@midscene/core 0.28.2-beta-20250910072710.0 → 0.28.2
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.
- package/dist/es/agent/agent.mjs +17 -27
- package/dist/es/agent/agent.mjs.map +1 -1
- package/dist/es/agent/tasks.mjs +31 -43
- package/dist/es/agent/tasks.mjs.map +1 -1
- package/dist/es/agent/utils.mjs +3 -4
- package/dist/es/agent/utils.mjs.map +1 -1
- package/dist/es/ai-model/common.mjs +26 -16
- package/dist/es/ai-model/common.mjs.map +1 -1
- package/dist/es/ai-model/index.mjs +3 -3
- package/dist/es/ai-model/inspect.mjs +34 -26
- package/dist/es/ai-model/inspect.mjs.map +1 -1
- package/dist/es/ai-model/llm-planning.mjs +15 -14
- package/dist/es/ai-model/llm-planning.mjs.map +1 -1
- package/dist/es/ai-model/prompt/common.mjs.map +1 -1
- package/dist/es/ai-model/prompt/llm-locator.mjs.map +1 -1
- package/dist/es/ai-model/prompt/llm-planning.mjs.map +1 -1
- package/dist/es/ai-model/prompt/llm-section-locator.mjs.map +1 -1
- package/dist/es/ai-model/prompt/playwright-generator.mjs +12 -6
- package/dist/es/ai-model/prompt/playwright-generator.mjs.map +1 -1
- package/dist/es/ai-model/prompt/util.mjs +3 -3
- package/dist/es/ai-model/prompt/util.mjs.map +1 -1
- package/dist/es/ai-model/prompt/yaml-generator.mjs +12 -6
- package/dist/es/ai-model/prompt/yaml-generator.mjs.map +1 -1
- package/dist/es/ai-model/service-caller/index.mjs +23 -28
- package/dist/es/ai-model/service-caller/index.mjs.map +1 -1
- package/dist/es/ai-model/ui-tars-planning.mjs +10 -10
- package/dist/es/ai-model/ui-tars-planning.mjs.map +1 -1
- package/dist/es/insight/index.mjs +26 -20
- package/dist/es/insight/index.mjs.map +1 -1
- package/dist/es/types.mjs.map +1 -1
- package/dist/es/utils.mjs +5 -4
- package/dist/es/utils.mjs.map +1 -1
- package/dist/lib/agent/agent.js +16 -26
- package/dist/lib/agent/agent.js.map +1 -1
- package/dist/lib/agent/tasks.js +31 -43
- package/dist/lib/agent/tasks.js.map +1 -1
- package/dist/lib/agent/utils.js +3 -4
- package/dist/lib/agent/utils.js.map +1 -1
- package/dist/lib/ai-model/common.js +31 -18
- package/dist/lib/ai-model/common.js.map +1 -1
- package/dist/lib/ai-model/index.js +15 -12
- package/dist/lib/ai-model/inspect.js +32 -24
- package/dist/lib/ai-model/inspect.js.map +1 -1
- package/dist/lib/ai-model/llm-planning.js +14 -13
- package/dist/lib/ai-model/llm-planning.js.map +1 -1
- package/dist/lib/ai-model/prompt/common.js.map +1 -1
- package/dist/lib/ai-model/prompt/llm-locator.js.map +1 -1
- package/dist/lib/ai-model/prompt/llm-planning.js.map +1 -1
- package/dist/lib/ai-model/prompt/llm-section-locator.js.map +1 -1
- package/dist/lib/ai-model/prompt/playwright-generator.js +11 -5
- package/dist/lib/ai-model/prompt/playwright-generator.js.map +1 -1
- package/dist/lib/ai-model/prompt/util.js +3 -3
- package/dist/lib/ai-model/prompt/util.js.map +1 -1
- package/dist/lib/ai-model/prompt/yaml-generator.js +11 -5
- package/dist/lib/ai-model/prompt/yaml-generator.js.map +1 -1
- package/dist/lib/ai-model/service-caller/index.js +31 -36
- package/dist/lib/ai-model/service-caller/index.js.map +1 -1
- package/dist/lib/ai-model/ui-tars-planning.js +8 -8
- package/dist/lib/ai-model/ui-tars-planning.js.map +1 -1
- package/dist/lib/insight/index.js +23 -17
- package/dist/lib/insight/index.js.map +1 -1
- package/dist/lib/types.js.map +1 -1
