@midscene/core 1.12.7 → 1.12.8-beta-20260916023839.0

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 (34) hide show
  1. package/dist/es/agent/task-builder.mjs +5 -10
  2. package/dist/es/agent/task-builder.mjs.map +1 -1
  3. package/dist/es/agent/utils.mjs +1 -1
  4. package/dist/es/ai-model/workflows/planning/custom-planning.mjs +1 -2
  5. package/dist/es/ai-model/workflows/planning/custom-planning.mjs.map +1 -1
  6. package/dist/es/ai-model/workflows/planning/locate-normalization.mjs +25 -26
  7. package/dist/es/ai-model/workflows/planning/locate-normalization.mjs.map +1 -1
  8. package/dist/es/ai-model/workflows/planning/parse-planning-actions.mjs +44 -0
  9. package/dist/es/ai-model/workflows/planning/parse-planning-actions.mjs.map +1 -0
  10. package/dist/es/ai-model/workflows/planning/standard-planning.mjs +8 -8
  11. package/dist/es/ai-model/workflows/planning/standard-planning.mjs.map +1 -1
  12. package/dist/es/common.mjs +4 -1
  13. package/dist/es/common.mjs.map +1 -1
  14. package/dist/es/report-html-template.mjs +1 -1
  15. package/dist/es/utils.mjs +1 -1
  16. package/dist/lib/agent/task-builder.js +4 -9
  17. package/dist/lib/agent/task-builder.js.map +1 -1
  18. package/dist/lib/agent/utils.js +1 -1
  19. package/dist/lib/ai-model/workflows/planning/custom-planning.js +1 -2
  20. package/dist/lib/ai-model/workflows/planning/custom-planning.js.map +1 -1
  21. package/dist/lib/ai-model/workflows/planning/locate-normalization.js +28 -26
  22. package/dist/lib/ai-model/workflows/planning/locate-normalization.js.map +1 -1
  23. package/dist/lib/ai-model/workflows/planning/parse-planning-actions.js +78 -0
  24. package/dist/lib/ai-model/workflows/planning/parse-planning-actions.js.map +1 -0
  25. package/dist/lib/ai-model/workflows/planning/standard-planning.js +8 -8
  26. package/dist/lib/ai-model/workflows/planning/standard-planning.js.map +1 -1
  27. package/dist/lib/common.js +7 -1
  28. package/dist/lib/common.js.map +1 -1
  29. package/dist/lib/report-html-template.js +1 -1
  30. package/dist/lib/utils.js +1 -1
  31. package/dist/types/ai-model/workflows/planning/locate-normalization.d.ts +5 -3
  32. package/dist/types/ai-model/workflows/planning/parse-planning-actions.d.ts +7 -0
  33. package/dist/types/common.d.ts +1 -0
  34. package/package.json +2 -2
@@ -1,5 +1,5 @@
1
1
  import { findAllMidsceneLocatorField, parseActionParam } from "../ai-model/index.mjs";
2
- import { findActionInActionSpaceOrThrow } from "../common.mjs";
2
+ import { findActionInActionSpaceOrThrow, validateRequiredLocateFields } from "../common.mjs";
3
3
  import { setTimingFieldOnce } from "../task-timing.mjs";
4
4
  import { ServiceError } from "../types.mjs";
5
5
  import { sleep } from "../utils.mjs";
@@ -99,19 +99,16 @@ class TaskBuilder {
99
99
  const action = findActionInActionSpaceOrThrow(planType, actionSpace);
100
100
  const param = plan.param;
101
101
  const locateFields = findAllMidsceneLocatorField(action.paramSchema);
102
- const requiredLocateFields = findAllMidsceneLocatorField(action.paramSchema, true);
102
+ validateRequiredLocateFields(param, action.paramSchema);
103
103
  locateFields.forEach((field)=>{
104
- if (param[field]) {
104
+ if (param?.[field]) {
105
105
  const locatePlan = locatePlanForLocate(param[field]);
106
106
  debug('will prepend locate param for field', `action.type=${planType}`, `param=${JSON.stringify(param[field])}`, `locatePlan=${JSON.stringify(locatePlan)}`, `hasLocatedPixelResult=${ifLocateParamHasLocatedPixelResult(param[field])}`);
107
107
  const locateTask = this.createLocateTask(locatePlan, param[field], context, (result)=>{
108
108
  param[field] = result;
109
109
  });
110
110
  context.tasks.push(locateTask);
111
- } else {
112
- assert(!requiredLocateFields.includes(field), `Required locate field '${field}' is not provided for action ${planType}`);
113
- debug(`field '${field}' is not provided for action ${planType}`);
114
- }
111
+ } else debug(`field '${field}' is not provided for action ${planType}`);
115
112
  });
116
113
  const task = {
117
114
  type: 'Action Space',
@@ -123,9 +120,7 @@ class TaskBuilder {
123
120
  debug('executing action', planType, param, `taskContext.element.center: ${taskContext.element?.center}`);
124
121
  const uiContext = taskContext.uiContext;
125
122
  assert(uiContext, 'uiContext is required for Action task');
126
- requiredLocateFields.forEach((field)=>{
127
- assert(param[field], `field '${field}' is required for action ${planType} but not provided. Cannot execute action ${planType}.`);
128
- });
123
+ validateRequiredLocateFields(param, action.paramSchema);
129
124
  setTimingFieldOnce(timing, 'beforeInvokeActionHookStart');
130
125
  const delayBeforeRunner = action.delayBeforeRunner ?? 200;
131
126
  try {
@@ -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 { ModelRuntime } from '@/ai-model/models';\nimport { findActionInActionSpaceOrThrow } from '@/common';\nimport type { AbstractInterface } from '@/device';\nimport type Service from '@/service';\nimport { setTimingFieldOnce } from '@/task-timing';\nimport type {\n AIUsageInfo,\n DetailedLocateParam,\n DeviceAction,\n ElementCacheFeature,\n ExecutionTaskActionApply,\n ExecutionTaskApply,\n ExecutionTaskHitBy,\n ExecutionTaskPlanningLocateApply,\n LocateResultElement,\n LocateResultWithDump,\n PlanningAction,\n Rect,\n ServiceDump,\n} from '@/types';\nimport { ServiceError } from '@/types';\nimport { sleep } from '@/utils';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type { TaskCache } from './task-cache';\nimport { withUsageIntent } from './usage-intent';\nimport {\n ifLocateParamHasLocatedPixelResult,\n matchElementFromCache,\n matchElementFromPlan,\n transformLogicalElementToScreenshot,\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\nfunction invalidLocateElementReason(\n element: LocateResultElement,\n): string | undefined {\n const values = [element.center?.[0], element.center?.[1]];\n if (\n element.center?.length !== 2 ||\n values.some((value) => typeof value !== 'number' || !Number.isFinite(value))\n ) {\n return `Invalid locate result coordinates: ${JSON.stringify(element)}`;\n }\n return undefined;\n}\n\ntype LocateParamWithDeprecatedAlias = DetailedLocateParam & {\n deepThink?: boolean;\n};\n\nfunction normalizeLocateParam(\n param: string | DetailedLocateParam,\n): DetailedLocateParam {\n if (typeof param === 'string') {\n return { prompt: param };\n }\n\n const { deepThink, ...rest } = param as LocateParamWithDeprecatedAlias;\n const deepLocate = rest.deepLocate ?? deepThink;\n\n return deepLocate === undefined ? rest : { ...rest, deepLocate };\n}\n\nexport function locatePlanForLocate(param: string | DetailedLocateParam) {\n const locate = normalizeLocateParam(param);\n const locatePlan: PlanningAction<DetailedLocateParam> = {\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 waitAfterAction?: number;\n}\n\ninterface BuildOptions {\n cacheable?: boolean;\n deepLocate?: boolean;\n abortSignal?: AbortSignal;\n}\n\ninterface PlanBuildContext {\n tasks: ExecutionTaskApply[];\n planningModel: ModelRuntime;\n defaultModel: ModelRuntime;\n cacheable?: boolean;\n deepLocate?: boolean;\n abortSignal?: AbortSignal;\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 private readonly waitAfterAction?: number;\n\n constructor({\n interfaceInstance,\n service,\n taskCache,\n actionSpace,\n waitAfterAction,\n }: TaskBuilderDeps) {\n this.interface = interfaceInstance;\n this.service = service;\n this.taskCache = taskCache;\n this.actionSpace = actionSpace;\n this.waitAfterAction = waitAfterAction;\n }\n\n public async build(\n plans: PlanningAction[],\n planningModel: ModelRuntime,\n defaultModel: ModelRuntime,\n options?: BuildOptions,\n ): Promise<{ tasks: ExecutionTaskApply[] }> {\n const tasks: ExecutionTaskApply[] = [];\n const cacheable = options?.cacheable;\n\n const context: PlanBuildContext = {\n tasks,\n planningModel,\n defaultModel,\n cacheable,\n deepLocate: options?.deepLocate,\n abortSignal: options?.abortSignal,\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<DetailedLocateParam>,\n context,\n ),\n ],\n ['Finished', (plan) => this.handleFinishedPlan(plan, context)],\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 executor: async () => {},\n };\n context.tasks.push(taskActionFinished);\n }\n\n private async handleLocatePlan(\n plan: PlanningAction<DetailedLocateParam>,\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 = findActionInActionSpaceOrThrow(planType, actionSpace);\n const param = plan.param;\n\n const locateFields = findAllMidsceneLocatorField(action.paramSchema);\n\n const requiredLocateFields = findAllMidsceneLocatorField(\n action.paramSchema,\n true,\n );\n\n locateFields.forEach((field) => {\n if (param[field]) {\n // Always use createLocateTask for all locate params.\n // This ensures cache writing happens even when locatedPixelResult 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 `hasLocatedPixelResult=${ifLocateParamHasLocatedPixelResult(param[field])}`,\n );\n const locateTask = this.createLocateTask(\n locatePlan,\n param[field],\n context,\n (result) => {\n param[field] = result;\n },\n );\n context.tasks.push(locateTask);\n } else {\n assert(\n !requiredLocateFields.includes(field),\n `Required locate field '${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 executor: async (taskContext) => {\n const timing = taskContext.task.timing;\n\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 setTimingFieldOnce(timing, 'beforeInvokeActionHookStart');\n const delayBeforeRunner = action.delayBeforeRunner ?? 200;\n try {\n await Promise.all([\n (async () => {\n if (this.interface.beforeInvokeAction) {\n debug(\n `will call \"beforeInvokeAction\" for interface with action name ${action.name}`,\n );\n await this.interface.beforeInvokeAction(action.name, param);\n debug(\n `called \"beforeInvokeAction\" for interface with action name ${action.name}`,\n );\n }\n })(),\n delayBeforeRunner > 0\n ? sleep(delayBeforeRunner)\n : Promise.resolve(),\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 setTimingFieldOnce(timing, 'beforeInvokeActionHookEnd');\n\n const { shrunkShotToLogicalRatio } = uiContext;\n if (shrunkShotToLogicalRatio === undefined) {\n throw new Error(\n 'shrunkShotToLogicalRatio is not defined in Action task',\n );\n }\n\n const parsedParam = (() => {\n if (!action.paramSchema) {\n return param;\n }\n\n try {\n return parseActionParam(param, action.paramSchema, {\n shrunkShotToLogicalRatio,\n });\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 setTimingFieldOnce(timing, 'callActionStart');\n\n debug('calling action', action.name);\n const actionFn = action.call.bind(this.interface);\n const actionResult = await actionFn(parsedParam, taskContext);\n setTimingFieldOnce(timing, 'callActionEnd');\n debug('called action', action.name, 'result:', actionResult);\n\n setTimingFieldOnce(timing, 'afterInvokeActionHookStart');\n\n const delayAfterRunner =\n action.delayAfterRunner ?? this.waitAfterAction ?? 300;\n if (delayAfterRunner > 0) {\n await sleep(delayAfterRunner);\n }\n\n try {\n if (this.interface.afterInvokeAction) {\n debug(\n `will call \"afterInvokeAction\" for interface with action name ${action.name}`,\n );\n await this.interface.afterInvokeAction(action.name, parsedParam);\n debug(\n `called \"afterInvokeAction\" for interface with action name ${action.name}`,\n );\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}: ${originalMessage}`,\n { cause: originalError },\n );\n }\n\n setTimingFieldOnce(timing, 'afterInvokeActionHookEnd');\n\n return {\n output: actionResult,\n };\n },\n };\n\n context.tasks.push(task);\n }\n\n private createLocateTask(\n plan: PlanningAction<DetailedLocateParam>,\n detailedLocateParam: DetailedLocateParam | string,\n context: PlanBuildContext,\n onResult?: (result: LocateResultElement) => void,\n ): ExecutionTaskPlanningLocateApply {\n const { cacheable, defaultModel, deepLocate, abortSignal } = context;\n\n let locateParam = normalizeLocateParam(detailedLocateParam);\n\n if (cacheable !== undefined) {\n locateParam = {\n ...locateParam,\n cacheable,\n };\n }\n\n if (deepLocate && !locateParam.deepLocate) {\n locateParam = {\n ...locateParam,\n deepLocate: true,\n };\n }\n\n const taskLocator: ExecutionTaskPlanningLocateApply = {\n type: 'Planning',\n subType: 'Locate',\n param: locateParam,\n thought: plan.thought,\n executor: async (taskContext) => {\n const { task } = taskContext;\n let { uiContext } = taskContext;\n const paramWithLocatedPixelResult = ifLocateParamHasLocatedPixelResult(\n locateParam,\n )\n ? locateParam\n : undefined;\n\n assert(\n locateParam?.prompt || paramWithLocatedPixelResult,\n `No prompt or id or position or locatedPixelResult to locate, param=${JSON.stringify(\n locateParam,\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 const { shrunkShotToLogicalRatio } = uiContext;\n\n if (shrunkShotToLogicalRatio === undefined) {\n throw new Error(\n 'shrunkShotToLogicalRatio is not defined in locate task',\n );\n }\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 rawResponse: dump.taskInfo?.rawResponse,\n rawChoiceMessage: dump.taskInfo?.rawChoiceMessage,\n searchAreaRawChoiceMessage:\n dump.taskInfo?.searchAreaRawChoiceMessage,\n };\n task.usage = withUsageIntent(dump.taskInfo?.usage, 'default');\n task.searchArea = dump.taskInfo?.searchArea;\n if (dump.taskInfo?.searchAreaUsage) {\n // The deepLocate search-area call belongs to the same locate task,\n // so it shares the companion usage intent.\n task.searchAreaUsage = withUsageIntent(\n dump.taskInfo.searchAreaUsage,\n 'default',\n );\n }\n if (dump.taskInfo?.reasoning_content) {\n task.reasoning_content = dump.taskInfo.reasoning_content;\n }\n };\n\n const planLocatedElement = paramWithLocatedPixelResult\n ? matchElementFromPlan(paramWithLocatedPixelResult)\n : undefined;\n\n // from locatedPixelResult (direct plan hit)\n // when deepLocate is enabled, locatedPixelResult should be used as search\n // area hint, not as a final direct hit\n const elementFromPlan = locateParam.deepLocate\n ? undefined\n : planLocatedElement;\n const isPlanDirectHit = !!elementFromPlan;\n\n // from xpath\n let rectFromXpath: Rect | undefined;\n if (\n !isPlanDirectHit &&\n locateParam.xpath &&\n this.interface.rectMatchesCacheFeature\n ) {\n try {\n rectFromXpath = await this.interface.rectMatchesCacheFeature({\n xpaths: [locateParam.xpath],\n });\n } catch {\n // xpath locate failed, allow fallback to cache or AI locate\n }\n }\n\n const elementFromXpath = rectFromXpath\n ? transformLogicalElementToScreenshot(\n {\n center: [\n rectFromXpath.left + rectFromXpath.width / 2,\n rectFromXpath.top + rectFromXpath.height / 2,\n ],\n description:\n typeof locateParam.prompt === 'string'\n ? locateParam.prompt\n : locateParam.prompt?.prompt || '',\n rect: rectFromXpath,\n },\n shrunkShotToLogicalRatio,\n )\n : undefined;\n\n const isXpathHit = !!elementFromXpath;\n\n const cachePrompt = locateParam.prompt;\n const locateCacheRecord = this.taskCache?.matchLocateCache(cachePrompt);\n const cacheEntry = locateCacheRecord?.cacheContent?.cache;\n\n const elementFromCacheResult =\n isPlanDirectHit || isXpathHit\n ? null\n : await matchElementFromCache(\n {\n taskCache: this.taskCache,\n interfaceInstance: this.interface,\n },\n cacheEntry,\n cachePrompt,\n locateParam.cacheable,\n );\n\n // elementFromCacheResult is in logical coordinates, which should be transformed to screenshot coordinates;\n const elementFromCache = elementFromCacheResult\n ? transformLogicalElementToScreenshot(\n elementFromCacheResult,\n shrunkShotToLogicalRatio,\n )\n : undefined;\n\n const isCacheHit = !!elementFromCache;\n\n let elementFromAiLocate: LocateResultElement | null | undefined;\n const timing = taskContext.task.timing;\n if (!isXpathHit && !isCacheHit && !isPlanDirectHit) {\n try {\n setTimingFieldOnce(timing, 'callAiStart');\n locateResult = await this.service.locate(\n locateParam,\n {\n context: uiContext,\n planLocatedElement,\n },\n defaultModel,\n abortSignal,\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 } finally {\n setTimingFieldOnce(timing, 'callAiEnd');\n }\n }\n\n const element =\n elementFromPlan ||\n elementFromXpath ||\n elementFromCache ||\n elementFromAiLocate;\n\n if (element) {\n const invalidElementReason = invalidLocateElementReason(element);\n if (invalidElementReason) {\n if (locateDump) {\n throw new ServiceError(invalidElementReason, locateDump);\n }\n throw new Error(invalidElementReason);\n }\n }\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 plan hit case (avoid redundant writes), OR allow update if cache validation failed\n // 5. cacheable is not explicitly false\n if (\n element &&\n this.taskCache &&\n !isCacheHit &&\n (!isPlanDirectHit || !locateCacheAlreadyExists) &&\n locateParam.cacheable !