@midscene/core 1.0.1-beta-20251208031823.0 → 1.0.1-beta-20251208033501.0

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 (62) hide show
  1. package/dist/es/agent/task-builder.mjs +23 -26
  2. package/dist/es/agent/task-builder.mjs.map +1 -1
  3. package/dist/es/agent/tasks.mjs +1 -4
  4. package/dist/es/agent/tasks.mjs.map +1 -1
  5. package/dist/es/agent/utils.mjs +1 -1
  6. package/dist/es/ai-model/prompt/llm-planning.mjs +95 -20
  7. package/dist/es/ai-model/prompt/llm-planning.mjs.map +1 -1
  8. package/dist/es/ai-model/service-caller/index.mjs +39 -54
  9. package/dist/es/ai-model/service-caller/index.mjs.map +1 -1
  10. package/dist/es/utils.mjs +2 -2
  11. package/dist/lib/agent/agent.js +2 -2
  12. package/dist/lib/agent/common.js +1 -1
  13. package/dist/lib/agent/execution-session.js +2 -2
  14. package/dist/lib/agent/index.js +2 -2
  15. package/dist/lib/agent/task-builder.js +25 -28
  16. package/dist/lib/agent/task-builder.js.map +1 -1
  17. package/dist/lib/agent/task-cache.js +2 -2
  18. package/dist/lib/agent/tasks.js +3 -6
  19. package/dist/lib/agent/tasks.js.map +1 -1
  20. package/dist/lib/agent/ui-utils.js +2 -2
  21. package/dist/lib/agent/utils.js +3 -3
  22. package/dist/lib/ai-model/conversation-history.js +2 -2
  23. package/dist/lib/ai-model/index.js +2 -2
  24. package/dist/lib/ai-model/inspect.js +2 -2
  25. package/dist/lib/ai-model/llm-planning.js +2 -2
  26. package/dist/lib/ai-model/prompt/assertion.js +2 -2
  27. package/dist/lib/ai-model/prompt/common.js +2 -2
  28. package/dist/lib/ai-model/prompt/describe.js +2 -2
  29. package/dist/lib/ai-model/prompt/extraction.js +2 -2
  30. package/dist/lib/ai-model/prompt/llm-locator.js +2 -2
  31. package/dist/lib/ai-model/prompt/llm-planning.js +100 -22
  32. package/dist/lib/ai-model/prompt/llm-planning.js.map +1 -1
  33. package/dist/lib/ai-model/prompt/llm-section-locator.js +2 -2
  34. package/dist/lib/ai-model/prompt/order-sensitive-judge.js +2 -2
  35. package/dist/lib/ai-model/prompt/playwright-generator.js +2 -2
  36. package/dist/lib/ai-model/prompt/ui-tars-locator.js +2 -2
  37. package/dist/lib/ai-model/prompt/ui-tars-planning.js +2 -2
  38. package/dist/lib/ai-model/prompt/util.js +2 -2
  39. package/dist/lib/ai-model/prompt/yaml-generator.js +2 -2
  40. package/dist/lib/ai-model/service-caller/index.js +43 -55
  41. package/dist/lib/ai-model/service-caller/index.js.map +1 -1
  42. package/dist/lib/ai-model/ui-tars-planning.js +2 -2
  43. package/dist/lib/common.js +2 -2
  44. package/dist/lib/device/device-options.js +1 -1
  45. package/dist/lib/device/index.js +2 -2
  46. package/dist/lib/image/index.js +2 -2
  47. package/dist/lib/index.js +2 -2
  48. package/dist/lib/report.js +2 -2
  49. package/dist/lib/service/index.js +2 -2
  50. package/dist/lib/service/utils.js +2 -2
  51. package/dist/lib/task-runner.js +2 -2
  52. package/dist/lib/tree.js +2 -2
  53. package/dist/lib/types.js +3 -3
  54. package/dist/lib/utils.js +4 -4
  55. package/dist/lib/yaml/builder.js +2 -2
  56. package/dist/lib/yaml/index.js +4 -4
  57. package/dist/lib/yaml/player.js +2 -2
  58. package/dist/lib/yaml/utils.js +2 -2
  59. package/dist/lib/yaml.js +1 -1
  60. package/dist/types/ai-model/prompt/llm-planning.d.ts +2 -0
  61. package/dist/types/ai-model/service-caller/index.d.ts +3 -1
  62. package/package.json +5 -5
@@ -16,9 +16,6 @@ function _define_property(obj, key, value) {
16
16
  return obj;
17
17
  }
18
18
  const debug = getDebug('agent:task-builder');
19
- function hasNonEmptyCache(cache) {
20
- return null != cache && 'object' == typeof cache && Object.keys(cache).length > 0;
21
- }
22
19
  function locatePlanForLocate(param) {
23
20
  const locate = 'string' == typeof param ? {
24
21
  prompt: param
@@ -106,14 +103,19 @@ class TaskBuilder {
106
103
  const locateFields = action ? findAllMidsceneLocatorField(action.paramSchema) : [];
107
104
  const requiredLocateFields = action ? findAllMidsceneLocatorField(action.paramSchema, true) : [];
108
105
  locateFields.forEach((field)=>{
109
- if (param[field]) {
106
+ if (param[field]) if (ifPlanLocateParamIsBbox(param[field])) {
107
+ debug('plan locate param is bbox, will match element from plan', param[field]);
108
+ const elementFromPlan = matchElementFromPlan(param[field]);
109
+ if (elementFromPlan) param[field] = elementFromPlan;
110
+ } else {
110
111
  const locatePlan = locatePlanForLocate(param[field]);
111
- debug('will prepend locate param for field', `action.type=${planType}`, `param=${JSON.stringify(param[field])}`, `locatePlan=${JSON.stringify(locatePlan)}`, `hasBbox=${ifPlanLocateParamIsBbox(param[field])}`);
112
+ debug('will prepend locate param for field', `action.type=${planType}`, `param=${JSON.stringify(param[field])}`, `locatePlan=${JSON.stringify(locatePlan)}`);
112
113
  const locateTask = this.createLocateTask(locatePlan, param[field], context, (result)=>{
113
114
  param[field] = result;
114
115
  });
115
116
  context.tasks.push(locateTask);
116
- } else {
117
+ }
118
+ else {
117
119
  assert(!requiredLocateFields.includes(field), `Required locate field '${field}' is not provided for action ${planType}`);
118
120
  debug(`field '${field}' is not provided for action ${planType}`);
119
121
  }
@@ -183,6 +185,7 @@ class TaskBuilder {
183
185
  createLocateTask(plan, detailedLocateParam, context, onResult) {
184
186
  const { cacheable, modelConfigForDefaultIntent } = context;
185
187
  let locateParam = detailedLocateParam;
188
+ debug('modelConfigForDefaultIntent-locateParam', modelConfigForDefaultIntent, locateParam);
186
189
  if ('string' == typeof locateParam) locateParam = {
187
190
  prompt: locateParam
188
191
  };
@@ -213,31 +216,32 @@ class TaskBuilder {
213
216
  task.usage = dump.taskInfo?.usage;
214
217
  if (dump.taskInfo?.searchAreaUsage) task.searchAreaUsage = dump.taskInfo.searchAreaUsage;
215
218
  };
216
- const elementFromBbox = ifPlanLocateParamIsBbox(param) ? matchElementFromPlan(param) : void 0;
217
- const isPlanHit = !!elementFromBbox;
218
219
  let rectFromXpath;
219
- if (!isPlanHit && param.xpath && this.interface.rectMatchesCacheFeature) try {
220
+ if (param.xpath && this.interface.rectMatchesCacheFeature) try {
220
221
  rectFromXpath = await this.interface.rectMatchesCacheFeature({
221
222
  xpaths: [
222
223
  param.xpath
223
224
  ]
224
225
  });
225
- } catch {}
226
+ } catch (error) {
227
+ rectFromXpath = void 0;
228
+ }
226
229
  const elementFromXpath = rectFromXpath ? generateElementByPosition({
227
230
  x: rectFromXpath.left + rectFromXpath.width / 2,
228
231
  y: rectFromXpath.top + rectFromXpath.height / 2
229
232
  }, 'string' == typeof param.prompt ? param.prompt : param.prompt?.prompt || '') : void 0;
230
- const isXpathHit = !!elementFromXpath;
233
+ const userExpectedPathHitFlag = !!elementFromXpath;
231
234
  const cachePrompt = param.prompt;
232
235
  const locateCacheRecord = this.taskCache?.matchLocateCache(cachePrompt);
233
236
  const cacheEntry = locateCacheRecord?.cacheContent?.cache;
234
- const elementFromCache = isPlanHit || isXpathHit ? null : await matchElementFromCache({
237
+ const elementFromCache = userExpectedPathHitFlag ? null : await matchElementFromCache({
235
238
  taskCache: this.taskCache,
236
239
  interfaceInstance: this.interface
237
240
  }, cacheEntry, cachePrompt, param.cacheable);
238
- const isCacheHit = !!elementFromCache;
241
+ const cacheHitFlag = !!elementFromCache;
242
+ const planHitFlag = false;
239
243
  let elementFromAiLocate;
240
- if (!isXpathHit && !isCacheHit && !isPlanHit) try {
244
+ if (!userExpectedPathHitFlag && !cacheHitFlag && !planHitFlag) try {
241
245
  locateResult = await this.service.locate(param, {
242
246
  context: uiContext
243
247
  }, modelConfigForDefaultIntent);
@@ -247,15 +251,14 @@ class TaskBuilder {
247
251
  if (error instanceof ServiceError) applyDump(error.dump);
248
252
  throw error;
249
253
  }
250
- const element = elementFromBbox || elementFromXpath || elementFromCache || elementFromAiLocate;
251
- const locateCacheAlreadyExists = hasNonEmptyCache(locateCacheRecord?.cacheContent?.cache);
254
+ const element = elementFromXpath || elementFromCache || elementFromAiLocate;
252
255
  let currentCacheEntry;
253
- if (element && this.taskCache && !isCacheHit && !locateCacheAlreadyExists && param?.cacheable !== false) if (this.interface.cacheFeatureForRect) try {
256
+ if (element && this.taskCache && !cacheHitFlag && param?.cacheable !== false) if (this.interface.cacheFeatureForRect) try {
254
257
  const feature = await this.interface.cacheFeatureForRect(element.rect, {
255
258
  targetDescription: 'string' == typeof param.prompt ? param.prompt : param.prompt?.prompt,
256
259
  modelConfig: modelConfigForDefaultIntent
257
260
  });
258
- if (hasNonEmptyCache(feature)) {
261
+ if (feature && Object.keys(feature).length > 0) {
259
262
  debug('update cache, prompt: %s, cache: %o', cachePrompt, feature);
260
263
  currentCacheEntry = feature;
261
264
  this.taskCache.updateOrAppendCacheRecord({
@@ -273,19 +276,13 @@ class TaskBuilder {
273
276
  throw new Error(`Element not found: ${param.prompt}`);
274
277
  }
275
278
  let hitBy;
276
- if (isPlanHit) hitBy = {
277
- from: 'Plan',
278
- context: {
279
- bbox: param.bbox
280
- }
281
- };
282
- else if (isXpathHit) hitBy = {
279
+ if (userExpectedPathHitFlag) hitBy = {
283
280
  from: 'User expected path',
284
281
  context: {
285
282
  xpath: param.xpath
286
283
  }
287
284
  };
288
- else if (isCacheHit) hitBy = {
285
+ else if (cacheHitFlag) hitBy = {
289
286
  from: 'Cache',
290
287
  context: {
291
288
  cacheEntry,
@@ -1 +1 @@
1
- {"version":3,"file":"agent/task-builder.mjs","sources":["../../../src/agent/task-builder.ts"],"sourcesContent":["import { findAllMidsceneLocatorField, parseActionParam } from '@/ai-model';\nimport type { AbstractInterface } from '@/device';\nimport type Service from '@/service';\nimport type {\n DetailedLocateParam,\n ElementCacheFeature,\n ExecutionTaskActionApply,\n ExecutionTaskApply,\n ExecutionTaskHitBy,\n ExecutionTaskPlanningLocateApply,\n LocateResultElement,\n LocateResultWithDump,\n PlanningAction,\n PlanningActionParamSleep,\n PlanningLocateParam,\n Rect,\n ServiceDump,\n} from '@/types';\nimport { ServiceError } from '@/types';\nimport { sleep } from '@/utils';\nimport type { IModelConfig } from '@midscene/shared/env';\nimport { generateElementByPosition } from '@midscene/shared/extractor';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type { TaskCache } from './task-cache';\nimport {\n ifPlanLocateParamIsBbox,\n matchElementFromCache,\n matchElementFromPlan,\n} from './utils';\n\nconst debug = getDebug('agent:task-builder');\n\n/**\n * Check if a cache object is non-empty\n */\nfunction hasNonEmptyCache(cache: unknown): boolean {\n return (\n cache !== null &&\n cache !== undefined &&\n typeof cache === 'object' &&\n Object.keys(cache).length > 0\n );\n}\n\nexport function locatePlanForLocate(param: string | DetailedLocateParam) {\n const locate = typeof param === 'string' ? { prompt: param } : param;\n const locatePlan: PlanningAction<PlanningLocateParam> = {\n type: 'Locate',\n param: locate,\n thought: '',\n };\n return locatePlan;\n}\n\ninterface TaskBuilderDeps {\n interfaceInstance: AbstractInterface;\n service: Service;\n taskCache?: TaskCache;\n}\n\ninterface BuildOptions {\n cacheable?: boolean;\n subTask?: boolean;\n}\n\ninterface PlanBuildContext {\n tasks: ExecutionTaskApply[];\n modelConfigForPlanning: IModelConfig;\n modelConfigForDefaultIntent: IModelConfig;\n cacheable?: boolean;\n subTask: boolean;\n}\n\nexport class TaskBuilder {\n private readonly interface: AbstractInterface;\n\n private readonly service: Service;\n\n private readonly taskCache?: TaskCache;\n\n constructor({ interfaceInstance, service, taskCache }: TaskBuilderDeps) {\n this.interface = interfaceInstance;\n this.service = service;\n this.taskCache = taskCache;\n }\n\n public async build(\n plans: PlanningAction[],\n modelConfigForPlanning: IModelConfig,\n modelConfigForDefaultIntent: IModelConfig,\n options?: BuildOptions,\n ): Promise<{ tasks: ExecutionTaskApply[] }> {\n const tasks: ExecutionTaskApply[] = [];\n const cacheable = options?.cacheable;\n\n const context: PlanBuildContext = {\n tasks,\n modelConfigForPlanning,\n modelConfigForDefaultIntent,\n cacheable,\n subTask: !!options?.subTask,\n };\n\n type PlanHandler = (plan: PlanningAction) => Promise<void> | void;\n\n const planHandlers = new Map<string, PlanHandler>([\n [\n 'Locate',\n (plan) =>\n this.handleLocatePlan(\n plan as PlanningAction<PlanningLocateParam>,\n context,\n ),\n ],\n ['Finished', (plan) => this.handleFinishedPlan(plan, context)],\n [\n 'Sleep',\n (plan) =>\n this.handleSleepPlan(\n plan as PlanningAction<PlanningActionParamSleep>,\n context,\n ),\n ],\n ]);\n\n const defaultHandler: PlanHandler = (plan) =>\n this.handleActionPlan(plan, context);\n\n for (const plan of plans) {\n const handler = planHandlers.get(plan.type) ?? defaultHandler;\n await handler(plan);\n }\n\n return {\n tasks,\n };\n }\n\n private handleFinishedPlan(\n plan: PlanningAction,\n context: PlanBuildContext,\n ): void {\n const taskActionFinished: ExecutionTaskActionApply<null> = {\n type: 'Action Space',\n subType: 'Finished',\n param: null,\n thought: plan.thought,\n subTask: context.subTask || undefined,\n executor: async () => {},\n };\n context.tasks.push(taskActionFinished);\n }\n\n private handleSleepPlan(\n plan: PlanningAction<PlanningActionParamSleep>,\n context: PlanBuildContext,\n ): void {\n const sleepTask = this.createSleepTask(plan.param, {\n thought: plan.thought,\n });\n if (context.subTask) {\n sleepTask.subTask = true;\n }\n context.tasks.push(sleepTask);\n }\n\n public createSleepTask(\n param: PlanningActionParamSleep,\n meta?: { thought?: string },\n ): ExecutionTaskActionApply<PlanningActionParamSleep> {\n return {\n type: 'Action Space',\n subType: 'Sleep',\n param,\n thought: meta?.thought,\n executor: async (taskParam) => {\n await sleep(taskParam?.timeMs || 3000);\n },\n };\n }\n\n private async handleLocatePlan(\n plan: PlanningAction<PlanningLocateParam>,\n context: