EAC1C;QACA,OAAO,IAAIC,iBACTF,OACA,IAAMG,QAAQ,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,kBAAkB,KACrD;YACE,aAAa,IAAI,CAAC,mBAAmB;YACrC,OAAOF,SAAS;YAChB,cAAc,IAAI,CAAC,KAAK,EAAE;QAC5B;IAEJ;IAEQ,iBAAiC;QACvC,OAAO,IAAI,CAAC,mBAAmB;IACjC;IAEA,MAAa,wBACXG,KAAuB,EACvBC,sBAAoC,EACpCC,2BAAyC,EACzCL,OAGC,EACD;QACA,OAAO,IAAI,CAAC,WAAW,CAAC,KAAK,CAC3BG,OACAC,wBACAC,6BACAL;IAEJ;IAEA,MAAM,uBAAuBM,eAAuB,EAAEC,UAAkB,EAAE;QACxE,MAAMC,UAAU,IAAI,CAAC,sBAAsB,CACzCC,aAAa,UAAUH;QAGzB,MAAMI,OAAmC;YACvC,MAAM;YACN,SAAS;YACT,OAAO;gBACLJ;YACF;YACA,UAAU,OAAOK,OAAOC;gBACtB,MAAM,EAAEC,SAAS,EAAE,GAAGD;gBACtBE,OAAOD,WAAW;gBAClB,OAAO;oBACL,QAAQ;wBACN,SAAS,EAAE;wBACX,oCAAoC;wBACpC,KAAK;wBACLN;oBACF;oBACA,OAAO;wBACL,KAAK;oBACP;oBACA,OAAO;wBACL,MAAM;wBACN,SAAS;4BACPA;wBACF;oBACF;gBACF;YACF;QACF;QACA,MAAMQ,SAASP,QAAQ,SAAS;QAChC,MAAMA,QAAQ,YAAY,CAACE;QAE3B,OAAO;YACLK;QACF;IACF;IAEA,MAAM,SACJhB,KAAa,EACbI,KAAuB,EACvBC,sBAAoC,EACpCC,2BAAyC,EACf;QAC1B,MAAMG,UAAU,IAAI,CAAC,sBAAsB,CAACT;QAC5C,MAAM,EAAEiB,KAAK,EAAE,GAAG,MAAM,IAAI,CAAC,uBAAuB,CAClDb,OACAC,wBACAC;QAEF,MAAMU,SAASP,QAAQ,SAAS;QAChC,MAAMS,SAAS,MAAMT,QAAQ,YAAY,CAACQ;QAC1C,MAAM,EAAEE,MAAM,EAAE,GAAGD,UAAU,CAAC;QAC9B,OAAO;YACLC;YACAH;QACF;IACF;IAEA,MAAM,OACJI,UAAkB,EAClBf,sBAAoC,EACpCC,2BAAyC,EACzCe,qBAA8B,EAC9BC,mBAA4B,EAC5BC,SAAmB,EACnBC,4BAAqC,EACrCC,kBAA2B,EAQ3B;QACA,IAAI,CAAC,mBAAmB,CAAC,KAAK;QAE9B,MAAMhB,UAAU,IAAI,CAAC,sBAAsB,CACzCC,aAAa,UAAUU;QAEzB,MAAMJ,SAASP,QAAQ,SAAS;QAEhC,IAAIiB,cAAc;QAClB,MAAMC,WAAmC,EAAE;QAC3C,MAAMC,uBACJJ,gCAAgC,IAAI,CAAC,oBAAoB;QAC3DT,OACEa,AAAyBC,WAAzBD,sBACA;QAGF,IAAIE,8BAA8B;QAGlC,MAAO,KAAM;YACX,MAAMZ,SAAS,MAAMT,QAAQ,YAAY,CACvC;gBACE,MAAM;gBACN,SAAS;gBACT,OAAO;oBACL,iBAAiBW;oBACjB,iBAAiBE;oBACjBG;gBACF;gBACA,UAAU,OAAOb,OAAOC;oBACtB,MAAMkB,YAAYC,KAAK,GAAG;oBAC1B,MAAM,EAAElB,SAAS,EAAE,GAAGD;oBACtBE,OAAOD,WAAW;oBAClB,MAAM,EAAEmB,MAAM,EAAE,GAAG5B;oBACnB,MAAM6B,qBACJD,AAAW,kBAAXA,SACI5B,uBAAuB,kBAAkB,GACzCwB;oBAEN,MAAMM,cAAc,IAAI,CAAC,cAAc;oBACvCvC,MACE,sCACAuC,YAAY,GAAG,CAAC,CAACC,SAAWA,OAAO,IAAI,EAAE,IAAI,CAAC;oBAEhDrB,OAAOsB,MAAM,OAAO,CAACF,cAAc;oBACnC,IAAIA,AAAuB,MAAvBA