== false\n ) {\n if (this.interface.cacheFeatureForPoint) {\n try {\n // Transform coordinates to logical space for cacheFeatureForPoint\n // cacheFeatureForPoint needs logical coordinates to locate elements in DOM\n const pointForCache: [number, number] = [\n Math.round(element.center[0] / shrunkShotToLogicalRatio),\n Math.round(element.center[1] / shrunkShotToLogicalRatio),\n ];\n\n const feature = await this.interface.cacheFeatureForPoint(\n pointForCache,\n {\n targetDescription:\n typeof locateParam.prompt === 'string'\n ? locateParam.prompt\n : locateParam.prompt?.prompt,\n modelRuntime: defaultModel,\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('cacheFeatureForPoint failed: %s', error);\n }\n } else {\n debug('cacheFeatureForPoint is not supported, skip cache update');\n }\n }\n\n if (!element) {\n if (locateDump) {\n throw new ServiceError(\n `Element not found : ${locateParam.prompt}`,\n locateDump,\n );\n }\n throw new Error(`Element not found: ${locateParam.prompt}`);\n }\n\n let hitBy: ExecutionTaskHitBy | undefined;\n\n if (isPlanDirectHit && paramWithLocatedPixelResult) {\n hitBy = {\n from: 'Plan',\n context: {\n locatedPixelResult:\n paramWithLocatedPixelResult.locatedPixelResult,\n },\n };\n } else if (isXpathHit) {\n hitBy = {\n from: 'User expected path',\n context: {\n xpath: locateParam.xpath,\n },\n };\n } else if (isCacheHit) {\n hitBy = {\n from: 'Cache',\n context: {\n cacheEntry,\n cacheToSave: currentCacheEntry,\n },\n };\n }\n\n const promptDisplay = locateParam.promptDisplay;\n const elementForAction = promptDisplay\n ? {\n ...element,\n description: promptDisplay,\n }\n : element;\n\n if (promptDisplay && locateDump?.matchedElement) {\n locateDump.matchedElement = locateDump.matchedElement.map(\n (matchedElement) => ({\n ...matchedElement,\n description: promptDisplay,\n }),\n );\n }\n\n onResult?.(elementForAction);\n\n return {\n output: {\n element: {\n ...elementForAction,\n // backward compatibility for aiLocate, which return value needs a dpr field\n dpr: uiContext.deprecatedDpr,\n },\n },\n hitBy,\n };\n },\n };\n\n return taskLocator;\n }\n}\n"],"names":["debug","getDebug","hasNonEmptyCache","cache","Object","invalidLocateElementReason","element","values","value","Number","JSON","normalizeLocateParam","param","deepThink","rest","deepLocate","undefined","locatePlanForLocate","locate","locatePlan","TaskBuilder","plans","planningModel","defaultModel","options","tasks","cacheable","context","planHandlers","Map","plan","defaultHandler","handler","taskActionFinished","taskLocate","planType","actionSpace","action","findActionInActionSpaceOrThrow","locateFields","findAllMidsceneLocatorField","requiredLocateFields","field","ifLocateParamHasLocatedPixelResult","locateTask","result","assert","task","taskContext","timing","uiContext","setTimingFieldOnce","delayBeforeRunner","Promise","sleep","originalError","originalMessage","String","Error","shrunkShotToLogicalRatio","parsedParam","parseActionParam","error","actionFn","actionResult","delayAfterRunner","detailedLocateParam","onResult","abortSignal","locateParam","taskLocator","paramWithLocatedPixelResult","locateDump","locateResult","applyDump","dump","withUsageIntent","planLocatedElement","matchElementFromPlan","elementFromPlan","isPlanDirectHit","rectFromXpath","elementFromXpath","transformLogicalElementToScreenshot","isXpathHit","cachePrompt","locateCacheRecord","cacheEntry","elementFromCacheResult","matchElementFromCache","elementFromCache","isCacheHit","elementFromAiLocate","ServiceError","invalidElementReason","locateCacheAlreadyExists","currentCacheEntry","pointForCache","Math","feature","hitBy","promptDisplay","elementForAction","matchedElement","interfaceInstance","service","taskCache","waitAfterAction"],"mappings":";;;;;;;;;;;;;;;;;;;AAkCA,MAAMA,QAAQC,SAAS;AAKvB,SAASC,iBAAiBC,KAAc;IACtC,OACEA,QAAAA,SAEA,AAAiB,YAAjB,OAAOA,SACPC,OAAO,IAAI,CAACD,OAAO,MAAM,GAAG;AAEhC;AAEA,SAASE,2BACPC,OAA4B;IAE5B,MAAMC,SAAS;QAACD,QAAQ,MAAM,EAAE,CAAC,EAAE;QAAEA,QAAQ,MAAM,EAAE,CAAC,EAAE;KAAC;IACzD,IACEA,QAAQ,MAAM,EAAE,WAAW,KAC3BC,OAAO,IAAI,CAAC,CAACC,QAAU,AAAiB,YAAjB,OAAOA,SAAsB,CAACC,OAAO,QAAQ,CAACD,SAErE,OAAO,CAAC,mCAAmC,EAAEE,KAAK,SAAS,CAACJ,UAAU;AAG1E;AAMA,SAASK,qBACPC,KAAmC;IAEnC,IAAI,AAAiB,YAAjB,OAAOA,OACT,OAAO;QAAE,QAAQA;IAAM;IAGzB,MAAM,EAAEC,SAAS,EAAE,GAAGC,MAAM,GAAGF;IAC/B,MAAMG,aAAaD,KAAK,UAAU,IAAID;IAEtC,OAAOE,AAAeC,WAAfD,aAA2BD,OAAO;QAAE,GAAGA,IAAI;QAAEC;IAAW;AACjE;AAEO,SAASE,oBAAoBL,KAAmC;IACrE,MAAMM,SAASP,qBAAqBC;IACpC,MAAMO,aAAkD;QACtD,MAAM;QACN,OAAOD;QACP,SAAS;IACX;IACA,OAAOC;AACT;AAyBO,MAAMC;IAyBX,MAAa,MACXC,KAAuB,EACvBC,aAA2B,EAC3BC,YAA0B,EAC1BC,OAAsB,EACoB;QAC1C,MAAMC,QAA8B,EAAE;QACtC,MAAMC,YAAYF,SAAS;QAE3B,MAAMG,UAA4B;YAChCF;YACAH;YACAC;YACAG;YACA,YAAYF,SAAS;YACrB,aAAaA,SAAS;QACxB;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;SAC/D;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,UAAU,WAAa;QACzB;QACAH,QAAQ,KAAK,CAAC,IAAI,CAACM;IACrB;IAEA,MAAc,iBACZH,IAAyC,EACzCH,OAAyB,EACV;QACf,MAAMO,aAAa,IAAI,CAAC,gBAAgB,CAACJ,MAAMA,KAAK,KAAK,EAAEH;QAC3DA,QAAQ,KAAK,CAAC,IAAI,CAACO;IACrB;IAEA,MAAc,iBACZJ,IAAoB,EACpBH,OAAyB,EACV;QACf,MAAMQ,WAAWL,KAAK,IAAI;QAC1B,MAAMM,cAAc,IAAI,CAAC,WAAW;QACpC,MAAMC,SAASC,+BAA+BH,UAAUC;QACxD,MAAMxB,QAAQkB,KAAK,KAAK;QAExB,MAAMS,eAAeC,4BAA4BH,OAAO,WAAW;QAEnE,MAAMI,uBAAuBD,4BAC3BH,OAAO,WAAW,EAClB;QAGFE,aAAa,OAAO,CAAC,CAACG;YACpB,IAAI9B,KAAK,CAAC8B,MAAM,EAAE;gBAGhB,MAAMvB,aAAaF,oBAAoBL,KAAK,CAAC8B,MAAM;gBACnD1C,MACE,uCACA,CAAC,YAAY,EAAEmC,UAAU,EACzB,CAAC,MAAM,EAAEzB,KAAK,SAAS,CAACE,KAAK,CAAC8B,MAAM,GAAG,EACvC,CAAC,WAAW,EAAEhC,KAAK,SAAS,CAACS,aAAa,EAC1C,CAAC,sBAAsB,EAAEwB,mCAAmC/B,KAAK,CAAC8B,MAAM,GAAG;gBAE7E,MAAME,aAAa,IAAI,CAAC,gBAAgB,CACtCzB,YACAP,KAAK,CAAC8B,MAAM,EACZf,SACA,CAACkB;oBACCjC,KAAK,CAAC8B,MAAM,GAAGG;gBACjB;gBAEFlB,QAAQ,KAAK,CAAC,IAAI,CAACiB;YACrB,OAAO;gBACLE,OACE,CAACL,qBAAqB,QAAQ,CAACC,QAC/B,CAAC,uBAAuB,EAAEA,MAAM,6BAA6B,EAAEP,UAAU;gBAE3EnC,MAAM,CAAC,OAAO,EAAE0C,MAAM,6BAA6B,EAAEP,UAAU;YACjE;QACF;QAEA,MAAMY,OAKF;YACF,MAAM;YACN,SAASZ;YACT,SAASL,KAAK,OAAO;YACrB,OAAOA,KAAK,KAAK;YACjB,UAAU,OAAOkB;gBACf,MAAMC,SAASD,YAAY,IAAI,CAAC,MAAM;gBAEtChD,MACE,oBACAmC,UACAvB,OACA,CAAC,4BAA4B,EAAEoC,YAAY,OAAO,EAAE,QAAQ;gBAG9D,MAAME,YAAYF,YAAY,SAAS;gBACvCF,OAAOI,WAAW;gBAElBT,qBAAqB,OAAO,CAAC,CAACC;oBAC5BI,OACElC,KAAK,CAAC8B,MAAM,EACZ,CAAC,OAAO,EAAEA,MAAM,yBAAyB,EAAEP,SAAS,yCAAyC,EAAEA,SAAS,CAAC,CAAC;gBAE9G;gBAEAgB,mBAAmBF,QAAQ;gBAC3B,MAAMG,oBAAoBf,OAAO,iBAAiB,IAAI;gBACtD,IAAI;oBACF,MAAMgB,QAAQ,GAAG,CAAC;wBACf;4BACC,IAAI,IAAI,CAAC,SAAS,CAAC,kBAAkB,EAAE;gCACrCrD,MACE,CAAC,8DAA8D,EAAEqC,OAAO,IAAI,EAAE;gCAEhF,MAAM,IAAI,CAAC,SAAS,CAAC,kBAAkB,CAACA,OAAO,IAAI,EAAEzB;gCACrDZ,MACE,CAAC,2DAA2D,EAAEqC,OAAO,IAAI,EAAE;4BAE/E;wBACF;wBACAe,oBAAoB,IAChBE,MAAMF,qBACNC,QAAQ,OAAO;qBACpB;gBACH,EAAE,OAAOE,eAAoB;oBAC3B,MAAMC,kBACJD,eAAe,WAAWE,OAAOF;oBACnC,MAAM,IAAIG,MACR,CAAC,wCAAwC,EAAErB,OAAO,IAAI,CAAC,EAAE,EAAEmB,iBAAiB,EAC5E;wBAAE,OAAOD;oBAAc;gBAE3B;gBACAJ,mBAAmBF,QAAQ;gBAE3B,MAAM,EAAEU,wBAAwB,EAAE,GAAGT;gBACrC,IAAIS,AAA6B3C,WAA7B2C,0BACF,MAAM,IAAID,MACR;gBAIJ,MAAME,cAAe,AAAC;oBACpB,IAAI,CAACvB,OAAO,WAAW,EACrB,OAAOzB;oBAGT,IAAI;wBACF,OAAOiD,iBAAiBjD,OAAOyB,OAAO,WAAW,EAAE;4BACjDsB;wBACF;oBACF,EAAE,OAAOG,OAAY;wBACnB,MAAM,IAAIJ,MACR,CAAC,8BAA8B,EAAErB,OAAO,IAAI,CAAC,EAAE,EAAEyB,MAAM,OAAO,CAAC,cAAc,EAAEpD,KAAK,SAAS,CAACE,QAAQ,EACtG;4BAAE,OAAOkD;wBAAM;oBAEnB;gBACF;gBAEAX,mBAAmBF,QAAQ;gBAE3BjD,MAAM,kBAAkBqC,OAAO,IAAI;gBACnC,MAAM0B,WAAW1B,OAAO,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS;gBAChD,MAAM2B,eAAe,MAAMD,SAASH,aAAaZ;gBACjDG,mBAAmBF,QAAQ;gBAC3BjD,MAAM,iBAAiBqC,OAAO,IAAI,EAAE,WAAW2B;gBAE/Cb,mBAAmBF,QAAQ;gBAE3B,MAAMgB,mBACJ5B,OAAO,gBAAgB,IAAI,IAAI,CAAC,eAAe,IAAI;gBACrD,IAAI4B,mBAAmB,GACrB,MAAMX,MAAMW;gBAGd,IAAI;oBACF,IAAI,IAAI,CAAC,SAAS,CAAC,iBAAiB,EAAE;wBACpCjE,MACE,CAAC,6DAA6D,EAAEqC,OAAO,IAAI,EAAE;wBAE/E,MAAM,IAAI,CAAC,SAAS,CAAC,iBAAiB,CAACA,OAAO,IAAI,EAAEuB;wBACpD5D,MACE,CAAC,0DAA0D,EAAEqC,OAAO,IAAI,EAAE;oBAE9E;gBACF,EAAE,OAAOkB,eAAoB;oBAC3B,MAAMC,kBACJD,eAAe,WAAWE,OAAOF;oBACnC,MAAM,IAAIG,MACR,CAAC,uCAAuC,EAAErB,OAAO,IAAI,CAAC,EAAE,EAAEmB,iBAAiB,EAC3E;wBAAE,OAAOD;oBAAc;gBAE3B;gBAEAJ,mBAAmBF,QAAQ;gBAE3B,OAAO;oBACL,QAAQe;gBACV;YACF;QACF;QAEArC,QAAQ,KAAK,CAAC,IAAI,CAACoB;IACrB;IAEQ,iBACNjB,IAAyC,EACzCoC,mBAAiD,EACjDvC,OAAyB,EACzBwC,QAAgD,EACd;QAClC,MAAM,EAAEzC,SAAS,EAAEH,YAAY,EAAER,UAAU,EAAEqD,WAAW,EAAE,GAAGzC;QAE7D,IAAI0C,cAAc1D,qBAAqBuD;QAEvC,IAAIxC,AAAcV,WAAdU,WACF2C,cAAc;YACZ,GAAGA,WAAW;YACd3C;QACF;QAGF,IAAIX,cAAc,CAACsD,YAAY,UAAU,EACvCA,cAAc;YACZ,GAAGA,WAAW;YACd,YAAY;QACd;QAGF,MAAMC,cAAgD;YACpD,MAAM;YACN,SAAS;YACT,OAAOD;YACP,SAASvC,KAAK,OAAO;YACrB,UAAU,OAAOkB;gBACf,MAAM,EAAED,IAAI,EAAE,GAAGC;gBACjB,IAAI,EAAEE,SAAS,EAAE,GAAGF;gBACpB,MAAMuB,8BAA8B5B,mCAClC0B,eAEEA,cACArD;gBAEJ8B,OACEuB,aAAa,UAAUE,6BACvB,CAAC,mEAAmE,EAAE7D,KAAK,SAAS,CAClF2D,cACC;gBAGL,IAAI,CAACnB,WACHA,YAAY,MAAM,IAAI,CAAC,OAAO,CAAC,kBAAkB;gBAGnDJ,OAAOI,WAAW;gBAElB,MAAM,EAAES,wBAAwB,EAAE,GAAGT;gBAErC,IAAIS,AAA6B3C,WAA7B2C,0BACF,MAAM,IAAID,MACR;gBAIJ,IAAIc;gBACJ,IAAIC;gBAEJ,MAAMC,YAAY,CAACC;oBACjB,IAAI,CAACA,MACH;oBAEFH,aAAaG;oBACb5B,KAAK,GAAG,GAAG;wBACT4B;wBACA,aAAaA,KAAK,QAAQ,EAAE;wBAC5B,kBAAkBA,KAAK,QAAQ,EAAE;wBACjC,4BACEA,KAAK,QAAQ,EAAE;oBACnB;oBACA5B,KAAK,KAAK,GAAG6B,gBAAgBD,KAAK,QAAQ,EAAE,OAAO;oBACnD5B,KAAK,UAAU,GAAG4B,KAAK,QAAQ,EAAE;oBACjC,IAAIA,KAAK,QAAQ,EAAE,iBAGjB5B,KAAK,eAAe,GAAG6B,gBACrBD,KAAK,QAAQ,CAAC,eAAe,EAC7B;oBAGJ,IAAIA,KAAK,QAAQ,EAAE,mBACjB5B,KAAK,iBAAiB,GAAG4B,KAAK,QAAQ,CAAC,iBAAiB;gBAE5D;gBAEA,MAAME,qBAAqBN,8BACvBO,qBAAqBP,+BACrBvD;gBAKJ,MAAM+D,kBAAkBV,YAAY,UAAU,GAC1CrD,SACA6D;gBACJ,MAAMG,kBAAkB,CAAC,CAACD;gBAG1B,IAAIE;gBACJ,IACE,CAACD,mBACDX,YAAY,KAAK,IACjB,IAAI,CAAC,SAAS,CAAC,uBAAuB,EAEtC,IAAI;oBACFY,gBAAgB,MAAM,IAAI,CAAC,SAAS,CAAC,uBAAuB,CAAC;wBAC3D,QAAQ;4BAACZ,YAAY,KAAK;yBAAC;oBAC7B;gBACF,EAAE,OAAM,CAER;gBAGF,MAAMa,mBAAmBD,gBACrBE,oCACE;oBACE,QAAQ;wBACNF,cAAc,IAAI,GAAGA,cAAc,KAAK,GAAG;wBAC3CA,cAAc,GAAG,GAAGA,cAAc,MAAM,GAAG;qBAC5C;oBACD,aACE,AAA8B,YAA9B,OAAOZ,YAAY,MAAM,GACrBA,YAAY,MAAM,GAClBA,YAAY,MAAM,EAAE,UAAU;oBACpC,MAAMY;gBACR,GACAtB,4BAEF3C;gBAEJ,MAAMoE,aAAa,CAAC,CAACF;gBAErB,MAAMG,cAAchB,YAAY,MAAM;gBACtC,MAAMiB,oBAAoB,IAAI,CAAC,SAAS,EAAE,iBAAiBD;gBAC3D,MAAME,aAAaD,mBAAmB,cAAc;gBAEpD,MAAME,yBACJR,mBAAmBI,aACf,OACA,MAAMK,sBACJ;oBACE,WAAW,IAAI,CAAC,SAAS;oBACzB,mBAAmB,IAAI,CAAC,SAAS;gBACnC,GACAF,YACAF,aACAhB,YAAY,SAAS;gBAI7B,MAAMqB,mBAAmBF,yBACrBL,oCACEK,wBACA7B,4BAEF3C;gBAEJ,MAAM2E,aAAa,CAAC,CAACD;gBAErB,IAAIE;gBACJ,MAAM3C,SAASD,YAAY,IAAI,CAAC,MAAM;gBACtC,IAAI,CAACoC,cAAc,CAACO,cAAc,CAACX,iBACjC,IAAI;oBACF7B,mBAAmBF,QAAQ;oBAC3BwB,eAAe,MAAM,IAAI,CAAC,OAAO,CAAC,MAAM,CACtCJ,aACA;wBACE,SAASnB;wBACT2B;oBACF,GACAtD,cACA6C;oBAEFM,UAAUD,aAAa,IAAI;oBAC3BmB,sBAAsBnB,aAAa,OAAO;gBAC5C,EAAE,OAAOX,OAAO;oBACd,IAAIA,iBAAiB+B,cACnBnB,UAAUZ,MAAM,IAAI;oBAEtB,MAAMA;gBACR,SAAU;oBACRX,mBAAmBF,QAAQ;gBAC7B;gBAGF,MAAM3C,UACJyE,mBACAG,oBACAQ,oBACAE;gBAEF,IAAItF,SAAS;oBACX,MAAMwF,uBAAuBzF,2BAA2BC;oBACxD,IAAIwF,sBAAsB;wBACxB,IAAItB,YACF,MAAM,IAAIqB,aAAaC,sBAAsBtB;wBAE/C,MAAM,IAAId,MAAMoC;oBAClB;gBACF;gBAGA,MAAMC,2BAA2B7F,iBAC/BoF,mBAAmB,cAAc;gBAGnC,IAAIU;gBAOJ,IACE1F,WACA,IAAI,CAAC,SAAS,IACd,CAACqF,cACA,EAACX,mBAAmB,CAACe,wBAAuB,KAC7C1B,AAA0B,UAA1BA,YAAY,SAAS,EAErB,IAAI,IAAI,CAAC,SAAS,CAAC,oBAAoB,EACrC,IAAI;oBAGF,MAAM4B,gBAAkC;wBACtCC,KAAK,KAAK,CAAC5F,QAAQ,MAAM,CAAC,EAAE,GAAGqD;wBAC/BuC,KAAK,KAAK,CAAC5F,QAAQ,MAAM,CAAC,EAAE,GAAGqD;qBAChC;oBAED,MAAMwC,UAAU,MAAM,IAAI,CAAC,SAAS,CAAC,oBAAoB,CACvDF,eACA;wBACE,mBACE,AAA8B,YAA9B,OAAO5B,YAAY,MAAM,GACrBA,YAAY,MAAM,GAClBA,YAAY,MAAM,EAAE;wBAC1B,cAAc9C;oBAChB;oBAEF,IAAIrB,iBAAiBiG,UAAU;wBAC7BnG,MACE,uCACAqF,aACAc;wBAEFH,oBAAoBG;wBACpB,IAAI,CAAC,SAAS,CAAC,yBAAyB,CACtC;4BACE,MAAM;4BACN,QAAQd;4BACR,OAAOc;wBACT,GACAb;oBAEJ,OACEtF,MACE,yDACAqF;gBAGN,EAAE,OAAOvB,OAAO;oBACd9D,MAAM,mCAAmC8D;gBAC3C;qBAEA9D,MAAM;gBAIV,IAAI,CAACM,SAAS;oBACZ,IAAIkE,YACF,MAAM,IAAIqB,aACR,CAAC,oBAAoB,EAAExB,YAAY,MAAM,EAAE,EAC3CG;oBAGJ,MAAM,IAAId,MAAM,CAAC,mBAAmB,EAAEW,YAAY,MAAM,EAAE;gBAC5D;gBAEA,IAAI+B;gBAEJ,IAAIpB,mBAAmBT,6BACrB6B,QAAQ;oBACN,MAAM;oBACN,SAAS;wBACP,oBACE7B,4BAA4B,kBAAkB;oBAClD;gBACF;qBACK,IAAIa,YACTgB,QAAQ;oBACN,MAAM;oBACN,SAAS;wBACP,OAAO/B,YAAY,KAAK;oBAC1B;gBACF;qBACK,IAAIsB,YACTS,QAAQ;oBACN,MAAM;oBACN,SAAS;wBACPb;wBACA,aAAaS;oBACf;gBACF;gBAGF,MAAMK,gBAAgBhC,YAAY,aAAa;gBAC/C,MAAMiC,mBAAmBD,gBACrB;oBACE,GAAG/F,OAAO;oBACV,aAAa+F;gBACf,IACA/F;gBAEJ,IAAI+F,iBAAiB7B,YAAY,gBAC/BA,WAAW,cAAc,GAAGA,WAAW,cAAc,CAAC,GAAG,CACvD,CAAC+B,iBAAoB;wBACnB,GAAGA,cAAc;wBACjB,aAAaF;oBACf;gBAIJlC,WAAWmC;gBAEX,OAAO;oBACL,QAAQ;wBACN,SAAS;4BACP,GAAGA,gBAAgB;4BAEnB,KAAKpD,UAAU,aAAa;wBAC9B;oBACF;oBACAkD;gBACF;YACF;QACF;QAEA,OAAO9B;IACT;IAtlBA,YAAY,EACVkC,iBAAiB,EACjBC,OAAO,EACPC,SAAS,EACTtE,WAAW,EACXuE,eAAe,EACC,CAAE;QAhBpB,uBAAiB,aAAjB;QAEA,uBAAiB,WAAjB;QAEA,uBAAiB,aAAjB;QAEA,uBAAiB,eAAjB;QAEA,uBAAiB,mBAAjB;QASE,IAAI,CAAC,SAAS,GAAGH;QACjB,IAAI,CAAC,OAAO,GAAGC;QACf,IAAI,CAAC,SAAS,GAAGC;QACjB,IAAI,CAAC,WAAW,GAAGtE;QACnB,IAAI,CAAC,eAAe,GAAGuE;IACzB;AA2kBF"}
1
+ {"version":3,"file":"agent/task-builder.mjs","sources":["../../../src/agent/task-builder.ts"],"sourcesContent":["import { findAllMidsceneLocatorField, parseActionParam } from '@/ai-model';\nimport type { ModelRuntime } from '@/ai-model/models';\nimport {\n findActionInActionSpaceOrThrow,\n validateRequiredLocateFields,\n} from '@/common';\nimport type { AbstractInterface } from '@/device';\nimport type Service from '@/service';\nimport { setTimingFieldOnce } from '@/task-timing';\nimport type {\n AIUsageInfo,\n DetailedLocateParam,\n DeviceAction,\n ElementCacheFeature,\n ExecutionTaskActionApply,\n ExecutionTaskApply,\n ExecutionTaskHitBy,\n ExecutionTaskPlanningLocateApply,\n LocateResultElement,\n LocateResultWithDump,\n PlanningAction,\n Rect,\n ServiceDump,\n} from '@/types';\nimport { ServiceError } from '@/types';\nimport { sleep } from '@/utils';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type { TaskCache } from './task-cache';\nimport { withUsageIntent } from './usage-intent';\nimport {\n ifLocateParamHasLocatedPixelResult,\n matchElementFromCache,\n matchElementFromPlan,\n transformLogicalElementToScreenshot,\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\nfunction invalidLocateElementReason(\n element: LocateResultElement,\n): string | undefined {\n const values = [element.center?.[0], element.center?.[1]];\n if (\n element.center?.length !== 2 ||\n values.some((value) => typeof value !== 'number' || !Number.isFinite(value))\n ) {\n return `Invalid locate result coordinates: ${JSON.stringify(element)}`;\n }\n return undefined;\n}\n\ntype LocateParamWithDeprecatedAlias = DetailedLocateParam & {\n deepThink?: boolean;\n};\n\nfunction normalizeLocateParam(\n param: string | DetailedLocateParam,\n): DetailedLocateParam {\n if (typeof param === 'string') {\n return { prompt: param };\n }\n\n const { deepThink, ...rest } = param as LocateParamWithDeprecatedAlias;\n const deepLocate = rest.deepLocate ?? deepThink;\n\n return deepLocate === undefined ? rest : { ...rest, deepLocate };\n}\n\nexport function locatePlanForLocate(param: string | DetailedLocateParam) {\n const locate = normalizeLocateParam(param);\n const locatePlan: PlanningAction<DetailedLocateParam> = {\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 waitAfterAction?: number;\n}\n\ninterface BuildOptions {\n cacheable?: boolean;\n deepLocate?: boolean;\n abortSignal?: AbortSignal;\n}\n\ninterface PlanBuildContext {\n tasks: ExecutionTaskApply[];\n planningModel: ModelRuntime;\n defaultModel: ModelRuntime;\n cacheable?: boolean;\n deepLocate?: boolean;\n abortSignal?: AbortSignal;\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 private readonly waitAfterAction?: number;\n\n constructor({\n interfaceInstance,\n service,\n taskCache,\n actionSpace,\n waitAfterAction,\n }: TaskBuilderDeps) {\n this.interface = interfaceInstance;\n this.service = service;\n this.taskCache = taskCache;\n this.actionSpace = actionSpace;\n this.waitAfterAction = waitAfterAction;\n }\n\n public async build(\n plans: PlanningAction[],\n planningModel: ModelRuntime,\n defaultModel: ModelRuntime,\n options?: BuildOptions,\n ): Promise<{ tasks: ExecutionTaskApply[] }> {\n const tasks: ExecutionTaskApply[] = [];\n const cacheable = options?.cacheable;\n\n const context: PlanBuildContext = {\n tasks,\n planningModel,\n defaultModel,\n cacheable,\n deepLocate: options?.deepLocate,\n abortSignal: options?.abortSignal,\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<DetailedLocateParam>,\n context,\n ),\n ],\n ['Finished', (plan) => this.handleFinishedPlan(plan, context)],\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 executor: async () => {},\n };\n context.tasks.push(taskActionFinished);\n }\n\n private async handleLocatePlan(\n plan: PlanningAction<DetailedLocateParam>,\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 = findActionInActionSpaceOrThrow(planType, actionSpace);\n const param = plan.param;\n\n const locateFields = findAllMidsceneLocatorField(action.paramSchema);\n\n validateRequiredLocateFields(param, action.paramSchema);\n\n locateFields.forEach((field) => {\n if (param?.