PlanBuildContext,\n ): Promise<void> {\n const taskLocate = this.createLocateTask(plan, plan.param, context);\n context.tasks.push(taskLocate);\n }\n\n private async handleActionPlan(\n plan: PlanningAction,\n context: PlanBuildContext,\n ): Promise<void> {\n const planType = plan.type;\n const actionSpace = await this.interface.actionSpace();\n const action = actionSpace.find((item) => item.name === planType);\n const param = plan.param;\n\n if (!action) {\n throw new Error(`Action type '${planType}' not found`);\n }\n\n const locateFields = action\n ? findAllMidsceneLocatorField(action.paramSchema)\n : [];\n\n const requiredLocateFields = action\n ? findAllMidsceneLocatorField(action.paramSchema, true)\n : [];\n\n locateFields.forEach((field) => {\n if (param[field]) {\n // Always use createLocateTask for all locate params (including bbox)\n // This ensures cache writing happens even when bbox is available\n const locatePlan = locatePlanForLocate(param[field]);\n debug(\n 'will prepend locate param for field',\n `action.type=${planType}`,\n `param=${JSON.stringify(param[field])}`,\n `locatePlan=${JSON.stringify(locatePlan)}`,\n `hasBbox=${ifPlanLocateParamIsBbox(param[field])}`,\n );\n const locateTask = this.createLocateTask(\n locatePlan,\n param[field],\n context,\n (result) => {\n param[field] = result;\n },\n );\n context.tasks.push(locateTask);\n } else {\n assert(\n !requiredLocateFields.includes(field),\n `Required locate field '${field}' is not provided for action ${planType}`,\n );\n debug(`field '${field}' is not provided for action ${planType}`);\n }\n });\n\n const task: ExecutionTaskApply<\n 'Action Space',\n any,\n { success: boolean; action: string; param: any },\n void\n > = {\n type: 'Action Space',\n subType: planType,\n thought: plan.thought,\n param: plan.param,\n subTask: context.subTask || undefined,\n executor: async (param, taskContext) => {\n debug(\n 'executing action',\n planType,\n param,\n `taskContext.element.center: ${taskContext.element?.center}`,\n );\n\n const uiContext = taskContext.uiContext;\n assert(uiContext, 'uiContext is required for Action task');\n\n requiredLocateFields.forEach((field) => {\n assert(\n param[field],\n `field '${field}' is required for action ${planType} but not provided. Cannot execute action ${planType}.`,\n );\n });\n\n try {\n await Promise.all([\n (async () => {\n if (this.interface.beforeInvokeAction) {\n debug('will call \"beforeInvokeAction\" for interface');\n await this.interface.beforeInvokeAction(action.name, param);\n debug('called \"beforeInvokeAction\" for interface');\n }\n })(),\n sleep(200),\n ]);\n } catch (originalError: any) {\n const originalMessage =\n originalError?.message || String(originalError);\n throw new Error(\n `error in running beforeInvokeAction for ${action.name}: ${originalMessage}`,\n { cause: originalError },\n );\n }\n\n if (action.paramSchema) {\n try {\n param = parseActionParam(param, action.paramSchema);\n } catch (error: any) {\n throw new Error(\n `Invalid parameters for action ${action.name}: ${error.message}\\nParameters: ${JSON.stringify(param)}`,\n { cause: error },\n );\n }\n }\n\n debug('calling action', action.name);\n const actionFn = action.call.bind(this.interface);\n const actionResult = await actionFn(param, taskContext);\n debug('called action', action.name, 'result:', actionResult);\n\n const delayAfterRunner = action.delayAfterRunner ?? 300;\n if (delayAfterRunner > 0) {\n await sleep(delayAfterRunner);\n }\n\n try {\n if (this.interface.afterInvokeAction) {\n debug('will call \"afterInvokeAction\" for interface');\n await this.interface.afterInvokeAction(action.name, param);\n debug('called \"afterInvokeAction\" for interface');\n }\n } catch (originalError: any) {\n const originalMessage =\n originalError?.message || String(originalError);\n throw new Error(\n `error in running afterInvokeAction for ${action.name}: ${originalMessage}`,\n { cause: originalError },\n );\n }\n\n return {\n output: actionResult,\n };\n },\n };\n\n context.tasks.push(task);\n }\n\n private createLocateTask(\n plan: PlanningAction<PlanningLocateParam>,\n detailedLocateParam: DetailedLocateParam | string,\n context: PlanBuildContext,\n onResult?: (result: LocateResultElement) => void,\n ): ExecutionTaskPlanningLocateApply {\n const { cacheable, modelConfigForDefaultIntent } = context;\n\n let locateParam = detailedLocateParam;\n\n if (typeof locateParam === 'string') {\n locateParam = {\n prompt: locateParam,\n };\n }\n\n if (cacheable !== undefined) {\n locateParam = {\n ...locateParam,\n cacheable,\n };\n }\n\n const taskLocator: ExecutionTaskPlanningLocateApply = {\n type: 'Planning',\n subType: 'Locate',\n subTask: context.subTask || undefined,\n param: locateParam,\n thought: plan.thought,\n executor: async (param, taskContext) => {\n const { task } = taskContext;\n let { uiContext } = taskContext;\n\n assert(\n param?.prompt || param?.bbox,\n `No prompt or id or position or bbox to locate, param=${JSON.stringify(\n param,\n )}`,\n );\n\n if (!uiContext) {\n uiContext = await this.service.contextRetrieverFn();\n }\n\n assert(uiContext, 'uiContext is required for Service task');\n\n let locateDump: ServiceDump | undefined;\n let locateResult: LocateResultWithDump | undefined;\n\n const applyDump = (dump?: ServiceDump) => {\n if (!dump) {\n return;\n }\n locateDump = dump;\n task.log = {\n dump,\n };\n task.usage = dump.taskInfo?.usage;\n if (dump.taskInfo?.searchAreaUsage) {\n task.searchAreaUsage = dump.taskInfo.searchAreaUsage;\n }\n };\n\n // from bbox (plan hit)\n const elementFromBbox = ifPlanLocateParamIsBbox(param)\n ? matchElementFromPlan(param)\n : undefined;\n const isPlanHit = !!elementFromBbox;\n\n // from xpath\n let rectFromXpath: Rect | undefined;\n if (\n !isPlanHit &&\n param.xpath &&\n this.interface.rectMatchesCacheFeature\n ) {\n try {\n rectFromXpath = await this.interface.rectMatchesCacheFeature({\n xpaths: [param.xpath],\n });\n } catch {\n // xpath locate failed, allow fallback to cache or AI locate\n }\n }\n const elementFromXpath = rectFromXpath\n ? generateElementByPosition(\n {\n x: rectFromXpath.left + rectFromXpath.width / 2,\n y: rectFromXpath.top + rectFromXpath.height / 2,\n },\n typeof param.prompt === 'string'\n ? param.prompt\n : param.prompt?.prompt || '',\n )\n : undefined;\n const isXpathHit = !!elementFromXpath;\n\n const cachePrompt = param.prompt;\n const locateCacheRecord = this.taskCache?.matchLocateCache(cachePrompt);\n const cacheEntry = locateCacheRecord?.cacheContent?.cache;\n\n const elementFromCache =\n isPlanHit || isXpathHit\n ? null\n : await matchElementFromCache(\n {\n taskCache: this.taskCache,\n interfaceInstance: this.interface,\n },\n cacheEntry,\n cachePrompt,\n param.cacheable,\n );\n const isCacheHit = !!elementFromCache;\n\n let elementFromAiLocate: LocateResultElement | null | undefined;\n if (!isXpathHit && !isCacheHit && !isPlanHit) {\n try {\n locateResult = await this.service.locate(\n param,\n {\n context: uiContext,\n },\n modelConfigForDefaultIntent,\n );\n applyDump(locateResult.dump);\n elementFromAiLocate = locateResult.element;\n } catch (error) {\n if (error instanceof ServiceError) {\n applyDump(error.dump);\n }\n throw error;\n }\n }\n\n const element =\n elementFromBbox ||\n elementFromXpath ||\n elementFromCache ||\n elementFromAiLocate;\n\n // Check if locate cache already exists (for planHitFlag case)\n const locateCacheAlreadyExists = hasNonEmptyCache(\n locateCacheRecord?.cacheContent?.cache,\n );\n\n let currentCacheEntry: ElementCacheFeature | undefined;\n // Write cache if:\n // 1. element found\n // 2. taskCache enabled\n // 3. not a cache hit (otherwise we'd be writing what we just read)\n // 4. not already cached (for bbox/plan hit case, avoid redundant writes)\n // 5. cacheable is not explicitly false\n if (\n element &&\n this.taskCache &&\n !isCacheHit &&\n !locateCacheAlreadyExists &&\n param?.cacheable !== false\n ) {\n if (this.interface.cacheFeatureForRect) {\n try {\n const feature = await this.interface.cacheFeatureForRect(\n element.rect,\n {\n targetDescription:\n typeof param.prompt === 'string'\n ? param.prompt\n : param.prompt?.prompt,\n modelConfig: modelConfigForDefaultIntent,\n },\n );\n if (hasNonEmptyCache(feature)) {\n debug(\n 'update cache, prompt: %s, cache: %o',\n cachePrompt,\n feature,\n );\n currentCacheEntry = feature;\n this.taskCache.updateOrAppendCacheRecord(\n {\n type: 'locate',\n prompt: cachePrompt,\n cache: feature,\n },\n locateCacheRecord,\n );\n } else {\n debug(\n 'no cache data returned, skip cache update, prompt: %s',\n cachePrompt,\n );\n }\n } catch (error) {\n debug('cacheFeatureForRect failed: %s', error);\n }\n } else {\n debug('cacheFeatureForRect is not supported, skip cache update');\n }\n }\n\n if (!element) {\n if (locateDump) {\n throw new ServiceError(\n `Element not found : ${param.prompt}`,\n locateDump,\n );\n }\n throw new Error(`Element not found: ${param.prompt}`);\n }\n\n let hitBy: ExecutionTaskHitBy | undefined;\n\n if (isPlanHit) {\n hitBy = {\n from: 'Plan',\n context: {\n bbox: param.bbox,\n },\n };\n } else if (isXpathHit) {\n hitBy = {\n from: 'User expected path',\n context: {\n xpath: param.xpath,\n },\n };\n } else if (isCacheHit) {\n hitBy = {\n from: 'Cache',\n context: {\n cacheEntry,\n cacheToSave: currentCacheEntry,\n },\n };\n }\n\n onResult?.(element);\n\n return {\n output: {\n element,\n },\n hitBy,\n };\n },\n };\n\n return taskLocator;\n }\n}\n"],"names":["debug","getDebug","hasNonEmptyCache","cache","Object","locatePlanForLocate","param","locate","locatePlan","TaskBuilder","plans","modelConfigForPlanning","modelConfigForDefaultIntent","options","tasks","cacheable","context","planHandlers","Map","plan","defaultHandler","handler","taskActionFinished","undefined","sleepTask","meta","taskParam","sleep","taskLocate","planType","actionSpace","action","item","Error","locateFields","findAllMidsceneLocatorField","requiredLocateFields","field","JSON","ifPlanLocateParamIsBbox","locateTask","result","assert","task","taskContext","uiContext","Promise","originalError","originalMessage","String","parseActionParam","error","actionFn","actionResult","delayAfterRunner","detailedLocateParam","onResult","locateParam","taskLocator","locateDump","locateResult","applyDump","dump","elementFromBbox","matchElementFromPlan","isPlanHit","rectFromXpath","elementFromXpath","generateElementByPosition","isXpathHit","cachePrompt","locateCacheRecord","cacheEntry","elementFromCache","matchElementFromCache","isCacheHit","elementFromAiLocate","ServiceError","element","locateCacheAlreadyExists","currentCacheEntry","feature","hitBy","interfaceInstance","service","taskCache"],"mappings":";;;;;;;;;;;;;;;;;AA+BA,MAAMA,QAAQC,SAAS;AAKvB,SAASC,iBAAiBC,KAAc;IACtC,OACEA,QAAAA,SAEA,AAAiB,YAAjB,OAAOA,SACPC,OAAO,IAAI,CAACD,OAAO,MAAM,GAAG;AAEhC;AAEO,SAASE,oBAAoBC,KAAmC;IACrE,MAAMC,SAAS,AAAiB,YAAjB,OAAOD,QAAqB;QAAE,QAAQA;IAAM,IAAIA;IAC/D,MAAME,aAAkD;QACtD,MAAM;QACN,OAAOD;QACP,SAAS;IACX;IACA,OAAOC;AACT;AAqBO,MAAMC;IAaX,MAAa,MACXC,KAAuB,EACvBC,sBAAoC,EACpCC,2BAAyC,EACzCC,OAAsB,EACoB;QAC1C,MAAMC,QAA8B,EAAE;QACtC,MAAMC,YAAYF,SAAS;QAE3B,MAAMG,UAA4B;YAChCF;YACAH;YACAC;YACAG;YACA,SAAS,CAAC,CAACF,SAAS;QACtB;QAIA,MAAMI,eAAe,IAAIC,IAAyB;YAChD;gBACE;gBACA,CAACC,OACC,IAAI,CAAC,gBAAgB,CACnBA,MACAH;aAEL;YACD;gBAAC;gBAAY,CAACG,OAAS,IAAI,CAAC,kBAAkB,CAACA,MAAMH;aAAS;YAC9D;gBACE;gBACA,CAACG,OACC,IAAI,CAAC,eAAe,CAClBA,MACAH;aAEL;SACF;QAED,MAAMI,iBAA8B,CAACD,OACnC,IAAI,CAAC,gBAAgB,CAACA,MAAMH;QAE9B,KAAK,MAAMG,QAAQT,MAAO;YACxB,MAAMW,UAAUJ,aAAa,GAAG,CAACE,KAAK,IAAI,KAAKC;YAC/C,MAAMC,QAAQF;QAChB;QAEA,OAAO;YACLL;QACF;IACF;IAEQ,mBACNK,IAAoB,EACpBH,OAAyB,EACnB;QACN,MAAMM,qBAAqD;YACzD,MAAM;YACN,SAAS;YACT,OAAO;YACP,SAASH,KAAK,OAAO;YACrB,SAASH,QAAQ,OAAO,IAAIO;YAC5B,UAAU,WAAa;QACzB;QACAP,QAAQ,KAAK,CAAC,IAAI,CAACM;IACrB;IAEQ,gBACNH,IAA8C,EAC9CH,OAAyB,EACnB;QACN,MAAMQ,YAAY,IAAI,CAAC,eAAe,CAACL,KAAK,KAAK,EAAE;YACjD,SAASA,KAAK,OAAO;QACvB;QACA,IAAIH,QAAQ,OAAO,EACjBQ,UAAU,OAAO,GAAG;QAEtBR,QAAQ,KAAK,CAAC,IAAI,CAACQ;IACrB;IAEO,gBACLlB,KAA+B,EAC/BmB,IAA2B,EACyB;QACpD,OAAO;YACL,MAAM;YACN,SAAS;YACTnB;YACA,SAASmB,MAAM;YACf,UAAU,OAAOC;gBACf,MAAMC,MAAMD,WAAW,UAAU;YACnC;QACF;IACF;IAEA,MAAc,iBACZP,IAAyC,EACzCH,OAAyB,EACV;QACf,MAAMY,aAAa,IAAI,CAAC,gBAAgB,CAACT,MAAMA,KAAK,KAAK,EAAEH;QAC3DA,QAAQ,KAAK,CAAC,IAAI,CAACY;IACrB;IAEA,MAAc,iBACZT,IAAoB,EACpBH,OAAyB,EACV;QACf,MAAMa,WAAWV,KAAK,IAAI;QAC1B,MAAMW,cAAc,MAAM,IAAI,CAAC,SAAS,CAAC,WAAW;QACpD,MAAMC,SAASD,YAAY,IAAI,CAAC,CAACE,OAASA,KAAK,IAAI,KAAKH;QACxD,MAAMvB,QAAQa,KAAK,KAAK;QAExB,IAAI,CAACY,QACH,MAAM,IAAIE,MAAM,CAAC,aAAa,EAAEJ,SAAS,WAAW,CAAC;QAGvD,MAAMK,eAAeH,SACjBI,4BAA4BJ,OAAO,WAAW,IAC9C,EAAE;QAEN,MAAMK,uBAAuBL,SACzBI,4BAA4BJ,OAAO,WAAW,EAAE,QAChD,EAAE;QAENG,aAAa,OAAO,CAAC,CAACG;YACpB,IAAI/B,KAAK,CAAC+B,MAAM,EAAE;gBAGhB,MAAM7B,aAAaH,oBAAoBC,KAAK,CAAC+B,MAAM;gBACnDrC,MACE,uCACA,CAAC,YAAY,EAAE6B,UAAU,EACzB,CAAC,MAAM,EAAES,KAAK,SAAS,CAAChC,KAAK,CAAC+B,MAAM,GAAG,EACvC,CAAC,WAAW,EAAEC,KAAK,SAAS,CAAC9B,aAAa,EAC1C,CAAC,QAAQ,EAAE+B,wBAAwBjC,KAAK,CAAC+B,MAAM,GAAG;gBAEpD,MAAMG,aAAa,IAAI,CAAC,gBAAgB,CACtChC,YACAF,KAAK,CAAC+B,MAAM,EACZrB,SACA,CAACyB;oBACCnC,KAAK,CAAC+B,MAAM,GAAGI;gBACjB;gBAEFzB,QAAQ,KAAK,CAAC,IAAI,CAACwB;YACrB,OAAO;gBACLE,OACE,CAACN,qBAAqB,QAAQ,CAACC,QAC/B,CAAC,uBAAuB,EAAEA,MAAM,6BAA6B,EAAER,UAAU;gBAE3E7B,MAAM,CAAC,OAAO,EAAEqC,MAAM,6BAA6B,EAAER,UAAU;YACjE;QACF;QAEA,MAAMc,OAKF;YACF,MAAM;YACN,SAASd;YACT,SAASV,KAAK,OAAO;YACrB,OAAOA,KAAK,KAAK;YACjB,SAASH,QAAQ,OAAO,IAAIO;YAC5B,UAAU,OAAOjB,OAAOsC;gBACtB5C,MACE,oBACA6B,UACAvB,OACA,CAAC,4BAA4B,EAAEsC,YAAY,OAAO,EAAE,QAAQ;gBAG9D,MAAMC,YAAYD,YAAY,SAAS;gBACvCF,OAAOG,WAAW;gBAElBT,qBAAqB,OAAO,CAAC,CAACC;oBAC5BK,OACEpC,KAAK,CAAC+B,MAAM,EACZ,CAAC,OAAO,EAAEA,MAAM,yBAAyB,EAAER,SAAS,yCAAyC,EAAEA,SAAS,CAAC,CAAC;gBAE9G;gBAEA,IAAI;oBACF,MAAMiB,QAAQ,GAAG,CAAC;wBACf;4BACC,IAAI,IAAI,CAAC,SAAS,CAAC,kBAAkB,EAAE;gCACrC9C,MAAM;gCACN,MAAM,IAAI,CAAC,SAAS,CAAC,kBAAkB,CAAC+B,OAAO,IAAI,EAAEzB;gCACrDN,MAAM;4BACR;wBACF;wBACA2B,MAAM;qBACP;gBACH,EAAE,OAAOoB,eAAoB;oBAC3B,MAAMC,kBACJD,eAAe,WAAWE,OAAOF;oBACnC,MAAM,IAAId,MACR,CAAC,wCAAwC,EAAEF,OAAO,IAAI,CAAC,EAAE,EAAEiB,iBAAiB,EAC5E;wBAAE,OAAOD;oBAAc;gBAE3B;gBAEA,IAAIhB,OAAO,WAAW,EACpB,IAAI;oBACFzB,QAAQ4C,iBAAiB5C,OAAOyB,OAAO,WAAW;gBACpD,EAAE,OAAOoB,OAAY;oBACnB,MAAM,IAAIlB,MACR,CAAC,8BAA8B,EAAEF,OAAO,IAAI,CAAC,EAAE,EAAEoB,MAAM,OAAO,CAAC,cAAc,EAAEb,KAAK,SAAS,CAAChC,QAAQ,EACtG;wBAAE,OAAO6C;oBAAM;gBAEnB;gBAGFnD,MAAM,kBAAkB+B,OAAO,IAAI;gBACnC,MAAMqB,WAAWrB,OAAO,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS;gBAChD,MAAMsB,eAAe,MAAMD,SAAS9C,OAAOsC;gBAC3C5C,MAAM,iBAAiB+B,OAAO,IAAI,EAAE,WAAWsB;gBAE/C,MAAMC,mBAAmBvB,OAAO,gBAAgB,IAAI;gBACpD,IAAIuB,mBAAmB,GACrB,MAAM3B,MAAM2B;gBAGd,IAAI;oBACF,IAAI,IAAI,CAAC,SAAS,CAAC,iBAAiB,EAAE;wBACpCtD,MAAM;wBACN,MAAM,IAAI,CAAC,SAAS,CAAC,iBAAiB,CAAC+B,OAAO,IAAI,EAAEzB;wBACpDN,MAAM;oBACR;gBACF,EAAE,OAAO+C,eAAoB;oBAC3B,MAAMC,kBACJD,eAAe,WAAWE,OAAOF;oBACnC,MAAM,IAAId,MACR,CAAC,uCAAuC,EAAEF,OAAO,IAAI,CAAC,EAAE,EAAEiB,iBAAiB,EAC3E;wBAAE,OAAOD;oBAAc;gBAE3B;gBAEA,OAAO;oBACL,QAAQM;gBACV;YACF;QACF;QAEArC,QAAQ,KAAK,CAAC,IAAI,CAAC2B;IACrB;IAEQ,iBACNxB,IAAyC,EACzCoC,mBAAiD,EACjDvC,OAAyB,EACzBwC,QAAgD,EACd;QAClC,MAAM,EAAEzC,SAAS,EAAEH,2BAA2B,EAAE,GAAGI;QAEnD,IAAIyC,cAAcF;QAElB,IAAI,AAAuB,YAAvB,OAAOE,aACTA,cAAc;YACZ,QAAQA;QACV;QAGF,IAAI1C,AAAcQ,WAAdR,WACF0C,cAAc;YACZ,GAAGA,WAAW;YACd1C;QACF;QAGF,MAAM2C,cAAgD;YACpD,MAAM;YACN,SAAS;YACT,SAAS1C,QAAQ,OAAO,IAAIO;YAC5B,OAAOkC;YACP,SAAStC,KAAK,OAAO;YACrB,UAAU,OAAOb,OAAOsC;gBACtB,MAAM,EAAED,IAAI,EAAE,GAAGC;gBACjB,IAAI,EAAEC,SAAS,EAAE,GAAGD;gBAEpBF,OACEpC,OAAO,UAAUA,OAAO,MACxB,CAAC,qDAAqD,EAAEgC,KAAK,SAAS,CACpEhC,QACC;gBAGL,IAAI,CAACuC,WACHA,YAAY,MAAM,IAAI,CAAC,OAAO,CAAC,kBAAkB;gBAGnDH,OAAOG,WAAW;gBAElB,IAAIc;gBACJ,IAAIC;gBAEJ,MAAMC,YAAY,CAACC;oBACjB,IAAI,CAACA,MACH;oBAEFH,aAAaG;oBACbnB,KAAK,GAAG,GAAG;wBACTmB;oBACF;oBACAnB,KAAK,KAAK,GAAGmB,KAAK,QAAQ,EAAE;oBAC5B,IAAIA,KAAK,QAAQ,EAAE,iBACjBnB,KAAK,eAAe,GAAGmB,KAAK,QAAQ,CAAC,eAAe;gBAExD;gBAGA,MAAMC,kBAAkBxB,wBAAwBjC,SAC5C0D,qBAAqB1D,SACrBiB;gBACJ,MAAM0C,YAAY,CAAC,CAACF;gBAGpB,IAAIG;gBACJ,IACE,CAACD,aACD3D,MAAM,KAAK,IACX,IAAI,CAAC,SAAS,CAAC,uBAAuB,EAEtC,IAAI;oBACF4D,gBAAgB,MAAM,IAAI,CAAC,SAAS,CAAC,uBAAuB,CAAC;wBAC3D,QAAQ;4BAAC5D,MAAM,KAAK;yBAAC;oBACvB;gBACF,EAAE,OAAM,CAER;gBAEF,MAAM6D,mBAAmBD,gBACrBE,0BACE;oBACE,GAAGF,cAAc,IAAI,GAAGA,cAAc,KAAK,GAAG;oBAC9C,GAAGA,cAAc,GAAG,GAAGA,cAAc,MAAM,GAAG;gBAChD,GACA,AAAwB,YAAxB,OAAO5D,MAAM,MAAM,GACfA,MAAM,MAAM,GACZA,MAAM,MAAM,EAAE,UAAU,MAE9BiB;gBACJ,MAAM8C,aAAa,CAAC,CAACF;gBAErB,MAAMG,cAAchE,MAAM,MAAM;gBAChC,MAAMiE,oBAAoB,IAAI,CAAC,SAAS,EAAE,iBAAiBD;gBAC3D,MAAME,aAAaD,mBAAmB,cAAc;gBAEpD,MAAME,mBACJR,aAAaI,aACT,OACA,MAAMK,sBACJ;oBACE,WAAW,IAAI,CAAC,SAAS;oBACzB,mBAAmB,IAAI,CAAC,SAAS;gBACnC,GACAF,YACAF,aACAhE,MAAM,SAAS;gBAEvB,MAAMqE,aAAa,CAAC,CAACF;gBAErB,IAAIG;gBACJ,IAAI,CAACP,cAAc,CAACM,cAAc,CAACV,WACjC,IAAI;oBACFL,eAAe,MAAM,IAAI,CAAC,OAAO,CAAC,MAAM,CACtCtD,OACA;wBACE,SAASuC;oBACX,GACAjC;oBAEFiD,UAAUD,aAAa,IAAI;oBAC3BgB,sBAAsBhB,aAAa,OAAO;gBAC5C,EAAE,OAAOT,OAAO;oBACd,IAAIA,iBAAiB0B,cACnBhB,UAAUV,MAAM,IAAI;oBAEtB,MAAMA;gBACR;gBAGF,MAAM2B,UACJf,mBACAI,oBACAM,oBACAG;gBAGF,MAAMG,2BAA2B7E,iBAC/BqE,mBAAmB,cAAc;gBAGnC,IAAIS;gBAOJ,IACEF,WACA,IAAI,CAAC,SAAS,IACd,CAACH,cACD,CAACI,4BACDzE,OAAO,cAAc,OAErB,IAAI,IAAI,CAAC,SAAS,CAAC,mBAAmB,EACpC,IAAI;oBACF,MAAM2E,UAAU,MAAM,IAAI,CAAC,SAAS,CAAC,mBAAmB,CACtDH,QAAQ,IAAI,EACZ;wBACE,mBACE,AAAwB,YAAxB,OAAOxE,MAAM,MAAM,GACfA,MAAM,MAAM,GACZA,MAAM,MAAM,EAAE;wBACpB,aAAaM;oBACf;oBAEF,IAAIV,iBAAiB+E,UAAU;wBAC7BjF,MACE,uCACAsE,aACAW;wBAEFD,oBAAoBC;wBACpB,IAAI,CAAC,SAAS,CAAC,yBAAyB,CACtC;4BACE,MAAM;4BACN,QAAQX;4BACR,OAAOW;wBACT,GACAV;oBAEJ,OACEvE,MACE,yDACAsE;gBAGN,EAAE,OAAOnB,OAAO;oBACdnD,MAAM,kCAAkCmD;gBAC1C;qBAEAnD,MAAM;gBAIV,IAAI,CAAC8E,SAAS;oBACZ,IAAInB,YACF,MAAM,IAAIkB,aACR,CAAC,oBAAoB,EAAEvE,MAAM,MAAM,EAAE,EACrCqD;oBAGJ,MAAM,IAAI1B,MAAM,CAAC,mBAAmB,EAAE3B,MAAM,MAAM,EAAE;gBACtD;gBAEA,IAAI4E;gBAEJ,IAAIjB,WACFiB,QAAQ;oBACN,MAAM;oBACN,SAAS;wBACP,MAAM5E,MAAM,IAAI;oBAClB;gBACF;qBACK,IAAI+D,YACTa,QAAQ;oBACN,MAAM;oBACN,SAAS;wBACP,OAAO5E,MAAM,KAAK;oBACpB;gBACF;qBACK,IAAIqE,YACTO,QAAQ;oBACN,MAAM;oBACN,SAAS;wBACPV;wBACA,aAAaQ;oBACf;gBACF;gBAGFxB,WAAWsB;gBAEX,OAAO;oBACL,QAAQ;wBACNA;oBACF;oBACAI;gBACF;YACF;QACF;QAEA,OAAOxB;IACT;IAvfA,YAAY,EAAEyB,iBAAiB,EAAEC,OAAO,EAAEC,SAAS,EAAmB,CAAE;QANxE,uBAAiB,aAAjB;QAEA,uBAAiB,WAAjB;QAEA,uBAAiB,aAAjB;QAGE,IAAI,CAAC,SAAS,GAAGF;QACjB,IAAI,CAAC,OAAO,GAAGC;QACf,IAAI,CAAC,SAAS,GAAGC;IACnB;AAofF"}
1
+ {"version":3,"file":"agent/task-builder.mjs","sources":["../../../src/agent/task-builder.ts"],"sourcesContent":["import { findAllMidsceneLocatorField, parseActionParam } from '@/ai-model';\nimport type { AbstractInterface } from '@/device';\nimport type Service from '@/service';\nimport type {\n DetailedLocateParam,\n ElementCacheFeature,\n ExecutionTaskActionApply,\n ExecutionTaskApply,\n ExecutionTaskHitBy,\n ExecutionTaskPlanningLocateApply,\n LocateResultElement,\n LocateResultWithDump,\n PlanningAction,\n PlanningActionParamSleep,\n PlanningLocateParam,\n Rect,\n ServiceDump,\n} from '@/types';\nimport { ServiceError } from '@/types';\nimport { sleep } from '@/utils';\nimport type { IModelConfig } from '@midscene/shared/env';\nimport { generateElementByPosition } from '@midscene/shared/extractor';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type { TaskCache } from './task-cache';\nimport {\n ifPlanLocateParamIsBbox,\n matchElementFromCache,\n matchElementFromPlan,\n} from './utils';\n\nconst debug = getDebug('agent:task-builder');\n\nexport function locatePlanForLocate(param: string | DetailedLocateParam) {\n const locate = typeof param === 'string' ? { prompt: param } : param;\n const locatePlan: PlanningAction<PlanningLocateParam> = {\n type: 'Locate',\n param: locate,\n thought: '',\n };\n return locatePlan;\n}\n\ninterface TaskBuilderDeps {\n interfaceInstance: AbstractInterface;\n service: Service;\n taskCache?: TaskCache;\n}\n\ninterface BuildOptions {\n cacheable?: boolean;\n subTask?: boolean;\n}\n\ninterface PlanBuildContext {\n tasks: ExecutionTaskApply[];\n modelConfigForPlanning: IModelConfig;\n modelConfigForDefaultIntent: IModelConfig;\n cacheable?: boolean;\n subTask: boolean;\n}\n\nexport class TaskBuilder {\n private readonly interface: AbstractInterface;\n\n private readonly service: Service;\n\n private readonly taskCache?: TaskCache;\n\n constructor({ interfaceInstance, service, taskCache }: TaskBuilderDeps) {\n this.interface = interfaceInstance;\n this.service = service;\n this.taskCache = taskCache;\n }\n\n public async build(\n plans: PlanningAction[],\n modelConfigForPlanning: IModelConfig,\n modelConfigForDefaultIntent: IModelConfig,\n options?: BuildOptions,\n ): Promise<{ tasks: ExecutionTaskApply[] }> {\n const tasks: ExecutionTaskApply[] = [];\n const cacheable = options?.cacheable;\n\n const context: PlanBuildContext = {\n tasks,\n modelConfigForPlanning,\n modelConfigForDefaultIntent,\n cacheable,\n subTask: !!options?.subTask,\n };\n\n type PlanHandler = (plan: PlanningAction) => Promise<void> | void;\n\n const planHandlers = new Map<string, PlanHandler>([\n [\n 'Locate',\n (plan) =>\n this.handleLocatePlan(\n plan as PlanningAction<PlanningLocateParam>,\n context,\n ),\n ],\n ['Finished', (plan) => this.handleFinishedPlan(plan, context)],\n [\n 'Sleep',\n (plan) =>\n this.handleSleepPlan(\n plan as PlanningAction<PlanningActionParamSleep>,\n context,\n ),\n ],\n ]);\n\n const defaultHandler: PlanHandler = (plan) =>\n this.handleActionPlan(plan, context);\n\n for (const plan of plans) {\n const handler = planHandlers.get(plan.type) ?? defaultHandler;\n await handler(plan);\n }\n\n return {\n tasks,\n };\n }\n\n private handleFinishedPlan(\n plan: PlanningAction,\n context: PlanBuildContext,\n ): void {\n const taskActionFinished: ExecutionTaskActionApply<null> = {\n type: 'Action Space',\n subType: 'Finished',\n param: null,\n thought: plan.thought,\n subTask: context.subTask || undefined,\n executor: async () => {},\n };\n context.tasks.push(taskActionFinished);\n }\n\n private handleSleepPlan(\n plan: PlanningAction<PlanningActionParamSleep>,\n context: PlanBuildContext,\n ): void {\n const sleepTask = this.createSleepTask(plan.param, {\n thought: plan.thought,\n });\n if (context.subTask) {\n sleepTask.subTask = true;\n }\n context.tasks.push(sleepTask);\n }\n\n public createSleepTask(\n param: PlanningActionParamSleep,\n meta?: { thought?: string },\n ): ExecutionTaskActionApply<PlanningActionParamSleep> {\n return {\n type: 'Action Space',\n subType: 'Sleep',\n param,\n thought: meta?.thought,\n executor: async (taskParam) => {\n await sleep(taskParam?.timeMs || 3000);\n },\n };\n }\n\n private async handleLocatePlan(\n plan: PlanningAction<PlanningLocateParam>,\n context: PlanBuildContext,\n ): Promise<void> {\n const taskLocate = this.createLocateTask(plan, plan.param, context);\n context.tasks.push(taskLocate);\n }\n\n private async handleActionPlan(\n plan: PlanningAction,\n context: PlanBuildContext,\n ): Promise<void> {\n const planType = plan.type;\n const actionSpace = await this.interface.actionSpace();\n const action = actionSpace.find((item) => item.name === planType);\n const param = plan.param;\n\n if (!action) {\n throw new Error(`Action type '${planType}' not found`);\n }\n\n const locateFields = action\n ? findAllMidsceneLocatorField(action.paramSchema)\n : [];\n\n const requiredLocateFields = action\n ? findAllMidsceneLocatorField(action.paramSchema, true)\n : [];\n\n locateFields.forEach((field) => {\n if (param[field]) {\n if (ifPlanLocateParamIsBbox(param[field])) {\n debug(\n 'plan locate param is bbox, will match element from plan',\n param[field],\n );\n const elementFromPlan = matchElementFromPlan(param[field]);\n if (elementFromPlan) {\n param[field] = elementFromPlan;\n }\n } else {\n const locatePlan = locatePlanForLocate(param[field]);\n debug(\n 'will prepend locate param for field',\n `action.type=${planType}`,\n `param=${JSON.stringify(param[field])}`,\n `locatePlan=${JSON.stringify(locatePlan)}`,\n );\n const locateTask = this.createLocateTask(\n locatePlan,\n param[field],\n context,\n (result) => {\n param[field] = result;\n },\n );\n context.tasks.push(locateTask);\n }\n } else {\n assert(\n !requiredLocateFields.includes(field),\n `Required locate field '${field}' is not provided for action ${planType}`,\n );\n debug(`field '${field}' is not provided for action ${planType}`);\n }\n });\n\n const task: ExecutionTaskApply<\n 'Action Space',\n any,\n { success: boolean; action: string; param: any },\n void\n > = {\n type: 'Action Space',\n subType: planType,\n thought: plan.thought,\n param: plan.param,\n subTask: context.subTask || undefined,\n executor: async (param, taskContext) => {\n debug(\n 'executing action',\n planType,\n param,\n `taskContext.element.center: ${taskContext.element?.center}`,\n );\n\n const uiContext = taskContext.uiContext;\n assert(uiContext, 'uiContext is required for Action task');\n\n requiredLocateFields.forEach((field) => {\n assert(\n param[field],\n `field '${field}' is required for action ${planType} but not provided. Cannot execute action ${planType}.`,\n );\n });\n\n try {\n await Promise.all([\n (async () => {\n if (this.interface.beforeInvokeAction) {\n debug('will call \"beforeInvokeAction\" for interface');\n await this.interface.beforeInvokeAction(action.name, param);\n debug('called \"beforeInvokeAction\" for interface');\n }\n })(),\n sleep(200),\n ]);\n } catch (originalError: any) {\n const originalMessage =\n originalError?.message || String(originalError);\n throw new Error(\n `error in running beforeInvokeAction for ${action.name}: ${originalMessage}`,\n { cause: originalError },\n );\n }\n\n if (action.paramSchema) {\n try {\n param = parseActionParam(param, action.paramSchema);\n } catch (error: any) {\n throw new Error(\n `Invalid parameters for action ${action.name}: ${error.message}\\nParameters: ${JSON.stringify(param)}`,\n { cause: error },\n );\n }\n }\n\n debug('calling action', action.name);\n const actionFn = action.call.bind(this.interface);\n const actionResult = await actionFn(param, taskContext);\n debug('called action', action.name, 'result:', actionResult);\n\n const delayAfterRunner = action.delayAfterRunner ?? 300;\n if (delayAfterRunner > 0) {\n await sleep(delayAfterRunner);\n }\n\n try {\n if (this.interface.afterInvokeAction) {\n debug('will call \"afterInvokeAction\" for interface');\n await this.interface.afterInvokeAction(action.name, param);\n debug('called \"afterInvokeAction\" for interface');\n }\n } catch (originalError: any) {\n const originalMessage =\n originalError?.message || String(originalError);\n throw new Error(\n `error in running afterInvokeAction for ${action.name}: ${originalMessage}`,\n { cause: originalError },\n );\n }\n\n return {\n output: actionResult,\n };\n },\n };\n\n context.tasks.push(task);\n }\n\n private createLocateTask(\n plan: PlanningAction<PlanningLocateParam>,\n detailedLocateParam: DetailedLocateParam | string,\n context: PlanBuildContext,\n onResult?: (result: LocateResultElement) => void,\n ): ExecutionTaskPlanningLocateApply {\n const { cacheable, modelConfigForDefaultIntent } = context;\n\n let locateParam = detailedLocateParam;\n debug(\n 'modelConfigForDefaultIntent-locateParam',\n modelConfigForDefaultIntent,\n locateParam,\n );\n\n if (typeof locateParam === 'string') {\n locateParam = {\n prompt: locateParam,\n };\n }\n\n if (cacheable !