,YAAY,MAAM,EACpBG,QAAQ,IAAI,CACV,CAAC,gBAAgB,EAAE,IAAI,CAAC,SAAS,CAAC,aAAa,CAAC,gDAAgD,CAAC;oBAIrG,MAAMC,aAAa,MAAOL,AAAAA,CAAAA,qBACtBM,iBACAC,IAAG,EAAG7B,MAAM,eAAe,EAAE;wBAC/B,SAASE;wBACT,eAAeF,MAAM,eAAe;wBACpC,eAAe,IAAI,CAAC,SAAS,CAAC,aAAa;wBAC3CuB;wBACA,aAAa9B;wBACb,qBAAqB,IAAI,CAAC,mBAAmB;wBAC7C,aAAagB;wBACbI;oBACF;oBACA7B,MAAM,cAAc8C,KAAK,SAAS,CAACH,YAAY,MAAM;oBAErD,MAAM,EACJI,OAAO,EACPC,GAAG,EACHC,kCAAkC,EAClCC,KAAK,EACLC,KAAK,EACLC,WAAW,EACXC,KAAK,EACN,GAAGV;oBAEJ1B,gBAAgB,IAAI,CAAC,GAAG,GAAG;wBACzB,GAAIA,gBAAgB,IAAI,CAAC,GAAG,IAAI,CAAC,CAAC;wBAClCmC;oBACF;oBACAnC,gBAAgB,IAAI,CAAC,KAAK,GAAGkC;oBAC7BlC,gBAAgB,IAAI,CAAC,MAAM,GAAG;wBAC5B,SAAS8B,WAAW,EAAE;wBACtBE;wBACAD;wBACA,UAAUL,WAAW,QAAQ;oBAC/B;oBACA1B,gBAAgB,SAAS,GAAGC;oBAE5B,MAAMoC,eAAe;2BAAKP,WAAW,EAAE;qBAAE;oBAEzC,IAAIM,OAAO;wBACT,MAAME,UAAUnB,KAAK,GAAG;wBACxB,MAAMoB,gBAAgBH,QAASE,CAAAA,UAAUpB,SAAQ;wBACjD,IAAIqB,gBAAgB,GAClBF,aAAa,IAAI,CAAC,IAAI,CAAC,SAAS,CAACE;oBAErC;oBAEArC,OAAO,CAAC+B,OAAO,CAAC,oBAAoB,EAAEA,MAAM,EAAE,EAAEF,OAAO,IAAI;oBAE3D,OAAO;wBACL,OAAO;4BACL,KAAK;wBACP;oBACF;gBACF;YACF,GACA;gBACE,gBAAgB;YAClB;YAGF,MAAML,aAAarB,QAAQ;YAG3B,MAAMd,QAAQmC,YAAY,WAAW,EAAE;YACvCZ,SAAS,IAAI,IAAKY,YAAY,YAAY,EAAE;YAE5C,IAAIc;YACJ,IAAI;gBACFA,cAAc,MAAM,IAAI,CAAC,uBAAuB,CAC9CjD,OACAC,wBACAC,6BACA;oBACEiB;oBACA,SAAS;gBACX;YAEJ,EAAE,OAAOuB,OAAO;gBACd,OAAOrC,QAAQ,eAAe,CAC5B,CAAC,4CAA4C,EAAEqC,MAAM,SAAS,EAAEJ,KAAK,SAAS,CAC5EtC,QACC;YAEP;YACA,IAAI,IAAI,CAAC,mBAAmB,CAAC,sBAAsB,EACjDkC,QAAQ,IAAI,CACV,8FACA,IAAI,CAAC,mBAAmB,CAAC,sBAAsB;YAGnD,IAAIgB,YAAY;YAChB,IAAI;gBACF,MAAM7C,QAAQ,YAAY,CAAC4C,YAAY,KAAK;YAC9C,EAAE,OAAOP,OAAY;gBACnBQ,YAAY;gBACZxB;gBACA,IAAI,CAAC,mBAAmB,CAAC,sBAAsB,GAAG,CAAC,+BAA+B,EAAEgB,OAAO,WAAWS,OAAOT,QAAQ;gBACrHlD,MACE,yFACAkD,iBAAiBU,QAAQV,MAAM,OAAO,GAAGS,OAAOT,QAChD,6CACAhB;YAEJ;YAEA,IAAIA,8BAA8BhC,uCAChC,OAAOW,QAAQ,eAAe,CAAC;YAIjC,IAAI,CAAC8B,YAAY,oCACf,IAAIe,WACF1D,MACE;iBAGF;YAKJ,EAAE8B;YAEF,IAAIA,cAAcE,sBAAsB;gBACtC,MAAM6B,WAAW,CAAC,UAAU,EAAE7B,qBAAqB,4JAA4J,CAAC;gBAChN,OAAOnB,QAAQ,eAAe,CAACgD;YACjC;YAEA,IAAI,CAAC,IAAI,CAAC,mBAAmB,CAAC,sBAAsB,EAClD,IAAI,CAAC,mBAAmB,CAAC,sBAAsB,GAC7C;QAEN;QAEA,MAAMC,cAAc;YAClB,QAAQ;gBACN/B;YACF;YACAX;QACF;QACA,OAAO0C;IACT;IAEQ,oBACNC,IAAsE,EACtEC,MAA2B,EAC3BC,WAAyB,EACzBC,GAA0B,EAC1BC,gBAAoC,EACpC;QACA,MAAMC,YAA4C;YAChD,MAAM;YACN,SAASL;YACT,OAAO;gBACL,YAAYI,mBACP;oBACCH;oBACAG;gBACF,IACAH;YACN;YACA,UAAU,OAAOhD,OAAOqD;gBACtB,MAAM,EAAEtD,IAAI,EAAE,GAAGsD;gBACjB,IAAIC;gBACJ,MAAMC,YAAY,CAACC;oBACjBF,YAAYE;oBACZzD,KAAK,GAAG,GAAG;wBACTyD;oBACF;gBACF;gBAGA,MAAMtD,YAAYmD,YAAY,SAAS;gBACvClD,OAAOD,WAAW;gBAElB,MAAMuD,mBAAmBV,AAAS,YAATA;gBACzB,IAAIW,cAAcV;gBAClB,IAAIW,cAAc;gBAClB,IAAIF,oBAAqBV,CAAAA,AAAS,aAATA,QAAqBA,AAAS,cAATA,IAAiB,GAAI;oBACjEY,cAAc;oBACd,MAAMC,gBACJb,AAAS,aAATA,OACI,CAAC,kDAAkD,EAAEC,QAAQ,GAC7D,CAAC,+GAA+G,EAAEA,QAAQ;oBAChIU,cAAc;wBACZ,CAACC,YAAY,EAAEC;oBACjB;gBACF,OAAO,IAAIH,kBACTC,cAAc;oBACZ,CAACC,YAAY,EAAE,GAAGZ,KAAK,EAAE,EAAEC,QAAQ;gBACrC;gBAGF,IAAIa;gBAEJ,IAAIC,uBAAuB;gBAC3B,IAAIZ,KAAK,eAAe,IAAI,CAAC,SAAS,CAAC,mBAAmB,EAAE;oBAC1DlE,MAAM;oBACN,MAAM+E,OAAO,MAAM,IAAI,CAAC,SAAS,CAAC,mBAAmB;oBACrDD,uBAAuB,MAAME,kBAC3BD,MACA,KACA,OACAb,KAAK,gBAAgB;gBAEzB;gBAEA,IAAI;oBACFW,gBAAgB,MAAM,IAAI,CAAC,OAAO,CAAC,OAAO,CACxCH,aACAT,aACAC,KACAY,sBACAX;gBAEJ,EAAE,OAAOjB,OAAO;oBACd,IAAIA,iBAAiB+B,cACnBV,UAAUrB,MAAM,IAAI;oBAEtB,MAAMA;gBACR;gBAEA,MAAM,EAAEgC,IAAI,EAAE/B,KAAK,EAAEgC,OAAO,EAAEX,IAAI,EAAE,GAAGK;gBACvCN,UAAUC;gBAEV,IAAIY,eAAeF;gBACnB,IAAIT,kBAEF,IAAI,AAAgB,YAAhB,OAAOS,MACTE,eAAeF;qBACV,IAAInB,AAAS,cAATA,MAEPqB,eADEF,QAAAA,OACa,QAECA,IAAY,CAACP,YAAY;qBAEtC,IAAIO,QAAAA,MACTE,eAAe;qBACV;oBACLjE,OACE+D,MAAM,CAACP,YAAY,KAAK1C,QACxB;oBAEFmD,eAAgBF,IAAY,CAACP,YAAY;gBAC3C;gBAGF,IAAIZ,AAAS,aAATA,QAAqB,CAACqB,cAAc;oBACtCrE,KAAK,KAAK,GAAGoC;oBACbpC,KAAK,OAAO,GAAGoE;oBACf,MAAM,IAAIvB,MAAM,CAAC,kBAAkB,EAAEuB,SAAS;gBAChD;gBAEA,OAAO;oBACL,QAAQC;oBACR,KAAKd;oBACLnB;oBACAgC;gBACF;YACF;QACF;QAEA,OAAOf;IACT;IACA,MAAM,yBACJL,IAA0D,EAC1DC,MAA2B,EAC3BC,WAAyB,EACzBC,GAA0B,EAC1BC,gBAAoC,EACP;QAC7B,MAAMtD,UAAU,IAAI,CAAC,sBAAsB,CACzCC,aACEiD,MACA,AAAkB,YAAlB,OAAOC,SAAsBA,SAASlB,KAAK,SAAS,CAACkB;QAIzD,MAAMI,YAAY,MAAM,IAAI,CAAC,mBAAmB,CAC9CL,MACAC,QACAC,aACAC,KACAC;QAGF,