[field]) {\n // Always use createLocateTask for all locate params.\n // This ensures cache writing happens even when locatedPixelResult 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 `hasLocatedPixelResult=${ifLocateParamHasLocatedPixelResult(param[field])}`,\n );\n const locateTask = this.createLocateTask(\n locatePlan,\n param[field],\n context,\n (result) => {\n param[field] = result;\n },\n );\n context.tasks.push(locateTask);\n } else {\n debug(`field '${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 executor: async (taskContext) => {\n const timing = taskContext.task.timing;\n\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 validateRequiredLocateFields(param, action.paramSchema);\n\n setTimingFieldOnce(timing, 'beforeInvokeActionHookStart');\n const delayBeforeRunner = action.delayBeforeRunner ?? 200;\n try {\n await Promise.all([\n (async () => {\n if (this.interface.beforeInvokeAction) {\n debug(\n `will call \"beforeInvokeAction\" for interface with action name ${action.name}`,\n );\n await this.interface.beforeInvokeAction(action.name, param);\n debug(\n `called \"beforeInvokeAction\" for interface with action name ${action.name}`,\n );\n }\n })(),\n delayBeforeRunner > 0\n ? sleep(delayBeforeRunner)\n : Promise.resolve(),\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 setTimingFieldOnce(timing, 'beforeInvokeActionHookEnd');\n\n const { shrunkShotToLogicalRatio } = uiContext;\n if (shrunkShotToLogicalRatio === undefined) {\n throw new Error(\n 'shrunkShotToLogicalRatio is not defined in Action task',\n );\n }\n\n const parsedParam = (() => {\n if (!action.paramSchema) {\n return param;\n }\n\n try {\n return parseActionParam(param, action.paramSchema, {\n shrunkShotToLogicalRatio,\n });\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 setTimingFieldOnce(timing, 'callActionStart');\n\n debug('calling action', action.name);\n const actionFn = action.call.bind(this.interface);\n const actionResult = await actionFn(parsedParam, taskContext);\n setTimingFieldOnce(timing, 'callActionEnd');\n debug('called action', action.name, 'result:', actionResult);\n\n setTimingFieldOnce(timing, 'afterInvokeActionHookStart');\n\n const delayAfterRunner =\n action.delayAfterRunner ?? this.waitAfterAction ?? 300;\n if (delayAfterRunner > 0) {\n await sleep(delayAfterRunner);\n }\n\n try {\n if (this.interface.afterInvokeAction) {\n debug(\n `will call \"afterInvokeAction\" for interface with action name ${action.name}`,\n );\n await this.interface.afterInvokeAction(action.name, parsedParam);\n debug(\n `called \"afterInvokeAction\" for interface with action name ${action.name}`,\n );\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}: ${originalMessage}`,\n { cause: originalError },\n );\n }\n\n setTimingFieldOnce(timing, 'afterInvokeActionHookEnd');\n\n return {\n output: actionResult,\n };\n },\n };\n\n context.tasks.push(task);\n }\n\n private createLocateTask(\n plan: PlanningAction<DetailedLocateParam>,\n detailedLocateParam: DetailedLocateParam | string,\n context: PlanBuildContext,\n onResult?: (result: LocateResultElement) => void,\n ): ExecutionTaskPlanningLocateApply {\n const { cacheable, defaultModel, deepLocate, abortSignal } = context;\n\n let locateParam = normalizeLocateParam(detailedLocateParam);\n\n if (cacheable !== undefined) {\n locateParam = {\n ...locateParam,\n cacheable,\n };\n }\n\n if (deepLocate && !locateParam.deepLocate) {\n locateParam = {\n ...locateParam,\n deepLocate: true,\n };\n }\n\n const taskLocator: ExecutionTaskPlanningLocateApply = {\n type: 'Planning',\n subType: 'Locate',\n param: locateParam,\n thought: plan.thought,\n executor: async (taskContext) => {\n const { task } = taskContext;\n let { uiContext } = taskContext;\n const paramWithLocatedPixelResult = ifLocateParamHasLocatedPixelResult(\n locateParam,\n )\n ? locateParam\n : undefined;\n\n assert(\n locateParam?.prompt || paramWithLocatedPixelResult,\n `No prompt or id or position or locatedPixelResult to locate, param=${JSON.stringify(\n locateParam,\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 const { shrunkShotToLogicalRatio } = uiContext;\n\n if (shrunkShotToLogicalRatio === undefined) {\n throw new Error(\n 'shrunkShotToLogicalRatio is not defined in locate task',\n );\n }\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 rawResponse: dump.taskInfo?.rawResponse,\n rawChoiceMessage: dump.taskInfo?.rawChoiceMessage,\n searchAreaRawChoiceMessage:\n dump.taskInfo?.searchAreaRawChoiceMessage,\n };\n task.usage = withUsageIntent(dump.taskInfo?.usage, 'default');\n task.searchArea = dump.taskInfo?.searchArea;\n if (dump.taskInfo?.searchAreaUsage) {\n // The deepLocate search-area call belongs to the same locate task,\n // so it shares the companion usage intent.\n task.searchAreaUsage = withUsageIntent(\n dump.taskInfo.searchAreaUsage,\n 'default',\n );\n }\n if (dump.taskInfo?.reasoning_content) {\n task.reasoning_content = dump.taskInfo.reasoning_content;\n }\n };\n\n const planLocatedElement = paramWithLocatedPixelResult\n ? matchElementFromPlan(paramWithLocatedPixelResult)\n : undefined;\n\n // from locatedPixelResult (direct plan hit)\n // when deepLocate is enabled, locatedPixelResult should be used as search\n // area hint, not as a final direct hit\n const elementFromPlan = locateParam.deepLocate\n ? undefined\n : planLocatedElement;\n const isPlanDirectHit = !!elementFromPlan;\n\n // from xpath\n let rectFromXpath: Rect | undefined;\n if (\n !isPlanDirectHit &&\n locateParam.xpath &&\n this.interface.rectMatchesCacheFeature\n ) {\n try {\n rectFromXpath = await this.interface.rectMatchesCacheFeature({\n xpaths: [locateParam.xpath],\n });\n } catch {\n // xpath locate failed, allow fallback to cache or AI locate\n }\n }\n\n const elementFromXpath = rectFromXpath\n ? transformLogicalElementToScreenshot(\n {\n center: [\n rectFromXpath.left + rectFromXpath.width / 2,\n rectFromXpath.top + rectFromXpath.height / 2,\n ],\n description:\n typeof locateParam.prompt === 'string'\n ? locateParam.prompt\n : locateParam.prompt?.prompt || '',\n rect: rectFromXpath,\n },\n shrunkShotToLogicalRatio,\n )\n : undefined;\n\n const isXpathHit = !!elementFromXpath;\n\n const cachePrompt = locateParam.prompt;\n const locateCacheRecord = this.taskCache?.matchLocateCache(cachePrompt);\n const cacheEntry = locateCacheRecord?.cacheContent?.cache;\n\n const elementFromCacheResult =\n isPlanDirectHit || isXpathHit\n ? null\n : await matchElementFromCache(\n {\n taskCache: this.taskCache,\n interfaceInstance: this.interface,\n },\n cacheEntry,\n cachePrompt,\n locateParam.cacheable,\n );\n\n // elementFromCacheResult is in logical coordinates, which should be transformed to screenshot coordinates;\n const elementFromCache = elementFromCacheResult\n ? transformLogicalElementToScreenshot(\n elementFromCacheResult,\n shrunkShotToLogicalRatio,\n )\n : undefined;\n\n const isCacheHit = !!elementFromCache;\n\n let elementFromAiLocate: LocateResultElement | null | undefined;\n const timing = taskContext.task.timing;\n if (!isXpathHit && !isCacheHit && !isPlanDirectHit) {\n try {\n setTimingFieldOnce(timing, 'callAiStart');\n locateResult = await this.service.locate(\n locateParam,\n {\n context: uiContext,\n planLocatedElement,\n },\n defaultModel,\n abortSignal,\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 } finally {\n setTimingFieldOnce(timing, 'callAiEnd');\n }\n }\n\n const element =\n elementFromPlan ||\n elementFromXpath ||\n elementFromCache ||\n elementFromAiLocate;\n\n if (element) {\n const invalidElementReason = invalidLocateElementReason(element);\n if (invalidElementReason) {\n if (locateDump) {\n throw new ServiceError(invalidElementReason, locateDump);\n }\n throw new Error(invalidElementReason);\n }\n }\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 plan hit case (avoid redundant writes), OR allow update if cache validation failed\n // 5. cacheable is not explicitly false\n if (\n element &&\n this.taskCache &&\n !isCacheHit &&\n (!isPlanDirectHit || !locateCacheAlreadyExists) &&\n locateParam.cacheable !== false\n ) {\n if (this.interface.cacheFeatureForPoint) {\n try {\n // Transform coordinates to logical space for cacheFeatureForPoint\n // cacheFeatureForPoint needs logical coordinates to locate elements in DOM\n const pointForCache: [number, number] = [\n Math.round(element.center[0] / shrunkShotToLogicalRatio),\n Math.round(element.center[1] / shrunkShotToLogicalRatio),\n ];\n\n const feature = await this.interface.cacheFeatureForPoint(\n pointForCache,\n {\n targetDescription:\n typeof locateParam.prompt === 'string'\n ? locateParam.prompt\n : locateParam.prompt?.prompt,\n modelRuntime: defaultModel,\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('cacheFeatureForPoint failed: %s', error);\n }\n } else {\n debug('cacheFeatureForPoint is not supported, skip cache update');\n }\n }\n\n if (!element) {\n if (locateDump) {\n throw new ServiceError(\n `Element not found : ${locateParam.prompt}`,\n locateDump,\n );\n }\n throw new Error(`Element not found: ${locateParam.prompt}`);\n }\n\n let hitBy: ExecutionTaskHitBy | undefined;\n\n if (isPlanDirectHit && paramWithLocatedPixelResult) {\n hitBy = {\n from: 'Plan',\n context: {\n locatedPixelResult:\n paramWithLocatedPixelResult.locatedPixelResult,\n },\n };\n } else if (isXpathHit) {\n hitBy = {\n from: 'User expected path',\n context: {\n xpath: locateParam.xpath,\n },\n };\n } else if (isCacheHit) {\n hitBy = {\n from: 'Cache',\n context: {\n cacheEntry,\n cacheToSave: currentCacheEntry,\n },\n };\n }\n\n const promptDisplay = locateParam.promptDisplay;\n const elementForAction = promptDisplay\n ? {\n ...element,\n description: promptDisplay,\n }\n : element;\n\n if (promptDisplay && locateDump?.matchedElement) {\n locateDump.matchedElement = locateDump.matchedElement.map(\n (matchedElement) => ({\n ...matchedElement,\n description: promptDisplay,\n }),\n );\n }\n\n onResult?.(elementForAction);\n\n return {\n output: {\n element: {\n ...elementForAction,\n // backward compatibility for aiLocate, which return value needs a dpr field\n dpr: uiContext.deprecatedDpr,\n },\n },\n hitBy,\n };\n },\n };\n\n return taskLocator;\n }\n}\n"],"names":["debug","getDebug","hasNonEmptyCache","cache","Object","invalidLocateElementReason","element","values","value","Number","JSON","normalizeLocateParam","param","deepThink","rest","deepLocate","undefined","locatePlanForLocate","locate","locatePlan","TaskBuilder","plans","planningModel","defaultModel","options","tasks","cacheable","context","planHandlers","Map","plan","defaultHandler","handler","taskActionFinished","taskLocate","planType","actionSpace","action","findActionInActionSpaceOrThrow","locateFields","findAllMidsceneLocatorField","validateRequiredLocateFields","field","ifLocateParamHasLocatedPixelResult","locateTask","result","task","taskContext","timing","uiContext","assert","setTimingFieldOnce","delayBeforeRunner","Promise","sleep","originalError","originalMessage","String","Error","shrunkShotToLogicalRatio","parsedParam","parseActionParam","error","actionFn","actionResult","delayAfterRunner","detailedLocateParam","onResult","abortSignal","locateParam","taskLocator","paramWithLocatedPixelResult","locateDump","locateResult","applyDump","dump","withUsageIntent","planLocatedElement","matchElementFromPlan","elementFromPlan","isPlanDirectHit","rectFromXpath","elementFromXpath","transformLogicalElementToScreenshot","isXpathHit","cachePrompt","locateCacheRecord","cacheEntry","elementFromCacheResult","matchElementFromCache","elementFromCache","isCacheHit","elementFromAiLocate","ServiceError","invalidElementReason","locateCacheAlreadyExists","currentCacheEntry","pointForCache","Math","feature","hitBy","promptDisplay","elementForAction","matchedElement","interfaceInstance","service","taskCache","waitAfterAction"],"mappings":";;;;;;;;;;;;;;;;;;;AAqCA,MAAMA,QAAQC,SAAS;AAKvB,SAASC,iBAAiBC,KAAc;IACtC,OACEA,QAAAA,SAEA,AAAiB,YAAjB,OAAOA,SACPC,OAAO,IAAI,CAACD,OAAO,MAAM,GAAG;AAEhC;AAEA,SAASE,2BACPC,OAA4B;IAE5B,MAAMC,SAAS;QAACD,QAAQ,MAAM,EAAE,CAAC,EAAE;QAAEA,QAAQ,MAAM,EAAE,CAAC,EAAE;KAAC;IACzD,IACEA,QAAQ,MAAM,EAAE,WAAW,KAC3BC,OAAO,IAAI,CAAC,CAACC,QAAU,AAAiB,YAAjB,OAAOA,SAAsB,CAACC,OAAO,QAAQ,CAACD,SAErE,OAAO,CAAC,mCAAmC,EAAEE,KAAK,SAAS,CAACJ,UAAU;AAG1E;AAMA,SAASK,qBACPC,KAAmC;IAEnC,IAAI,AAAiB,YAAjB,OAAOA,OACT,OAAO;QAAE,QAAQA;IAAM;IAGzB,MAAM,EAAEC,SAAS,EAAE,GAAGC,MAAM,GAAGF;IAC/B,MAAMG,aAAaD,KAAK,UAAU,IAAID;IAEtC,OAAOE,AAAeC,WAAfD,aAA2BD,OAAO;QAAE,GAAGA,IAAI;QAAEC;IAAW;AACjE;AAEO,SAASE,oBAAoBL,KAAmC;IACrE,MAAMM,SAASP,qBAAqBC;IACpC,MAAMO,aAAkD;QACtD,MAAM;QACN,OAAOD;QACP,SAAS;IACX;IACA,OAAOC;AACT;AAyBO,MAAMC;IAyBX,MAAa,MACXC,KAAuB,EACvBC,aAA2B,EAC3BC,YAA0B,EAC1BC,OAAsB,EACoB;QAC1C,MAAMC,QAA8B,EAAE;QACtC,MAAMC,YAAYF,SAAS;QAE3B,MAAMG,UAA4B;YAChCF;YACAH;YACAC;YACAG;YACA,YAAYF,SAAS;YACrB,aAAaA,SAAS;QACxB;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;SAC/D;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,UAAU,WAAa;QACzB;QACAH,QAAQ,KAAK,CAAC,IAAI,CAACM;IACrB;IAEA,MAAc,iBACZH,IAAyC,EACzCH,OAAyB,EACV;QACf,MAAMO,aAAa,IAAI,CAAC,gBAAgB,CAACJ,MAAMA,KAAK,KAAK,EAAEH;QAC3DA,QAAQ,KAAK,CAAC,IAAI,CAACO;IACrB;IAEA,MAAc,iBACZJ,IAAoB,EACpBH,OAAyB,EACV;QACf,MAAMQ,WAAWL,KAAK,IAAI;QAC1B,MAAMM,cAAc,IAAI,CAAC,WAAW;QACpC,MAAMC,SAASC,+BAA+BH,UAAUC;QACxD,MAAMxB,QAAQkB,KAAK,KAAK;QAExB,MAAMS,eAAeC,4BAA4BH,OAAO,WAAW;QAEnEI,6BAA6B7B,OAAOyB,OAAO,WAAW;QAEtDE,aAAa,OAAO,CAAC,CAACG;YACpB,IAAI9B,OAAO,CAAC8B,MAAM,EAAE;gBAGlB,MAAMvB,aAAaF,oBAAoBL,KAAK,CAAC8B,MAAM;gBACnD1C,MACE,uCACA,CAAC,YAAY,EAAEmC,UAAU,EACzB,CAAC,MAAM,EAAEzB,KAAK,SAAS,CAACE,KAAK,CAAC8B,MAAM,GAAG,EACvC,CAAC,WAAW,EAAEhC,KAAK,SAAS,CAACS,aAAa,EAC1C,CAAC,sBAAsB,EAAEwB,mCAAmC/B,KAAK,CAAC8B,MAAM,GAAG;gBAE7E,MAAME,aAAa,IAAI,CAAC,gBAAgB,CACtCzB,YACAP,KAAK,CAAC8B,MAAM,EACZf,SACA,CAACkB;oBACCjC,KAAK,CAAC8B,MAAM,GAAGG;gBACjB;gBAEFlB,QAAQ,KAAK,CAAC,IAAI,CAACiB;YACrB,OACE5C,MAAM,CAAC,OAAO,EAAE0C,MAAM,6BAA6B,EAAEP,UAAU;QAEnE;QAEA,MAAMW,OAKF;YACF,MAAM;YACN,SAASX;YACT,SAASL,KAAK,OAAO;YACrB,OAAOA,KAAK,KAAK;YACjB,UAAU,OAAOiB;gBACf,MAAMC,SAASD,YAAY,IAAI,CAAC,MAAM;gBAEtC/C,MACE,oBACAmC,UACAvB,OACA,CAAC,4BAA4B,EAAEmC,YAAY,OAAO,EAAE,QAAQ;gBAG9D,MAAME,YAAYF,YAAY,SAAS;gBACvCG,OAAOD,WAAW;gBAElBR,6BAA6B7B,OAAOyB,OAAO,WAAW;gBAEtDc,mBAAmBH,QAAQ;gBAC3B,MAAMI,oBAAoBf,OAAO,iBAAiB,IAAI;gBACtD,IAAI;oBACF,MAAMgB,QAAQ,GAAG,CAAC;wBACf;4BACC,IAAI,IAAI,CAAC,SAAS,CAAC,kBAAkB,EAAE;gCACrCrD,MACE,CAAC,8DAA8D,EAAEqC,OAAO,IAAI,EAAE;gCAEhF,MAAM,IAAI,CAAC,SAAS,CAAC,kBAAkB,CAACA,OAAO,IAAI,EAAEzB;gCACrDZ,MACE,CAAC,2DAA2D,EAAEqC,OAAO,IAAI,EAAE;4BAE/E;wBACF;wBACAe,oBAAoB,IAChBE,MAAMF,qBACNC,QAAQ,OAAO;qBACpB;gBACH,EAAE,OAAOE,eAAoB;oBAC3B,MAAMC,kBACJD,eAAe,WAAWE,OAAOF;oBACnC,MAAM,IAAIG,MACR,CAAC,wCAAwC,EAAErB,OAAO,IAAI,CAAC,EAAE,EAAEmB,iBAAiB,EAC5E;wBAAE,OAAOD;oBAAc;gBAE3B;gBACAJ,mBAAmBH,QAAQ;gBAE3B,MAAM,EAAEW,wBAAwB,EAAE,GAAGV;gBACrC,IAAIU,AAA6B3C,WAA7B2C,0BACF,MAAM,IAAID,MACR;gBAIJ,MAAME,cAAe,AAAC;oBACpB,IAAI,CAACvB,OAAO,WAAW,EACrB,OAAOzB;oBAGT,IAAI;wBACF,OAAOiD,iBAAiBjD,OAAOyB,OAAO,WAAW,EAAE;4BACjDsB;wBACF;oBACF,EAAE,OAAOG,OAAY;wBACnB,MAAM,IAAIJ,MACR,CAAC,8BAA8B,EAAErB,OAAO,IAAI,CAAC,EAAE,EAAEyB,MAAM,OAAO,CAAC,cAAc,EAAEpD,KAAK,SAAS,CAACE,QAAQ,EACtG;4BAAE,OAAOkD;wBAAM;oBAEnB;gBACF;gBAEAX,mBAAmBH,QAAQ;gBAE3BhD,MAAM,kBAAkBqC,OAAO,IAAI;gBACnC,MAAM0B,WAAW1B,OAAO,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS;gBAChD,MAAM2B,eAAe,MAAMD,SAASH,aAAab;gBACjDI,mBAAmBH,QAAQ;gBAC3BhD,MAAM,iBAAiBqC,OAAO,IAAI,EAAE,WAAW2B;gBAE/Cb,mBAAmBH,QAAQ;gBAE3B,MAAMiB,mBACJ5B,OAAO,gBAAgB,IAAI,IAAI,CAAC,eAAe,IAAI;gBACrD,IAAI4B,mBAAmB,GACrB