== undefined) {\n locateParam = {\n ...locateParam,\n cacheable,\n };\n }\n\n const taskLocator: ExecutionTaskPlanningLocateApply = {\n type: 'Planning',\n subType: 'Locate',\n subTask: context.subTask || undefined,\n param: locateParam,\n thought: plan.thought,\n executor: async (param, taskContext) => {\n const { task } = taskContext;\n let { uiContext } = taskContext;\n\n assert(\n param?.prompt || param?.bbox,\n `No prompt or id or position or bbox to locate, param=${JSON.stringify(\n param,\n )}`,\n );\n\n if (!uiContext) {\n uiContext = await this.service.contextRetrieverFn();\n }\n\n assert(uiContext, 'uiContext is required for Service task');\n\n let locateDump: ServiceDump | undefined;\n let locateResult: LocateResultWithDump | undefined;\n\n const applyDump = (dump?: ServiceDump) => {\n if (!dump) {\n return;\n }\n locateDump = dump;\n task.log = {\n dump,\n };\n task.usage = dump.taskInfo?.usage;\n if (dump.taskInfo?.searchAreaUsage) {\n task.searchAreaUsage = dump.taskInfo.searchAreaUsage;\n }\n };\n\n // from xpath\n let rectFromXpath: Rect | undefined;\n if (param.xpath && this.interface.rectMatchesCacheFeature) {\n try {\n rectFromXpath = await this.interface.rectMatchesCacheFeature({\n xpaths: [param.xpath],\n });\n } catch (error) {\n // xpath locate failed, allow fallback to cache or AI locate\n rectFromXpath = undefined;\n }\n }\n const elementFromXpath = rectFromXpath\n ? generateElementByPosition(\n {\n x: rectFromXpath.left + rectFromXpath.width / 2,\n y: rectFromXpath.top + rectFromXpath.height / 2,\n },\n typeof param.prompt === 'string'\n ? param.prompt\n : param.prompt?.prompt || '',\n )\n : undefined;\n const userExpectedPathHitFlag = !!elementFromXpath;\n\n const cachePrompt = param.prompt;\n const locateCacheRecord = this.taskCache?.matchLocateCache(cachePrompt);\n const cacheEntry = locateCacheRecord?.cacheContent?.cache;\n\n const elementFromCache = userExpectedPathHitFlag\n ? null\n : await matchElementFromCache(\n {\n taskCache: this.taskCache,\n interfaceInstance: this.interface,\n },\n cacheEntry,\n cachePrompt,\n param.cacheable,\n );\n const cacheHitFlag = !!elementFromCache;\n\n const planHitFlag = false;\n\n let elementFromAiLocate: LocateResultElement | null | undefined;\n if (!userExpectedPathHitFlag && !cacheHitFlag && !planHitFlag) {\n try {\n locateResult = await this.service.locate(\n param,\n {\n context: uiContext,\n },\n modelConfigForDefaultIntent,\n );\n applyDump(locateResult.dump);\n elementFromAiLocate = locateResult.element;\n } catch (error) {\n if (error instanceof ServiceError) {\n applyDump(error.dump);\n }\n throw error;\n }\n }\n\n const element =\n elementFromXpath || elementFromCache || elementFromAiLocate;\n\n let currentCacheEntry: ElementCacheFeature | undefined;\n if (\n element &&\n this.taskCache &&\n !cacheHitFlag &&\n param?.cacheable !== false\n ) {\n if (this.interface.cacheFeatureForRect) {\n try {\n const feature = await this.interface.cacheFeatureForRect(\n element.rect,\n {\n targetDescription:\n typeof param.prompt === 'string'\n ? param.prompt\n : param.prompt?.prompt,\n modelConfig: modelConfigForDefaultIntent,\n },\n );\n if (feature && Object.keys(feature).length > 0) {\n debug(\n 'update cache, prompt: %s, cache: %o',\n cachePrompt,\n feature,\n );\n currentCacheEntry = feature;\n this.taskCache.updateOrAppendCacheRecord(\n {\n type: 'locate',\n prompt: cachePrompt,\n cache: feature,\n },\n locateCacheRecord,\n );\n } else {\n debug(\n 'no cache data returned, skip cache update, prompt: %s',\n cachePrompt,\n );\n }\n } catch (error) {\n debug('cacheFeatureForRect failed: %s', error);\n }\n } else {\n debug('cacheFeatureForRect is not supported, skip cache update');\n }\n }\n\n if (!element) {\n if (locateDump) {\n throw new ServiceError(\n `Element not found : ${param.prompt}`,\n locateDump,\n );\n }\n throw new Error(`Element not found: ${param.prompt}`);\n }\n\n let hitBy: ExecutionTaskHitBy | undefined;\n\n if (userExpectedPathHitFlag) {\n hitBy = {\n from: 'User expected path',\n context: {\n xpath: param.xpath,\n },\n };\n } else if (cacheHitFlag) {\n hitBy = {\n from: 'Cache',\n context: {\n cacheEntry,\n cacheToSave: currentCacheEntry,\n },\n };\n }\n\n onResult?.(element);\n\n return {\n output: {\n element,\n },\n hitBy,\n };\n },\n };\n\n return taskLocator;\n }\n}\n"],"names":["debug","getDebug","locatePlanForLocate","param","locate","locatePlan","TaskBuilder","plans","modelConfigForPlanning","modelConfigForDefaultIntent","options","tasks","cacheable","context","planHandlers","Map","plan","defaultHandler","handler","taskActionFinished","undefined","sleepTask","meta","taskParam","sleep","taskLocate","planType","actionSpace","action","item","Error","locateFields","findAllMidsceneLocatorField","requiredLocateFields","field","ifPlanLocateParamIsBbox","elementFromPlan","matchElementFromPlan","JSON","locateTask","result","assert","task","taskContext","uiContext","Promise","originalError","originalMessage","String","parseActionParam","error","actionFn","actionResult","delayAfterRunner","detailedLocateParam","onResult","locateParam","taskLocator","locateDump","locateResult","applyDump","dump","rectFromXpath","elementFromXpath","generateElementByPosition","userExpectedPathHitFlag","cachePrompt","locateCacheRecord","cacheEntry","elementFromCache","matchElementFromCache","cacheHitFlag","planHitFlag","elementFromAiLocate","ServiceError","element","currentCacheEntry","feature","Object","hitBy","interfaceInstance","service","taskCache"],"mappings":";;;;;;;;;;;;;;;;;AA+BA,MAAMA,QAAQC,SAAS;AAEhB,SAASC,oBAAoBC,KAAmC;IACrE,MAAMC,SAAS,AAAiB,YAAjB,OAAOD,QAAqB;QAAE,QAAQA;IAAM,IAAIA;IAC/D,MAAME,aAAkD;QACtD,MAAM;QACN,OAAOD;QACP,SAAS;IACX;IACA,OAAOC;AACT;AAqBO,MAAMC;IAaX,MAAa,MACXC,KAAuB,EACvBC,sBAAoC,EACpCC,2BAAyC,EACzCC,OAAsB,EACoB;QAC1C,MAAMC,QAA8B,EAAE;QACtC,MAAMC,YAAYF,SAAS;QAE3B,MAAMG,UAA4B;YAChCF;YACAH;YACAC;YACAG;YACA,SAAS,CAAC,CAACF,SAAS;QACtB;QAIA,MAAMI,eAAe,IAAIC,IAAyB;YAChD;gBACE;gBACA,CAACC,OACC,IAAI,CAAC,gBAAgB,CACnBA,MACAH;aAEL;YACD;gBAAC;gBAAY,CAACG,OAAS,IAAI,CAAC,kBAAkB,CAACA,MAAMH;aAAS;YAC9D;gBACE;gBACA,CAACG,OACC,IAAI,CAAC,eAAe,CAClBA,MACAH;aAEL;SACF;QAED,MAAMI,iBAA8B,CAACD,OACnC,IAAI,CAAC,gBAAgB,CAACA,MAAMH;QAE9B,KAAK,MAAMG,QAAQT,MAAO;YACxB,MAAMW,UAAUJ,aAAa,GAAG,CAACE,KAAK,IAAI,KAAKC;YAC/C,MAAMC,QAAQF;QAChB;QAEA,OAAO;YACLL;QACF;IACF;IAEQ,mBACNK,IAAoB,EACpBH,OAAyB,EACnB;QACN,MAAMM,qBAAqD;YACzD,MAAM;YACN,SAAS;YACT,OAAO;YACP,SAASH,KAAK,OAAO;YACrB,SAASH,QAAQ,OAAO,IAAIO;YAC5B,UAAU,WAAa;QACzB;QACAP,QAAQ,KAAK,CAAC,IAAI,CAACM;IACrB;IAEQ,gBACNH,IAA8C,EAC9CH,OAAyB,EACnB;QACN,MAAMQ,YAAY,IAAI,CAAC,eAAe,CAACL,KAAK,KAAK,EAAE;YACjD,SAASA,KAAK,OAAO;QACvB;QACA,IAAIH,QAAQ,OAAO,EACjBQ,UAAU,OAAO,GAAG;QAEtBR,QAAQ,KAAK,CAAC,IAAI,CAACQ;IACrB;IAEO,gBACLlB,KAA+B,EAC/BmB,IAA2B,EACyB;QACpD,OAAO;YACL,MAAM;YACN,SAAS;YACTnB;YACA,SAASmB,MAAM;YACf,UAAU,OAAOC;gBACf,MAAMC,MAAMD,WAAW,UAAU;YACnC;QACF;IACF;IAEA,MAAc,iBACZP,IAAyC,EACzCH,OAAyB,EACV;QACf,MAAMY,aAAa,IAAI,CAAC,gBAAgB,CAACT,MAAMA,KAAK,KAAK,EAAEH;QAC3DA,QAAQ,KAAK,CAAC,IAAI,CAACY;IACrB;IAEA,MAAc,iBACZT,IAAoB,EACpBH,OAAyB,EACV;QACf,MAAMa,WAAWV,KAAK,IAAI;QAC1B,MAAMW,cAAc,MAAM,IAAI,CAAC,SAAS,CAAC,WAAW;QACpD,MAAMC,SAASD,YAAY,IAAI,CAAC,CAACE,OAASA,KAAK,IAAI,KAAKH;QACxD,MAAMvB,QAAQa,KAAK,KAAK;QAExB,IAAI,CAACY,QACH,MAAM,IAAIE,MAAM,CAAC,aAAa,EAAEJ,SAAS,WAAW,CAAC;QAGvD,MAAMK,eAAeH,SACjBI,4BAA4BJ,OAAO,WAAW,IAC9C,EAAE;QAEN,MAAMK,uBAAuBL,SACzBI,4BAA4BJ,OAAO,WAAW,EAAE,QAChD,EAAE;QAENG,aAAa,OAAO,CAAC,CAACG;YACpB,IAAI/B,KAAK,CAAC+B,MAAM,EACd,IAAIC,wBAAwBhC,KAAK,CAAC+B,MAAM,GAAG;gBACzClC,MACE,2DACAG,KAAK,CAAC+B,MAAM;gBAEd,MAAME,kBAAkBC,qBAAqBlC,KAAK,CAAC+B,MAAM;gBACzD,IAAIE,iBACFjC,KAAK,CAAC+B,MAAM,GAAGE;YAEnB,OAAO;gBACL,MAAM/B,aAAaH,oBAAoBC,KAAK,CAAC+B,MAAM;gBACnDlC,MACE,uCACA,CAAC,YAAY,EAAE0B,UAAU,EACzB,CAAC,MAAM,EAAEY,KAAK,SAAS,CAACnC,KAAK,CAAC+B,MAAM,GAAG,EACvC,CAAC,WAAW,EAAEI,KAAK,SAAS,CAACjC,aAAa;gBAE5C,MAAMkC,aAAa,IAAI,CAAC,gBAAgB,CACtClC,YACAF,KAAK,CAAC+B,MAAM,EACZrB,SACA,CAAC2B;oBACCrC,KAAK,CAAC+B,MAAM,GAAGM;gBACjB;gBAEF3B,QAAQ,KAAK,CAAC,IAAI,CAAC0B;YACrB;iBACK;gBACLE,OACE,CAACR,qBAAqB,QAAQ,CAACC,QAC/B,CAAC,uBAAuB,EAAEA,MAAM,6BAA6B,EAAER,UAAU;gBAE3E1B,MAAM,CAAC,OAAO,EAAEkC,MAAM,6BAA6B,EAAER,UAAU;YACjE;QACF;QAEA,MAAMgB,OAKF;YACF,MAAM;YACN,SAAShB;YACT,SAASV,KAAK,OAAO;YACrB,OAAOA,KAAK,KAAK;YACjB,SAASH,QAAQ,OAAO,IAAIO;YAC5B,UAAU,OAAOjB,OAAOwC;gBACtB3C,MACE,oBACA0B,UACAvB,OACA,CAAC,4BAA4B,EAAEwC,YAAY,OAAO,EAAE,QAAQ;gBAG9D,MAAMC,YAAYD,YAAY,SAAS;gBACvCF,OAAOG,WAAW;gBAElBX,qBAAqB,OAAO,CAAC,CAACC;oBAC5BO,OACEtC,KAAK,CAAC+B,MAAM,EACZ,CAAC,OAAO,EAAEA,MAAM,yBAAyB,EAAER,SAAS,yCAAyC,EAAEA,SAAS,CAAC,CAAC;gBAE9G;gBAEA,IAAI;oBACF,MAAMmB,QAAQ,GAAG,CAAC;wBACf;4BACC,IAAI,IAAI,CAAC,SAAS,CAAC,kBAAkB,EAAE;gCACrC7C,MAAM;gCACN,MAAM,IAAI,CAAC,SAAS,CAAC,kBAAkB,CAAC4B,OAAO,IAAI,EAAEzB;gCACrDH,MAAM;4BACR;wBACF;wBACAwB,MAAM;qBACP;gBACH,EAAE,OAAOsB,eAAoB;oBAC3B,MAAMC,kBACJD,eAAe,WAAWE,OAAOF;oBACnC,MAAM,IAAIhB,MACR,CAAC,wCAAwC,EAAEF,OAAO,IAAI,CAAC,EAAE,EAAEmB,iBAAiB,EAC5E;wBAAE,OAAOD;oBAAc;gBAE3B;gBAEA,IAAIlB,OAAO,WAAW,EACpB,IAAI;oBACFzB,QAAQ8C,iBAAiB9C,OAAOyB,OAAO,WAAW;gBACpD,EAAE,OAAOsB,OAAY;oBACnB,MAAM,IAAIpB,MACR,CAAC,8BAA8B,EAAEF,OAAO,IAAI,CAAC,EAAE,EAAEsB,MAAM,OAAO,CAAC,cAAc,EAAEZ,KAAK,SAAS,CAACnC,QAAQ,EACtG;wBAAE,OAAO+C;oBAAM;gBAEnB;gBAGFlD,MAAM,kBAAkB4B,OAAO,IAAI;gBACnC,MAAMuB,WAAWvB,OAAO,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS;gBAChD,MAAMwB,eAAe,MAAMD,SAAShD,OAAOwC;gBAC3C3C,MAAM,iBAAiB4B,OAAO,IAAI,EAAE,WAAWwB;gBAE/C,MAAMC,mBAAmBzB,OAAO,gBAAgB,IAAI;gBACpD,IAAIyB,mBAAmB,GACrB,MAAM7B,MAAM6B;gBAGd,IAAI;oBACF,IAAI,IAAI,CAAC,SAAS,CAAC,iBAAiB,EAAE;wBACpCrD,MAAM;wBACN,MAAM,IAAI,CAAC,SAAS,CAAC,iBAAiB,CAAC4B,OAAO,IAAI,EAAEzB;wBACpDH,MAAM;oBACR;gBACF,EAAE,OAAO8C,eAAoB;oBAC3B,MAAMC,kBACJD,eAAe,WAAWE,OAAOF;oBACnC,MAAM,IAAIhB,MACR,CAAC,uCAAuC,EAAEF,OAAO,IAAI,CAAC,EAAE,EAAEmB,iBAAiB,EAC3E;wBAAE,OAAOD;oBAAc;gBAE3B;gBAEA,OAAO;oBACL,QAAQM;gBACV;YACF;QACF;QAEAvC,QAAQ,KAAK,CAAC,IAAI,CAAC6B;IACrB;IAEQ,iBACN1B,IAAyC,EACzCsC,mBAAiD,EACjDzC,OAAyB,EACzB0C,QAAgD,EACd;QAClC,MAAM,EAAE3C,SAAS,EAAEH,2BAA2B,EAAE,GAAGI;QAEnD,IAAI2C,cAAcF;QAClBtD,MACE,2CACAS,6BACA+C;QAGF,IAAI,AAAuB,YAAvB,OAAOA,aACTA,cAAc;YACZ,QAAQA;QACV;QAGF,IAAI5C,AAAcQ,WAAdR,WACF4C,cAAc;YACZ,GAAGA,WAAW;YACd5C;QACF;QAGF,MAAM6C,cAAgD;YACpD,MAAM;YACN,SAAS;YACT,SAAS5C,QAAQ,OAAO,IAAIO;YAC5B,OAAOoC;YACP,SAASxC,KAAK,OAAO;YACrB,UAAU,OAAOb,OAAOwC;gBACtB,MAAM,EAAED,IAAI,EAAE,GAAGC;gBACjB,IAAI,EAAEC,SAAS,EAAE,GAAGD;gBAEpBF,OACEtC,OAAO,UAAUA,OAAO,MACxB,CAAC,qDAAqD,EAAEmC,KAAK,SAAS,CACpEnC,QACC;gBAGL,IAAI,CAACyC,WACHA,YAAY,MAAM,IAAI,CAAC,OAAO,CAAC,kBAAkB;gBAGnDH,OAAOG,WAAW;gBAElB,IAAIc;gBACJ,IAAIC;gBAEJ,MAAMC,YAAY,CAACC;oBACjB,IAAI,CAACA,MACH;oBAEFH,aAAaG;oBACbnB,KAAK,GAAG,GAAG;wBACTmB;oBACF;oBACAnB,KAAK,KAAK,GAAGmB,KAAK,QAAQ,EAAE;oBAC5B,IAAIA,KAAK,QAAQ,EAAE,iBACjBnB,KAAK,eAAe,GAAGmB,KAAK,QAAQ,CAAC,eAAe;gBAExD;gBAGA,IAAIC;gBACJ,IAAI3D,MAAM,KAAK,IAAI,IAAI,CAAC,SAAS,CAAC,uBAAuB,EACvD,IAAI;oBACF2D,gBAAgB,MAAM,IAAI,CAAC,SAAS,CAAC,uBAAuB,CAAC;wBAC3D,QAAQ;4BAAC3D,MAAM,KAAK;yBAAC;oBACvB;gBACF,EAAE,OAAO+C,OAAO;oBAEdY,gBAAgB1C;gBAClB;gBAEF,MAAM2C,mBAAmBD,gBACrBE,0BACE;oBACE,GAAGF,cAAc,IAAI,GAAGA,cAAc,KAAK,GAAG;oBAC9C,GAAGA,cAAc,GAAG,GAAGA,cAAc,MAAM,GAAG;gBAChD,GACA,AAAwB,YAAxB,OAAO3D,MAAM,MAAM,GACfA,MAAM,MAAM,GACZA,MAAM,MAAM,EAAE,UAAU,MAE9BiB;gBACJ,MAAM6C,0BAA0B,CAAC,CAACF;gBAElC,MAAMG,cAAc/D,MAAM,MAAM;gBAChC,MAAMgE,oBAAoB,IAAI,CAAC,SAAS,EAAE,iBAAiBD;gBAC3D,MAAME,aAAaD,mBAAmB,cAAc;gBAEpD,MAAME,mBAAmBJ,0BACrB,OACA,MAAMK,sBACJ;oBACE,WAAW,IAAI,CAAC,SAAS;oBACzB,mBAAmB,IAAI,CAAC,SAAS;gBACnC,GACAF,YACAF,aACA/D,MAAM,SAAS;gBAErB,MAAMoE,eAAe,CAAC,CAACF;gBAEvB,MAAMG,cAAc;gBAEpB,IAAIC;gBACJ,IAAI,CAACR,2BAA2B,CAACM,gBAAgB,CAACC,aAChD,IAAI;oBACFb,eAAe,MAAM,IAAI,CAAC,OAAO,CAAC,MAAM,CACtCxD,OACA;wBACE,SAASyC;oBACX,GACAnC;oBAEFmD,UAAUD,aAAa,IAAI;oBAC3Bc,sBAAsBd,aAAa,OAAO;gBAC5C,EAAE,OAAOT,OAAO;oBACd,IAAIA,iBAAiBwB,cACnBd,UAAUV,MAAM,IAAI;oBAEtB,MAAMA;gBACR;gBAGF,MAAMyB,UACJZ,oBAAoBM,oBAAoBI;gBAE1C,IAAIG;gBACJ,IACED,WACA,IAAI,CAAC,SAAS,IACd,CAACJ,gBACDpE,OAAO,cAAc,OAErB,IAAI,IAAI,CAAC,SAAS,CAAC,mBAAmB,EACpC,IAAI;oBACF,MAAM0E,UAAU,MAAM,IAAI,CAAC,SAAS,CAAC,mBAAmB,CACtDF,QAAQ,IAAI,EACZ;wBACE,mBACE,AAAwB,YAAxB,OAAOxE,MAAM,MAAM,GACfA,MAAM,MAAM,GACZA,MAAM,MAAM,EAAE;wBACpB,aAAaM;oBACf;oBAEF,IAAIoE,WAAWC,OAAO,IAAI,CAACD,SAAS,MAAM,GAAG,GAAG;wBAC9C7E,MACE,uCACAkE,aACAW;wBAEFD,oBAAoBC;wBACpB,IAAI,CAAC,SAAS,CAAC,yBAAyB,CACtC;4BACE,MAAM;4BACN,QAAQX;4BACR,OAAOW;wBACT,GACAV;oBAEJ,OACEnE,MACE,yDACAkE;gBAGN,EAAE,OAAOhB,OAAO;oBACdlD,MAAM,kCAAkCkD;gBAC1C;qBAEAlD,MAAM;gBAIV,IAAI,CAAC2E,SAAS;oBACZ,IAAIjB,YACF,MAAM,IAAIgB,aACR,CAAC,oBAAoB,EAAEvE,MAAM,MAAM,EAAE,EACrCuD;oBAGJ,MAAM,IAAI5B,MAAM,CAAC,mBAAmB,EAAE3B,MAAM,MAAM,EAAE;gBACtD;gBAEA,IAAI4E;gBAEJ,IAAId,yBACFc,QAAQ;oBACN,MAAM;oBACN,SAAS;wBACP,OAAO5E,MAAM,KAAK;oBACpB;gBACF;qBACK,IAAIoE,cACTQ,QAAQ;oBACN,MAAM;oBACN,SAAS;wBACPX;wBACA,aAAaQ;oBACf;gBACF;gBAGFrB,WAAWoB;gBAEX,OAAO;oBACL,QAAQ;wBACNA;oBACF;oBACAI;gBACF;YACF;QACF;QAEA,OAAOtB;IACT;IAteA,YAAY,EAAEuB,iBAAiB,EAAEC,OAAO,EAAEC,SAAS,EAAmB,CAAE;QANxE,uBAAiB,aAAjB;QAEA,uBAAiB,WAAjB;QAEA,uBAAiB,aAAjB;QAGE,IAAI,CAAC,SAAS,GAAGF;QACjB,IAAI,CAAC,OAAO,GAAGC;QACf,IAAI,CAAC,SAAS,GAAGC;IACnB;AAmeF"}
@@ -163,18 +163,15 @@ class TaskExecutor {
163
163
  return session.appendErrorPlan(`Error converting plans to executable tasks: ${error}, plans: ${JSON.stringify(plans)}`);
164
164
  }
165
165
  if (this.conversationHistory.pendingFeedbackMessage) console.warn('unconsumed pending feedback message detected, this may lead to unexpected planning result:', this.conversationHistory.pendingFeedbackMessage);
166
- let errorFlag = false;
167
166
  try {
168
167
  await session.appendAndRun(executables.tasks);
169
168
  } catch (error) {
170
- errorFlag = true;
171
169
  errorCountInOnePlanningLoop++;
172
170
  this.conversationHistory.pendingFeedbackMessage = `Error executing running tasks: ${error?.message || String(error)}`;
173
171
  debug('error when executing running tasks, but continue to run if it is not too many errors:', error instanceof Error ? error.message : String(error), 'current error count in one planning loop:', errorCountInOnePlanningLoop);
174
172
  }
175
173
  if (errorCountInOnePlanningLoop > maxErrorCountAllowedInOnePlanningLoop) return session.appendErrorPlan('Too many errors in one planning loop');
176
- if (!planResult?.more_actions_needed_by_instruction) if (errorFlag) debug('more_actions_needed_by_instruction is false, but there are errors in one planning loop, continue to run');
177
- else break;
174
+ if (!planResult?.more_actions_needed_by_instruction) break;
178
175
  ++replanCount;
179
176
  if (replanCount > replanningCycleLimit) {
180
177
  const errorMsg = `Replanned ${replanningCycleLimit} times, exceeding the limit. Please configure a larger value for replanningCycleLimit (or use MIDSCENE_REPLANNING_CYCLE_LIMIT) to handle more complex tasks.`;
@@ -1 +1 @@
1