MAAM/C,SAASP,QAAQ,SAAS;QAChC,MAAMS,SAAS,MAAMT,QAAQ,YAAY,CAACuD;QAE1C,IAAI,CAAC9C,QACH,MAAM,IAAIsC,MACR;QAIJ,MAAM,EAAErC,MAAM,EAAE4D,OAAO,EAAE,GAAG7D;QAE5B,OAAO;YACLC;YACA4D;YACA/D;QACF;IACF;IAEQ,UAAUiE,MAAc,EAA4C;QAC1E,OAAO;YACL,MAAM;YACN,OAAO;gBACLA;YACF;QACF;IACF;IAEA,MAAM,aAAaA,MAAc,EAAEC,YAA0B,EAAE;QAC7D,OAAO,IAAI,CAAC,WAAW,CAAC,eAAe,CAAC;YACtCD;QACF;IACF;IAEA,MAAM,QACJE,SAAsB,EACtBrB,GAA+B,EAC/BD,WAAyB,EACO;QAChC,MAAM,EAAEuB,UAAU,EAAErB,gBAAgB,EAAE,GAAGsB,YAAYF;QAErD,MAAMG,cAAc,CAAC,SAAS,EAAEF,YAAY;QAC5C,MAAM3E,UAAU,IAAI,CAAC,sBAAsB,CACzCC,aAAa,WAAW4E;QAE1B,MAAMtE,SAASP,QAAQ,SAAS;QAChC,MAAM,EAAE8E,SAAS,EAAEC,eAAe,EAAE,GAAG1B;QAEvC/C,OAAOoE,WAAW;QAClBpE,OAAOwE,WAAW;QAClBxE,OAAOyE,iBAAiB;QAExBzE,OACEyE,mBAAmBD,WACnB,CAAC,iGAAiG,EAAEC,gBAAgB,aAAa,EAAED,UAAU,CAAC,CAAC;QAGjJ,MAAME,mBAAmBzD,KAAK,GAAG;QACjC,IAAI0D,iBAAiBD;QACrB,IAAIE,eAAe;QAEnB,MAAOD,iBAAiBD,oBAAoBF,UAAW;YACrD,MAAMK,oBAAoB5D,KAAK,GAAG;YAClC0D,iBAAiBE;YACjB,MAAM5B,YAAY,MAAM,IAAI,CAAC,mBAAmB,CAC9C,WACAoB,YACAvB,aACAhC,QACAkC;YAGF,MAAM7C,SAAU,MAAMT,QAAQ,YAAY,CAACuD;YAO3C,IAAI9C,QAAQ,QACV,OAAO;gBACL,QAAQW;gBACRb;YACF;YAGF2E,eACEzE,QAAQ,WACP,CAACA,UAAU,CAAC,0BAA0B,EAAEkE,YAAY,IACrD,CAAC,0CAA0C,EAAEA,YAAY;YAC3D,MAAMS,MAAM7D,KAAK,GAAG;YACpB,IAAI6D,MAAMD,oBAAoBJ,iBAAiB;gBAC7C,MAAMpC,gBAAgBoC,kBAAmBK,CAAAA,MAAMD,iBAAgB;gBAC/D,MAAME,YAAY,IAAI,CAAC,WAAW,CAAC,eAAe,CAAC;oBACjD,QAAQ1C;gBACV;gBACA,MAAM3C,QAAQ,MAAM,CAACqF;YACvB;QACF;QAEA,OAAOrF,QAAQ,eAAe,CAAC,CAAC,iBAAiB,EAAEkF,cAAc;IACnE;IApkBA,YACEI,iBAAoC,EACpCC,OAAgB,EAChBC,IAMC,CACD;QAjCF;QAEA;QAEA;QAEA,uBAAiB,uBAAjB;QAEA,uBAAiB,eAAjB;QAEA,uBAAQ,uBAAR;QAEA;QAEA,uBAAiB,SAAjB;QAEA;QAkBE,IAAI,CAAC,SAAS,GAAGF;QACjB,IAAI,CAAC,OAAO,GAAGC;QACf,IAAI,CAAC,SAAS,GAAGC,KAAK,SAAS;QAC/B,IAAI,CAAC,mBAAmB,GAAGA,MAAM;QACjC,IAAI,CAAC,oBAAoB,GAAGA,KAAK,oBAAoB;QACrD,IAAI,CAAC,KAAK,GAAGA,KAAK,KAAK;QACvB,IAAI,CAAC,mBAAmB,GAAG,IAAIC;QAC/B,IAAI,CAAC,mBAAmB,GAAGD,KAAK,WAAW;QAC3C,IAAI,CAAC,WAAW,GAAG,IAAIE,YAAY;YACjCJ;YACAC;YACA,WAAWC,KAAK,SAAS;YACzB,aAAa,IAAI,CAAC,cAAc;QAClC;IACF;AA4iBF"}
@@ -100,7 +100,7 @@ async function matchElementFromCache(context, cacheEntry, cachePrompt, cacheable