,MAAMX,MAAMW;gBAGd,IAAI;oBACF,IAAI,IAAI,CAAC,SAAS,CAAC,iBAAiB,EAAE;wBACpCjE,MACE,CAAC,6DAA6D,EAAEqC,OAAO,IAAI,EAAE;wBAE/E,MAAM,IAAI,CAAC,SAAS,CAAC,iBAAiB,CAACA,OAAO,IAAI,EAAEuB;wBACpD5D,MACE,CAAC,0DAA0D,EAAEqC,OAAO,IAAI,EAAE;oBAE9E;gBACF,EAAE,OAAOkB,eAAoB;oBAC3B,MAAMC,kBACJD,eAAe,WAAWE,OAAOF;oBACnC,MAAM,IAAIG,MACR,CAAC,uCAAuC,EAAErB,OAAO,IAAI,CAAC,EAAE,EAAEmB,iBAAiB,EAC3E;wBAAE,OAAOD;oBAAc;gBAE3B;gBAEAJ,mBAAmBH,QAAQ;gBAE3B,OAAO;oBACL,QAAQgB;gBACV;YACF;QACF;QAEArC,QAAQ,KAAK,CAAC,IAAI,CAACmB;IACrB;IAEQ,iBACNhB,IAAyC,EACzCoC,mBAAiD,EACjDvC,OAAyB,EACzBwC,QAAgD,EACd;QAClC,MAAM,EAAEzC,SAAS,EAAEH,YAAY,EAAER,UAAU,EAAEqD,WAAW,EAAE,GAAGzC;QAE7D,IAAI0C,cAAc1D,qBAAqBuD;QAEvC,IAAIxC,AAAcV,WAAdU,WACF2C,cAAc;YACZ,GAAGA,WAAW;YACd3C;QACF;QAGF,IAAIX,cAAc,CAACsD,YAAY,UAAU,EACvCA,cAAc;YACZ,GAAGA,WAAW;YACd,YAAY;QACd;QAGF,MAAMC,cAAgD;YACpD,MAAM;YACN,SAAS;YACT,OAAOD;YACP,SAASvC,KAAK,OAAO;YACrB,UAAU,OAAOiB;gBACf,MAAM,EAAED,IAAI,EAAE,GAAGC;gBACjB,IAAI,EAAEE,SAAS,EAAE,GAAGF;gBACpB,MAAMwB,8BAA8B5B,mCAClC0B,eAEEA,cACArD;gBAEJkC,OACEmB,aAAa,UAAUE,6BACvB,CAAC,mEAAmE,EAAE7D,KAAK,SAAS,CAClF2D,cACC;gBAGL,IAAI,CAACpB,WACHA,YAAY,MAAM,IAAI,CAAC,OAAO,CAAC,kBAAkB;gBAGnDC,OAAOD,WAAW;gBAElB,MAAM,EAAEU,wBAAwB,EAAE,GAAGV;gBAErC,IAAIU,AAA6B3C,WAA7B2C,0BACF,MAAM,IAAID,MACR;gBAIJ,IAAIc;gBACJ,IAAIC;gBAEJ,MAAMC,YAAY,CAACC;oBACjB,IAAI,CAACA,MACH;oBAEFH,aAAaG;oBACb7B,KAAK,GAAG,GAAG;wBACT6B;wBACA,aAAaA,KAAK,QAAQ,EAAE;wBAC5B,kBAAkBA,KAAK,QAAQ,EAAE;wBACjC,4BACEA,KAAK,QAAQ,EAAE;oBACnB;oBACA7B,KAAK,KAAK,GAAG8B,gBAAgBD,KAAK,QAAQ,EAAE,OAAO;oBACnD7B,KAAK,UAAU,GAAG6B,KAAK,QAAQ,EAAE;oBACjC,IAAIA,KAAK,QAAQ,EAAE,iBAGjB7B,KAAK,eAAe,GAAG8B,gBACrBD,KAAK,QAAQ,CAAC,eAAe,EAC7B;oBAGJ,IAAIA,KAAK,QAAQ,EAAE,mBACjB7B,KAAK,iBAAiB,GAAG6B,KAAK,QAAQ,CAAC,iBAAiB;gBAE5D;gBAEA,MAAME,qBAAqBN,8BACvBO,qBAAqBP,+BACrBvD;gBAKJ,MAAM+D,kBAAkBV,YAAY,UAAU,GAC1CrD,SACA6D;gBACJ,MAAMG,kBAAkB,CAAC,CAACD;gBAG1B,IAAIE;gBACJ,IACE,CAACD,mBACDX,YAAY,KAAK,IACjB,IAAI,CAAC,SAAS,CAAC,uBAAuB,EAEtC,IAAI;oBACFY,gBAAgB,MAAM,IAAI,CAAC,SAAS,CAAC,uBAAuB,CAAC;wBAC3D,QAAQ;4BAACZ,YAAY,KAAK;yBAAC;oBAC7B;gBACF,EAAE,OAAM,CAER;gBAGF,MAAMa,mBAAmBD,gBACrBE,oCACE;oBACE,QAAQ;wBACNF,cAAc,IAAI,GAAGA,cAAc,KAAK,GAAG;wBAC3CA,cAAc,GAAG,GAAGA,cAAc,MAAM,GAAG;qBAC5C;oBACD,aACE,AAA8B,YAA9B,OAAOZ,YAAY,MAAM,GACrBA,YAAY,MAAM,GAClBA,YAAY,MAAM,EAAE,UAAU;oBACpC,MAAMY;gBACR,GACAtB,4BAEF3C;gBAEJ,MAAMoE,aAAa,CAAC,CAACF;gBAErB,MAAMG,cAAchB,YAAY,MAAM;gBACtC,MAAMiB,oBAAoB,IAAI,CAAC,SAAS,EAAE,iBAAiBD;gBAC3D,MAAME,aAAaD,mBAAmB,cAAc;gBAEpD,MAAME,yBACJR,mBAAmBI,aACf,OACA,MAAMK,sBACJ;oBACE,WAAW,IAAI,CAAC,SAAS;oBACzB,mBAAmB,IAAI,CAAC,SAAS;gBACnC,GACAF,YACAF,aACAhB,YAAY,SAAS;gBAI7B,MAAMqB,mBAAmBF,yBACrBL,oCACEK,wBACA7B,4BAEF3C;gBAEJ,MAAM2E,aAAa,CAAC,CAACD;gBAErB,IAAIE;gBACJ,MAAM5C,SAASD,YAAY,IAAI,CAAC,MAAM;gBACtC,IAAI,CAACqC,cAAc,CAACO,cAAc,CAACX,iBACjC,IAAI;oBACF7B,mBAAmBH,QAAQ;oBAC3ByB,eAAe,MAAM,IAAI,CAAC,OAAO,CAAC,MAAM,CACtCJ,aACA;wBACE,SAASpB;wBACT4B;oBACF,GACAtD,cACA6C;oBAEFM,UAAUD,aAAa,IAAI;oBAC3BmB,sBAAsBnB,aAAa,OAAO;gBAC5C,EAAE,OAAOX,OAAO;oBACd,IAAIA,iBAAiB+B,cACnBnB,UAAUZ,MAAM,IAAI;oBAEtB,MAAMA;gBACR,SAAU;oBACRX,mBAAmBH,QAAQ;gBAC7B;gBAGF,MAAM1C,UACJyE,mBACAG,oBACAQ,oBACAE;gBAEF,IAAItF,SAAS;oBACX,MAAMwF,uBAAuBzF,2BAA2BC;oBACxD,IAAIwF,sBAAsB;wBACxB,IAAItB,YACF,MAAM,IAAIqB,aAAaC,sBAAsBtB;wBAE/C,MAAM,IAAId,MAAMoC;oBAClB;gBACF;gBAGA,MAAMC,2BAA2B7F,iBAC/BoF,mBAAmB,cAAc;gBAGnC,IAAIU;gBAOJ,IACE1F,WACA,IAAI,CAAC,SAAS,IACd,CAACqF,cACA,EAACX,mBAAmB,CAACe,wBAAuB,KAC7C1B,AAA0B,UAA1BA,YAAY,SAAS,EAErB,IAAI,IAAI,CAAC,SAAS,CAAC,oBAAoB,EACrC,IAAI;oBAGF,MAAM4B,gBAAkC;wBACtCC,KAAK,KAAK,CAAC5F,QAAQ,MAAM,CAAC,EAAE,GAAGqD;wBAC/BuC,KAAK,KAAK,CAAC5F,QAAQ,MAAM,CAAC,EAAE,GAAGqD;qBAChC;oBAED,MAAMwC,UAAU,MAAM,IAAI,CAAC,SAAS,CAAC,oBAAoB,CACvDF,eACA;wBACE,mBACE,AAA8B,YAA9B,OAAO5B,YAAY,MAAM,GACrBA,YAAY,MAAM,GAClBA,YAAY,MAAM,EAAE;wBAC1B,cAAc9C;oBAChB;oBAEF,IAAIrB,iBAAiBiG,UAAU;wBAC7BnG,MACE,uCACAqF,aACAc;wBAEFH,oBAAoBG;wBACpB,IAAI,CAAC,SAAS,CAAC,yBAAyB,CACtC;4BACE,MAAM;4BACN,QAAQd;4BACR,OAAOc;wBACT,GACAb;oBAEJ,OACEtF,MACE,yDACAqF;gBAGN,EAAE,OAAOvB,OAAO;oBACd9D,MAAM,mCAAmC8D;gBAC3C;qBAEA9D,MAAM;gBAIV,IAAI,CAACM,SAAS;oBACZ,IAAIkE,YACF,MAAM,IAAIqB,aACR,CAAC,oBAAoB,EAAExB,YAAY,MAAM,EAAE,EAC3CG;oBAGJ,MAAM,IAAId,MAAM,CAAC,mBAAmB,EAAEW,YAAY,MAAM,EAAE;gBAC5D;gBAEA,IAAI+B;gBAEJ,IAAIpB,mBAAmBT,6BACrB6B,QAAQ;oBACN,MAAM;oBACN,SAAS;wBACP,oBACE7B,4BAA4B,kBAAkB;oBAClD;gBACF;qBACK,IAAIa,YACTgB,QAAQ;oBACN,MAAM;oBACN,SAAS;wBACP,OAAO/B,YAAY,KAAK;oBAC1B;gBACF;qBACK,IAAIsB,YACTS,QAAQ;oBACN,MAAM;oBACN,SAAS;wBACPb;wBACA,aAAaS;oBACf;gBACF;gBAGF,MAAMK,gBAAgBhC,YAAY,aAAa;gBAC/C,MAAMiC,mBAAmBD,gBACrB;oBACE,GAAG/F,OAAO;oBACV,aAAa+F;gBACf,IACA/F;gBAEJ,IAAI+F,iBAAiB7B,YAAY,gBAC/BA,WAAW,cAAc,GAAGA,WAAW,cAAc,CAAC,GAAG,CACvD,CAAC+B,iBAAoB;wBACnB,GAAGA,cAAc;wBACjB,aAAaF;oBACf;gBAIJlC,WAAWmC;gBAEX,OAAO;oBACL,QAAQ;wBACN,SAAS;4BACP,GAAGA,gBAAgB;4BAEnB,KAAKrD,UAAU,aAAa;wBAC9B;oBACF;oBACAmD;gBACF;YACF;QACF;QAEA,OAAO9B;IACT;IA1kBA,YAAY,EACVkC,iBAAiB,EACjBC,OAAO,EACPC,SAAS,EACTtE,WAAW,EACXuE,eAAe,EACC,CAAE;QAhBpB,uBAAiB,aAAjB;QAEA,uBAAiB,WAAjB;QAEA,uBAAiB,aAAjB;QAEA,uBAAiB,eAAjB;QAEA,uBAAiB,mBAAjB;QASE,IAAI,CAAC,SAAS,GAAGH;QACjB,IAAI,CAAC,OAAO,GAAGC;QACf,IAAI,CAAC,SAAS,GAAGC;QACjB,IAAI,CAAC,WAAW,GAAGtE;QACnB,IAAI,CAAC,eAAe,GAAGuE;IACzB;AA+jBF"}
@@ -153,7 +153,7 @@ async function matchElementFromCache(context, cacheEntry, cachePrompt, cacheable
153
153
  return;
154
154
  }
155
155
  }
156
- const getMidsceneVersion = ()=>"1.12.7";
156
+ const getMidsceneVersion = ()=>"1.12.8-beta-20260916023839.0";
157
157
  const parsePrompt = (prompt)=>{
158
158
  if ('string' == typeof prompt) return {
159
159
  textPrompt: prompt,
@@ -102,8 +102,7 @@ async function runCustomPlanning(userInstruction, options, config) {
102
102
  preparedSize: preparedImage.preparedSize,
103
103
  contentSize: preparedImage.contentSize
104
104
  },
105
- acceptBbox2dAlias: preparedOptions.modelRuntime.adapter.acceptBbox2dAlias,
106
- parseRawLocateParameter: (value)=>value
105
+ acceptBbox2dAlias: preparedOptions.modelRuntime.adapter.acceptBbox2dAlias
107
106
  });
108
107
  shouldContinuePlanning = config.shouldContinuePlanning(parsed, actions);
109
108
  } catch (parseError) {
@@ -1 +1 @@
1
- {"version":3,"file":"ai-model/workflows/planning/custom-planning.mjs","sources":["../../../../../src/ai-model/workflows/planning/custom-planning.ts"],"sourcesContent":["import type { PlanningAIResponse, PlanningAction } from '@/types';\nimport type { ChatCompletionMessageParam } from 'openai/resources/index';\nimport { ScreenshotItem } from '../../../screenshot-item';\nimport type {\n CustomPlanningInput,\n CustomPlanningMessageConfig,\n ResolvedCustomPlanningDefinition,\n} from '../../model-adapter/custom-planning-types';\nimport { prepareModelImage } from '../../model-adapter/image-preprocess';\nimport type { ParsedPlanningLocateParameter } from '../../model-adapter/planning-protocol';\nimport {\n AIResponseParseError,\n callAIWithStringResponse,\n} from '../../service-caller/index';\nimport {\n type PreparedUserPrompt,\n preparedReferenceImagesToChatMessages,\n} from '../../shared/multimodal-prompt';\nimport { normalizePlanningActionLocateFields } from './locate-normalization';\nimport type { PlanOptions } from './types';\n\nfunction appendActionContext(\n systemPrompt: string,\n actionContext?: string,\n): string {\n return systemPrompt + (actionContext ? `${actionContext}\\n` : '');\n}\n\nexport function buildCustomPlanningMessages<TParsed>(\n input: CustomPlanningInput,\n config: CustomPlanningMessageConfig<TParsed>,\n): ChatCompletionMessageParam[] {\n const { options } = input;\n const { conversationHistory, context, actionContext } = options;\n const systemPrompt = appendActionContext(\n config.buildSystemPrompt(),\n actionContext,\n );\n const userInstructionText = input.userInstruction.text;\n const userInstruction = config.buildUserInstruction\n ? config.buildUserInstruction(userInstructionText)\n : userInstructionText;\n const referenceImageMessages = preparedReferenceImagesToChatMessages(\n input.userInstruction.referenceImages,\n );\n\n if (conversationHistory.pendingFeedbackMessage) {\n conversationHistory.append({\n role: 'user',\n content: [\n {\n type: 'text',\n text: `${conversationHistory.pendingFeedbackMessage}. The previous action has been executed, here is the latest screenshot. Please continue according to the instruction.`,\n },\n ],\n });\n conversationHistory.resetPendingFeedbackMessageIfExists();\n }\n\n conversationHistory.append({\n role: 'user',\n content: [\n {\n type: 'image_url',\n image_url: { url: context.screenshot.base64 },\n },\n ],\n });\n\n if (config.systemPromptPlacement === 'system-message') {\n return [\n { role: 'system', content: systemPrompt },\n {\n role: 'user',\n content: [{ type: 'text', text: userInstruction }],\n },\n ...referenceImageMessages,\n ...conversationHistory.snapshot(config.historyImageLimit),\n ];\n }\n\n return [\n {\n role: 'user',\n content: `${systemPrompt}${userInstruction}`,\n },\n ...referenceImageMessages,\n ...conversationHistory.snapshot(config.historyImageLimit),\n ];\n}\n\nexport async function runCustomPlanning<TParsed>(\n userInstruction: PreparedUserPrompt,\n options: PlanOptions,\n config: ResolvedCustomPlanningDefinition<TParsed>,\n): Promise<PlanningAIResponse> {\n const { context } = options;\n const preparedImage = await prepareModelImage({\n imageBase64: context.screenshot.base64,\n width: context.shotSize.width,\n height: context.shotSize.height,\n policy: options.modelRuntime.adapter.imagePreprocess,\n });\n const preparedOptions: PlanOptions = {\n ...options,\n context: {\n ...context,\n screenshot: ScreenshotItem.create(\n preparedImage.imageBase64,\n context.screenshot.capturedAt,\n ),\n shotSize: preparedImage.preparedSize,\n },\n };\n const input: CustomPlanningInput = {\n userInstruction,\n options: preparedOptions,\n coordinateSystem: config.coordinateSystem,\n };\n\n const messages = buildCustomPlanningMessages(input, config.messages);\n const { content, usage, rawChoiceMessage } = await callAIWithStringResponse(\n messages,\n preparedOptions.modelRuntime,\n {\n abortSignal: preparedOptions.abortSignal,\n requiresOriginalImageDetail: preparedOptions.includeLocateInPlanning,\n },\n );\n\n let parsed: TParsed;\n let actions: PlanningAction[];\n let shouldContinuePlanning: boolean;\n\n try {\n parsed = config.parseResponse(content, input);\n actions = config.transformActions(parsed, input);\n normalizePlanningActionLocateFields(actions, {\n actionSpace: preparedOptions.actionSpace,\n includeLocateInPlanning: preparedOptions.includeLocateInPlanning,\n locateResultCodec: config.coordinateNormalizer,\n locateResultContext: {\n preparedSize: preparedImage.preparedSize,\n contentSize: preparedImage.contentSize,\n },\n acceptBbox2dAlias: preparedOptions.modelRuntime.adapter.acceptBbox2dAlias,\n parseRawLocateParameter: (value) =>\n value as ParsedPlanningLocateParameter,\n });\n shouldContinuePlanning = config.shouldContinuePlanning(parsed, actions);\n } catch (parseError) {\n const errorMessage = `Parse error: ${\n parseError instanceof Error ? parseError.message : String(parseError)\n }`;\n throw new AIResponseParseError(\n errorMessage,\n JSON.stringify(content, undefined, 2),\n usage,\n rawChoiceMessage,\n );\n }\n\n const assistantContent = config.messages.buildAssistantContent?.(\n parsed,\n content,\n input,\n );\n // UI-TARS and Auto-GLM define their own assistant history format.\n if (assistantContent) {\n options.conversationHistory.append({\n role: 'assistant',\n content: assistantContent,\n });\n }\n\n return {\n actions,\n log: config.buildResponseLog(parsed, content),\n usage,\n shouldContinuePlanning,\n rawResponse: JSON.stringify(content, undefined, 2),\n rawChoiceMessage,\n };\n}\n"],"names":["appendActionContext","systemPrompt","actionContext","buildCustomPlanningMessages","input","config","options","conversationHistory","context","userInstructionText","userInstruction","referenceImageMessages","preparedReferenceImagesToChatMessages","runCustomPlanning","preparedImage","prepareModelImage","preparedOptions","ScreenshotItem","messages","content","usage","rawChoiceMessage","callAIWithStringResponse","parsed","actions","shouldContinuePlanning","normalizePlanningActionLocateFields","value","parseError","errorMessage","Error","String","AIResponseParseError","JSON","undefined","assistantContent"],"mappings":";;;;;AAqBA,SAASA,oBACPC,YAAoB,EACpBC,aAAsB;IAEtB,OAAOD,eAAgBC,CAAAA,gBAAgB,GAAGA,cAAc,EAAE,CAAC,GAAG,EAAC;AACjE;AAEO,SAASC,4BACdC,KAA0B,EAC1BC,MAA4C;IAE5C,MAAM,EAAEC,OAAO,EAAE,GAAGF;IACpB,MAAM,EAAEG,mBAAmB,EAAEC,OAAO,EAAEN,aAAa,EAAE,GAAGI;IACxD,MAAML,eAAeD,oBACnBK,OAAO,iBAAiB,IACxBH;IAEF,MAAMO,sBAAsBL,MAAM,eAAe,CAAC,IAAI;IACtD,MAAMM,kBAAkBL,OAAO,oBAAoB,GAC/CA,OAAO,oBAAoB,CAACI,uBAC5BA;IACJ,MAAME,yBAAyBC,sCAC7BR,MAAM,eAAe,CAAC,eAAe;IAGvC,IAAIG,oBAAoB,sBAAsB,EAAE;QAC9CA,oBAAoB,MAAM,CAAC;YACzB,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,MAAM,GAAGA,oBAAoB,sBAAsB,CAAC,qHAAqH,CAAC;gBAC5K;aACD;QACH;QACAA,oBAAoB,mCAAmC;IACzD;IAEAA,oBAAoB,MAAM,CAAC;QACzB,MAAM;QACN,SAAS;YACP;gBACE,MAAM;gBACN,WAAW;oBAAE,KAAKC,QAAQ,UAAU,CAAC,MAAM;gBAAC;YAC9C;SACD;IACH;IAEA,IAAIH,AAAiC,qBAAjCA,OAAO,qBAAqB,EAC9B,OAAO;QACL;YAAE,MAAM;YAAU,SAASJ;QAAa;QACxC;YACE,MAAM;YACN,SAAS;gBAAC;oBAAE,MAAM;oBAAQ,MAAMS;gBAAgB;aAAE;QACpD;WACGC;WACAJ,oBAAoB,QAAQ,CAACF,OAAO,iBAAiB;KACzD;IAGH,OAAO;QACL;YACE,MAAM;YACN,SAAS,GAAGJ,eAAeS,iBAAiB;QAC9C;WACGC;WACAJ,oBAAoB,QAAQ,CAACF,OAAO,iBAAiB;KACzD;AACH;AAEO,eAAeQ,kBACpBH,eAAmC,EACnCJ,OAAoB,EACpBD,MAAiD;IAEjD,MAAM,EAAEG,OAAO,EAAE,GAAGF;IACpB,MAAMQ,gBAAgB,MAAMC,kBAAkB;QAC5C,aAAaP,QAAQ,UAAU,CAAC,MAAM;QACtC,OAAOA,QAAQ,QAAQ,CAAC,KAAK;QAC7B,QAAQA,QAAQ,QAAQ,CAAC,MAAM;QAC/B,QAAQF,QAAQ,YAAY,CAAC,OAAO,CAAC,eAAe;IACtD;IACA,MAAMU,kBAA+B;QACnC,GAAGV,OAAO;QACV,SAAS;YACP,GAAGE,OAAO;YACV,YAAYS,eAAe,MAAM,CAC/BH,cAAc,WAAW,EACzBN,QAAQ,UAAU,CAAC,UAAU;YAE/B,UAAUM,cAAc,YAAY;QACtC;IACF;IACA,MAAMV,QAA6B;QACjCM;QACA,SAASM;QACT,kBAAkBX,OAAO,gBAAgB;IAC3C;IAEA,MAAMa,WAAWf,4BAA4BC,OAAOC,OAAO,QAAQ;IACnE,MAAM,EAAEc,OAAO,EAAEC,KAAK,EAAEC,gBAAgB,EAAE,GAAG,MAAMC,yBACjDJ,UACAF,gBAAgB,YAAY,EAC5B;QACE,aAAaA,gBAAgB,WAAW;QACxC,6BAA6BA,gBAAgB,uBAAuB;IACtE;IAGF,IAAIO;IACJ,IAAIC;IACJ,IAAIC;IAEJ,IAAI;QACFF,SAASlB,OAAO,aAAa,CAACc,SAASf;QACvCoB,UAAUnB,OAAO,gBAAgB,CAACkB,QAAQnB;QAC1CsB,oCAAoCF,SAAS;YAC3C,aAAaR,gBAAgB,WAAW;YACxC,yBAAyBA,gBAAgB,uBAAuB;YAChE,mBAAmBX,OAAO,oBAAoB;YAC9C,qBAAqB;gBACnB,cAAcS,cAAc,YAAY;gBACxC,aAAaA,cAAc,WAAW;YACxC;YACA,mBAAmBE,gBAAgB,YAAY,CAAC,OAAO,CAAC,iBAAiB;YACzE,yBAAyB,CAACW,QACxBA;QACJ;QACAF,yBAAyBpB,OAAO,sBAAsB,CAACkB,QAAQC;IACjE,EAAE,OAAOI,YAAY;QACnB,MAAMC,eAAe,CAAC,aAAa,EACjCD,sBAAsBE,QAAQF,WAAW,OAAO,GAAGG,OAAOH,aAC1D;QACF,MAAM,IAAII,qBACRH,cACAI,KAAK,SAAS,CAACd,SAASe,QAAW,IACnCd,OACAC;IAEJ;IAEA,MAAMc,mBAAmB9B,OAAO,QAAQ,CAAC,qBAAqB,GAC5DkB,QACAJ,SACAf;IAGF,IAAI+B,kBACF7B,QAAQ,mBAAmB,CAAC,MAAM,CAAC;QACjC,MAAM;QACN,SAAS6B;IACX;IAGF,OAAO;QACLX;QACA,KAAKnB,OAAO,gBAAgB,CAACkB,QAAQJ;QACrCC;QACAK;QACA,aAAaQ,KAAK,SAAS,CAACd,SAASe,QAAW;QAChDb;IACF;AACF"}
1