- {"version":3,"file":"agent/tasks.mjs","sources":["../../../src/agent/tasks.ts"],"sourcesContent":["import { ConversationHistory, plan, uiTarsPlanning } from '@/ai-model';\nimport type { TMultimodalPrompt, TUserPrompt } from '@/common';\nimport type { AbstractInterface } from '@/device';\nimport type Service from '@/service';\nimport type { TaskRunner } from '@/task-runner';\nimport { TaskExecutionError } from '@/task-runner';\nimport type {\n ExecutionTaskApply,\n ExecutionTaskInsightQueryApply,\n ExecutionTaskPlanningApply,\n ExecutionTaskProgressOptions,\n InterfaceType,\n MidsceneYamlFlowItem,\n PlanningAIResponse,\n PlanningAction,\n PlanningActionParamSleep,\n PlanningActionParamWaitFor,\n ServiceDump,\n ServiceExtractOption,\n ServiceExtractParam,\n} from '@/types';\nimport { ServiceError } from '@/types';\nimport type { IModelConfig } from '@midscene/shared/env';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport { ExecutionSession } from './execution-session';\nimport { TaskBuilder } from './task-builder';\nimport type { TaskCache } from './task-cache';\nexport { locatePlanForLocate } from './task-builder';\nimport { descriptionOfTree } from '@midscene/shared/extractor';\nimport { taskTitleStr } from './ui-utils';\nimport { parsePrompt } from './utils';\n\ninterface ExecutionResult<OutputType = any> {\n output: OutputType;\n thought?: string;\n runner: TaskRunner;\n}\n\ninterface TaskExecutorHooks {\n onTaskUpdate?: (\n runner: TaskRunner,\n error?: TaskExecutionError,\n ) => Promise<void> | void;\n}\n\nconst debug = getDebug('device-task-executor');\nconst maxErrorCountAllowedInOnePlanningLoop = 5;\n\nexport { TaskExecutionError };\n\nexport class TaskExecutor {\n interface: AbstractInterface;\n\n service: Service;\n\n taskCache?: TaskCache;\n\n private readonly taskBuilder: TaskBuilder;\n\n private conversationHistory: ConversationHistory;\n\n onTaskStartCallback?: ExecutionTaskProgressOptions['onTaskStart'];\n\n private readonly hooks?: TaskExecutorHooks;\n\n replanningCycleLimit?: number;\n\n // @deprecated use .interface instead\n get page() {\n return this.interface;\n }\n\n constructor(\n interfaceInstance: AbstractInterface,\n service: Service,\n opts: {\n taskCache?: TaskCache;\n onTaskStart?: ExecutionTaskProgressOptions['onTaskStart'];\n replanningCycleLimit?: number;\n hooks?: TaskExecutorHooks;\n },\n ) {\n this.interface = interfaceInstance;\n this.service = service;\n this.taskCache = opts.taskCache;\n this.onTaskStartCallback = opts?.onTaskStart;\n this.replanningCycleLimit = opts.replanningCycleLimit;\n this.hooks = opts.hooks;\n this.conversationHistory = new ConversationHistory();\n this.taskBuilder = new TaskBuilder({\n interfaceInstance,\n service,\n taskCache: opts.taskCache,\n });\n }\n\n private createExecutionSession(\n title: string,\n options?: { tasks?: ExecutionTaskApply[] },\n ) {\n return new ExecutionSession(\n title,\n () => Promise.resolve(this.service.contextRetrieverFn()),\n {\n onTaskStart: this.onTaskStartCallback,\n tasks: options?.tasks,\n onTaskUpdate: this.hooks?.onTaskUpdate,\n },\n );\n }\n\n public async convertPlanToExecutable(\n plans: PlanningAction[],\n modelConfigForPlanning: IModelConfig,\n modelConfigForDefaultIntent: IModelConfig,\n options?: {\n cacheable?: boolean;\n subTask?: boolean;\n },\n ) {\n return this.taskBuilder.build(\n plans,\n modelConfigForPlanning,\n modelConfigForDefaultIntent,\n options,\n );\n }\n\n async loadYamlFlowAsPlanning(userInstruction: string, yamlString: string) {\n const session = this.createExecutionSession(\n taskTitleStr('Action', userInstruction),\n );\n\n const task: ExecutionTaskPlanningApply = {\n type: 'Planning',\n subType: 'LoadYaml',\n param: {\n userInstruction,\n },\n executor: async (param, executorContext) => {\n const { uiContext } = executorContext;\n assert(uiContext, 'uiContext is required for Planning task');\n return {\n output: {\n actions: [],\n more_actions_needed_by_instruction: false,\n log: '',\n yamlString,\n },\n cache: {\n hit: true,\n },\n hitBy: {\n from: 'Cache',\n context: {\n yamlString,\n },\n },\n };\n },\n };\n const runner = session.getRunner();\n await session.appendAndRun(task);\n\n return {\n runner,\n };\n }\n\n async runPlans(\n title: string,\n plans: PlanningAction[],\n modelConfigForPlanning: IModelConfig,\n modelConfigForDefaultIntent: IModelConfig,\n ): Promise<ExecutionResult> {\n const session = this.createExecutionSession(title);\n const { tasks } = await this.convertPlanToExecutable(\n plans,\n modelConfigForPlanning,\n modelConfigForDefaultIntent,\n );\n const runner = session.getRunner();\n const result = await session.appendAndRun(tasks);\n const { output } = result ?? {};\n return {\n output,\n runner,\n };\n }\n\n async action(\n userPrompt: string,\n modelConfigForPlanning: IModelConfig,\n modelConfigForDefaultIntent: IModelConfig,\n includeBboxInPlanning: boolean,\n backgroundKnowledge?: string,\n cacheable?: boolean,\n replanningCycleLimitOverride?: number,\n ): Promise<\n ExecutionResult<\n | {\n yamlFlow?: MidsceneYamlFlowItem[]; // for cache use\n }\n | undefined\n >\n > {\n this.conversationHistory.reset();\n\n const session = this.createExecutionSession(\n taskTitleStr('Action', userPrompt),\n );\n const runner = session.getRunner();\n\n let replanCount = 0;\n const yamlFlow: MidsceneYamlFlowItem[] = [];\n const replanningCycleLimit =\n replanningCycleLimitOverride ?? this.replanningCycleLimit;\n assert(\n replanningCycleLimit !== undefined,\n 'replanningCycleLimit is required for TaskExecutor.action',\n );\n\n let errorCountInOnePlanningLoop = 0; // count the number of errors in one planning loop\n\n // Main planning loop - unified plan/replan logic\n while (true) {\n const result = await session.appendAndRun(\n {\n type: 'Planning',\n subType: 'Plan',\n param: {\n userInstruction: userPrompt,\n aiActionContext: backgroundKnowledge,\n },\n executor: async (param, executorContext) => {\n const startTime = Date.now();\n const { uiContext } = executorContext;\n assert(uiContext, 'uiContext is required for Planning task');\n const { vlMode } = modelConfigForPlanning;\n const uiTarsModelVersion =\n vlMode === 'vlm-ui-tars'\n ? modelConfigForPlanning.uiTarsModelVersion\n : undefined;\n\n assert(\n this.interface.actionSpace,\n 'actionSpace for device is not implemented',\n );\n const actionSpace = await this.interface.actionSpace();\n debug(\n 'actionSpace for this interface is:',\n actionSpace.map((action) => action.name).join(', '),\n );\n assert(Array.isArray(actionSpace), 'actionSpace must be an array');\n if (actionSpace.length === 0) {\n console.warn(\n `ActionSpace for ${this.interface.interfaceType} is empty. This may lead to unexpected behavior.`,\n );\n }\n\n const planResult = await (uiTarsModelVersion\n ? uiTarsPlanning\n : plan)(param.userInstruction, {\n context: uiContext,\n actionContext: param.aiActionContext,\n interfaceType: this.interface.interfaceType as InterfaceType,\n actionSpace,\n modelConfig: modelConfigForPlanning,\n conversationHistory: this.conversationHistory,\n includeBbox: includeBboxInPlanning,\n });\n debug('planResult', JSON.stringify(planResult, null, 2));\n\n const {\n actions,\n log,\n more_actions_needed_by_instruction,\n error,\n usage,\n rawResponse,\n sleep,\n } = planResult;\n\n executorContext.task.log = {\n ...(executorContext.task.log || {}),\n rawResponse,\n };\n executorContext.task.usage = usage;\n executorContext.task.output = {\n actions: actions || [],\n more_actions_needed_by_instruction,\n log,\n yamlFlow: planResult.yamlFlow,\n };\n executorContext.uiContext = uiContext;\n\n const finalActions = [...(actions || [])];\n\n if (sleep) {\n const timeNow = Date.now();\n const timeRemaining = sleep - (timeNow - startTime);\n if (timeRemaining > 0) {\n finalActions.push(this.sleepPlan(timeRemaining));\n }\n }\n\n assert(!error, `Failed to continue: ${error}\\n${log || ''}`);\n\n return {\n cache: {\n hit: false,\n },\n } as any;\n },\n },\n {\n allowWhenError: true,\n },\n );\n\n const planResult = result?.output as PlanningAIResponse | undefined;\n\n // Execute planned actions\n const plans = planResult?.actions || [];\n yamlFlow.push(...(planResult?.yamlFlow || []));\n\n let executables: Awaited<ReturnType<typeof this.convertPlanToExecutable>>;\n try {\n executables = await this.convertPlanToExecutable(\n plans,\n modelConfigForPlanning,\n modelConfigForDefaultIntent,\n {\n cacheable,\n subTask: true,\n },\n );\n } catch (error) {\n return session.appendErrorPlan(\n `Error converting plans to executable tasks: ${error}, plans: ${JSON.stringify(\n plans,\n )}`,\n );\n }\n if (this.conversationHistory.pendingFeedbackMessage) {\n console.warn(\n 'unconsumed pending feedback message detected, this may lead to unexpected planning result:',\n this.conversationHistory.pendingFeedbackMessage,\n );\n }\n let errorFlag = false;\n try {\n await session.appendAndRun(executables.tasks);\n } catch (error: any) {\n errorFlag = true;\n errorCountInOnePlanningLoop++;\n this.conversationHistory.pendingFeedbackMessage = `Error executing running tasks: ${error?.message || String(error)}`;\n debug(\n 'error when executing running tasks, but continue to run if it is not too many errors:',\n error instanceof Error ? error.message : String(error),\n 'current error count in one planning loop:',\n errorCountInOnePlanningLoop,\n );\n }\n\n if (errorCountInOnePlanningLoop > maxErrorCountAllowedInOnePlanningLoop) {\n return session.appendErrorPlan('Too many errors in one planning loop');\n }\n\n // Check if task is complete\n if (!planResult?.more_actions_needed_by_instruction) {\n if (errorFlag) {\n debug(\n 'more_actions_needed_by_instruction is false, but there are errors in one planning loop, continue to run',\n );\n } else {\n break;\n }\n }\n\n // Increment replan count for next iteration\n ++replanCount;\n\n if (replanCount > replanningCycleLimit) {\n const errorMsg = `Replanned ${replanningCycleLimit} times, exceeding the limit. Please configure a larger value for replanningCycleLimit (or use MIDSCENE_REPLANNING_CYCLE_LIMIT) to handle more complex tasks.`;\n return session.appendErrorPlan(errorMsg);\n }\n\n if (!this.conversationHistory.pendingFeedbackMessage) {\n this.conversationHistory.pendingFeedbackMessage =\n 'I have finished the action previously planned.';\n }\n }\n\n const finalResult = {\n output: {\n yamlFlow,\n },\n runner,\n };\n return finalResult;\n }\n\n private createTypeQueryTask(\n type: 'Query' | 'Boolean' | 'Number' | 'String' | 'Assert' | 'WaitFor',\n demand: ServiceExtractParam,\n modelConfig: IModelConfig,\n opt?: ServiceExtractOption,\n multimodalPrompt?: TMultimodalPrompt,\n ) {\n const queryTask: ExecutionTaskInsightQueryApply = {\n type: 'Insight',\n subType: type,\n param: {\n dataDemand: multimodalPrompt\n ? ({\n demand,\n multimodalPrompt,\n } as never)\n : demand, // for user param presentation in report right sidebar\n },\n executor: async (param, taskContext) => {\n const { task } = taskContext;\n let queryDump: ServiceDump | undefined;\n const applyDump = (dump: ServiceDump) => {\n queryDump = dump;\n task.log = {\n dump,\n };\n };\n\n // Get context for query operations\n const uiContext = taskContext.uiContext;\n assert(uiContext, 'uiContext is required for Query task');\n\n const ifTypeRestricted = type !== 'Query';\n let demandInput = demand;\n let keyOfResult = 'result';\n if (ifTypeRestricted && (type === 'Assert' || type === 'WaitFor')) {\n keyOfResult = 'StatementIsTruthy';\n const booleanPrompt =\n type === 'Assert'\n ? `Boolean, whether the following statement is true: ${demand}`\n : `Boolean, the user wants to do some 'wait for' operation, please check whether the following statement is true: ${demand}`;\n demandInput = {\n [keyOfResult]: booleanPrompt,\n };\n } else if (ifTypeRestricted) {\n demandInput = {\n [keyOfResult]: `${type}, ${demand}`,\n };\n }\n\n let extractResult;\n\n let extraPageDescription = '';\n if (opt?.domIncluded && this.interface.getElementsNodeTree) {\n debug('appending tree info for page');\n const tree = await this.interface.getElementsNodeTree();\n extraPageDescription = await descriptionOfTree(\n tree,\n 200,\n false,\n opt?.domIncluded === 'visible-only',\n );\n }\n\n try {\n extractResult = await this.service.extract<any>(\n demandInput,\n modelConfig,\n opt,\n extraPageDescription,\n multimodalPrompt,\n );\n } catch (error) {\n if (error instanceof ServiceError) {\n applyDump(error.dump);\n }\n throw error;\n }\n\n const { data, usage, thought, dump } = extractResult;\n applyDump(dump);\n\n let outputResult = data;\n if (ifTypeRestricted) {\n // If AI returned a plain string instead of structured format, use it directly\n if (typeof data === 'string') {\n outputResult = data;\n } else if (type === 'WaitFor') {\n if (data === null || data === undefined) {\n outputResult = false;\n } else {\n outputResult = (data as any)[keyOfResult];\n }\n } else if (data === null || data === undefined) {\n outputResult = null;\n } else {\n assert(\n data?.[keyOfResult] !== undefined,\n 'No result in query data',\n );\n outputResult = (data as any)[keyOfResult];\n }\n }\n\n if (type === 'Assert' && !outputResult) {\n task.usage = usage;\n task.thought = thought;\n throw new Error(`Assertion failed: ${thought}`);\n }\n\n return {\n output: outputResult,\n log: queryDump,\n usage,\n thought,\n };\n },\n };\n\n return queryTask;\n }\n async createTypeQueryExecution<T>(\n type: 'Query' | 'Boolean' | 'Number' | 'String' | 'Assert',\n demand: ServiceExtractParam,\n modelConfig: IModelConfig,\n opt?: ServiceExtractOption,\n multimodalPrompt?: TMultimodalPrompt,\n ): Promise<ExecutionResult<T>> {\n const session = this.createExecutionSession(\n taskTitleStr(\n type,\n typeof demand === 'string' ? demand : JSON.stringify(demand),\n ),\n );\n\n const queryTask = await this.createTypeQueryTask(\n type,\n demand,\n modelConfig,\n opt,\n multimodalPrompt,\n );\n\n const runner = session.getRunner();\n const result = await session.appendAndRun(queryTask);\n\n if (!result) {\n throw new Error(\n 'result of taskExecutor.flush() is undefined in function createTypeQueryTask',\n );\n }\n\n const { output, thought } = result;\n\n return {\n output,\n thought,\n runner,\n };\n }\n\n private sleepPlan(timeMs: number): PlanningAction<PlanningActionParamSleep> {\n return {\n type: 'Sleep',\n param: {\n timeMs,\n },\n };\n }\n\n async taskForSleep(timeMs: number, _modelConfig: IModelConfig) {\n return this.taskBuilder.createSleepTask({\n timeMs,\n });\n }\n\n async waitFor(\n assertion: TUserPrompt,\n opt: PlanningActionParamWaitFor,\n modelConfig: IModelConfig,\n ): Promise<ExecutionResult<void>> {\n const { textPrompt, multimodalPrompt } = parsePrompt(assertion);\n\n const description = `waitFor: ${textPrompt}`;\n const session = this.createExecutionSession(\n taskTitleStr('WaitFor', description),\n );\n const runner = session.getRunner();\n const { timeoutMs, checkIntervalMs } = opt;\n\n assert(assertion, 'No assertion for waitFor');\n assert(timeoutMs, 'No timeoutMs for waitFor');\n assert(checkIntervalMs, 'No checkIntervalMs for waitFor');\n\n assert(\n checkIntervalMs <= timeoutMs,\n `wrong config for waitFor: checkIntervalMs must be less than timeoutMs, config: {checkIntervalMs: ${checkIntervalMs}, timeoutMs: ${timeoutMs}}`,\n );\n\n const overallStartTime = Date.now();\n let lastCheckStart = overallStartTime;\n let errorThought = '';\n // Continue checking as long as the previous iteration began within the timeout window.