100
100
  return;
101
101
  }
102
102
  }
103
- const getMidsceneVersion = ()=>"1.0.1-beta-20251208075922.0";
103
+ const getMidsceneVersion = ()=>"1.0.1-beta-20251208112226.0";
104
104
  const parsePrompt = (prompt)=>{
105
105
  if ('string' == typeof prompt) return {
106
106
  textPrompt: prompt,
@@ -27,10 +27,27 @@ class ConversationHistory {
27
27
  reset() {
28
28
  this.messages.length = 0;
29
29
  }
30
- snapshot() {
31
- return [
30
+ snapshot(maxImages) {
31
+ if (void 0 === maxImages) return [
32
32
  ...this.messages
33
33
  ];
34
+ const clonedMessages = structuredClone(this.messages);
35
+ let imageCount = 0;
36
+ for(let i = clonedMessages.length - 1; i >= 0; i--){
37
+ const message = clonedMessages[i];
38
+ const content = message.content;
39
+ if (Array.isArray(content)) for(let j = 0; j < content.length; j++){
40
+ const item = content[j];
41
+ if ('image_url' === item.type) {
42
+ imageCount++;
43
+ if (imageCount > maxImages) content[j] = {
44
+ type: 'text',
45
+ text: '(image ignored due to size optimization)'
46
+ };
47
+ }
48
+ }
49
+ }
50
+ return clonedMessages;
34
51
  }
35
52
  get length() {
36
53
  return this.messages.length;
@@ -1 +1 @@
1
- {"version":3,"file":"ai-model/conversation-history.mjs","sources":["../../../src/ai-model/conversation-history.ts"],"sourcesContent":["import type { ChatCompletionMessageParam } from 'openai/resources/index';\n\nexport interface ConversationHistoryOptions {\n initialMessages?: ChatCompletionMessageParam[];\n}\n\nexport class ConversationHistory {\n private readonly messages: ChatCompletionMessageParam[] = [];\n\n public pendingFeedbackMessage: string;\n\n constructor(options?: ConversationHistoryOptions) {\n if (options?.initialMessages?.length) {\n this.seed(options.initialMessages);\n }\n this.pendingFeedbackMessage = '';\n }\n\n resetPendingFeedbackMessageIfExists() {\n if (this.pendingFeedbackMessage) {\n this.pendingFeedbackMessage = '';\n }\n }\n\n append(message: ChatCompletionMessageParam) {\n this.messages.push(message);\n }\n\n seed(messages: ChatCompletionMessageParam[]) {\n this.reset();\n messages.forEach((message) => {\n this.append(message);\n });\n }\n\n reset() {\n this.messages.length = 0;\n }\n\n snapshot(): ChatCompletionMessageParam[] {\n return [...this.messages];\n }\n\n get length(): number {\n return this.messages.length;\n }\n\n [Symbol.iterator](): IterableIterator<ChatCompletionMessageParam> {\n return this.messages[Symbol.iterator]();\n }\n\n toJSON(): ChatCompletionMessageParam[] {\n return this.snapshot();\n }\n}\n"],"names":["Symbol","ConversationHistory","message","messages","options"],"mappings":";;;;;;;;;;;eA+CGA,OAAO,QAAQ;;AAzCX,MAAMC;IAYX,sCAAsC;QACpC,IAAI,IAAI,CAAC,sBAAsB,EAC7B,IAAI,CAAC,sBAAsB,GAAG;IAElC;IAEA,OAAOC,OAAmC,EAAE;QAC1C,IAAI,CAAC,QAAQ,CAAC,IAAI,CAACA;IACrB;IAEA,KAAKC,QAAsC,EAAE;QAC3C,IAAI,CAAC,KAAK;QACVA,SAAS,OAAO,CAAC,CAACD;YAChB,IAAI,CAAC,MAAM,CAACA;QACd;IACF;IAEA,QAAQ;QACN,IAAI,CAAC,QAAQ,CAAC,MAAM,GAAG;IACzB;IAEA,WAAyC;QACvC,OAAO;eAAI,IAAI,CAAC,QAAQ;SAAC;IAC3B;IAEA,IAAI,SAAiB;QACnB,OAAO,IAAI,CAAC,QAAQ,CAAC,MAAM;IAC7B;IAEA,CAAC,cAAD,GAAkE;QAChE,OAAO,IAAI,CAAC,QAAQ,CAACF,OAAO,QAAQ,CAAC;IACvC;IAEA,SAAuC;QACrC,OAAO,IAAI,CAAC,QAAQ;IACtB;IA1CA,YAAYI,OAAoC,CAAE;QAJlD,uBAAiB,YAAyC,EAAE;QAE5D,uBAAO,0BAAP;QAGE,IAAIA,SAAS,iBAAiB,QAC5B,IAAI,CAAC,IAAI,CAACA,QAAQ,eAAe;QAEnC,IAAI,CAAC,sBAAsB,GAAG;IAChC;AAsCF"}
1
+ {"version":3,"file":"ai-model/conversation-history.mjs","sources":["../../../src/ai-model/conversation-history.ts"],"sourcesContent":["import type { ChatCompletionMessageParam } from 'openai/resources/index';\n\nexport interface ConversationHistoryOptions {\n initialMessages?: ChatCompletionMessageParam[];\n}\n\nexport class ConversationHistory {\n private readonly messages: ChatCompletionMessageParam[] = [];\n\n public pendingFeedbackMessage: string;\n\n constructor(options?: ConversationHistoryOptions) {\n if (options?.initialMessages?.length) {\n this.seed(options.initialMessages);\n }\n this.pendingFeedbackMessage = '';\n }\n\n resetPendingFeedbackMessageIfExists() {\n if (this.pendingFeedbackMessage) {\n