+ {"version":3,"file":"ai-model/workflows/planning/custom-planning.mjs","sources":["../../../../../src/ai-model/workflows/planning/custom-planning.ts"],"sourcesContent":["import type { PlanningAIResponse, PlanningAction } from '@/types';\nimport type { ChatCompletionMessageParam } from 'openai/resources/index';\nimport { ScreenshotItem } from '../../../screenshot-item';\nimport type {\n CustomPlanningInput,\n CustomPlanningMessageConfig,\n ResolvedCustomPlanningDefinition,\n} from '../../model-adapter/custom-planning-types';\nimport { prepareModelImage } from '../../model-adapter/image-preprocess';\nimport {\n AIResponseParseError,\n callAIWithStringResponse,\n} from '../../service-caller/index';\nimport {\n type PreparedUserPrompt,\n preparedReferenceImagesToChatMessages,\n} from '../../shared/multimodal-prompt';\nimport { normalizePlanningActionLocateFields } from './locate-normalization';\nimport type { PlanOptions } from './types';\n\nfunction appendActionContext(\n systemPrompt: string,\n actionContext?: string,\n): string {\n return systemPrompt + (actionContext ? `${actionContext}\\n` : '');\n}\n\nexport function buildCustomPlanningMessages<TParsed>(\n input: CustomPlanningInput,\n config: CustomPlanningMessageConfig<TParsed>,\n): ChatCompletionMessageParam[] {\n const { options } = input;\n const { conversationHistory, context, actionContext } = options;\n const systemPrompt = appendActionContext(\n config.buildSystemPrompt(),\n actionContext,\n );\n const userInstructionText = input.userInstruction.text;\n const userInstruction = config.buildUserInstruction\n ? config.buildUserInstruction(userInstructionText)\n : userInstructionText;\n const referenceImageMessages = preparedReferenceImagesToChatMessages(\n input.userInstruction.referenceImages,\n );\n\n if (conversationHistory.pendingFeedbackMessage) {\n conversationHistory.append({\n role: 'user',\n content: [\n {\n type: 'text',\n text: `${conversationHistory.pendingFeedbackMessage}. The previous action has been executed, here is the latest screenshot. Please continue according to the instruction.`,\n },\n ],\n });\n conversationHistory.resetPendingFeedbackMessageIfExists();\n }\n\n conversationHistory.append({\n role: 'user',\n content: [\n {\n type: 'image_url',\n image_url: { url: context.screenshot.base64 },\n },\n ],\n });\n\n if (config.systemPromptPlacement === 'system-message') {\n return [\n { role: 'system', content: systemPrompt },\n {\n role: 'user',\n content: [{ type: 'text', text: userInstruction }],\n },\n ...referenceImageMessages,\n ...conversationHistory.snapshot(config.historyImageLimit),\n ];\n }\n\n return [\n {\n role: 'user',\n content: `${systemPrompt}${userInstruction}`,\n },\n ...referenceImageMessages,\n ...conversationHistory.snapshot(config.historyImageLimit),\n ];\n}\n\nexport async function runCustomPlanning<TParsed>(\n userInstruction: PreparedUserPrompt,\n options: PlanOptions,\n config: ResolvedCustomPlanningDefinition<TParsed>,\n): Promise<PlanningAIResponse> {\n const { context } = options;\n const preparedImage = await prepareModelImage({\n imageBase64: context.screenshot.base64,\n width: context.shotSize.width,\n height: context.shotSize.height,\n policy: options.modelRuntime.adapter.imagePreprocess,\n });\n const preparedOptions: PlanOptions = {\n ...options,\n context: {\n ...context,\n screenshot: ScreenshotItem.create(\n preparedImage.imageBase64,\n context.screenshot.capturedAt,\n ),\n shotSize: preparedImage.preparedSize,\n },\n };\n const input: CustomPlanningInput = {\n userInstruction,\n options: preparedOptions,\n coordinateSystem: config.coordinateSystem,\n };\n\n const messages = buildCustomPlanningMessages(input, config.messages);\n const { content, usage, rawChoiceMessage } = await callAIWithStringResponse(\n messages,\n preparedOptions.modelRuntime,\n {\n abortSignal: preparedOptions.abortSignal,\n requiresOriginalImageDetail: preparedOptions.includeLocateInPlanning,\n },\n );\n\n let parsed: TParsed;\n let actions: PlanningAction[];\n let shouldContinuePlanning: boolean;\n\n try {\n parsed = config.parseResponse(content, input);\n actions = config.transformActions(parsed, input);\n normalizePlanningActionLocateFields(actions, {\n actionSpace: preparedOptions.actionSpace,\n includeLocateInPlanning: preparedOptions.includeLocateInPlanning,\n locateResultCodec: config.coordinateNormalizer,\n locateResultContext: {\n preparedSize: preparedImage.preparedSize,\n contentSize: preparedImage.contentSize,\n },\n acceptBbox2dAlias: preparedOptions.modelRuntime.adapter.acceptBbox2dAlias,\n });\n shouldContinuePlanning = config.shouldContinuePlanning(parsed, actions);\n } catch (parseError) {\n const errorMessage = `Parse error: ${\n parseError instanceof Error ? parseError.message : String(parseError)\n }`;\n throw new AIResponseParseError(\n errorMessage,\n JSON.stringify(content, undefined, 2),\n usage,\n rawChoiceMessage,\n );\n }\n\n const assistantContent = config.messages.buildAssistantContent?.(\n parsed,\n content,\n input,\n );\n // UI-TARS and Auto-GLM define their own assistant history format.\n if (assistantContent) {\n options.conversationHistory.append({\n role: 'assistant',\n content: assistantContent,\n });\n }\n\n return {\n actions,\n log: config.buildResponseLog(parsed, content),\n usage,\n shouldContinuePlanning,\n rawResponse: JSON.stringify(content, undefined, 2),\n rawChoiceMessage,\n };\n}\n"],"names":["appendActionContext","systemPrompt","actionContext","buildCustomPlanningMessages","input","config","options","conversationHistory","context","userInstructionText","userInstruction","referenceImageMessages","preparedReferenceImagesToChatMessages","runCustomPlanning","preparedImage","prepareModelImage","preparedOptions","ScreenshotItem","messages","content","usage","rawChoiceMessage","callAIWithStringResponse","parsed","actions","shouldContinuePlanning","normalizePlanningActionLocateFields","parseError","errorMessage","Error","String","AIResponseParseError","JSON","undefined","assistantContent"],"mappings":";;;;;AAoBA,SAASA,oBACPC,YAAoB,EACpBC,aAAsB;IAEtB,OAAOD,eAAgBC,CAAAA,gBAAgB,GAAGA,cAAc,EAAE,CAAC,GAAG,EAAC;AACjE;AAEO,SAASC,4BACdC,KAA0B,EAC1BC,MAA4C;IAE5C,MAAM,EAAEC,OAAO,EAAE,GAAGF;IACpB,MAAM,EAAEG,mBAAmB,EAAEC,OAAO,EAAEN,aAAa,EAAE,GAAGI;IACxD,MAAML,eAAeD,oBACnBK,OAAO,iBAAiB,IACxBH;IAEF,MAAMO,sBAAsBL,MAAM,eAAe,CAAC,IAAI;IACtD,MAAMM,kBAAkBL,OAAO,oBAAoB,GAC/CA,OAAO,oBAAoB,CAACI,uBAC5BA;IACJ,MAAME,yBAAyBC,sCAC7BR,MAAM,eAAe,CAAC,eAAe;IAGvC,IAAIG,oBAAoB,sBAAsB,EAAE;QAC9CA,oBAAoB,MAAM,CAAC;YACzB,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,MAAM,GAAGA,oBAAoB,sBAAsB,CAAC,qHAAqH,CAAC;gBAC5K;aACD;QACH;QACAA,oBAAoB,mCAAmC;IACzD;IAEAA,oBAAoB,MAAM,CAAC;QACzB,MAAM;QACN,SAAS;YACP;gBACE,MAAM;gBACN,WAAW;oBAAE,KAAKC,QAAQ,UAAU,CAAC,MAAM;gBAAC;YAC9C;SACD;IACH;IAEA,IAAIH,AAAiC,qBAAjCA,OAAO,qBAAqB,EAC9B,OAAO;QACL;YAAE,MAAM;YAAU,SAASJ;QAAa;QACxC;YACE,MAAM;YACN,SAAS;gBAAC;oBAAE,MAAM;oBAAQ,MAAMS;gBAAgB;aAAE;QACpD;WACGC;WACAJ,oBAAoB,QAAQ,CAACF,OAAO,iBAAiB;KACzD;IAGH,OAAO;QACL;YACE,MAAM;YACN,SAAS,GAAGJ,eAAeS,iBAAiB;QAC9C;WACGC;WACAJ,oBAAoB,QAAQ,CAACF,OAAO,iBAAiB;KACzD;AACH;AAEO,eAAeQ,kBACpBH,eAAmC,EACnCJ,OAAoB,EACpBD,MAAiD;IAEjD,MAAM,EAAEG,OAAO,EAAE,GAAGF;IACpB,MAAMQ,gBAAgB,MAAMC,kBAAkB;QAC5C,aAAaP,QAAQ,UAAU,CAAC,MAAM;QACtC,OAAOA,QAAQ,QAAQ,CAAC,KAAK;QAC7B,QAAQA,QAAQ,QAAQ,CAAC,MAAM;QAC/B,QAAQF,QAAQ,YAAY,CAAC,OAAO,CAAC,eAAe;IACtD;IACA,MAAMU,kBAA+B;QACnC,GAAGV,OAAO;QACV,SAAS;YACP,GAAGE,OAAO;YACV,YAAYS,eAAe,MAAM,CAC/BH,cAAc,WAAW,EACzBN,QAAQ,UAAU,CAAC,UAAU;YAE/B,UAAUM,cAAc,YAAY;QACtC;IACF;IACA,MAAMV,QAA6B;QACjCM;QACA,SAASM;QACT,kBAAkBX,OAAO,gBAAgB;IAC3C;IAEA,MAAMa,WAAWf,4BAA4BC,OAAOC,OAAO,QAAQ;IACnE,MAAM,EAAEc,OAAO,EAAEC,KAAK,EAAEC,gBAAgB,EAAE,GAAG,MAAMC,yBACjDJ,UACAF,gBAAgB,YAAY,EAC5B;QACE,aAAaA,gBAAgB,WAAW;QACxC,6BAA6BA,gBAAgB,uBAAuB;IACtE;IAGF,IAAIO;IACJ,IAAIC;IACJ,IAAIC;IAEJ,IAAI;QACFF,SAASlB,OAAO,aAAa,CAACc,SAASf;QACvCoB,UAAUnB,OAAO,gBAAgB,CAACkB,QAAQnB;QAC1CsB,oCAAoCF,SAAS;YAC3C,aAAaR,gBAAgB,WAAW;YACxC,yBAAyBA,gBAAgB,uBAAuB;YAChE,mBAAmBX,OAAO,oBAAoB;YAC9C,qBAAqB;gBACnB,cAAcS,cAAc,YAAY;gBACxC,aAAaA,cAAc,WAAW;YACxC;YACA,mBAAmBE,gBAAgB,YAAY,CAAC,OAAO,CAAC,iBAAiB;QAC3E;QACAS,yBAAyBpB,OAAO,sBAAsB,CAACkB,QAAQC;IACjE,EAAE,OAAOG,YAAY;QACnB,MAAMC,eAAe,CAAC,aAAa,EACjCD,sBAAsBE,QAAQF,WAAW,OAAO,GAAGG,OAAOH,aAC1D;QACF,MAAM,IAAII,qBACRH,cACAI,KAAK,SAAS,CAACb,SAASc,QAAW,IACnCb,OACAC;IAEJ;IAEA,MAAMa,mBAAmB7B,OAAO,QAAQ,CAAC,qBAAqB,GAC5DkB,QACAJ,SACAf;IAGF,IAAI8B,kBACF5B,QAAQ,mBAAmB,CAAC,MAAM,CAAC;QACjC,MAAM;QACN,SAAS4B;IACX;IAGF,OAAO;QACLV;QACA,KAAKnB,OAAO,gBAAgB,CAACkB,QAAQJ;QACrCC;QACAK;QACA,aAAaO,KAAK,SAAS,CAACb,SAASc,QAAW;QAChDZ;IACF;AACF"}
@@ -2,7 +2,7 @@ import { findAllMidsceneLocatorField } from "../../../common.mjs";
2
2
  import { getDebug } from "@midscene/shared/logger";
3
3
  import { assert } from "@midscene/shared/utils";
4
4
  const debug = getDebug('planning');
5
- function normalizePlanningActionLocateFields(actions, { actionSpace, includeLocateInPlanning, locateResultCodec, locateResultContext, acceptBbox2dAlias = false, parseRawLocateParameter }) {
5
+ function normalizePlanningActionLocateFields(actions, { actionSpace, ...options }) {
6
6
  actions.forEach((action)=>{
7
7
  const actionInActionSpace = actionSpace.find((actionInSpace)=>actionInSpace.name === action.type);
8
8
  if (!actionInActionSpace) return void debug('skip locate normalization for action outside actionSpace', action);
@@ -10,33 +10,32 @@ function normalizePlanningActionLocateFields(actions, { actionSpace, includeLoca
10
10
  const locateFields = findAllMidsceneLocatorField(actionInActionSpace.paramSchema);
11
11
  debug('locateFields', locateFields);
12
12
  locateFields.forEach((field)=>{
13
- const rawLocateParameter = action.param?.[field];
14
- if (!rawLocateParameter) return;
15
- const locateParameter = parseRawLocateParameter(rawLocateParameter);
16
- if (!includeLocateInPlanning) {
17
- action.param[field] = {
18
- prompt: locateParameter.prompt
19
- };
20
- return;
21
- }
22
- assert(locateResultCodec, 'planning locate normalization requires a locate result codec');
23
- const resultKey = locateResultCodec.promptSpec.resultKey;
24
- const rawLocateValue = void 0 !== locateParameter[resultKey] ? locateParameter[resultKey] : acceptBbox2dAlias && 'bbox' === resultKey ? locateParameter.bbox_2d : void 0;
25
- const rawCoordinateKeys = new Set([
26
- resultKey,
27
- ...acceptBbox2dAlias && 'bbox' === resultKey ? [
28
- 'bbox_2d'
29
- ] : []
30
- ]);
31
- const locateParamWithoutRawCoordinates = Object.fromEntries(Object.entries(locateParameter).filter(([key])=>!rawCoordinateKeys.has(key)));
32
- const result = locateResultCodec.toPixelResult(rawLocateValue, locateResultContext);
33
- action.param[field] = {
34
- ...locateParamWithoutRawCoordinates,
35
- locatedPixelResult: result
36
- };
13
+ const locateParameter = action.param?.[field];
14
+ if (!locateParameter) return;
15
+ action.param[field] = normalizePlanningLocateParameter(locateParameter, options);
37
16
  });
38
17
  });
39
18
  }
40
- export { normalizePlanningActionLocateFields };
19
+ function normalizePlanningLocateParameter(locateParameter, { includeLocateInPlanning, locateResultCodec, locateResultContext, acceptBbox2dAlias = false }) {
20
+ if (!includeLocateInPlanning) return {
21
+ prompt: locateParameter.prompt
22
+ };
23
+ assert(locateResultCodec, 'planning locate normalization requires a locate result codec');
24
+ const resultKey = locateResultCodec.promptSpec.resultKey;
25
+ const rawLocateValue = void 0 !== locateParameter[resultKey] ? locateParameter[resultKey] : acceptBbox2dAlias && 'bbox' === resultKey ? locateParameter.bbox_2d : void 0;
26
+ const rawCoordinateKeys = new Set([
27
+ resultKey,
28
+ ...acceptBbox2dAlias && 'bbox' === resultKey ? [
29
+ 'bbox_2d'
30
+ ] : []
31
+ ]);
32
+ const locateParamWithoutRawCoordinates = Object.fromEntries(Object.entries(locateParameter).filter(([key])=>!rawCoordinateKeys.has(key)));
33
+ const result = locateResultCodec.toPixelResult(rawLocateValue, locateResultContext);
34
+ return {
35
+ ...locateParamWithoutRawCoordinates,
36
+ locatedPixelResult: result
37
+ };
38
+ }
39
+ export { normalizePlanningActionLocateFields, normalizePlanningLocateParameter };
41
40
 
42
41
  //# sourceMappingURL=locate-normalization.mjs.map
@@ -1 +1 @@
1
- {"version":3,"file":"ai-model/workflows/planning/locate-normalization.mjs","sources":["../../../../../src/ai-model/workflows/planning/locate-normalization.ts"],"sourcesContent":["import { findAllMidsceneLocatorField } from '@/common';\nimport type { DeviceAction } from '@/device';\nimport type { PlanningAction } from '@/types';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type { ParsedPlanningLocateParameter } from '../../model-adapter/planning-protocol';\nimport type {\n LocateResultCodec,\n LocateResultContext,\n} from '../../shared/model-locate-result/types';\n\nconst debug = getDebug('planning');\n\nexport function normalizePlanningActionLocateFields(\n actions: PlanningAction[],\n {\n actionSpace,\n includeLocateInPlanning,\n locateResultCodec,\n locateResultContext,\n acceptBbox2dAlias = false,\n parseRawLocateParameter,\n }: {\n actionSpace: DeviceAction[];\n includeLocateInPlanning: boolean;\n locateResultCodec?: LocateResultCodec;\n locateResultContext: LocateResultContext;\n acceptBbox2dAlias?: boolean;\n parseRawLocateParameter: (value: unknown) => ParsedPlanningLocateParameter;\n },\n): void {\n actions.forEach((action) => {\n const actionInActionSpace = actionSpace.find(\n (actionInSpace) => actionInSpace.name === action.type,\n );\n if (!actionInActionSpace) {\n debug('skip locate normalization for action outside actionSpace', action);\n return;\n }\n\n debug('actionInActionSpace matched', actionInActionSpace);\n const locateFields = findAllMidsceneLocatorField(\n actionInActionSpace.paramSchema,\n );\n\n debug('locateFields', locateFields);\n\n locateFields.forEach((field) => {\n const rawLocateParameter = action.param?.[field];\n if (!rawLocateParameter) {\n return;\n }\n\n const locateParameter = parseRawLocateParameter(rawLocateParameter);\n\n if (!includeLocateInPlanning) {\n // In prompt-only planning mode, ignore any accidental coordinates from the model.\n action.param[field] = { prompt: locateParameter.prompt };\n return;\n }\n\n assert(\n locateResultCodec,\n 'planning locate normalization requires a locate result codec',\n );\n\n const resultKey = locateResultCodec.promptSpec.resultKey;\n const rawLocateValue =\n locateParameter[resultKey] !== undefined\n ? locateParameter[resultKey]\n : acceptBbox2dAlias && resultKey === 'bbox'\n ? locateParameter.bbox_2d\n : undefined;\n\n // The raw result field is replaced by locatedPixelResult, so it should not\n // remain in the normalized locate parameter.\n const rawCoordinateKeys = new Set([\n resultKey,\n ...(acceptBbox2dAlias && resultKey === 'bbox' ? ['bbox_2d'] : []),\n ]);\n\n const locateParamWithoutRawCoordinates = Object.fromEntries(\n Object.entries(locateParameter).filter(\n ([key]) => !rawCoordinateKeys.has(key),\n ),\n );\n\n const result = locateResultCodec.toPixelResult(\n rawLocateValue,\n locateResultContext,\n );\n action.param[field] = {\n ...locateParamWithoutRawCoordinates,\n locatedPixelResult: result,\n };\n });\n });\n}\n"],"names":["debug","getDebug","normalizePlanningActionLocateFields","actions","actionSpace","includeLocateInPlanning","locateResultCodec","locateResultContext","acceptBbox2dAlias","parseRawLocateParameter","action","actionInActionSpace","actionInSpace","locateFields","findAllMidsceneLocatorField","field","rawLocateParameter","locateParameter","assert","resultKey","rawLocateValue","undefined","rawCoordinateKeys","Set","locateParamWithoutRawCoordinates","Object","key","result"],"mappings":";;;AAWA,MAAMA,QAAQC,SAAS;AAEhB,SAASC,oCACdC,OAAyB,EACzB,EACEC,WAAW,EACXC,uBAAuB,EACvBC,iBAAiB,EACjBC,mBAAmB,EACnBC,oBAAoB,KAAK,EACzBC,uBAAuB,EAQxB;IAEDN,QAAQ,OAAO,CAAC,CAACO;QACf,MAAMC,sBAAsBP,YAAY,IAAI,CAC1C,CAACQ,gBAAkBA,cAAc,IAAI,KAAKF,OAAO,IAAI;QAEvD,IAAI,CAACC,qBAAqB,YACxBX,MAAM,4DAA4DU;QAIpEV,MAAM,+BAA+BW;QACrC,MAAME,eAAeC,4BACnBH,oBAAoB,WAAW;QAGjCX,MAAM,gBAAgBa;QAEtBA,aAAa,OAAO,CAAC,CAACE;YACpB,MAAMC,qBAAqBN,OAAO,KAAK,EAAE,CAACK,MAAM;YAChD,IAAI,CAACC,oBACH;YAGF,MAAMC,kBAAkBR,wBAAwBO;YAEhD,IAAI,CAACX,yBAAyB;gBAE5BK,OAAO,KAAK,CAACK,MAAM,GAAG;oBAAE,QAAQE,gBAAgB,MAAM;gBAAC;gBACvD;YACF;YAEAC,OACEZ,mBACA;YAGF,MAAMa,YAAYb,kBAAkB,UAAU,CAAC,SAAS;YACxD,MAAMc,iBACJH,AAA+BI,WAA/BJ,eAAe,CAACE,UAAU,GACtBF,eAAe,CAACE,UAAU,GAC1BX,qBAAqBW,AAAc,WAAdA,YACnBF,gBAAgB,OAAO,GACvBI;YAIR,MAAMC,oBAAoB,IAAIC,IAAI;gBAChCJ;mBACIX,qBAAqBW,AAAc,WAAdA,YAAuB;oBAAC;iBAAU,GAAG,EAAE;aACjE;YAED,MAAMK,mCAAmCC,OAAO,WAAW,CACzDA,OAAO,OAAO,CAACR,iBAAiB,MAAM,CACpC,CAAC,CAACS,IAAI,GAAK,CAACJ,kBAAkB,GAAG,CAACI;YAItC,MAAMC,SAASrB,kBAAkB,aAAa,CAC5Cc,gBACAb;YAEFG,OAAO,KAAK,CAACK,MAAM,GAAG;gBACpB,GAAGS,gCAAgC;gBACnC,oBAAoBG;YACtB;QACF;IACF;AACF"}
1