\n while (lastCheckStart - overallStartTime <= timeoutMs) {\n const currentCheckStart = Date.now();\n lastCheckStart = currentCheckStart;\n const queryTask = await this.createTypeQueryTask(\n 'WaitFor',\n textPrompt,\n modelConfig,\n undefined,\n multimodalPrompt,\n );\n\n const result = (await session.appendAndRun(queryTask)) as\n | {\n output: boolean;\n thought?: string;\n }\n | undefined;\n\n if (result?.output) {\n return {\n output: undefined,\n runner,\n };\n }\n\n errorThought =\n result?.thought ||\n (!result && `No result from assertion: ${textPrompt}`) ||\n `unknown error when waiting for assertion: ${textPrompt}`;\n const now = Date.now();\n if (now - currentCheckStart < checkIntervalMs) {\n const timeRemaining = checkIntervalMs - (now - currentCheckStart);\n const sleepTask = this.taskBuilder.createSleepTask({\n timeMs: timeRemaining,\n });\n await session.append(sleepTask);\n }\n }\n\n return session.appendErrorPlan(`waitFor timeout: ${errorThought}`);\n }\n}\n"],"names":["debug","getDebug","maxErrorCountAllowedInOnePlanningLoop","TaskExecutor","title","options","ExecutionSession","Promise","plans","modelConfigForPlanning","modelConfigForDefaultIntent","userInstruction","yamlString","session","taskTitleStr","task","param","executorContext","uiContext","assert","runner","tasks","result","output","userPrompt","includeBboxInPlanning","backgroundKnowledge","cacheable","replanningCycleLimitOverride","replanCount","yamlFlow","replanningCycleLimit","undefined","errorCountInOnePlanningLoop","startTime","Date","vlMode","uiTarsModelVersion","actionSpace","action","Array","console","planResult","uiTarsPlanning","plan","JSON","actions","log","more_actions_needed_by_instruction","error","usage","rawResponse","sleep","finalActions","timeNow","timeRemaining","executables","errorFlag","String","Error","errorMsg","finalResult","type","demand","modelConfig","opt","multimodalPrompt","queryTask","taskContext","queryDump","applyDump","dump","ifTypeRestricted","demandInput","keyOfResult","booleanPrompt","extractResult","extraPageDescription","tree","descriptionOfTree","ServiceError","data","thought","outputResult","timeMs","_modelConfig","assertion","textPrompt","parsePrompt","description","timeoutMs","checkIntervalMs","overallStartTime","lastCheckStart","errorThought","currentCheckStart","now","sleepTask","interfaceInstance","service","opts","ConversationHistory","TaskBuilder"],"mappings":";;;;;;;;;;;;;;;;;;;;AA8CA,MAAMA,QAAQC,SAAS;AACvB,MAAMC,wCAAwC;AAIvC,MAAMC;IAkBX,IAAI,OAAO;QACT,OAAO,IAAI,CAAC,SAAS;IACvB;IA0BQ,uBACNC,KAAa,EACbC,OAA0C,EAC1C;QACA,OAAO,IAAIC,iBACTF,OACA,IAAMG,QAAQ,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,kBAAkB,KACrD;YACE,aAAa,IAAI,CAAC,mBAAmB;YACrC,OAAOF,SAAS;YAChB,cAAc,IAAI,CAAC,KAAK,EAAE;QAC5B;IAEJ;IAEA,MAAa,wBACXG,KAAuB,EACvBC,sBAAoC,EACpCC,2BAAyC,EACzCL,OAGC,EACD;QACA,OAAO,IAAI,CAAC,WAAW,CAAC,KAAK,CAC3BG,OACAC,wBACAC,6BACAL;IAEJ;IAEA,MAAM,uBAAuBM,eAAuB,EAAEC,UAAkB,EAAE;QACxE,MAAMC,UAAU,IAAI,CAAC,sBAAsB,CACzCC,aAAa,UAAUH;QAGzB,MAAMI,OAAmC;YACvC,MAAM;YACN,SAAS;YACT,OAAO;gBACLJ;YACF;YACA,UAAU,OAAOK,OAAOC;gBACtB,MAAM,EAAEC,SAAS,EAAE,GAAGD;gBACtBE,OAAOD,WAAW;gBAClB,OAAO;oBACL,QAAQ;wBACN,SAAS,EAAE;wBACX,oCAAoC;wBACpC,KAAK;wBACLN;oBACF;oBACA,OAAO;wBACL,KAAK;oBACP;oBACA,OAAO;wBACL,MAAM;wBACN,SAAS;4BACPA;wBACF;oBACF;gBACF;YACF;QACF;QACA,MAAMQ,SAASP,QAAQ,SAAS;QAChC,MAAMA,QAAQ,YAAY,CAACE;QAE3B,OAAO;YACLK;QACF;IACF;IAEA,MAAM,SACJhB,KAAa,EACbI,KAAuB,EACvBC,sBAAoC,EACpCC,2BAAyC,EACf;QAC1B,MAAMG,UAAU,IAAI,CAAC,sBAAsB,CAACT;QAC5C,MAAM,EAAEiB,KAAK,EAAE,GAAG,MAAM,IAAI,CAAC,uBAAuB,CAClDb,OACAC,wBACAC;QAEF,MAAMU,SAASP,QAAQ,SAAS;QAChC,MAAMS,SAAS,MAAMT,QAAQ,YAAY,CAACQ;QAC1C,MAAM,EAAEE,MAAM,EAAE,GAAGD,UAAU,CAAC;QAC9B,OAAO;YACLC;YACAH;QACF;IACF;IAEA,MAAM,OACJI,UAAkB,EAClBf,sBAAoC,EACpCC,2BAAyC,EACzCe,qBAA8B,EAC9BC,mBAA4B,EAC5BC,SAAmB,EACnBC,4BAAqC,EAQrC;QACA,IAAI,CAAC,mBAAmB,CAAC,KAAK;QAE9B,MAAMf,UAAU,IAAI,CAAC,sBAAsB,CACzCC,aAAa,UAAUU;QAEzB,MAAMJ,SAASP,QAAQ,SAAS;QAEhC,IAAIgB,cAAc;QAClB,MAAMC,WAAmC,EAAE;QAC3C,MAAMC,uBACJH,gCAAgC,IAAI,CAAC,oBAAoB;QAC3DT,OACEY,AAAyBC,WAAzBD,sBACA;QAGF,IAAIE,8BAA8B;QAGlC,MAAO,KAAM;YACX,MAAMX,SAAS,MAAMT,QAAQ,YAAY,CACvC;gBACE,MAAM;gBACN,SAAS;gBACT,OAAO;oBACL,iBAAiBW;oBACjB,iBAAiBE;gBACnB;gBACA,UAAU,OAAOV,OAAOC;oBACtB,MAAMiB,YAAYC,KAAK,GAAG;oBAC1B,MAAM,EAAEjB,SAAS,EAAE,GAAGD;oBACtBE,OAAOD,WAAW;oBAClB,MAAM,EAAEkB,MAAM,EAAE,GAAG3B;oBACnB,MAAM4B,qBACJD,AAAW,kBAAXA,SACI3B,uBAAuB,kBAAkB,GACzCuB;oBAENb,OACE,IAAI,CAAC,SAAS,CAAC,WAAW,EAC1B;oBAEF,MAAMmB,cAAc,MAAM,IAAI,CAAC,SAAS,CAAC,WAAW;oBACpDtC,MACE,sCACAsC,YAAY,GAAG,CAAC,CAACC,SAAWA,OAAO,IAAI,EAAE,IAAI,CAAC;oBAEhDpB,OAAOqB,MAAM,OAAO,CAACF,cAAc;oBACnC,IAAIA,AAAuB,MAAvBA,YAAY,MAAM,EACpBG,QAAQ,IAAI,CACV,CAAC,gBAAgB,EAAE,IAAI,CAAC,SAAS,CAAC,aAAa,CAAC,gDAAgD,CAAC;oBAIrG,MAAMC,aAAa,MAAOL,AAAAA,CAAAA,qBACtBM,iBACAC,IAAG,EAAG5B,MAAM,eAAe,EAAE;wBAC/B,SAASE;wBACT,eAAeF,MAAM,eAAe;wBACpC,eAAe,IAAI,CAAC,SAAS,CAAC,aAAa;wBAC3CsB;wBACA,aAAa7B;wBACb,qBAAqB,IAAI,CAAC,mBAAmB;wBAC7C,aAAagB;oBACf;oBACAzB,MAAM,cAAc6C,KAAK,SAAS,CAACH,YAAY,MAAM;oBAErD,MAAM,EACJI,OAAO,EACPC,GAAG,EACHC,kCAAkC,EAClCC,KAAK,EACLC,KAAK,EACLC,WAAW,EACXC,KAAK,EACN,GAAGV;oBAEJzB,gBAAgB,IAAI,CAAC,GAAG,GAAG;wBACzB,GAAIA,gBAAgB,IAAI,CAAC,GAAG,IAAI,CAAC,CAAC;wBAClCkC;oBACF;oBACAlC,gBAAgB,IAAI,CAAC,KAAK,GAAGiC;oBAC7BjC,gBAAgB,IAAI,CAAC,MAAM,GAAG;wBAC5B,SAAS6B,WAAW,EAAE;wBACtBE;wBACAD;wBACA,UAAUL,WAAW,QAAQ;oBAC/B;oBACAzB,gBAAgB,SAAS,GAAGC;oBAE5B,MAAMmC,eAAe;2BAAKP,WAAW,EAAE;qBAAE;oBAEzC,IAAIM,OAAO;wBACT,MAAME,UAAUnB,KAAK,GAAG;wBACxB,MAAMoB,gBAAgBH,QAASE,CAAAA,UAAUpB,SAAQ;wBACjD,IAAIqB,gBAAgB,GAClBF,aAAa,IAAI,CAAC,IAAI,CAAC,SAAS,CAACE;oBAErC;oBAEApC,OAAO,CAAC8B,OAAO,CAAC,oBAAoB,EAAEA,MAAM,EAAE,EAAEF,OAAO,IAAI;oBAE3D,OAAO;wBACL,OAAO;4BACL,KAAK;wBACP;oBACF;gBACF;YACF,GACA;gBACE,gBAAgB;YAClB;YAGF,MAAML,aAAapB,QAAQ;YAG3B,MAAMd,QAAQkC,YAAY,WAAW,EAAE;YACvCZ,SAAS,IAAI,IAAKY,YAAY,YAAY,EAAE;YAE5C,IAAIc;YACJ,IAAI;gBACFA,cAAc,MAAM,IAAI,CAAC,uBAAuB,CAC9ChD,OACAC,wBACAC,6BACA;oBACEiB;oBACA,SAAS;gBACX;YAEJ,EAAE,OAAOsB,OAAO;gBACd,OAAOpC,QAAQ,eAAe,CAC5B,CAAC,4CAA4C,EAAEoC,MAAM,SAAS,EAAEJ,KAAK,SAAS,CAC5ErC,QACC;YAEP;YACA,IAAI,IAAI,CAAC,mBAAmB,CAAC,sBAAsB,EACjDiC,QAAQ,IAAI,CACV,8FACA,IAAI,CAAC,mBAAmB,CAAC,sBAAsB;YAGnD,IAAIgB,YAAY;YAChB,IAAI;gBACF,MAAM5C,QAAQ,YAAY,CAAC2C,YAAY,KAAK;YAC9C,EAAE,OAAOP,OAAY;gBACnBQ,YAAY;gBACZxB;gBACA,IAAI,CAAC,mBAAmB,CAAC,sBAAsB,GAAG,CAAC,+BAA+B,EAAEgB,OAAO,WAAWS,OAAOT,QAAQ;gBACrHjD,MACE,yFACAiD,iBAAiBU,QAAQV,MAAM,OAAO,GAAGS,OAAOT,QAChD,6CACAhB;YAEJ;YAEA,IAAIA,8BAA8B/B,uCAChC,OAAOW,QAAQ,eAAe,CAAC;YAIjC,IAAI,CAAC6B,YAAY,oCACf,IAAIe,WACFzD,MACE;iBAGF;YAKJ,EAAE6B;YAEF,IAAIA,cAAcE,sBAAsB;gBACtC,MAAM6B,WAAW,CAAC,UAAU,EAAE7B,qBAAqB,4JAA4J,CAAC;gBAChN,OAAOlB,QAAQ,eAAe,CAAC+C;YACjC;YAEA,IAAI,CAAC,IAAI,CAAC,mBAAmB,CAAC,sBAAsB,EAClD,IAAI,CAAC,mBAAmB,CAAC,sBAAsB,GAC7C;QAEN;QAEA,MAAMC,cAAc;YAClB,QAAQ;gBACN/B;YACF;YACAV;QACF;QACA,OAAOyC;IACT;IAEQ,oBACNC,IAAsE,EACtEC,MAA2B,EAC3BC,WAAyB,EACzBC,GAA0B,EAC1BC,gBAAoC,EACpC;QACA,MAAMC,YAA4C;YAChD,MAAM;YACN,SAASL;YACT,OAAO;gBACL,YAAYI,mBACP;oBACCH;oBACAG;gBACF,IACAH;YACN;YACA,UAAU,OAAO/C,OAAOoD;gBACtB,MAAM,EAAErD,IAAI,EAAE,GAAGqD;gBACjB,IAAIC;gBACJ,MAAMC,YAAY,CAACC;oBACjBF,YAAYE;oBACZxD,KAAK,GAAG,GAAG;wBACTwD;oBACF;gBACF;gBAGA,MAAMrD,YAAYkD,YAAY,SAAS;gBACvCjD,OAAOD,WAAW;gBAElB,MAAMsD,mBAAmBV,AAAS,YAATA;gBACzB,IAAIW,cAAcV;gBAClB,IAAIW,cAAc;gBAClB,IAAIF,oBAAqBV,CAAAA,AAAS,aAATA,QAAqBA,AAAS,cAATA,IAAiB,GAAI;oBACjEY,cAAc;oBACd,MAAMC,gBACJb,AAAS,aAATA,OACI,CAAC,kDAAkD,EAAEC,QAAQ,GAC7D,CAAC,+GAA+G,EAAEA,QAAQ;oBAChIU,cAAc;wBACZ,CAACC,YAAY,EAAEC;oBACjB;gBACF,OAAO,IAAIH,kBACTC,cAAc;oBACZ,CAACC,YAAY,EAAE,GAAGZ,KAAK,EAAE,EAAEC,QAAQ;gBACrC;gBAGF,IAAIa;gBAEJ,IAAIC,uBAAuB;gBAC3B,IAAIZ,KAAK,eAAe,IAAI,CAAC,SAAS,CAAC,mBAAmB,EAAE;oBAC1DjE,MAAM;oBACN,MAAM8E,OAAO,MAAM,IAAI,CAAC,SAAS,CAAC,mBAAmB;oBACrDD,uBAAuB,MAAME,kBAC3BD,MACA,KACA,OACAb,KAAK,gBAAgB;gBAEzB;gBAEA,IAAI;oBACFW,gBAAgB,MAAM,IAAI,CAAC,OAAO,CAAC,OAAO,CACxCH,aACAT,aACAC,KACAY,sBACAX;gBAEJ,EAAE,OAAOjB,OAAO;oBACd,IAAIA,iBAAiB+B,cACnBV,UAAUrB,MAAM,IAAI;oBAEtB,MAAMA;gBACR;gBAEA,MAAM,EAAEgC,IAAI,EAAE/B,KAAK,EAAEgC,OAAO,EAAEX,IAAI,EAAE,GAAGK;gBACvCN,UAAUC;gBAEV,IAAIY,eAAeF;gBACnB,IAAIT,kBAEF,IAAI,AAAgB,YAAhB,OAAOS,MACTE,eAAeF;qBACV,IAAInB,AAAS,cAATA,MAEPqB,eADEF,QAAAA,OACa,QAECA,IAAY,CAACP,YAAY;qBAEtC,IAAIO,QAAAA,MACTE,eAAe;qBACV;oBACLhE,OACE8D,MAAM,CAACP,YAAY,KAAK1C,QACxB;oBAEFmD,eAAgBF,IAAY,CAACP,YAAY;gBAC3C;gBAGF,IAAIZ,AAAS,aAATA,QAAqB,CAACqB,cAAc;oBACtCpE,KAAK,KAAK,GAAGmC;oBACbnC,KAAK,OAAO,GAAGmE;oBACf,MAAM,IAAIvB,MAAM,CAAC,kBAAkB,EAAEuB,SAAS;gBAChD;gBAEA,OAAO;oBACL,QAAQC;oBACR,KAAKd;oBACLnB;oBACAgC;gBACF;YACF;QACF;QAEA,OAAOf;IACT;IACA,MAAM,yBACJL,IAA0D,EAC1DC,MAA2B,EAC3BC,WAAyB,EACzBC,GAA0B,EAC1BC,gBAAoC,EACP;QAC7B,MAAMrD,UAAU,IAAI,CAAC,sBAAsB,CACzCC,aACEgD,MACA,AAAkB,YAAlB,OAAOC,SAAsBA,SAASlB,KAAK,SAAS,CAACkB;QAIzD,MAAMI,YAAY,MAAM,IAAI,CAAC,mBAAmB,CAC9CL,MACAC,QACAC,aACAC,KACAC;QAGF,MAAM9C,SAASP,QAAQ,SAAS;QAChC,MAAMS,SAAS,MAAMT,QAAQ,YAAY,CAACsD;QAE1C,IAAI,CAAC7C,QACH,MAAM,IAAIqC,MACR;QAIJ,MAAM,EAAEpC,MAAM,EAAE2D,OAAO,EAAE,GAAG5D;QAE5B,OAAO;YACLC;YACA2D;YACA9D;QACF;IACF;IAEQ,UAAUgE,MAAc,EAA4C;QAC1E,OAAO;YACL,MAAM;YACN,OAAO;gBACLA;YACF;QACF;IACF;IAEA,MAAM,aAAaA,MAAc,EAAEC,YAA0B,EAAE;QAC7D,OAAO,IAAI,CAAC,WAAW,CAAC,eAAe,CAAC;YACtCD;QACF;IACF;IAEA,MAAM,QACJE,SAAsB,EACtBrB,GAA+B,EAC/BD,WAAyB,EACO;QAChC,MAAM,EAAEuB,UAAU,EAAErB,gBAAgB,EAAE,GAAGsB,YAAYF;QAErD,MAAMG,cAAc,CAAC,SAAS,EAAEF,YAAY;QAC5C,MAAM1E,UAAU,IAAI,CAAC,sBAAsB,CACzCC,aAAa,WAAW2E;QAE1B,MAAMrE,SAASP,QAAQ,SAAS;QAChC,MAAM,EAAE6E,SAAS,EAAEC,eAAe,EAAE,GAAG1B;QAEvC9C,OAAOmE,WAAW;QAClBnE,OAAOuE,WAAW;QAClBvE,OAAOwE,iBAAiB;QAExBxE,OACEwE,mBAAmBD,WACnB,CAAC,iGAAiG,EAAEC,gBAAgB,aAAa,EAAED,UAAU,CAAC,CAAC;QAGjJ,MAAME,mBAAmBzD,KAAK,GAAG;QACjC,IAAI0D,iBAAiBD;QACrB,IAAIE,eAAe;QAEnB,MAAOD,iBAAiBD,oBAAoBF,UAAW;YACrD,MAAMK,oBAAoB5D,KAAK,GAAG;YAClC0D,iBAAiBE;YACjB,MAAM5B,YAAY,MAAM,IAAI,CAAC,mBAAmB,CAC9C,WACAoB,YACAvB,aACAhC,QACAkC;YAGF,MAAM5C,SAAU,MAAMT,QAAQ,YAAY,CAACsD;YAO3C,IAAI7C,QAAQ,QACV,OAAO;gBACL,QAAQU;gBACRZ;YACF;YAGF0E,eACExE,QAAQ,WACP,CAACA,UAAU,CAAC,0BAA0B,EAAEiE,YAAY,IACrD,CAAC,0CAA0C,EAAEA,YAAY;YAC3D,MAAMS,MAAM7D,KAAK,GAAG;YACpB,IAAI6D,MAAMD,oBAAoBJ,iBAAiB;gBAC7C,MAAMpC,gBAAgBoC,kBAAmBK,CAAAA,MAAMD,iBAAgB;gBAC/D,MAAME,YAAY,IAAI,CAAC,WAAW,CAAC,eAAe,CAAC;oBACjD,QAAQ1C;gBACV;gBACA,MAAM1C,QAAQ,MAAM,CAACoF;YACvB;QACF;QAEA,OAAOpF,QAAQ,eAAe,CAAC,CAAC,iBAAiB,EAAEiF,cAAc;IACnE;IA9jBA,YACEI,iBAAoC,EACpCC,OAAgB,EAChBC,IAKC,CACD;QA9BF;QAEA;QAEA;QAEA,uBAAiB,eAAjB;QAEA,uBAAQ,uBAAR;QAEA;QAEA,uBAAiB,SAAjB;QAEA;QAiBE,IAAI,CAAC,SAAS,GAAGF;QACjB,IAAI,CAAC,OAAO,GAAGC;QACf,IAAI,CAAC,SAAS,GAAGC,KAAK,SAAS;QAC/B,IAAI,CAAC,mBAAmB,GAAGA,MAAM;QACjC,IAAI,CAAC,oBAAoB,GAAGA,KAAK,oBAAoB;QACrD,IAAI,CAAC,KAAK,GAAGA,KAAK,KAAK;QACvB,IAAI,CAAC,mBAAmB,GAAG,IAAIC;QAC/B,IAAI,CAAC,WAAW,GAAG,IAAIC,YAAY;YACjCJ;YACAC;YACA,WAAWC,KAAK,SAAS;QAC3B;IACF;AAyiBF"}
1
+ {"version":3,"file":"agent/tasks.mjs","sources":["../../../src/agent/tasks.ts"],"sourcesContent":["import { ConversationHistory, plan, uiTarsPlanning } from '@/ai-model';\nimport type { TMultimodalPrompt, TUserPrompt } from '@/common';\nimport type { AbstractInterface } from '@/device';\nimport type Service from '@/service';\nimport type { TaskRunner } from '@/task-runner';\nimport { TaskExecutionError } from '@/task-runner';\nimport type {\n ExecutionTaskApply,\n ExecutionTaskInsightQueryApply,\n ExecutionTaskPlanningApply,\n ExecutionTaskProgressOptions,\n InterfaceType,\n MidsceneYamlFlowItem,\n PlanningAIResponse,\n PlanningAction,\n PlanningActionParamSleep,\n PlanningActionParamWaitFor,\n ServiceDump,\n ServiceExtractOption,\n ServiceExtractParam,\n} from '@/types';\nimport { ServiceError } from '@/types';\nimport type { IModelConfig } from '@midscene/shared/env';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport { ExecutionSession } from './execution-session';\nimport { TaskBuilder } from './task-builder';\nimport type { TaskCache } from './task-cache';\nexport { locatePlanForLocate } from './task-builder';\nimport { descriptionOfTree } from '@midscene/shared/extractor';\nimport { taskTitleStr } from './ui-utils';\nimport { parsePrompt } from './utils';\n\ninterface ExecutionResult<OutputType = any> {\n output: OutputType;\n thought?: string;\n runner: TaskRunner;\n}\n\ninterface TaskExecutorHooks {\n onTaskUpdate?: (\n runner: TaskRunner,\n error?: TaskExecutionError,\n ) => Promise<void> | void;\n}\n\nconst debug = getDebug('device-task-executor');\nconst maxErrorCountAllowedInOnePlanningLoop = 5;\n\nexport { TaskExecutionError };\n\nexport class TaskExecutor {\n interface: AbstractInterface;\n\n service: Service;\n\n taskCache?: TaskCache;\n\n private readonly taskBuilder: TaskBuilder;\n\n private conversationHistory: ConversationHistory;\n\n onTaskStartCallback?: ExecutionTaskProgressOptions['onTaskStart'];\n\n private readonly hooks?: TaskExecutorHooks;\n\n replanningCycleLimit?: number;\n\n // @deprecated use .interface instead\n get page() {\n return this.interface;\n }\n\n constructor(\n interfaceInstance: AbstractInterface,\n service: Service,\n opts: {\n taskCache?: TaskCache;\n onTaskStart?: ExecutionTaskProgressOptions['onTaskStart'];\n replanningCycleLimit?: number;\n hooks?: TaskExecutorHooks;\n },\n ) {\n this.interface = interfaceInstance;\n this.service = service;\n this.taskCache = opts.taskCache;\n this.onTaskStartCallback = opts?.onTaskStart;\n this.replanningCycleLimit = opts.replanningCycleLimit;\n this.hooks = opts.hooks;\n