this.pendingFeedbackMessage = '';\n }\n }\n\n append(message: ChatCompletionMessageParam) {\n this.messages.push(message);\n }\n\n seed(messages: ChatCompletionMessageParam[]) {\n this.reset();\n messages.forEach((message) => {\n this.append(message);\n });\n }\n\n reset() {\n this.messages.length = 0;\n }\n\n /**\n * Snapshot the conversation history, and replace the images with text if the number of images exceeds the limit.\n * @param maxImages - The maximum number of images to include in the snapshot. Undefined means no limit.\n * @returns The snapshot of the conversation history.\n */\n snapshot(maxImages?: number): ChatCompletionMessageParam[] {\n if (maxImages === undefined) {\n return [...this.messages];\n }\n\n const clonedMessages = structuredClone(this.messages);\n let imageCount = 0;\n\n // Traverse from the end to the beginning\n for (let i = clonedMessages.length - 1; i >= 0; i--) {\n const message = clonedMessages[i];\n const content = message.content;\n\n // Only process if content is an array\n if (Array.isArray(content)) {\n for (let j = 0; j < content.length; j++) {\n const item = content[j];\n\n // Check if this is an image\n if (item.type === 'image_url') {\n imageCount++;\n\n // If we've exceeded the limit, replace with text\n if (imageCount > maxImages) {\n content[j] = {\n type: 'text',\n text: '(image ignored due to size optimization)',\n };\n }\n }\n }\n }\n }\n\n return clonedMessages;\n }\n\n get length(): number {\n return this.messages.length;\n }\n\n [Symbol.iterator](): IterableIterator<ChatCompletionMessageParam> {\n return this.messages[Symbol.iterator]();\n }\n\n toJSON(): ChatCompletionMessageParam[] {\n return this.snapshot();\n }\n}\n"],"names":["Symbol","ConversationHistory","message","messages","maxImages","undefined","clonedMessages","structuredClone","imageCount","i","content","Array","j","item","options"],"mappings":";;;;;;;;;;;eAqFGA,OAAO,QAAQ;;AA/EX,MAAMC;IAYX,sCAAsC;QACpC,IAAI,IAAI,CAAC,sBAAsB,EAC7B,IAAI,CAAC,sBAAsB,GAAG;IAElC;IAEA,OAAOC,OAAmC,EAAE;QAC1C,IAAI,CAAC,QAAQ,CAAC,IAAI,CAACA;IACrB;IAEA,KAAKC,QAAsC,EAAE;QAC3C,IAAI,CAAC,KAAK;QACVA,SAAS,OAAO,CAAC,CAACD;YAChB,IAAI,CAAC,MAAM,CAACA;QACd;IACF;IAEA,QAAQ;QACN,IAAI,CAAC,QAAQ,CAAC,MAAM,GAAG;IACzB;IAOA,SAASE,SAAkB,EAAgC;QACzD,IAAIA,AAAcC,WAAdD,WACF,OAAO;eAAI,IAAI,CAAC,QAAQ;SAAC;QAG3B,MAAME,iBAAiBC,gBAAgB,IAAI,CAAC,QAAQ;QACpD,IAAIC,aAAa;QAGjB,IAAK,IAAIC,IAAIH,eAAe,MAAM,GAAG,GAAGG,KAAK,GAAGA,IAAK;YACnD,MAAMP,UAAUI,cAAc,CAACG,EAAE;YACjC,MAAMC,UAAUR,QAAQ,OAAO;YAG/B,IAAIS,MAAM,OAAO,CAACD,UAChB,IAAK,IAAIE,IAAI,GAAGA,IAAIF,QAAQ,MAAM,EAAEE,IAAK;gBACvC,MAAMC,OAAOH,OAAO,CAACE,EAAE;gBAGvB,IAAIC,AAAc,gBAAdA,KAAK,IAAI,EAAkB;oBAC7BL;oBAGA,IAAIA,aAAaJ,WACfM,OAAO,CAACE,EAAE,GAAG;wBACX,MAAM;wBACN,MAAM;oBACR;gBAEJ;YACF;QAEJ;QAEA,OAAON;IACT;IAEA,IAAI,SAAiB;QACnB,OAAO,IAAI,CAAC,QAAQ,CAAC,MAAM;IAC7B;IAEA,CAAC,cAAD,GAAkE;QAChE,OAAO,IAAI,CAAC,QAAQ,CAACN,OAAO,QAAQ,CAAC;IACvC;IAEA,SAAuC;QACrC,OAAO,IAAI,CAAC,QAAQ;IACtB;IAhFA,YAAYc,OAAoC,CAAE;QAJlD,uBAAiB,YAAyC,EAAE;QAE5D,uBAAO,0BAAP;QAGE,IAAIA,SAAS,iBAAiB,QAC5B,IAAI,CAAC,IAAI,CAACA,QAAQ,eAAe;QAEnC,IAAI,CAAC,sBAAsB,GAAG;IAChC;AA4EF"}
@@ -25,13 +25,14 @@ async function plan(userInstruction, opts) {
25
25
  imageHeight = paddedResult.height;
26
26
  imagePayload = paddedResult.imageBase64;
27
27
  }
28
+ const actionContext = opts.actionContext ? `<high_priority_knowledge>${opts.actionContext}</high_priority_knowledge>\n` : '';
28
29
  const instruction = [
29
30
  {
30
31
  role: 'user',
31
32
  content: [
32
33
  {
33
34
  type: 'text',
34
- text: `<high_priority_knowledge>${opts.actionContext}</high_priority_knowledge>\n<user_instruction>${userInstruction}</user_instruction>`
35
+ text: `${actionContext}<user_instruction>${userInstruction}</user_instruction>`
35
36
  }
36
37
  ]
37
38
  }
@@ -72,7 +73,7 @@ async function plan(userInstruction, opts) {
72
73
  ]
73
74
  };
74
75
  conversationHistory.append(latestFeedbackMessage);
75
- const historyLog = conversationHistory.snapshot();
76
+ const historyLog = conversationHistory.snapshot(opts.imagesIncludeCount);
76
77
  const msgs = [
77
78
  {
78
79
  role: 'system',