+ {"version":3,"file":"ai-model/workflows/planning/locate-normalization.mjs","sources":["../../../../../src/ai-model/workflows/planning/locate-normalization.ts"],"sourcesContent":["import { findAllMidsceneLocatorField } from '@/common';\nimport type { DeviceAction } from '@/device';\nimport type { PlanningAction } from '@/types';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type { ParsedPlanningLocateParameter } from '../../model-adapter/planning-protocol';\nimport type {\n LocateResultCodec,\n LocateResultContext,\n} from '../../shared/model-locate-result/types';\n\nconst debug = getDebug('planning');\n\nexport function normalizePlanningActionLocateFields(\n actions: PlanningAction[],\n {\n actionSpace,\n ...options\n }: PlanningLocateNormalizationOptions & {\n actionSpace: DeviceAction[];\n },\n): void {\n actions.forEach((action) => {\n const actionInActionSpace = actionSpace.find(\n (actionInSpace) => actionInSpace.name === action.type,\n );\n if (!actionInActionSpace) {\n debug('skip locate normalization for action outside actionSpace', action);\n return;\n }\n\n debug('actionInActionSpace matched', actionInActionSpace);\n const locateFields = findAllMidsceneLocatorField(\n actionInActionSpace.paramSchema,\n );\n\n debug('locateFields', locateFields);\n\n locateFields.forEach((field) => {\n const locateParameter = action.param?.[field];\n if (!locateParameter) {\n return;\n }\n\n action.param[field] = normalizePlanningLocateParameter(\n locateParameter,\n options,\n );\n });\n });\n}\n\nexport type PlanningLocateNormalizationOptions = {\n includeLocateInPlanning: boolean;\n locateResultCodec?: LocateResultCodec;\n locateResultContext: LocateResultContext;\n acceptBbox2dAlias?: boolean;\n};\n\nexport function normalizePlanningLocateParameter(\n locateParameter: ParsedPlanningLocateParameter,\n {\n includeLocateInPlanning,\n locateResultCodec,\n locateResultContext,\n acceptBbox2dAlias = false,\n }: PlanningLocateNormalizationOptions,\n): Record<string, unknown> {\n if (!includeLocateInPlanning) {\n // In prompt-only planning mode, ignore any accidental coordinates from the model.\n return { prompt: locateParameter.prompt };\n }\n\n assert(\n locateResultCodec,\n 'planning locate normalization requires a locate result codec',\n );\n\n const resultKey = locateResultCodec.promptSpec.resultKey;\n const rawLocateValue =\n locateParameter[resultKey] !== undefined\n ? locateParameter[resultKey]\n : acceptBbox2dAlias && resultKey === 'bbox'\n ? locateParameter.bbox_2d\n : undefined;\n\n // The raw result field is replaced by locatedPixelResult, so it should not\n // remain in the normalized locate parameter.\n const rawCoordinateKeys = new Set([\n resultKey,\n ...(acceptBbox2dAlias && resultKey === 'bbox' ? ['bbox_2d'] : []),\n ]);\n\n const locateParamWithoutRawCoordinates = Object.fromEntries(\n Object.entries(locateParameter).filter(\n ([key]) => !rawCoordinateKeys.has(key),\n ),\n );\n\n const result = locateResultCodec.toPixelResult(\n rawLocateValue,\n locateResultContext,\n );\n return {\n ...locateParamWithoutRawCoordinates,\n locatedPixelResult: result,\n };\n}\n"],"names":["debug","getDebug","normalizePlanningActionLocateFields","actions","actionSpace","options","action","actionInActionSpace","actionInSpace","locateFields","findAllMidsceneLocatorField","field","locateParameter","normalizePlanningLocateParameter","includeLocateInPlanning","locateResultCodec","locateResultContext","acceptBbox2dAlias","assert","resultKey","rawLocateValue","undefined","rawCoordinateKeys","Set","locateParamWithoutRawCoordinates","Object","key","result"],"mappings":";;;AAWA,MAAMA,QAAQC,SAAS;AAEhB,SAASC,oCACdC,OAAyB,EACzB,EACEC,WAAW,EACX,GAAGC,SAGJ;IAEDF,QAAQ,OAAO,CAAC,CAACG;QACf,MAAMC,sBAAsBH,YAAY,IAAI,CAC1C,CAACI,gBAAkBA,cAAc,IAAI,KAAKF,OAAO,IAAI;QAEvD,IAAI,CAACC,qBAAqB,YACxBP,MAAM,4DAA4DM;QAIpEN,MAAM,+BAA+BO;QACrC,MAAME,eAAeC,4BACnBH,oBAAoB,WAAW;QAGjCP,MAAM,gBAAgBS;QAEtBA,aAAa,OAAO,CAAC,CAACE;YACpB,MAAMC,kBAAkBN,OAAO,KAAK,EAAE,CAACK,MAAM;YAC7C,IAAI,CAACC,iBACH;YAGFN,OAAO,KAAK,CAACK,MAAM,GAAGE,iCACpBD,iBACAP;QAEJ;IACF;AACF;AASO,SAASQ,iCACdD,eAA8C,EAC9C,EACEE,uBAAuB,EACvBC,iBAAiB,EACjBC,mBAAmB,EACnBC,oBAAoB,KAAK,EACU;IAErC,IAAI,CAACH,yBAEH,OAAO;QAAE,QAAQF,gBAAgB,MAAM;IAAC;IAG1CM,OACEH,mBACA;IAGF,MAAMI,YAAYJ,kBAAkB,UAAU,CAAC,SAAS;IACxD,MAAMK,iBACJR,AAA+BS,WAA/BT,eAAe,CAACO,UAAU,GACtBP,eAAe,CAACO,UAAU,GAC1BF,qBAAqBE,AAAc,WAAdA,YACnBP,gBAAgB,OAAO,GACvBS;IAIR,MAAMC,oBAAoB,IAAIC,IAAI;QAChCJ;WACIF,qBAAqBE,AAAc,WAAdA,YAAuB;YAAC;SAAU,GAAG,EAAE;KACjE;IAED,MAAMK,mCAAmCC,OAAO,WAAW,CACzDA,OAAO,OAAO,CAACb,iBAAiB,MAAM,CACpC,CAAC,CAACc,IAAI,GAAK,CAACJ,kBAAkB,GAAG,CAACI;IAItC,MAAMC,SAASZ,kBAAkB,aAAa,CAC5CK,gBACAJ;IAEF,OAAO;QACL,GAAGQ,gCAAgC;QACnC,oBAAoBG;IACtB;AACF"}
@@ -0,0 +1,44 @@
1
+ import { TUserPromptSchema, findActionInActionSpaceOrThrow, findAllMidsceneLocatorField, parseActionParam, validateRequiredLocateFields } from "../../../common.mjs";
2
+ import { isPlainObject } from "@midscene/shared/utils";
3
+ import { z } from "zod";
4
+ import { normalizePlanningLocateParameter } from "./locate-normalization.mjs";
5
+ function parsePlanningActions(actions, { actionSpace, parseRawLocateParameter, ...normalizationOptions }) {
6
+ for (const action of actions){
7
+ const definition = findActionInActionSpaceOrThrow(action.type, actionSpace);
8
+ if (!definition.paramSchema) continue;
9
+ const locateFields = findAllMidsceneLocatorField(definition.paramSchema);
10
+ try {
11
+ for (const field of locateFields){
12
+ const rawLocate = action.param?.[field];
13
+ if (null != rawLocate) action.param[field] = parseRawLocateParameter(rawLocate);
14
+ }
15
+ validateRequiredLocateFields(action.param, definition.paramSchema);
16
+ parseActionParam(action.param, definition.paramSchema);
17
+ for (const field of locateFields){
18
+ const locate = action.param?.[field];
19
+ if (void 0 === locate) continue;
20
+ if (!isPlainObject(locate)) throw new Error(`${field}: Expected an object with a prompt field`);
21
+ const prompt = locate.prompt;
22
+ if (void 0 === prompt) throw new Error(`${field}.prompt: Required`);
23
+ const result = TUserPromptSchema.safeParse(prompt);
24
+ if (!result.success) throw new z.ZodError(result.error.issues.map((issue)=>({
25
+ ...issue,
26
+ path: [
27
+ field,
28
+ 'prompt',
29
+ ...issue.path
30
+ ]
31
+ })));
32
+ action.param[field] = normalizePlanningLocateParameter(locate, normalizationOptions);
33
+ }
34
+ } catch (error) {
35
+ const details = error instanceof z.ZodError ? error.issues.map((issue)=>`${issue.path.join('.') || 'param'}: ${issue.message}`).join('; ') : error instanceof Error ? error.message : String(error);
36
+ throw new Error(`Invalid parameters for action ${action.type}: ${details}`, {
37
+ cause: error
38
+ });
39
+ }
40
+ }
41
+ }
42
+ export { parsePlanningActions };
43
+
44
+ //# sourceMappingURL=parse-planning-actions.mjs.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"ai-model/workflows/planning/parse-planning-actions.mjs","sources":["../../../../../src/ai-model/workflows/planning/parse-planning-actions.ts"],"sourcesContent":["import {\n TUserPromptSchema,\n findActionInActionSpaceOrThrow,\n findAllMidsceneLocatorField,\n parseActionParam,\n validateRequiredLocateFields,\n} from '@/common';\nimport type { DeviceAction, PlanningAction } from '@/types';\nimport { isPlainObject } from '@midscene/shared/utils';\nimport { z } from 'zod';\n\nimport type { ParsedPlanningLocateParameter } from '../../model-adapter/planning-protocol';\nimport {\n type PlanningLocateNormalizationOptions,\n normalizePlanningLocateParameter,\n} from './locate-normalization';\n\n// Mutates locator fields in place; ordinary parameters retain their model values.\nexport function parsePlanningActions(\n actions: PlanningAction[],\n {\n actionSpace,\n parseRawLocateParameter,\n ...normalizationOptions\n }: PlanningLocateNormalizationOptions & {\n actionSpace: DeviceAction[];\n parseRawLocateParameter: (value: unknown) => ParsedPlanningLocateParameter;\n },\n): void {\n for (const action of actions) {\n const definition = findActionInActionSpaceOrThrow(action.type, actionSpace);\n if (!definition.paramSchema) {\n continue;\n }\n\n const locateFields = findAllMidsceneLocatorField(definition.paramSchema);\n try {\n for (const field of locateFields) {\n const rawLocate = action.param?.[field];\n if (rawLocate !== undefined && rawLocate !== null) {\n action.param[field] = parseRawLocateParameter(rawLocate);\n }\n }\n validateRequiredLocateFields(action.param, definition.paramSchema);\n // Only preflight validation: keep ordinary parameters unchanged. Parsing can\n // evaluate defaults/transforms, but its output is consumed only at execution.\n parseActionParam(action.param, definition.paramSchema);\n\n for (const field of locateFields) {\n const locate = action.param?.[field];\n if (locate === undefined) {\n continue;\n }\n // Validate only the model-facing prompt; coordinates are normalized later.\n if (!isPlainObject(locate)) {\n throw new Error(`${field}: Expected an object with a prompt field`);\n }\n const prompt = locate.prompt;\n if (prompt === undefined) {\n throw new Error(`${field}.prompt: Required`);\n }\n const result = TUserPromptSchema.safeParse(prompt);\n if (!result.success) {\n throw new z.ZodError(\n result.error.issues.map((issue) => ({\n ...issue,\n path: [field, 'prompt', ...issue.path],\n })),\n );\n }\n action.param[field] = normalizePlanningLocateParameter(\n locate,\n normalizationOptions,\n );\n }\n } catch (error) {\n const details =\n error instanceof z.ZodError\n ? error.issues\n .map(\n (issue) =>\n `${issue.path.join('.') || 'param'}: ${issue.message}`,\n )\n .join('; ')\n : error instanceof Error\n ? error.message\n : String(error);\n throw new Error(\n `Invalid parameters for action ${action.type}: ${details}`,\n { cause: error },\n );\n }\n }\n}\n"],"names":["parsePlanningActions","actions","actionSpace","parseRawLocateParameter","normalizationOptions","action","definition","findActionInActionSpaceOrThrow","locateFields","findAllMidsceneLocatorField","field","rawLocate","validateRequiredLocateFields","parseActionParam","locate","undefined","isPlainObject","Error","prompt","result","TUserPromptSchema","z","issue","normalizePlanningLocateParameter","error","details","String"],"mappings":";;;;AAkBO,SAASA,qBACdC,OAAyB,EACzB,EACEC,WAAW,EACXC,uBAAuB,EACvB,GAAGC,sBAIJ;IAED,KAAK,MAAMC,UAAUJ,QAAS;QAC5B,MAAMK,aAAaC,+BAA+BF,OAAO,IAAI,EAAEH;QAC/D,IAAI,CAACI,WAAW,WAAW,EACzB;QAGF,MAAME,eAAeC,4BAA4BH,WAAW,WAAW;QACvE,IAAI;YACF,KAAK,MAAMI,SAASF,aAAc;gBAChC,MAAMG,YAAYN,OAAO,KAAK,EAAE,CAACK,MAAM;gBACvC,IAAIC,QAAAA,WACFN,OAAO,KAAK,CAACK,MAAM,GAAGP,wBAAwBQ;YAElD;YACAC,6BAA6BP,OAAO,KAAK,EAAEC,WAAW,WAAW;YAGjEO,iBAAiBR,OAAO,KAAK,EAAEC,WAAW,WAAW;YAErD,KAAK,MAAMI,SAASF,aAAc;gBAChC,MAAMM,SAAST,OAAO,KAAK,EAAE,CAACK,MAAM;gBACpC,IAAII,AAAWC,WAAXD,QACF;gBAGF,IAAI,CAACE,cAAcF,SACjB,MAAM,IAAIG,MAAM,GAAGP,MAAM,wCAAwC,CAAC;gBAEpE,MAAMQ,SAASJ,OAAO,MAAM;gBAC5B,IAAII,AAAWH,WAAXG,QACF,MAAM,IAAID,MAAM,GAAGP,MAAM,iBAAiB,CAAC;gBAE7C,MAAMS,SAASC,kBAAkB,SAAS,CAACF;gBAC3C,IAAI,CAACC,OAAO,OAAO,EACjB,MAAM,IAAIE,EAAE,QAAQ,CAClBF,OAAO,KAAK,CAAC,MAAM,CAAC,GAAG,CAAC,CAACG,QAAW;wBAClC,GAAGA,KAAK;wBACR,MAAM;4BAACZ;4BAAO;+BAAaY,MAAM,IAAI;yBAAC;oBACxC;gBAGJjB,OAAO,KAAK,CAACK,MAAM,GAAGa,iCACpBT,QACAV;YAEJ;QACF,EAAE,OAAOoB,OAAO;YACd,MAAMC,UACJD,iBAAiBH,EAAE,QAAQ,GACvBG,MAAM,MAAM,CACT,GAAG,CACF,CAACF,QACC,GAAGA,MAAM,IAAI,CAAC,IAAI,CAAC,QAAQ,QAAQ,EAAE,EAAEA,MAAM,OAAO,EAAE,EAEzD,IAAI,CAAC,QACRE,iBAAiBP,QACfO,MAAM,OAAO,GACbE,OAAOF;YACf,MAAM,IAAIP,MACR,CAAC,8BAA8B,EAAEZ,OAAO,IAAI,CAAC,EAAE,EAAEoB,SAAS,EAC1D;gBAAE,OAAOD;YAAM;QAEnB;IACF;AACF"}
@@ -7,7 +7,7 @@ import { AIResponseParseError, callAI } from "../../service-caller/index.mjs";
7
7
  import { callAiAndParseWithRetry, withSemanticRetryFeedback } from "../../service-caller/semantic-retry.mjs";
8
8
  import { planningModelFamilyRequiredForLocateMessage } from "../../shared/model-locate-result/errors.mjs";
9
9
  import { preparedReferenceImagesToChatMessages } from "../../shared/multimodal-prompt.mjs";
10
- import { normalizePlanningActionLocateFields } from "./locate-normalization.mjs";
10
+ import { parsePlanningActions } from "./parse-planning-actions.mjs";
11
11
  import { parseStandardPlanningResponse } from "./standard-planning-parser.mjs";
12
12
  const debug = getDebug('planning');
13
13
  const warnLog = getDebug('planning', {
@@ -31,6 +31,7 @@ async function callAndParsePlanningResponse(options) {
31
31
  actionSpace,
32
32
  logSource: includeLog ? 'model' : 'action'
33
33
  });
34
+ if (!planFromAI.action && void 0 === planFromAI.finalizeSuccess && !planFromAI.error) throw new Error('Incomplete planning response: provide an action, <complete>, or <error>. A <planning> explanation alone is not sufficient.');
34
35
  if (planFromAI.action && void 0 !== planFromAI.finalizeSuccess) {
35
36
  warnLog('Planning response included both an action and <complete>; ignoring <complete> output.');
36
37
  planFromAI.finalizeMessage = void 0;
@@ -39,20 +40,18 @@ async function callAndParsePlanningResponse(options) {
39
40
  const actions = planFromAI.action ? [
40
41
  planFromAI.action
41
42
  ] : [];
42
- normalizePlanningActionLocateFields(actions, {
43
+ parsePlanningActions(actions, {
44
+ parseRawLocateParameter: actionOutputProtocol.parseRawLocateParameter,
43
45
  actionSpace,
44
46
  includeLocateInPlanning,
45
47
  locateResultCodec,
46
48
  locateResultContext,
47
- acceptBbox2dAlias: modelRuntime.adapter.acceptBbox2dAlias,
48
- parseRawLocateParameter: actionOutputProtocol.parseRawLocateParameter
49
+ acceptBbox2dAlias: modelRuntime.adapter.acceptBbox2dAlias
49
50
  });
50
- const yamlFlow = buildYamlFlowFromPlans(actions, actionSpace);
51
51
  return {
52
52
  response,
53
53
  planFromAI,
54
- actions,
55
- yamlFlow
54
+ actions
56
55
  };
57
56
  },
58
57
  toParseError: (parseError, response)=>{
@@ -165,7 +164,7 @@ async function standardPlan(userInstruction, opts) {
165
164
  ...instruction,
166
165
  ...historyLog
167
166
  ];
168
- const { response: { content: rawResponse, usage, reasoning_content, rawChoiceMessage }, planFromAI, actions, yamlFlow } = await callAndParsePlanningResponse({
167
+ const { response: { content: rawResponse, usage, reasoning_content, rawChoiceMessage }, planFromAI, actions } = await callAndParsePlanningResponse({
169
168
  messages: msgs,
170
169
  modelRuntime,
171
170
  abortSignal: opts.abortSignal,
@@ -179,6 +178,7 @@ async function standardPlan(userInstruction, opts) {
179
178
  includeThought,
180
179
  includeLog
181
180
  });
181
+ const yamlFlow = buildYamlFlowFromPlans(actions, opts.actionSpace);
182
182
  let shouldContinuePlanning = true;
183
183
  if (void 0 !== planFromAI.finalizeSuccess) {
184
184
  debug('task completed via <complete> tag, stop planning');
@@ -1 +1 @@
1
- {"version":3,"file":"ai-model/workflows/planning/standard-planning.mjs","sources":["../../../../../src/ai-model/workflows/planning/standard-planning.ts"],"sourcesContent":["import type {\n PlanningAIResponse,\n PlanningAction,\n RawResponsePlanningAIResponse,\n} from '@/types';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type { ChatCompletionMessageParam } from 'openai/resources/index';\nimport { buildYamlFlowFromPlans } from '../../../common';\nimport { prepareModelImage } from '../../model-adapter/image-preprocess';\nimport { buildStandardPlanningSystemPrompt } from '../../prompt/planning';\nimport { AIResponseParseError, callAI } from '../../service-caller/index';\nimport {\n callAiAndParseWithRetry,\n withSemanticRetryFeedback,\n} from '../../service-caller/semantic-retry';\nimport type {\n LocateResultCodec,\n LocateResultContext,\n} from '../../shared/model-locate-result';\nimport { planningModelFamilyRequiredForLocateMessage } from '../../shared/model-locate-result/errors';\nimport {\n type PreparedUserPrompt,\n preparedReferenceImagesToChatMessages,\n} from '../../shared/multimodal-prompt';\nimport { normalizePlanningActionLocateFields } from './locate-normalization';\nimport { parseStandardPlanningResponse } from './standard-planning-parser';\nimport type { PlanOptions } from './types';\n\nconst debug = getDebug('planning');\nconst warnLog = getDebug('planning', { console: true });\n\nconst noPreviousActionsText =\n 'No previous actions have been executed in this aiAct execution yet. If the instruction asks for actions, choose the first action to execute.';\ntype PlanningCallResponse = Awaited<ReturnType<typeof callAI>>;\n\ntype CallAndParsePlanningResponseOptions = {\n messages: ChatCompletionMessageParam[];\n modelRuntime: PlanOptions['modelRuntime'];\n abortSignal?: AbortSignal;\n includeLocateInPlanning: boolean;\n actionSpace: PlanOptions['actionSpace'];\n locateResultCodec?: LocateResultCodec;\n locateResultContext: LocateResultContext;\n includeThought: boolean;\n includeLog: boolean;\n};\n\nasync function callAndParsePlanningResponse(\n options: CallAndParsePlanningResponseOptions,\n): Promise<{\n response: PlanningCallResponse;\n planFromAI: RawResponsePlanningAIResponse;\n actions: PlanningAction[];\n yamlFlow: ReturnType<typeof buildYamlFlowFromPlans>;\n}> {\n const {\n messages,\n modelRuntime,\n abortSignal,\n includeLocateInPlanning,\n actionSpace,\n locateResultCodec,\n locateResultContext,\n includeThought,\n includeLog,\n } = options;\n assert(\n modelRuntime.adapter.planning.kind === 'standard',\n 'callAndParsePlanningResponse requires a standard planning adapter',\n );\n const actionOutputProtocol =\n modelRuntime.adapter.planning.protocol.actionOutputProtocol;\n\n return callAiAndParseWithRetry({\n callAi: (retryAttempt, previousParseError) =>\n callAI(\n withSemanticRetryFeedback(messages, previousParseError),\n modelRuntime,\n {\n abortSignal,\n requiresOriginalImageDetail: includeLocateInPlanning,\n semanticRetryAttempt: retryAttempt,\n },\n ),\n parseResponse: (response) => {\n const planFromAI = parseStandardPlanningResponse(response.content, {\n includeThought,\n actionOutputProtocol,\n actionSpace,\n logSource: includeLog ? 