this.conversationHistory = new ConversationHistory();\n this.taskBuilder = new TaskBuilder({\n interfaceInstance,\n service,\n taskCache: opts.taskCache,\n });\n }\n\n private createExecutionSession(\n title: string,\n options?: { tasks?: ExecutionTaskApply[] },\n ) {\n return new ExecutionSession(\n title,\n () => Promise.resolve(this.service.contextRetrieverFn()),\n {\n onTaskStart: this.onTaskStartCallback,\n tasks: options?.tasks,\n onTaskUpdate: this.hooks?.onTaskUpdate,\n },\n );\n }\n\n public async convertPlanToExecutable(\n plans: PlanningAction[],\n modelConfigForPlanning: IModelConfig,\n modelConfigForDefaultIntent: IModelConfig,\n options?: {\n cacheable?: boolean;\n subTask?: boolean;\n },\n ) {\n return this.taskBuilder.build(\n plans,\n modelConfigForPlanning,\n modelConfigForDefaultIntent,\n options,\n );\n }\n\n async loadYamlFlowAsPlanning(userInstruction: string, yamlString: string) {\n const session = this.createExecutionSession(\n taskTitleStr('Action', userInstruction),\n );\n\n const task: ExecutionTaskPlanningApply = {\n type: 'Planning',\n subType: 'LoadYaml',\n param: {\n userInstruction,\n },\n executor: async (param, executorContext) => {\n const { uiContext } = executorContext;\n assert(uiContext, 'uiContext is required for Planning task');\n return {\n output: {\n actions: [],\n more_actions_needed_by_instruction: false,\n log: '',\n yamlString,\n },\n cache: {\n hit: true,\n },\n hitBy: {\n from: 'Cache',\n context: {\n yamlString,\n },\n },\n };\n },\n };\n const runner = session.getRunner();\n await session.appendAndRun(task);\n\n return {\n runner,\n };\n }\n\n async runPlans(\n title: string,\n plans: PlanningAction[],\n modelConfigForPlanning: IModelConfig,\n modelConfigForDefaultIntent: IModelConfig,\n ): Promise<ExecutionResult> {\n const session = this.createExecutionSession(title);\n const { tasks } = await this.convertPlanToExecutable(\n plans,\n modelConfigForPlanning,\n modelConfigForDefaultIntent,\n );\n const runner = session.getRunner();\n const result = await session.appendAndRun(tasks);\n const { output } = result ?? {};\n return {\n output,\n runner,\n };\n }\n\n async action(\n userPrompt: string,\n modelConfigForPlanning: IModelConfig,\n modelConfigForDefaultIntent: IModelConfig,\n includeBboxInPlanning: boolean,\n backgroundKnowledge?: string,\n cacheable?: boolean,\n replanningCycleLimitOverride?: number,\n ): Promise<\n ExecutionResult<\n | {\n yamlFlow?: MidsceneYamlFlowItem[]; // for cache use\n }\n | undefined\n >\n > {\n this.conversationHistory.reset();\n\n const session = this.createExecutionSession(\n taskTitleStr('Action', userPrompt),\n );\n const runner = session.getRunner();\n\n let replanCount = 0;\n const yamlFlow: MidsceneYamlFlowItem[] = [];\n const replanningCycleLimit =\n replanningCycleLimitOverride ?? this.replanningCycleLimit;\n assert(\n replanningCycleLimit !== undefined,\n 'replanningCycleLimit is required for TaskExecutor.action',\n );\n\n let errorCountInOnePlanningLoop = 0; // count the number of errors in one planning loop\n\n // Main planning loop - unified plan/replan logic\n while (true) {\n const result = await session.appendAndRun(\n {\n type: 'Planning',\n subType: 'Plan',\n param: {\n userInstruction: userPrompt,\n aiActionContext: backgroundKnowledge,\n },\n executor: async (param, executorContext) => {\n const startTime = Date.now();\n const { uiContext } = executorContext;\n assert(uiContext, 'uiContext is required for Planning task');\n const { vlMode } = modelConfigForPlanning;\n const uiTarsModelVersion =\n vlMode === 'vlm-ui-tars'\n ? modelConfigForPlanning.uiTarsModelVersion\n : undefined;\n\n assert(\n this.interface.actionSpace,\n 'actionSpace for device is not implemented',\n );\n const actionSpace = await this.interface.actionSpace();\n debug(\n 'actionSpace for this interface is:',\n actionSpace.map((action) => action.name).join(', '),\n );\n assert(Array.isArray(actionSpace), 'actionSpace must be an array');\n if (actionSpace.length === 0) {\n console.warn(\n `ActionSpace for ${this.interface.interfaceType} is empty. This may lead to unexpected behavior.`,\n );\n }\n\n const planResult = await (uiTarsModelVersion\n ? uiTarsPlanning\n : plan)(param.userInstruction, {\n context: uiContext,\n actionContext: param.aiActionContext,\n interfaceType: this.interface.interfaceType as InterfaceType,\n actionSpace,\n modelConfig: modelConfigForPlanning,\n conversationHistory: this.conversationHistory,\n includeBbox: includeBboxInPlanning,\n });\n debug('planResult', JSON.stringify(planResult, null, 2));\n\n const {\n actions,\n log,\n more_actions_needed_by_instruction,\n error,\n usage,\n rawResponse,\n sleep,\n } = planResult;\n\n executorContext.task.log = {\n ...(executorContext.task.log || {}),\n rawResponse,\n };\n executorContext.task.usage = usage;\n executorContext.task.output = {\n actions: actions || [],\n more_actions_needed_by_instruction,\n log,\n yamlFlow: planResult.yamlFlow,\n };\n executorContext.uiContext = uiContext;\n\n const finalActions = [...(actions || [])];\n\n if (sleep) {\n const timeNow = Date.now();\n const timeRemaining = sleep - (timeNow - startTime);\n if (timeRemaining > 0) {\n finalActions.push(this.sleepPlan(timeRemaining));\n }\n }\n\n assert(!error, `Failed to continue: ${error}\\n${log || ''}`);\n\n return {\n cache: {\n hit: false,\n },\n } as any;\n },\n },\n {\n allowWhenError: true,\n },\n );\n\n const planResult = result?.output as PlanningAIResponse | undefined;\n\n // Execute planned actions\n const plans = planResult?.actions || [];\n yamlFlow.push(...(planResult?.yamlFlow || []));\n\n let executables: Awaited<ReturnType<typeof this.convertPlanToExecutable>>;\n try {\n executables = await this.convertPlanToExecutable(\n plans,\n modelConfigForPlanning,\n modelConfigForDefaultIntent,\n {\n cacheable,\n subTask: true,\n },\n );\n } catch (error) {\n return session.appendErrorPlan(\n `Error converting plans to executable tasks: ${error}, plans: ${JSON.stringify(\n plans,\n )}`,\n );\n }\n if (this.conversationHistory.pendingFeedbackMessage) {\n console.warn(\n 'unconsumed pending feedback message detected, this may lead to unexpected planning result:',\n this.conversationHistory.pendingFeedbackMessage,\n );\n }\n try {\n await session.appendAndRun(executables.tasks);\n } catch (error: any) {\n errorCountInOnePlanningLoop++;\n this.conversationHistory.pendingFeedbackMessage = `Error executing running tasks: ${error?.message || String(error)}`;\n debug(\n 'error when executing running tasks, but continue to run if it is not too many errors:',\n error instanceof Error ? error.message : String(error),\n 'current error count in one planning loop:',\n errorCountInOnePlanningLoop,\n );\n }\n\n if (errorCountInOnePlanningLoop > maxErrorCountAllowedInOnePlanningLoop) {\n return session.appendErrorPlan('Too many errors in one planning loop');\n }\n\n // Check if task is complete\n if (!planResult?.more_actions_needed_by_instruction) {\n break;\n }\n\n // Increment replan count for next iteration\n ++replanCount;\n\n if (replanCount > replanningCycleLimit) {\n const errorMsg = `Replanned ${replanningCycleLimit} times, exceeding the limit. Please configure a larger value for replanningCycleLimit (or use MIDSCENE_REPLANNING_CYCLE_LIMIT) to handle more complex tasks.`;\n return session.appendErrorPlan(errorMsg);\n }\n\n if (!this.conversationHistory.pendingFeedbackMessage) {\n this.conversationHistory.pendingFeedbackMessage =\n 'I have finished the action previously planned.';\n }\n }\n\n const finalResult = {\n output: {\n yamlFlow,\n },\n runner,\n };\n return finalResult;\n }\n\n private createTypeQueryTask(\n type: 'Query' | 'Boolean' | 'Number' | 'String' | 'Assert' | 'WaitFor',\n demand: ServiceExtractParam,\n modelConfig: IModelConfig,\n opt?: ServiceExtractOption,\n multimodalPrompt?: TMultimodalPrompt,\n ) {\n const queryTask: ExecutionTaskInsightQueryApply = {\n type: 'Insight',\n subType: type,\n param: {\n dataDemand: multimodalPrompt\n ? ({\n demand,\n multimodalPrompt,\n } as never)\n : demand, // for user param presentation in report right sidebar\n },\n executor: async (param, taskContext) => {\n const { task } = taskContext;\n let queryDump: ServiceDump | undefined;\n const applyDump = (dump: ServiceDump) => {\n queryDump = dump;\n task.log = {\n dump,\n };\n };\n\n // Get context for query operations\n const uiContext = taskContext.uiContext;\n assert(uiContext, 'uiContext is required for Query task');\n\n const ifTypeRestricted = type !== 'Query';\n let demandInput = demand;\n let keyOfResult = 'result';\n if (ifTypeRestricted && (type === 'Assert' || type === 'WaitFor')) {\n keyOfResult = 'StatementIsTruthy';\n const booleanPrompt =\n type === 'Assert'\n ? `Boolean, whether the following statement is true: ${demand}`\n : `Boolean, the user wants to do some 'wait for' operation, please check whether the following statement is true: ${demand}`;\n demandInput = {\n [keyOfResult]: booleanPrompt,\n };\n } else if (ifTypeRestricted) {\n demandInput = {\n [keyOfResult]: `${type}, ${demand}`,\n };\n }\n\n let extractResult;\n\n let extraPageDescription = '';\n if (opt?.domIncluded && this.interface.getElementsNodeTree) {\n debug('appending tree info for page');\n const tree = await this.interface.getElementsNodeTree();\n extraPageDescription = await descriptionOfTree(\n tree,\n 200,\n false,\n opt?.domIncluded === 'visible-only',\n );\n }\n\n try {\n extractResult = await this.service.extract<any>(\n demandInput,\n modelConfig,\n opt,\n extraPageDescription,\n multimodalPrompt,\n );\n } catch (error) {\n if (error instanceof ServiceError) {\n applyDump(error.dump);\n }\n throw error;\n }\n\n const { data, usage, thought, dump } = extractResult;\n applyDump(dump);\n\n let outputResult = data;\n if (ifTypeRestricted) {\n // If AI returned a plain string instead of structured format, use it directly\n if (typeof data === 'string') {\n outputResult = data;\n } else if (type === 'WaitFor') {\n if (data === null || data === undefined) {\n outputResult = false;\n } else {\n outputResult = (data as any)[keyOfResult];\n }\n } else if (data === null || data === undefined) {\n outputResult = null;\n } else {\n assert(\n data?.[keyOfResult] !== undefined,\n 'No result in query data',\n );\n outputResult = (data as any)[keyOfResult];\n }\n }\n\n if (type === 'Assert' && !outputResult) {\n task.usage = usage;\n task.thought = thought;\n throw new Error(`Assertion failed: ${thought}`);\n }\n\n return {\n output: outputResult,\n log: queryDump,\n usage,\n thought,\n };\n },\n };\n\n return queryTask;\n }\n async createTypeQueryExecution<T>(\n type: 'Query' | 'Boolean' | 'Number' | 'String' | 'Assert',\n demand: ServiceExtractParam,\n modelConfig: IModelConfig,\n opt?: ServiceExtractOption,\n multimodalPrompt?: TMultimodalPrompt,\n ): Promise<ExecutionResult<T>> {\n const session = this.createExecutionSession(\n taskTitleStr(\n type,\n typeof demand === 'string' ? demand : JSON.stringify(demand),\n ),\n );\n\n const queryTask = await this.createTypeQueryTask(\n type,\n demand,\n modelConfig,\n opt,\n multimodalPrompt,\n );\n\n const runner = session.getRunner();\n const result = await session.appendAndRun(queryTask);\n\n if (!result) {\n throw new Error(\n 'result of taskExecutor.flush() is undefined in function createTypeQueryTask',\n );\n }\n\n const { output, thought } = result;\n\n return {\n output,\n thought,\n runner,\n };\n }\n\n private sleepPlan(timeMs: number): PlanningAction<PlanningActionParamSleep> {\n return {\n type: 'Sleep',\n param: {\n timeMs,\n },\n };\n }\n\n async taskForSleep(timeMs: number, _modelConfig: IModelConfig) {\n return this.taskBuilder.createSleepTask({\n timeMs,\n });\n }\n\n async waitFor(\n assertion: TUserPrompt,\n opt: PlanningActionParamWaitFor,\n modelConfig: IModelConfig,\n ): Promise<ExecutionResult<void>> {\n const { textPrompt, multimodalPrompt } = parsePrompt(assertion);\n\n const description = `waitFor: ${textPrompt}`;\n const session = this.createExecutionSession(\n taskTitleStr('WaitFor', description),\n );\n const runner = session.getRunner();\n const { timeoutMs, checkIntervalMs } = opt;\n\n assert(assertion, 'No assertion for waitFor');\n assert(timeoutMs, 'No timeoutMs for waitFor');\n assert(checkIntervalMs, 'No checkIntervalMs for waitFor');\n\n assert(\n checkIntervalMs <= timeoutMs,\n `wrong config for waitFor: checkIntervalMs must be less than timeoutMs, config: {checkIntervalMs: ${checkIntervalMs}, timeoutMs: ${timeoutMs}}`,\n );\n\n const overallStartTime = Date.now();\n let lastCheckStart = overallStartTime;\n let errorThought = '';\n // Continue checking as long as the previous iteration began within the timeout window.\n while (lastCheckStart - overallStartTime <= timeoutMs) {\n const currentCheckStart = Date.now();\n lastCheckStart = currentCheckStart;\n const queryTask = await this.createTypeQueryTask(\n 'WaitFor',\n textPrompt,\n modelConfig,\n undefined,\n multimodalPrompt,\n );\n\n const result = (await session.appendAndRun(queryTask)) as\n | {\n output: boolean;\n thought?: string;\n }\n | undefined;\n\n if (result?.output) {\n return {\n output: undefined,\n runner,\n };\n }\n\n errorThought =\n result?.thought ||\n (!result && `No result from assertion: ${textPrompt}`) ||\n `unknown error when waiting for assertion: ${textPrompt}`;\n const now = Date.now();\n if (now - currentCheckStart < checkIntervalMs) {\n const timeRemaining = checkIntervalMs - (now - currentCheckStart);\n const sleepTask = this.taskBuilder.createSleepTask({\n timeMs: timeRemaining,\n });\n await session.append(sleepTask);\n }\n }\n\n return session.appendErrorPlan(`waitFor timeout: ${errorThought}`);\n }\n}\n"],"names":["debug","getDebug","maxErrorCountAllowedInOnePlanningLoop","TaskExecutor","title","options","ExecutionSession","Promise","plans","modelConfigForPlanning","modelConfigForDefaultIntent","userInstruction","yamlString","session","taskTitleStr","task","param","executorContext","uiContext","assert","runner","tasks","result","output","userPrompt","includeBboxInPlanning","backgroundKnowledge","cacheable","replanningCycleLimitOverride","replanCount","yamlFlow","replanningCycleLimit","undefined","errorCountInOnePlanningLoop","startTime","Date","vlMode","uiTarsModelVersion","actionSpace","action","Array","console","planResult","uiTarsPlanning","plan","JSON","actions","log","more_actions_needed_by_instruction","error","usage","rawResponse","sleep","finalActions","timeNow","timeRemaining","executables","String","Error","errorMsg","finalResult","type","demand","modelConfig","opt","multimodalPrompt","queryTask","taskContext","queryDump","applyDump","dump","ifTypeRestricted","demandInput","keyOfResult","booleanPrompt","extractResult","extraPageDescription","tree","descriptionOfTree","ServiceError","data","thought","outputResult","timeMs","_modelConfig","assertion","textPrompt","parsePrompt","description","timeoutMs","checkIntervalMs","overallStartTime","lastCheckStart","errorThought","currentCheckStart","now","sleepTask","interfaceInstance","service","opts","ConversationHistory","TaskBuilder"],"mappings":";;;;;;;;;;;;;;;;;;;;AA8CA,MAAMA,QAAQC,SAAS;AACvB,MAAMC,wCAAwC;AAIvC,MAAMC;IAkBX,IAAI,OAAO;QACT,OAAO,IAAI,CAAC,SAAS;IACvB;IA0BQ,uBACNC,KAAa,EACbC,OAA0C,EAC1C;QACA,OAAO,IAAIC,iBACTF,OACA,IAAMG,QAAQ,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,kBAAkB,KACrD;YACE,aAAa,IAAI,CAAC,mBAAmB;YACrC,OAAOF,SAAS;YAChB,cAAc,IAAI,CAAC,KAAK,EAAE;QAC5B;IAEJ;IAEA,MAAa,wBACXG,KAAuB,EACvBC,sBAAoC,EACpCC,2BAAyC,EACzCL,OAGC,EACD;QACA,OAAO,IAAI,CAAC,WAAW,CAAC,KAAK,CAC3BG,OACAC,wBACAC,6BACAL;IAEJ;IAEA,MAAM,uBAAuBM,eAAuB,EAAEC,UAAkB,EAAE;QACxE,MAAMC,UAAU,IAAI,CAAC,sBAAsB,CACzCC,aAAa,UAAUH;QAGzB,MAAMI,OAAmC;YACvC,MAAM;YACN,SAAS;YACT,OAAO;gBACLJ;YACF;YACA,UAAU,OAAOK,OAAOC;gBACtB,MAAM,EAAEC,SAAS,EAAE,GAAGD;gBACtBE,OAAOD,WAAW;gBAClB,OAAO;oBACL,QAAQ;wBACN,SAAS,EAAE;wBACX,oCAAoC;wBACpC,KAAK;wBACLN;oBACF;oBACA,OAAO;wBACL,KAAK;oBACP;oBACA,OAAO;wBACL,MAAM;wBACN,SAAS;4BACPA;wBACF;oBACF;gBACF;YACF;QACF;QACA,MAAMQ,SAASP,QAAQ,SAAS;QAChC,MAAMA,QAAQ,YAAY,CAACE;QAE3B,OAAO;YACLK;QACF;IACF;IAEA,MAAM,SACJhB,KAAa,EACbI,KAAuB,EACvBC,sBAAoC,EACpCC,2BAAyC,EACf;QAC1B,MAAMG,UAAU,IAAI,CAAC,sBAAsB,CAACT;QAC5C,MAAM,EAAEiB,KAAK,EAAE,GAAG,MAAM,IAAI,CAAC,uBAAuB,CAClDb,OACAC,wBACAC;QAEF,MAAMU,SAASP,QAAQ,SAAS;QAChC,MAAMS,SAAS,MAAMT,QAAQ,YAAY,CAACQ;QAC1C,MAAM,EAAEE,MAAM,EAAE,GAAGD,UAAU,CAAC;QAC9B,OAAO;YACLC;YACAH;QACF;IACF;IAEA,MAAM,OACJI,UAAkB,EAClBf,sBAAoC,EACpCC,2BAAyC,EACzCe,qBAA8B,EAC9BC,mBAA4B,EAC5BC,SAAmB,EACnBC,4BAAqC,EAQrC;QACA,IAAI,CAAC,mBAAmB,CAAC,KAAK;QAE9B,MAAMf,UAAU,IAAI,CAAC,sBAAsB,CACzCC,aAAa,UAAUU;QAEzB,MAAMJ,SAASP,QAAQ,SAAS;QAEhC,IAAIgB,cAAc;QAClB,MAAMC,WAAmC,EAAE;QAC3C,MAAMC,uBACJH,gCAAgC,IAAI,CAAC,oBAAoB;QAC3DT,OACEY,AAAyBC,WAAzBD,sBACA;QAGF,IAAIE,8BAA8B;QAGlC,MAAO,KAAM;YACX,MAAMX,SAAS,MAAMT,QAAQ,YAAY,CACvC;gBACE,MAAM;gBACN,SAAS;gBACT,OAAO;oBACL,iBAAiBW;oBACjB,iBAAiBE;gBACnB;gBACA,UAAU,OAAOV,OAAOC;oBACtB,MAAMiB,YAAYC,KAAK,GAAG;oBAC1B,MAAM,EAAEjB,SAAS,EAAE,GAAGD;oBACtBE,OAAOD,WAAW;oBAClB,MAAM,EAAEkB,MAAM,EAAE,GAAG3B;oBACnB,MAAM4B,qBACJD,AAAW,kBAAXA,SACI3B,uBAAuB,kBAAkB,GACzCuB;oBAENb,OACE,IAAI,CAAC,SAAS,CAAC,WAAW,EAC1B;oBAEF,MAAMmB,cAAc,MAAM,IAAI,CAAC,SAAS,CAAC,WAAW;oBACpDtC,MACE,sCACAsC,YAAY,GAAG,CAAC,CAACC,SAAWA,OAAO,IAAI,EAAE,IAAI,CAAC;oBAEhDpB,OAAOqB,MAAM,OAAO,CAACF,cAAc;oBACnC,IAAIA,AAAuB,MAAvBA,YAAY,MAAM,EACpBG,QAAQ,IAAI,CACV,CAAC,gBAAgB,EAAE,IAAI,CAAC,SAAS,CAAC,aAAa,CAAC,gDAAgD,CAAC;oBAIrG,MAAMC,aAAa,MAAOL,AAAAA,CAAAA,qBACtBM,iBACAC,IAAG,EAAG5B,MAAM,eAAe,EAAE;wBAC/B,SAASE;wBACT,eAAeF,MAAM,eAAe;wBACpC,eAAe,IAAI,CAAC,SAAS,CAAC,aAAa;wBAC3CsB;wBACA,aAAa7B;wBACb,qBAAqB,IAAI,CAAC,mBAAmB;wBAC7C,aAAagB;oBACf;oBACAzB,MAAM,cAAc6C,KAAK,SAAS,CAACH,YAAY,MAAM;oBAErD,MAAM,EACJI,OAAO,EACPC,GAAG,EACHC,kCAAkC,EAClCC,KAAK,EACLC,KAAK,EACLC,WAAW,EACXC,KAAK,EACN,GAAGV;oBAEJzB,gBAAgB,IAAI,CAAC,GAAG,GAAG;wBACzB,GAAIA,gBAAgB,IAAI,CAAC,GAAG,IAAI,CAAC,CAAC;wBAClCkC;oBACF;oBACAlC,gBAAgB,IAAI,CAAC,KAAK,GAAGiC;oBAC7BjC,gBAAgB,IAAI,CAAC,MAAM,GAAG;wBAC5B,SAAS6B,WAAW,EAAE;wBACtBE;wBACAD;wBACA,UAAUL,WAAW,QAAQ;oBAC/B;oBACAzB,gBAAgB,SAAS,GAAGC;oBAE5B,MAAMmC,eAAe;2BAAKP,WAAW,EAAE;qBAAE;oBAEzC,IAAIM,OAAO;wBACT,MAAME,UAAUnB,KAAK,GAAG;wBACxB,MAAMoB,gBAAgBH,QAASE,CAAAA,UAAUpB,SAAQ;wBACjD,IAAIqB,gBAAgB,GAClBF,aAAa,IAAI,CAAC,IAAI,CAAC,SAAS,CAACE;oBAErC;oBAEApC,OAAO,CAAC8B,OAAO,CAAC,oBAAoB,EAAEA,MAAM,EAAE,EAAEF,OAAO,IAAI;oBAE3D,OAAO;wBACL,OAAO;4BACL,KAAK;wBACP;oBACF;gBACF;YACF,GACA;gBACE,gBAAgB;YAClB;YAGF,MAAML,aAAapB,QAAQ;YAG3B,MAAMd,QAAQkC,YAAY,WAAW,EAAE;YACvCZ,SAAS,IAAI,IAAKY,YAAY,YAAY,EAAE;YAE5C,IAAIc;YACJ,IAAI;gBACFA,cAAc,MAAM,IAAI,CAAC,uBAAuB,CAC9ChD,OACAC,wBACAC,6BACA;oBACEiB;oBACA,SAAS;gBACX;YAEJ,EAAE,OAAOsB,OAAO;gBACd,OAAOpC,QAAQ,eAAe,CAC5B,CAAC,4CAA4C,EAAEoC,MAAM,SAAS,EAAEJ,KAAK,SAAS,CAC5ErC,QACC;YAEP;YACA,IAAI,IAAI,CAAC,mBAAmB,CAAC,sBAAsB,EACjDiC,QAAQ,IAAI,CACV,8FACA,IAAI,CAAC,mBAAmB,CAAC,sBAAsB;YAGnD,IAAI;gBACF,MAAM5B,QAAQ,YAAY,CAAC2C,YAAY,KAAK;YAC9C,EAAE,OAAOP,OAAY;gBACnBhB;gBACA,IAAI,CAAC,mBAAmB,CAAC,sBAAsB,GAAG,CAAC,+BAA+B,EAAEgB,OAAO,WAAWQ,OAAOR,QAAQ;gBACrHjD,MACE,yFACAiD,iBAAiBS,QAAQT,MAAM,OAAO,GAAGQ,OAAOR,QAChD,6CACAhB;YAEJ;YAEA,IAAIA,8BAA8B/B,uCAChC,OAAOW,QAAQ,eAAe,CAAC;YAIjC,IAAI,CAAC6B,YAAY,oCACf;YAIF,EAAEb;YAEF,IAAIA,cAAcE,sBAAsB;gBACtC,MAAM4B,WAAW,CAAC,UAAU,EAAE5B,qBAAqB,4JAA4J,CAAC;gBAChN,OAAOlB,QAAQ,eAAe,CAAC8C;YACjC;YAEA,IAAI,CAAC,IAAI,CAAC,mBAAmB,CAAC,sBAAsB,EAClD,IAAI,CAAC,mBAAmB,CAAC,sBAAsB,GAC7C;QAEN;QAEA,MAAMC,cAAc;YAClB,QAAQ;gBACN9B;YACF;YACAV;QACF;QACA,OAAOwC;IACT;IAEQ,oBACNC,IAAsE,EACtEC,MAA2B,EAC3BC,WAAyB,EACzBC,GAA0B,EAC1BC,gBAAoC,EACpC;QACA,MAAMC,YAA4C;YAChD,MAAM;YACN,SAASL;YACT,OAAO;gBACL,YAAYI,mBACP;oBACCH;oBACAG;gBACF,IACAH;YACN;YACA,UAAU,OAAO9C,OAAOmD;gBACtB,MAAM,EAAEpD,IAAI,EAAE,GAAGoD;gBACjB,IAAIC;gBACJ,MAAMC,YAAY,CAACC;oBACjBF,YAAYE;oBACZvD,KAAK,GAAG,GAAG;wBACTuD;oBACF;gBACF;gBAGA,MAAMpD,YAAYiD,YAAY,SAAS;gBACvChD,OAAOD,WAAW;gBAElB,MAAMqD,mBAAmBV,AAAS,YAATA;gBACzB,IAAIW,cAAcV;gBAClB,IAAIW,cAAc;gBAClB,IAAIF,oBAAqBV,CAAAA,AAAS,aAATA,QAAqBA,AAAS,cAATA,IAAiB,GAAI;oBACjEY,cAAc;oBACd,MAAMC,gBACJb,AAAS,aAATA,OACI,CAAC,kDAAkD,EAAEC,QAAQ,GAC7D,CAAC,+GAA+G,EAAEA,QAAQ;oBAChIU,cAAc;wBACZ,CAACC,YAAY,EAAEC;oBACjB;gBACF,OAAO,IAAIH,kBACTC,cAAc;oBACZ,CAACC,YAAY,EAAE,GAAGZ,KAAK,EAAE,EAAEC,QAAQ;gBACrC;gBAGF,IAAIa;gBAEJ,IAAIC,uBAAuB;gBAC3B,IAAIZ,KAAK,eAAe,IAAI,CAAC,SAAS,CAAC,mBAAmB,EAAE;oBAC1DhE,MAAM;oBACN,MAAM6E,OAAO,MAAM,IAAI,CAAC,SAAS,CAAC,mBAAmB;oBACrDD,uBAAuB,MAAME,kBAC3BD,MACA,KACA,OACAb,KAAK,gBAAgB;gBAEzB;gBAEA,IAAI;oBACFW,gBAAgB,MAAM,IAAI,CAAC,OAAO,CAAC,OAAO,CACxCH,aACAT,aACAC,KACAY,sBACAX;gBAEJ,EAAE,OAAOhB,OAAO;oBACd,IAAIA,iBAAiB8B,cACnBV,UAAUpB,MAAM,IAAI;oBAEtB,MAAMA;gBACR;gBAEA,MAAM,EAAE+B,IAAI,EAAE9B,KAAK,EAAE+B,OAAO,EAAEX,IAAI,EAAE,GAAGK;gBACvCN,UAAUC;gBAEV,IAAIY,eAAeF;gBACnB,IAAIT,kBAEF,IAAI,AAAgB,YAAhB,OAAOS,MACTE,eAAeF;qBACV,IAAInB,AAAS,cAATA,MAEPqB,eADEF,QAAAA,OACa,QAECA,IAAY,CAACP,YAAY;qBAEtC,IAAIO,QAAAA,MACTE,eAAe;qBACV;oBACL/D,OACE6D,MAAM,CAACP,YAAY,KAAKzC,QACxB;oBAEFkD,eAAgBF,IAAY,CAACP,YAAY;gBAC3C;gBAGF,IAAIZ,AAAS,aAATA,QAAqB,CAACqB,cAAc;oBACtCnE,KAAK,KAAK,GAAGmC;oBACbnC,KAAK,OAAO,GAAGkE;oBACf,MAAM,IAAIvB,MAAM,CAAC,kBAAkB,EAAEuB,SAAS;gBAChD;gBAEA,OAAO;oBACL,QAAQC;oBACR,KAAKd;oBACLlB;oBACA+B;gBACF;YACF;QACF;QAEA,OAAOf;IACT;IACA,MAAM,yBACJL,IAA0D,EAC1DC,MAA2B,EAC3BC,WAAyB,EACzBC,GAA0B,EAC1BC,gBAAoC,EACP;QAC7B,MAAMpD,UAAU,IAAI,CAAC,sBAAsB,CACzCC,aACE+C,MACA,AAAkB,YAAlB,OAAOC,SAAsBA,SAASjB,KAAK,SAAS,CAACiB;QAIzD,MAAMI,YAAY,MAAM,IAAI,CAAC,mBAAmB,CAC9CL,MACAC,QACAC,aACAC,KACAC;QAGF,MAAM7C,SAASP,QAAQ,SAAS;QAChC,MAAMS,SAAS,MAAMT,QAAQ,YAAY,CAACqD;QAE1C,IAAI,CAAC5C,QACH,MAAM,IAAIoC,MACR;QAIJ,MAAM,EAAEnC,MAAM,EAAE0D,OAAO,EAAE,GAAG3D;QAE5B,OAAO;YACLC;YACA0D;YACA7D;QACF;IACF;IAEQ,UAAU+D,MAAc,EAA4C;QAC1E,OAAO;YACL,MAAM;YACN,OAAO;gBACLA;YACF;QACF;IACF;IAEA,MAAM,aAAaA,MAAc,EAAEC,YAA0B,EAAE;QAC7D,OAAO,IAAI,CAAC,WAAW,CAAC,eAAe,CAAC;YACtCD;QACF;IACF;IAEA,MAAM,QACJE,SAAsB,EACtBrB,GAA+B,EAC/BD,WAAyB,EACO;QAChC,MAAM,EAAEuB,UAAU,EAAErB,gBAAgB,EAAE,GAAGsB,YAAYF;QAErD,MAAMG,cAAc,CAAC,SAAS,EAAEF,YAAY;QAC5C,MAAMzE,UAAU,IAAI,CAAC,sBAAsB,CACzCC,aAAa,WAAW0E;QAE1B,MAAMpE,SAASP,QAAQ,SAAS;QAChC,MAAM,EAAE4E,SAAS,EAAEC,eAAe,EAAE,GAAG1B;QAEvC7C,OAAOkE,WAAW;QAClBlE,OAAOsE,WAAW;QAClBtE,OAAOuE,iBAAiB;QAExBvE,OACEuE,mBAAmBD,WACnB,CAAC,iGAAiG,EAAEC,gBAAgB,aAAa,EAAED,UAAU,CAAC,CAAC;QAGjJ,MAAME,mBAAmBxD,KAAK,GAAG;QACjC,IAAIyD,iBAAiBD;QACrB,IAAIE,eAAe;QAEnB,MAAOD,iBAAiBD,oBAAoBF,UAAW;YACrD,MAAMK,oBAAoB3D,KAAK,GAAG;YAClCyD,iBAAiBE;YACjB,MAAM5B,YAAY,MAAM,IAAI,CAAC,mBAAmB,CAC9C,WACAoB,YACAvB,aACA/B,QACAiC;YAGF,MAAM3C,SAAU,MAAMT,QAAQ,YAAY,CAACqD;YAO3C,IAAI5C,QAAQ,QACV,OAAO;gBACL,QAAQU;gBACRZ;YACF;YAGFyE,eACEvE,QAAQ,WACP,CAACA,UAAU,CAAC,0BAA0B,EAAEgE,YAAY,IACrD,CAAC,0CAA0C,EAAEA,YAAY;YAC3D,MAAMS,MAAM5D,KAAK,GAAG;YACpB,IAAI4D,MAAMD,oBAAoBJ,iBAAiB;gBAC7C,MAAMnC,gBAAgBmC,kBAAmBK,CAAAA,MAAMD,iBAAgB;gBAC/D,MAAME,YAAY,IAAI,CAAC,WAAW,CAAC,eAAe,CAAC;oBACjD,QAAQzC;gBACV;gBACA,MAAM1C,QAAQ,MAAM,CAACmF;YACvB;QACF;QAEA,OAAOnF,QAAQ,eAAe,CAAC,CAAC,iBAAiB,EAAEgF,cAAc;IACnE;IAtjBA,YACEI,iBAAoC,EACpCC,OAAgB,EAChBC,IAKC,CACD;QA9BF;QAEA;QAEA;QAEA,uBAAiB,eAAjB;QAEA,uBAAQ,uBAAR;QAEA;QAEA,uBAAiB,SAAjB;QAEA;QAiBE,IAAI,CAAC,SAAS,GAAGF;QACjB,IAAI,CAAC,OAAO,GAAGC;QACf,IAAI,CAAC,SAAS,GAAGC,KAAK,SAAS;QAC/B,IAAI,CAAC,mBAAmB,GAAGA,MAAM;QACjC,IAAI,CAAC,oBAAoB,GAAGA,KAAK,oBAAoB;QACrD,IAAI,CAAC,KAAK,GAAGA,KAAK,KAAK;QACvB,IAAI,CAAC,mBAAmB,GAAG,IAAIC;QAC/B,IAAI,CAAC,WAAW,GAAG,IAAIC,YAAY;YACjCJ;YACAC;YACA,WAAWC,KAAK,SAAS;QAC3B;IACF;AAiiBF"}
@@ -100,7 +100,7 @@ async function matchElementFromCache(context, cacheEntry, cachePrompt, cacheable
100
100
  return;
101
101
  }
102
102
  }
103
- const getMidsceneVersion = ()=>"1.0.1-beta-20251208031823.0";
103
+ const getMidsceneVersion = ()=>"1.0.1-beta-20251208033501.0";
104
104
  const parsePrompt = (prompt)=>{
105
105
  if ('string' == typeof prompt) return {
106
106
  textPrompt: prompt,
@@ -134,7 +134,7 @@ Please tell what the next one action is (or null if no action should be done) to
134
134
  - Make sure the previous actions are completed successfully before performing the next step
135
135
  - If there are some error messages reported by the previous actions, don't give up, try parse a new action to recover. If the error persists for more than 5 times, you should think this is an error and set the "error" field to the error message.
136
136
  - If there is nothing to do but waiting, set the "sleep" field to the positive waiting time in milliseconds and null for the "action" field.
137
- - Assertions are also important steps. When getting the assertion instruction, a solid conclusion is required. You should explicitly state your conclusion in the "log" field like this: "Assert: <condition>. I can see <...>, and I think <...>, so the assertion is <truthy / falsy>". If the assertion is falsy, think this an fatal error and set the reason into the "error" field.
137
+ - When the next step is to assert something, this is a very important step, you should think about it carefully and give a solid result. Write your result in the "log" field like this: "Assert: <condition>. I think <...>, so the result is <true / false>". You don't need to give the next one action when you are asserting something. If the assertion result is false, think this an fatal error and set the reason into the "error" field. If the assertion result is true, you can continue to the next step.
138
138
 
139
139
  ## Supporting actions
140
140
  ${actionList}
@@ -149,30 +149,105 @@ Return in JSON format:
149
149
  ${commonOutputFields}
150
150
  "action":
151
151
  {
152
- "type": string, // the type of the action
153
- "param"?: { // The parameter of the action, if any
154
- // k-v style parameter fields
155
- },
152
+ // one of the supporting actions
156
153
  } | null,
157
154
  ,
158
155
  "sleep"?: number, // The sleep time after the action, in milliseconds.
159
156
  }
160
-
161
- For example, if the instruction is to login and the form has already been filled, this is a valid return value:
162
-
163
- {
164
- "log": "Click the login button",
165
- "more_actions_needed_by_instruction": false,
166
- "action": {
167
- "type": "Tap",
168
- "param": {
169
- "locate": {
170
- "prompt": "The login button"${vlMode ? ', "bbox": [100, 200, 300, 400]' : ''}
171
- }
172
- }
173
- }
174
157
  `;
175
158
  }
176
- export { descriptionForAction, systemPromptToTaskPlanning };
159
+ const planSchema = {
160
+ type: 'json_schema',
161
+ json_schema: {
162
+ name: 'action_items',
163
+ strict: false,
164
+ schema: {
165
+ type: 'object',
166
+ strict: false,
167
+ properties: {
168
+ actions: {
169
+ type: 'array',
170
+ items: {
171
+ type: 'object',
172
+ strict: false,
173
+ properties: {
174
+ thought: {
175
+ type: 'string',
176
+ description: 'Reasons for generating this task, and why this task is feasible on this page'
177
+ },
178
+ type: {
179
+ type: 'string',
180
+ description: 'Type of action'
181
+ },
182
+ param: {
183
+ anyOf: [
184
+ {
185
+ type: 'null'
186
+ },
187
+ {
188
+ type: 'object',
189
+ additionalProperties: true
190
+ }
191
+ ],
192
+ description: 'Parameter of the action'
193
+ },
194
+ locate: {
195
+ type: [
196
+ 'object',
197
+ 'null'
198
+ ],
199
+ properties: {
200
+ id: {
201
+ type: 'string'
202
+ },
203
+ prompt: {
204
+ type: 'string'
205
+ }
206
+ },
207
+ required: [
208
+ 'id',
209
+ 'prompt'
210
+ ],
211
+ additionalProperties: false,
212
+ description: 'Location information for the target element'
213
+ }
214
+ },
215
+ required: [
216
+ 'thought',
217
+ 'type',
218
+ 'param',
219
+ 'locate'
220
+ ],
221
+ additionalProperties: false
222
+ },
223
+ description: 'List of actions to be performed'
224
+ },
225
+ more_actions_needed_by_instruction: {
226
+ type: 'boolean',
227
+ description: 'If all the actions described in the instruction have been covered by this action and logs, set this field to false.'
228
+ },
229
+ log: {
230
+ type: 'string',
231
+ description: 'Log what these planned actions do. Do not include further actions that have not been planned.'
232
+ },
233
+ error: {
234
+ type: [
235
+ 'string',
236
+ 'null'
237
+ ],
238
+ description: 'Error messages about unexpected situations'
239
+ }
240
+ },
241
+ required: [
242
+ 'actions',
243
+ 'more_actions_needed_by_instruction',
244
+ 'log',
245
+ 'error'
246
+ ],
247
+ additionalProperties: false
248
+ }
249
+ }
250
+ };
251
+ export { descriptionForAction, planSchema, systemPromptToTaskPlanning };
177
252
 
178
253
  //# sourceMappingURL=llm-planning.mjs.map