'model' : 'action',\n });\n if (planFromAI.action && planFromAI.finalizeSuccess !== undefined) {\n warnLog(\n 'Planning response included both an action and <complete>; ignoring <complete> output.',\n );\n planFromAI.finalizeMessage = undefined;\n planFromAI.finalizeSuccess = undefined;\n }\n\n const actions = planFromAI.action ? [planFromAI.action] : [];\n normalizePlanningActionLocateFields(actions, {\n actionSpace,\n includeLocateInPlanning,\n locateResultCodec,\n locateResultContext,\n acceptBbox2dAlias: modelRuntime.adapter.acceptBbox2dAlias,\n parseRawLocateParameter: actionOutputProtocol.parseRawLocateParameter,\n });\n // dumpActionParam keeps only the locator prompt, so runtime-only\n // locatedPixelResult fields added during normalization are not serialized.\n const yamlFlow = buildYamlFlowFromPlans(actions, actionSpace);\n return { response, planFromAI, actions, yamlFlow };\n },\n toParseError: (parseError, response) => {\n const errorMessage =\n parseError instanceof Error ? parseError.message : String(parseError);\n return new AIResponseParseError(\n `XML parse error: ${errorMessage}`,\n response.content,\n response.usage,\n response.rawChoiceMessage,\n response.reasoning_content,\n );\n },\n parseRetryTimes: modelRuntime.config.retryCount,\n parseRetryInterval: modelRuntime.config.retryInterval,\n abortSignal,\n onParseRetry: (parseError) => {\n debug(\n 'retrying plan after response parsing failed: %s',\n parseError instanceof Error ? parseError.message : String(parseError),\n );\n },\n });\n}\n\nexport async function standardPlan(\n userInstruction: PreparedUserPrompt,\n opts: PlanOptions,\n): Promise<PlanningAIResponse> {\n const { context, conversationHistory } = opts;\n const modelRuntime = opts.modelRuntime;\n const { adapter } = modelRuntime;\n const { shotSize } = context;\n const screenshotBase64 = context.screenshot.base64;\n assert(\n adapter.planning.kind === 'standard',\n 'standardPlan requires a standard planning adapter',\n );\n const planningProtocol = adapter.planning.protocol;\n\n if (opts.includeLocateInPlanning && !modelRuntime.config.modelFamily) {\n throw new Error(\n planningModelFamilyRequiredForLocateMessage(modelRuntime.config.slot),\n );\n }\n\n const locateResultCodec = modelRuntime.config.modelFamily\n ? adapter.planning.locateResultCodec\n : undefined;\n\n // Only enable sub-goals when aiAct is in deep-thinking planning mode.\n const includeSubGoals = opts.effort === 'deepThink';\n const includeThought = opts.effort !== 'fast';\n const includeLog = opts.effort !== 'fast';\n\n if (opts.includeLocateInPlanning && !locateResultCodec) {\n throw new Error(\n planningModelFamilyRequiredForLocateMessage(modelRuntime.config.slot),\n );\n }\n\n const systemPrompt = await buildStandardPlanningSystemPrompt({\n actionSpace: opts.actionSpace,\n includeThought,\n includeLog,\n includeSubGoals,\n planningProtocol,\n ...(opts.includeLocateInPlanning && locateResultCodec\n ? {\n includeLocateInPlanning: true,\n locatePromptSpec: locateResultCodec.promptSpec,\n }\n : { includeLocateInPlanning: false }),\n });\n\n const preparedImage = await prepareModelImage({\n imageBase64: screenshotBase64,\n width: shotSize.width,\n height: shotSize.height,\n policy: adapter.imagePreprocess,\n });\n const imagePayload = preparedImage.imageBase64;\n\n const actionContext = opts.actionContext ? `${opts.actionContext}\\n` : '';\n\n const referenceImageMessages = preparedReferenceImagesToChatMessages(\n userInstruction.referenceImages,\n );\n const instruction: ChatCompletionMessageParam[] = [\n {\n role: 'user',\n content: [\n {\n type: 'text',\n text: `${actionContext}<user_instruction>${userInstruction.text}</user_instruction>`,\n },\n ],\n },\n ...referenceImageMessages,\n ];\n\n let latestFeedbackMessage: ChatCompletionMessageParam;\n\n // Build sub-goal status text to include in the message\n // In planning deep-think mode: show full sub-goals with logs\n // Otherwise: show historical execution logs\n const executionProgressText = includeSubGoals\n ? conversationHistory.subGoalsToText()\n : conversationHistory.historicalLogsToText();\n const executionProgressSection = executionProgressText\n ? `\\n\\n${executionProgressText}`\n : conversationHistory.pendingFeedbackMessage\n ? ''\n : `\\n\\n${noPreviousActionsText}`;\n\n // Build memories text to include in the message\n const memoriesText = conversationHistory.memoriesToText();\n const memoriesSection = memoriesText ? `\\n\\n${memoriesText}` : '';\n\n if (conversationHistory.pendingFeedbackMessage) {\n latestFeedbackMessage = {\n role: 'user',\n content: [\n {\n type: 'text',\n text: `${conversationHistory.pendingFeedbackMessage}. The previous action has been executed, here is the latest screenshot. Please continue according to the instruction.${memoriesSection}${executionProgressSection}`,\n },\n {\n type: 'image_url',\n image_url: {\n url: imagePayload,\n detail: 'high',\n },\n },\n ],\n };\n\n conversationHistory.resetPendingFeedbackMessageIfExists();\n } else {\n latestFeedbackMessage = {\n role: 'user',\n content: [\n {\n type: 'text',\n text: `This is the current screenshot.${memoriesSection}${executionProgressSection}`,\n },\n {\n type: 'image_url',\n image_url: {\n url: imagePayload,\n detail: 'high',\n },\n },\n ],\n };\n }\n conversationHistory.append(latestFeedbackMessage);\n\n // Compress history if it exceeds the threshold to avoid context overflow\n conversationHistory.compressHistory(50, 20);\n\n const historyLog = conversationHistory.snapshot(opts.imagesIncludeCount);\n\n const msgs: ChatCompletionMessageParam[] = [\n { role: 'system', content: systemPrompt },\n ...instruction,\n ...historyLog,\n ];\n\n const {\n response: {\n content: rawResponse,\n usage,\n reasoning_content,\n rawChoiceMessage,\n },\n planFromAI,\n actions,\n yamlFlow,\n } = await callAndParsePlanningResponse({\n messages: msgs,\n modelRuntime,\n abortSignal: opts.abortSignal,\n includeLocateInPlanning: opts.includeLocateInPlanning,\n actionSpace: opts.actionSpace,\n locateResultCodec,\n locateResultContext: {\n preparedSize: preparedImage.preparedSize,\n contentSize: preparedImage.contentSize,\n },\n includeThought,\n includeLog,\n });\n\n let shouldContinuePlanning = true;\n\n // Check if task is completed via <complete> tag\n if (planFromAI.finalizeSuccess !== undefined) {\n debug('task completed via <complete> tag, stop planning');\n shouldContinuePlanning = false;\n // Mark all sub-goals as finished when goal is completed in planning deep-think mode.\n if (includeSubGoals) {\n conversationHistory.markAllSubGoalsFinished();\n }\n }\n\n const returnValue: PlanningAIResponse = {\n ...planFromAI,\n actions,\n rawResponse,\n rawChoiceMessage,\n usage,\n reasoning_content,\n yamlFlow,\n shouldContinuePlanning,\n };\n\n assert(planFromAI, \"can't get plans from AI\");\n\n // TODO: The plan log is recorded before its action has executed, so a failed\n // action may still appear in the next round as an action already performed.\n // Move this write to the successful action execution path in TaskExecutor.action.\n // Update sub-goals in conversation history only in planning deep-think mode.\n if (includeSubGoals) {\n if (planFromAI.updateSubGoals?.length) {\n conversationHistory.mergeSubGoals(planFromAI.updateSubGoals);\n }\n if (planFromAI.markFinishedIndexes?.length) {\n for (const index of planFromAI.markFinishedIndexes) {\n conversationHistory.markSubGoalFinished(index);\n }\n }\n // Append the planning log to the currently running sub-goal\n if (planFromAI.log) {\n conversationHistory.appendSubGoalLog(planFromAI.log);\n }\n } else {\n // Without planning deep-think mode, accumulate logs as historical execution steps.\n if (planFromAI.log) {\n conversationHistory.appendHistoricalLog(planFromAI.log);\n }\n }\n\n // Append memory to conversation history if present\n if (planFromAI.memory) {\n conversationHistory.appendMemory(planFromAI.memory);\n }\n\n // Some model providers require opaque assistant fields to be replayed\n // verbatim in later turns. Keep this opt-in per model adapter so that an\n // unverified provider does not receive non-standard response fields.\n if (\n modelRuntime.adapter.chatCompletion.replayRawAssistantMessage &&\n rawChoiceMessage\n ) {\n conversationHistory.append(rawChoiceMessage as ChatCompletionMessageParam);\n } else {\n conversationHistory.append({\n role: 'assistant',\n content: [\n {\n type: 'text',\n text: rawResponse,\n },\n ],\n });\n }\n\n return returnValue;\n}\n"],"names":["debug","getDebug","warnLog","noPreviousActionsText","callAndParsePlanningResponse","options","messages","modelRuntime","abortSignal","includeLocateInPlanning","actionSpace","locateResultCodec","locateResultContext","includeThought","includeLog","assert","actionOutputProtocol","callAiAndParseWithRetry","retryAttempt","previousParseError","callAI","withSemanticRetryFeedback","response","planFromAI","parseStandardPlanningResponse","undefined","actions","normalizePlanningActionLocateFields","yamlFlow","buildYamlFlowFromPlans","parseError","errorMessage","Error","String","AIResponseParseError","standardPlan","userInstruction","opts","context","conversationHistory","adapter","shotSize","screenshotBase64","planningProtocol","planningModelFamilyRequiredForLocateMessage","includeSubGoals","systemPrompt","buildStandardPlanningSystemPrompt","preparedImage","prepareModelImage","imagePayload","actionContext","referenceImageMessages","preparedReferenceImagesToChatMessages","instruction","latestFeedbackMessage","executionProgressText","executionProgressSection","memoriesText","memoriesSection","historyLog","msgs","rawResponse","usage","reasoning_content","rawChoiceMessage","shouldContinuePlanning","returnValue","index"],"mappings":";;;;;;;;;;;AA6BA,MAAMA,QAAQC,SAAS;AACvB,MAAMC,UAAUD,SAAS,YAAY;IAAE,SAAS;AAAK;AAErD,MAAME,wBACJ;AAeF,eAAeC,6BACbC,OAA4C;IAO5C,MAAM,EACJC,QAAQ,EACRC,YAAY,EACZC,WAAW,EACXC,uBAAuB,EACvBC,WAAW,EACXC,iBAAiB,EACjBC,mBAAmB,EACnBC,cAAc,EACdC,UAAU,EACX,GAAGT;IACJU,OACER,AAAuC,eAAvCA,aAAa,OAAO,CAAC,QAAQ,CAAC,IAAI,EAClC;IAEF,MAAMS,uBACJT,aAAa,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAC,oBAAoB;IAE7D,OAAOU,wBAAwB;QAC7B,QAAQ,CAACC,cAAcC,qBACrBC,OACEC,0BAA0Bf,UAAUa,qBACpCZ,cACA;gBACEC;gBACA,6BAA6BC;gBAC7B,sBAAsBS;YACxB;QAEJ,eAAe,CAACI;YACd,MAAMC,aAAaC,8BAA8BF,SAAS,OAAO,EAAE;gBACjET;gBACAG;gBACAN;gBACA,WAAWI,aAAa,UAAU;YACpC;YACA,IAAIS,WAAW,MAAM,IAAIA,AAA+BE,WAA/BF,WAAW,eAAe,EAAgB;gBACjErB,QACE;gBAEFqB,WAAW,eAAe,GAAGE;gBAC7BF,WAAW,eAAe,GAAGE;YAC/B;YAEA,MAAMC,UAAUH,WAAW,MAAM,GAAG;gBAACA,WAAW,MAAM;aAAC,GAAG,EAAE;YAC5DI,oCAAoCD,SAAS;gBAC3ChB;gBACAD;gBACAE;gBACAC;gBACA,mBAAmBL,aAAa,OAAO,CAAC,iBAAiB;gBACzD,yBAAyBS,qBAAqB,uBAAuB;YACvE;YAGA,MAAMY,WAAWC,uBAAuBH,SAAShB;YACjD,OAAO;gBAAEY;gBAAUC;gBAAYG;gBAASE;YAAS;QACnD;QACA,cAAc,CAACE,YAAYR;YACzB,MAAMS,eACJD,sBAAsBE,QAAQF,WAAW,OAAO,GAAGG,OAAOH;YAC5D,OAAO,IAAII,qBACT,CAAC,iBAAiB,EAAEH,cAAc,EAClCT,SAAS,OAAO,EAChBA,SAAS,KAAK,EACdA,SAAS,gBAAgB,EACzBA,SAAS,iBAAiB;QAE9B;QACA,iBAAiBf,aAAa,MAAM,CAAC,UAAU;QAC/C,oBAAoBA,aAAa,MAAM,CAAC,aAAa;QACrDC;QACA,cAAc,CAACsB;YACb9B,MACE,mDACA8B,sBAAsBE,QAAQF,WAAW,OAAO,GAAGG,OAAOH;QAE9D;IACF;AACF;AAEO,eAAeK,aACpBC,eAAmC,EACnCC,IAAiB;IAEjB,MAAM,EAAEC,OAAO,EAAEC,mBAAmB,EAAE,GAAGF;IACzC,MAAM9B,eAAe8B,KAAK,YAAY;IACtC,MAAM,EAAEG,OAAO,EAAE,GAAGjC;IACpB,MAAM,EAAEkC,QAAQ,EAAE,GAAGH;IACrB,MAAMI,mBAAmBJ,QAAQ,UAAU,CAAC,MAAM;IAClDvB,OACEyB,AAA0B,eAA1BA,QAAQ,QAAQ,CAAC,IAAI,EACrB;IAEF,MAAMG,mBAAmBH,QAAQ,QAAQ,CAAC,QAAQ;IAElD,IAAIH,KAAK,uBAAuB,IAAI,CAAC9B,aAAa,MAAM,CAAC,WAAW,EAClE,MAAM,IAAIyB,MACRY,4CAA4CrC,aAAa,MAAM,CAAC,IAAI;IAIxE,MAAMI,oBAAoBJ,aAAa,MAAM,CAAC,WAAW,GACrDiC,QAAQ,QAAQ,CAAC,iBAAiB,GAClCf;IAGJ,MAAMoB,kBAAkBR,AAAgB,gBAAhBA,KAAK,MAAM;IACnC,MAAMxB,iBAAiBwB,AAAgB,WAAhBA,KAAK,MAAM;IAClC,MAAMvB,aAAauB,AAAgB,WAAhBA,KAAK,MAAM;IAE9B,IAAIA,KAAK,uBAAuB,IAAI,CAAC1B,mBACnC,MAAM,IAAIqB,MACRY,4CAA4CrC,aAAa,MAAM,CAAC,IAAI;IAIxE,MAAMuC,eAAe,MAAMC,kCAAkC;QAC3D,aAAaV,KAAK,WAAW;QAC7BxB;QACAC;QACA+B;QACAF;QACA,GAAIN,KAAK,uBAAuB,IAAI1B,oBAChC;YACE,yBAAyB;YACzB,kBAAkBA,kBAAkB,UAAU;QAChD,IACA;YAAE,yBAAyB;QAAM,CAAC;IACxC;IAEA,MAAMqC,gBAAgB,MAAMC,kBAAkB;QAC5C,aAAaP;QACb,OAAOD,SAAS,KAAK;QACrB,QAAQA,SAAS,MAAM;QACvB,QAAQD,QAAQ,eAAe;IACjC;IACA,MAAMU,eAAeF,cAAc,WAAW;IAE9C,MAAMG,gBAAgBd,KAAK,aAAa,GAAG,GAAGA,KAAK,aAAa,CAAC,EAAE,CAAC,GAAG;IAEvE,MAAMe,yBAAyBC,sCAC7BjB,gBAAgB,eAAe;IAEjC,MAAMkB,cAA4C;QAChD;YACE,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,MAAM,GAAGH,cAAc,kBAAkB,EAAEf,gBAAgB,IAAI,CAAC,mBAAmB,CAAC;gBACtF;aACD;QACH;WACGgB;KACJ;IAED,IAAIG;IAKJ,MAAMC,wBAAwBX,kBAC1BN,oBAAoB,cAAc,KAClCA,oBAAoB,oBAAoB;IAC5C,MAAMkB,2BAA2BD,wBAC7B,CAAC,IAAI,EAAEA,uBAAuB,GAC9BjB,oBAAoB,sBAAsB,GACxC,KACA,CAAC,IAAI,EAAEpC,uBAAuB;IAGpC,MAAMuD,eAAenB,oBAAoB,cAAc;IACvD,MAAMoB,kBAAkBD,eAAe,CAAC,IAAI,EAAEA,cAAc,GAAG;IAE/D,IAAInB,oBAAoB,sBAAsB,EAAE;QAC9CgB,wBAAwB;YACtB,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,MAAM,GAAGhB,oBAAoB,sBAAsB,CAAC,qHAAqH,EAAEoB,kBAAkBF,0BAA0B;gBACzN;gBACA;oBACE,MAAM;oBACN,WAAW;wBACT,KAAKP;wBACL,QAAQ;oBACV;gBACF;aACD;QACH;QAEAX,oBAAoB,mCAAmC;IACzD,OACEgB,wBAAwB;QACtB,MAAM;QACN,SAAS;YACP;gBACE,MAAM;gBACN,MAAM,CAAC,+BAA+B,EAAEI,kBAAkBF,0BAA0B;YACtF;YACA;gBACE,MAAM;gBACN,WAAW;oBACT,KAAKP;oBACL,QAAQ;gBACV;YACF;SACD;IACH;IAEFX,oBAAoB,MAAM,CAACgB;IAG3BhB,oBAAoB,eAAe,CAAC,IAAI;IAExC,MAAMqB,aAAarB,oBAAoB,QAAQ,CAACF,KAAK,kBAAkB;IAEvE,MAAMwB,OAAqC;QACzC;YAAE,MAAM;YAAU,SAASf;QAAa;WACrCQ;WACAM;KACJ;IAED,MAAM,EACJ,UAAU,EACR,SAASE,WAAW,EACpBC,KAAK,EACLC,iBAAiB,EACjBC,gBAAgB,EACjB,EACD1C,UAAU,EACVG,OAAO,EACPE,QAAQ,EACT,GAAG,MAAMxB,6BAA6B;QACrC,UAAUyD;QACVtD;QACA,aAAa8B,KAAK,WAAW;QAC7B,yBAAyBA,KAAK,uBAAuB;QACrD,aAAaA,KAAK,WAAW;QAC7B1B;QACA,qBAAqB;YACnB,cAAcqC,cAAc,YAAY;YACxC,aAAaA,cAAc,WAAW;QACxC;QACAnC;QACAC;IACF;IAEA,IAAIoD,yBAAyB;IAG7B,IAAI3C,AAA+BE,WAA/BF,WAAW,eAAe,EAAgB;QAC5CvB,MAAM;QACNkE,yBAAyB;QAEzB,IAAIrB,iBACFN,oBAAoB,uBAAuB;IAE/C;IAEA,MAAM4B,cAAkC;QACtC,GAAG5C,UAAU;QACbG;QACAoC;QACAG;QACAF;QACAC;QACApC;QACAsC;IACF;IAEAnD,OAAOQ,YAAY;IAMnB,IAAIsB,iBAAiB;QACnB,IAAItB,WAAW,cAAc,EAAE,QAC7BgB,oBAAoB,aAAa,CAAChB,WAAW,cAAc;QAE7D,IAAIA,WAAW,mBAAmB,EAAE,QAClC,KAAK,MAAM6C,SAAS7C,WAAW,mBAAmB,CAChDgB,oBAAoB,mBAAmB,CAAC6B;QAI5C,IAAI7C,WAAW,GAAG,EAChBgB,oBAAoB,gBAAgB,CAAChB,WAAW,GAAG;IAEvD,OAEE,IAAIA,WAAW,GAAG,EAChBgB,oBAAoB,mBAAmB,CAAChB,WAAW,GAAG;IAK1D,IAAIA,WAAW,MAAM,EACnBgB,oBAAoB,YAAY,CAAChB,WAAW,MAAM;IAMpD,IACEhB,aAAa,OAAO,CAAC,cAAc,CAAC,yBAAyB,IAC7D0D,kBAEA1B,oBAAoB,MAAM,CAAC0B;SAE3B1B,oBAAoB,MAAM,CAAC;QACzB,MAAM;QACN,SAAS;YACP;gBACE,MAAM;gBACN,MAAMuB;YACR;SACD;IACH;IAGF,OAAOK;AACT"}
1
+ {"version":3,"file":"ai-model/workflows/planning/standard-planning.mjs","sources":["../../../../../src/ai-model/workflows/planning/standard-planning.ts"],"sourcesContent":["import type {\n PlanningAIResponse,\n PlanningAction,\n RawResponsePlanningAIResponse,\n} from '@/types';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type { ChatCompletionMessageParam } from 'openai/resources/index';\nimport { buildYamlFlowFromPlans } from '../../../common';\nimport { prepareModelImage } from '../../model-adapter/image-preprocess';\nimport { buildStandardPlanningSystemPrompt } from '../../prompt/planning';\nimport { AIResponseParseError, callAI } from '../../service-caller/index';\nimport {\n callAiAndParseWithRetry,\n withSemanticRetryFeedback,\n} from '../../service-caller/semantic-retry';\nimport type {\n LocateResultCodec,\n LocateResultContext,\n} from '../../shared/model-locate-result';\nimport { planningModelFamilyRequiredForLocateMessage } from '../../shared/model-locate-result/errors';\nimport {\n type PreparedUserPrompt,\n preparedReferenceImagesToChatMessages,\n} from '../../shared/multimodal-prompt';\nimport { parsePlanningActions } from './parse-planning-actions';\nimport { parseStandardPlanningResponse } from './standard-planning-parser';\nimport type { PlanOptions } from './types';\n\nconst debug = getDebug('planning');\nconst warnLog = getDebug('planning', { console: true });\n\nconst noPreviousActionsText =\n 'No previous actions have been executed in this aiAct execution yet. If the instruction asks for actions, choose the first action to execute.';\ntype PlanningCallResponse = Awaited<ReturnType<typeof callAI>>;\n\ntype CallAndParsePlanningResponseOptions = {\n messages: ChatCompletionMessageParam[];\n modelRuntime: PlanOptions['modelRuntime'];\n abortSignal?: AbortSignal;\n includeLocateInPlanning: boolean;\n actionSpace: PlanOptions['actionSpace'];\n locateResultCodec?: LocateResultCodec;\n locateResultContext: LocateResultContext;\n includeThought: boolean;\n includeLog: boolean;\n};\n\nasync function callAndParsePlanningResponse(\n options: CallAndParsePlanningResponseOptions,\n): Promise<{\n response: PlanningCallResponse;\n planFromAI: RawResponsePlanningAIResponse;\n actions: PlanningAction[];\n}> {\n const {\n messages,\n modelRuntime,\n abortSignal,\n includeLocateInPlanning,\n actionSpace,\n locateResultCodec,\n locateResultContext,\n includeThought,\n includeLog,\n } = options;\n assert(\n modelRuntime.adapter.planning.kind === 'standard',\n 'callAndParsePlanningResponse requires a standard planning adapter',\n );\n const actionOutputProtocol =\n modelRuntime.adapter.planning.protocol.actionOutputProtocol;\n\n return callAiAndParseWithRetry({\n callAi: (retryAttempt, previousParseError) =>\n callAI(\n withSemanticRetryFeedback(messages, previousParseError),\n modelRuntime,\n {\n abortSignal,\n requiresOriginalImageDetail: includeLocateInPlanning,\n semanticRetryAttempt: retryAttempt,\n },\n ),\n parseResponse: (response) => {\n const planFromAI = parseStandardPlanningResponse(response.content, {\n includeThought,\n actionOutputProtocol,\n actionSpace,\n logSource: includeLog ? 'model' : 'action',\n });\n if (\n !planFromAI.action &&\n planFromAI.finalizeSuccess === undefined &&\n !planFromAI.error\n ) {\n throw new Error(\n 'Incomplete planning response: provide an action, <complete>, or <error>. A <planning> explanation alone is not sufficient.',\n );\n }\n if (planFromAI.action && planFromAI.finalizeSuccess !== undefined) {\n warnLog(\n 'Planning response included both an action and <complete>; ignoring <complete> output.',\n );\n planFromAI.finalizeMessage = undefined;\n planFromAI.finalizeSuccess = undefined;\n }\n\n const actions = planFromAI.action ? [planFromAI.action] : [];\n parsePlanningActions(actions, {\n parseRawLocateParameter: actionOutputProtocol.parseRawLocateParameter,\n actionSpace,\n includeLocateInPlanning,\n locateResultCodec,\n locateResultContext,\n acceptBbox2dAlias: modelRuntime.adapter.acceptBbox2dAlias,\n });\n return { response, planFromAI, actions };\n },\n toParseError: (parseError, response) => {\n const errorMessage =\n parseError instanceof Error ? parseError.message : String(parseError);\n return new AIResponseParseError(\n `XML parse error: ${errorMessage}`,\n response.content,\n response.usage,\n response.rawChoiceMessage,\n response.reasoning_content,\n );\n },\n parseRetryTimes: modelRuntime.config.retryCount,\n parseRetryInterval: modelRuntime.config.retryInterval,\n abortSignal,\n onParseRetry: (parseError) => {\n debug(\n 'retrying plan after response parsing failed: %s',\n parseError instanceof Error ? parseError.message : String(parseError),\n );\n },\n });\n}\n\nexport async function standardPlan(\n userInstruction: PreparedUserPrompt,\n opts: PlanOptions,\n): Promise<PlanningAIResponse> {\n const { context, conversationHistory } = opts;\n const modelRuntime = opts.modelRuntime;\n const { adapter } = modelRuntime;\n const { shotSize } = context;\n const screenshotBase64 = context.screenshot.base64;\n assert(\n adapter.planning.kind === 'standard',\n 'standardPlan requires a standard planning adapter',\n );\n const planningProtocol = adapter.planning.protocol;\n\n if (opts.includeLocateInPlanning && !modelRuntime.config.modelFamily) {\n throw new Error(\n planningModelFamilyRequiredForLocateMessage(modelRuntime.config.slot),\n );\n }\n\n const locateResultCodec = modelRuntime.config.modelFamily\n ? adapter.planning.locateResultCodec\n : undefined;\n\n // Only enable sub-goals when aiAct is in deep-thinking planning mode.\n const includeSubGoals = opts.effort === 'deepThink';\n const includeThought = opts.effort !== 'fast';\n const includeLog = opts.effort !== 'fast';\n\n if (opts.includeLocateInPlanning && !locateResultCodec) {\n throw new Error(\n planningModelFamilyRequiredForLocateMessage(modelRuntime.config.slot),\n );\n }\n\n const systemPrompt = await buildStandardPlanningSystemPrompt({\n actionSpace: opts.actionSpace,\n includeThought,\n includeLog,\n includeSubGoals,\n planningProtocol,\n ...(opts.includeLocateInPlanning && locateResultCodec\n ? {\n includeLocateInPlanning: true,\n locatePromptSpec: locateResultCodec.promptSpec,\n }\n : { includeLocateInPlanning: false }),\n });\n\n const preparedImage = await prepareModelImage({\n imageBase64: screenshotBase64,\n width: shotSize.width,\n height: shotSize.height,\n policy: adapter.imagePreprocess,\n });\n const imagePayload = preparedImage.imageBase64;\n\n const actionContext = opts.actionContext ? `${opts.actionContext}\\n` : '';\n\n const referenceImageMessages = preparedReferenceImagesToChatMessages(\n userInstruction.referenceImages,\n );\n const instruction: ChatCompletionMessageParam[] = [\n {\n role: 'user',\n content: [\n {\n type: 'text',\n text: `${actionContext}<user_instruction>${userInstruction.text}</user_instruction>`,\n },\n ],\n },\n ...referenceImageMessages,\n ];\n\n let latestFeedbackMessage: ChatCompletionMessageParam;\n\n // Build sub-goal status text to include in the message\n // In planning deep-think mode: show full sub-goals with logs\n // Otherwise: show historical execution logs\n const executionProgressText = includeSubGoals\n ? conversationHistory.subGoalsToText()\n : conversationHistory.historicalLogsToText();\n const executionProgressSection = executionProgressText\n ? `\\n\\n${executionProgressText}`\n : conversationHistory.pendingFeedbackMessage\n ? ''\n : `\\n\\n${noPreviousActionsText}`;\n\n // Build memories text to include in the message\n const memoriesText = conversationHistory.memoriesToText();\n const memoriesSection = memoriesText ? `\\n\\n${memoriesText}` : '';\n\n if (conversationHistory.pendingFeedbackMessage) {\n latestFeedbackMessage = {\n role: 'user',\n content: [\n {\n type: 'text',\n text: `${conversationHistory.pendingFeedbackMessage}. The previous action has been executed, here is the latest screenshot. Please continue according to the instruction.${memoriesSection}${executionProgressSection}`,\n },\n {\n type: 'image_url',\n image_url: {\n url: imagePayload,\n detail: 'high',\n },\n },\n ],\n };\n\n conversationHistory.resetPendingFeedbackMessageIfExists();\n } else {\n latestFeedbackMessage = {\n role: 'user',\n content: [\n {\n type: 'text',\n text: `This is the current screenshot.${memoriesSection}${executionProgressSection}`,\n },\n {\n type: 'image_url',\n image_url: {\n url: imagePayload,\n detail: 'high',\n },\n },\n ],\n };\n }\n conversationHistory.append(latestFeedbackMessage);\n\n // Compress history if it exceeds the threshold to avoid context overflow\n conversationHistory.compressHistory(50, 20);\n\n const historyLog = conversationHistory.snapshot(opts.imagesIncludeCount);\n\n const msgs: ChatCompletionMessageParam[] = [\n { role: 'system', content: systemPrompt },\n ...instruction,\n ...historyLog,\n ];\n\n const {\n response: {\n content: rawResponse,\n usage,\n reasoning_content,\n rawChoiceMessage,\n },\n planFromAI,\n actions,\n } = await callAndParsePlanningResponse({\n messages: msgs,\n modelRuntime,\n abortSignal: opts.abortSignal,\n includeLocateInPlanning: opts.includeLocateInPlanning,\n actionSpace: opts.actionSpace,\n locateResultCodec,\n locateResultContext: {\n preparedSize: preparedImage.preparedSize,\n contentSize: preparedImage.contentSize,\n },\n includeThought,\n includeLog,\n });\n\n // YAML is derived from validated actions outside the model-response retry scope.\n // dumpActionParam omits runtime-only locatedPixelResult fields.\n const yamlFlow = buildYamlFlowFromPlans(actions, opts.actionSpace);\n\n let shouldContinuePlanning = true;\n\n // Check if task is completed via <complete> tag\n if (planFromAI.finalizeSuccess !== undefined) {\n debug('task completed via <complete> tag, stop planning');\n shouldContinuePlanning = false;\n // Mark all sub-goals as finished when goal is completed in planning deep-think mode.\n if (includeSubGoals) {\n conversationHistory.markAllSubGoalsFinished();\n }\n }\n\n const returnValue: PlanningAIResponse = {\n ...planFromAI,\n actions,\n rawResponse,\n rawChoiceMessage,\n usage,\n reasoning_content,\n yamlFlow,\n shouldContinuePlanning,\n };\n\n assert(planFromAI, \"can't get plans from AI\");\n\n // TODO: The plan log is recorded before its action has executed, so a failed\n // action may still appear in the next round as an action already performed.\n // Move this write to the successful action execution path in TaskExecutor.action.\n // Update sub-goals in conversation history only in planning deep-think mode.\n if (includeSubGoals) {\n if (planFromAI.updateSubGoals?.length) {\n conversationHistory.mergeSubGoals(planFromAI.updateSubGoals);\n }\n if (planFromAI.markFinishedIndexes?.length) {\n for (const index of planFromAI.markFinishedIndexes) {\n conversationHistory.markSubGoalFinished(index);\n }\n }\n // Append the planning log to the currently running sub-goal\n if (planFromAI.log) {\n conversationHistory.appendSubGoalLog(planFromAI.log);\n }\n } else {\n // Without planning deep-think mode, accumulate logs as historical execution steps.\n if (planFromAI.log) {\n conversationHistory.appendHistoricalLog(planFromAI.log);\n }\n }\n\n // Append memory to conversation history if present\n if (planFromAI.memory) {\n conversationHistory.appendMemory(planFromAI.memory);\n }\n\n // Some model providers require opaque assistant fields to be replayed\n // verbatim in later turns. Keep this opt-in per model adapter so that an\n // unverified provider does not receive non-standard response fields.\n if (\n modelRuntime.adapter.chatCompletion.replayRawAssistantMessage &&\n rawChoiceMessage\n ) {\n conversationHistory.append(rawChoiceMessage as ChatCompletionMessageParam);\n } else {\n conversationHistory.append({\n role: 'assistant',\n content: [\n {\n type: 'text',\n text: rawResponse,\n },\n ],\n });\n }\n\n return returnValue;\n}\n"],"names":["debug","getDebug","warnLog","noPreviousActionsText","callAndParsePlanningResponse","options","messages","modelRuntime","abortSignal","includeLocateInPlanning","actionSpace","locateResultCodec","locateResultContext","includeThought","includeLog","assert","actionOutputProtocol","callAiAndParseWithRetry","retryAttempt","previousParseError","callAI","withSemanticRetryFeedback","response","planFromAI","parseStandardPlanningResponse","undefined","Error","actions","parsePlanningActions","parseError","errorMessage","String","AIResponseParseError","standardPlan","userInstruction","opts","context","conversationHistory","adapter","shotSize","screenshotBase64","planningProtocol","planningModelFamilyRequiredForLocateMessage","includeSubGoals","systemPrompt","buildStandardPlanningSystemPrompt","preparedImage","prepareModelImage","imagePayload","actionContext","referenceImageMessages","preparedReferenceImagesToChatMessages","instruction","latestFeedbackMessage","executionProgressText","executionProgressSection","memoriesText","memoriesSection","historyLog","msgs","rawResponse","usage","reasoning_content","rawChoiceMessage","yamlFlow","buildYamlFlowFromPlans","shouldContinuePlanning","returnValue","index"],"mappings":";;;;;;;;;;;AA6BA,MAAMA,QAAQC,SAAS;AACvB,MAAMC,UAAUD,SAAS,YAAY;IAAE,SAAS;AAAK;AAErD,MAAME,wBACJ;AAeF,eAAeC,6BACbC,OAA4C;IAM5C,MAAM,EACJC,QAAQ,EACRC,YAAY,EACZC,WAAW,EACXC,uBAAuB,EACvBC,WAAW,EACXC,iBAAiB,EACjBC,mBAAmB,EACnBC,cAAc,EACdC,UAAU,EACX,GAAGT;IACJU,OACER,AAAuC,eAAvCA,aAAa,OAAO,CAAC,QAAQ,CAAC,IAAI,EAClC;IAEF,MAAMS,uBACJT,aAAa,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAC,oBAAoB;IAE7D,OAAOU,wBAAwB;QAC7B,QAAQ,CAACC,cAAcC,qBACrBC,OACEC,0BAA0Bf,UAAUa,qBACpCZ,cACA;gBACEC;gBACA,6BAA6BC;gBAC7B,sBAAsBS;YACxB;QAEJ,eAAe,CAACI;YACd,MAAMC,aAAaC,8BAA8BF,SAAS,OAAO,EAAE;gBACjET;gBACAG;gBACAN;gBACA,WAAWI,aAAa,UAAU;YACpC;YACA,IACE,CAACS,WAAW,MAAM,IAClBA,AAA+BE,WAA/BF,WAAW,eAAe,IAC1B,CAACA,WAAW,KAAK,EAEjB,MAAM,IAAIG,MACR;YAGJ,IAAIH,WAAW,MAAM,IAAIA,AAA+BE,WAA/BF,WAAW,eAAe,EAAgB;gBACjErB,QACE;gBAEFqB,WAAW,eAAe,GAAGE;gBAC7BF,WAAW,eAAe,GAAGE;YAC/B;YAEA,MAAME,UAAUJ,WAAW,MAAM,GAAG;gBAACA,WAAW,MAAM;aAAC,GAAG,EAAE;YAC5DK,qBAAqBD,SAAS;gBAC5B,yBAAyBX,qBAAqB,uBAAuB;gBACrEN;gBACAD;gBACAE;gBACAC;gBACA,mBAAmBL,aAAa,OAAO,CAAC,iBAAiB;YAC3D;YACA,OAAO;gBAAEe;gBAAUC;gBAAYI;YAAQ;QACzC;QACA,cAAc,CAACE,YAAYP;YACzB,MAAMQ,eACJD,sBAAsBH,QAAQG,WAAW,OAAO,GAAGE,OAAOF;YAC5D,OAAO,IAAIG,qBACT,CAAC,iBAAiB,EAAEF,cAAc,EAClCR,SAAS,OAAO,EAChBA,SAAS,KAAK,EACdA,SAAS,gBAAgB,EACzBA,SAAS,iBAAiB;QAE9B;QACA,iBAAiBf,aAAa,MAAM,CAAC,UAAU;QAC/C,oBAAoBA,aAAa,MAAM,CAAC,aAAa;QACrDC;QACA,cAAc,CAACqB;YACb7B,MACE,mDACA6B,sBAAsBH,QAAQG,WAAW,OAAO,GAAGE,OAAOF;QAE9D;IACF;AACF;AAEO,eAAeI,aACpBC,eAAmC,EACnCC,IAAiB;IAEjB,MAAM,EAAEC,OAAO,EAAEC,mBAAmB,EAAE,GAAGF;IACzC,MAAM5B,eAAe4B,KAAK,YAAY;IACtC,MAAM,EAAEG,OAAO,EAAE,GAAG/B;IACpB,MAAM,EAAEgC,QAAQ,EAAE,GAAGH;IACrB,MAAMI,mBAAmBJ,QAAQ,UAAU,CAAC,MAAM;IAClDrB,OACEuB,AAA0B,eAA1BA,QAAQ,QAAQ,CAAC,IAAI,EACrB;IAEF,MAAMG,mBAAmBH,QAAQ,QAAQ,CAAC,QAAQ;IAElD,IAAIH,KAAK,uBAAuB,IAAI,CAAC5B,aAAa,MAAM,CAAC,WAAW,EAClE,MAAM,IAAImB,MACRgB,4CAA4CnC,aAAa,MAAM,CAAC,IAAI;IAIxE,MAAMI,oBAAoBJ,aAAa,MAAM,CAAC,WAAW,GACrD+B,QAAQ,QAAQ,CAAC,iBAAiB,GAClCb;IAGJ,MAAMkB,kBAAkBR,AAAgB,gBAAhBA,KAAK,MAAM;IACnC,MAAMtB,iBAAiBsB,AAAgB,WAAhBA,KAAK,MAAM;IAClC,MAAMrB,aAAaqB,AAAgB,WAAhBA,KAAK,MAAM;IAE9B,IAAIA,KAAK,uBAAuB,IAAI,CAACxB,mBACnC,MAAM,IAAIe,MACRgB,4CAA4CnC,aAAa,MAAM,CAAC,IAAI;IAIxE,MAAMqC,eAAe,MAAMC,kCAAkC;QAC3D,aAAaV,KAAK,WAAW;QAC7BtB;QACAC;QACA6B;QACAF;QACA,GAAIN,KAAK,uBAAuB,IAAIxB,oBAChC;YACE,yBAAyB;YACzB,kBAAkBA,kBAAkB,UAAU;QAChD,IACA;YAAE,yBAAyB;QAAM,CAAC;IACxC;IAEA,MAAMmC,gBAAgB,MAAMC,kBAAkB;QAC5C,aAAaP;QACb,OAAOD,SAAS,KAAK;QACrB,QAAQA,SAAS,MAAM;QACvB,QAAQD,QAAQ,eAAe;IACjC;IACA,MAAMU,eAAeF,cAAc,WAAW;IAE9C,MAAMG,gBAAgBd,KAAK,aAAa,GAAG,GAAGA,KAAK,aAAa,CAAC,EAAE,CAAC,GAAG;IAEvE,MAAMe,yBAAyBC,sCAC7BjB,gBAAgB,eAAe;IAEjC,MAAMkB,cAA4C;QAChD;YACE,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,MAAM,GAAGH,cAAc,kBAAkB,EAAEf,gBAAgB,IAAI,CAAC,mBAAmB,CAAC;gBACtF;aACD;QACH;WACGgB;KACJ;IAED,IAAIG;IAKJ,MAAMC,wBAAwBX,kBAC1BN,oBAAoB,cAAc,KAClCA,oBAAoB,oBAAoB;IAC5C,MAAMkB,2BAA2BD,wBAC7B,CAAC,IAAI,EAAEA,uBAAuB,GAC9BjB,oBAAoB,sBAAsB,GACxC,KACA,CAAC,IAAI,EAAElC,uBAAuB;IAGpC,MAAMqD,eAAenB,oBAAoB,cAAc;IACvD,MAAMoB,kBAAkBD,eAAe,CAAC,IAAI,EAAEA,cAAc,GAAG;IAE/D,IAAInB,oBAAoB,sBAAsB,EAAE;QAC9CgB,wBAAwB;YACtB,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,MAAM,GAAGhB,oBAAoB,sBAAsB,CAAC,qHAAqH,EAAEoB,kBAAkBF,0BAA0B;gBACzN;gBACA;oBACE,MAAM;oBACN,WAAW;wBACT,KAAKP;wBACL,QAAQ;oBACV;gBACF;aACD;QACH;QAEAX,oBAAoB,mCAAmC;IACzD,OACEgB,wBAAwB;QACtB,MAAM;QACN,SAAS;YACP;gBACE,MAAM;gBACN,MAAM,CAAC,+BAA+B,EAAEI,kBAAkBF,0BAA0B;YACtF;YACA;gBACE,MAAM;gBACN,WAAW;oBACT,KAAKP;oBACL,QAAQ;gBACV;YACF;SACD;IACH;IAEFX,oBAAoB,MAAM,CAACgB;IAG3BhB,oBAAoB,eAAe,CAAC,IAAI;IAExC,MAAMqB,aAAarB,oBAAoB,QAAQ,CAACF,KAAK,kBAAkB;IAEvE,MAAMwB,OAAqC;QACzC;YAAE,MAAM;YAAU,SAASf;QAAa;WACrCQ;WACAM;KACJ;IAED,MAAM,EACJ,UAAU,EACR,SAASE,WAAW,EACpBC,KAAK,EACLC,iBAAiB,EACjBC,gBAAgB,EACjB,EACDxC,UAAU,EACVI,OAAO,EACR,GAAG,MAAMvB,6BAA6B;QACrC,UAAUuD;QACVpD;QACA,aAAa4B,KAAK,WAAW;QAC7B,yBAAyBA,KAAK,uBAAuB;QACrD,aAAaA,KAAK,WAAW;QAC7BxB;QACA,qBAAqB;YACnB,cAAcmC,cAAc,YAAY;YACxC,aAAaA,cAAc,WAAW;QACxC;QACAjC;QACAC;IACF;IAIA,MAAMkD,WAAWC,uBAAuBtC,SAASQ,KAAK,WAAW;IAEjE,IAAI+B,yBAAyB;IAG7B,IAAI3C,AAA+BE,WAA/BF,WAAW,eAAe,EAAgB;QAC5CvB,MAAM;QACNkE,yBAAyB;QAEzB,IAAIvB,iBACFN,oBAAoB,uBAAuB;IAE/C;IAEA,MAAM8B,cAAkC;QACtC,GAAG5C,UAAU;QACbI;QACAiC;QACAG;QACAF;QACAC;QACAE;QACAE;IACF;IAEAnD,OAAOQ,YAAY;IAMnB,IAAIoB,iBAAiB;QACnB,IAAIpB,WAAW,cAAc,EAAE,QAC7Bc,oBAAoB,aAAa,CAACd,WAAW,cAAc;QAE7D,IAAIA,WAAW,mBAAmB,EAAE,QAClC,KAAK,MAAM6C,SAAS7C,WAAW,mBAAmB,CAChDc,oBAAoB,mBAAmB,CAAC+B;QAI5C,IAAI7C,WAAW,GAAG,EAChBc,oBAAoB,gBAAgB,CAACd,WAAW,GAAG;IAEvD,OAEE,IAAIA,WAAW,GAAG,EAChBc,oBAAoB,mBAAmB,CAACd,WAAW,GAAG;IAK1D,IAAIA,WAAW,MAAM,EACnBc,oBAAoB,YAAY,CAACd,WAAW,MAAM;IAMpD,IACEhB,aAAa,OAAO,CAAC,cAAc,CAAC,yBAAyB,IAC7DwD,kBAEA1B,oBAAoB,MAAM,CAAC0B;SAE3B1B,oBAAoB,MAAM,CAAC;QACzB,MAAM;QACN,SAAS;YACP;gBACE,MAAM;gBACN,MAAMuB;YACR;SACD;IACH;IAGF,OAAOO;AACT"}