- package/dist/lib/utils.js +4 -3
- package/dist/lib/utils.js.map +1 -1
- package/dist/types/agent/agent.d.ts +0 -2
- package/dist/types/agent/tasks.d.ts +7 -8
- package/dist/types/agent/utils.d.ts +1 -3
- package/dist/types/ai-model/common.d.ts +11 -7
- package/dist/types/ai-model/index.d.ts +2 -2
- package/dist/types/ai-model/inspect.d.ts +6 -7
- package/dist/types/ai-model/llm-planning.d.ts +2 -2
- package/dist/types/ai-model/prompt/common.d.ts +2 -2
- package/dist/types/ai-model/prompt/llm-locator.d.ts +2 -2
- package/dist/types/ai-model/prompt/llm-planning.d.ts +3 -3
- package/dist/types/ai-model/prompt/llm-section-locator.d.ts +2 -2
- package/dist/types/ai-model/prompt/playwright-generator.d.ts +2 -3
- package/dist/types/ai-model/prompt/util.d.ts +2 -3
- package/dist/types/ai-model/prompt/yaml-generator.d.ts +2 -3
- package/dist/types/ai-model/service-caller/index.d.ts +5 -5
- package/dist/types/ai-model/ui-tars-planning.d.ts +3 -3
- package/dist/types/device/index.d.ts +2 -2
- package/dist/types/insight/index.d.ts +7 -12
- package/dist/types/types.d.ts +9 -0
- package/dist/types/utils.d.ts +1 -2
- package/package.json +3 -3
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"agent/utils.mjs","sources":["webpack://@midscene/core/./src/agent/utils.ts"],"sourcesContent":["import { elementByPositionWithElementInfo } from '@/ai-model';\nimport type { AbstractInterface } from '@/device';\nimport type {\n BaseElement,\n ElementTreeNode,\n ExecutionDump,\n ExecutionTask,\n ExecutorContext,\n PlanningLocateParam,\n TMultimodalPrompt,\n TUserPrompt,\n UIContext,\n} from '@/index';\nimport { uploadTestInfoToServer } from '@/utils';\nimport {\n MIDSCENE_REPORT_TAG_NAME,\n globalConfigManager,\n} from '@midscene/shared/env';\nimport {\n generateElementByPosition,\n getNodeFromCacheList,\n} from '@midscene/shared/extractor';\nimport { resizeImgBase64 } from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { _keyDefinitions } from '@midscene/shared/us-keyboard-layout';\nimport { assert, logMsg, uuid } from '@midscene/shared/utils';\nimport dayjs from 'dayjs';\nimport { debug as cacheDebug } from './task-cache';\nimport type { TaskExecutor } from './tasks';\n\nconst debugProfile = getDebug('web:tool:profile');\n\nexport async function commonContextParser(\n interfaceInstance: AbstractInterface,\n _opt: { uploadServerUrl?: string },\n): Promise<UIContext> {\n assert(interfaceInstance, 'interfaceInstance is required');\n\n debugProfile('Getting interface description');\n const description = interfaceInstance.describe?.() || '';\n debugProfile('Interface description end');\n\n debugProfile('Uploading test info to server');\n uploadTestInfoToServer({\n testUrl: description,\n serverUrl: _opt.uploadServerUrl,\n });\n debugProfile('UploadTestInfoToServer end');\n\n let screenshotBase64 = await interfaceInstance.screenshotBase64();\n assert(screenshotBase64!, 'screenshotBase64 is required');\n\n const size = await interfaceInstance.size();\n debugProfile(`size: ${size.width}x${size.height} dpr: ${size.dpr}`);\n\n if (size.dpr && size.dpr > 1) {\n debugProfile('Resizing screenshot for high DPR display');\n screenshotBase64 = await resizeImgBase64(screenshotBase64, {\n width: size.width,\n height: size.height,\n });\n debugProfile('ResizeImgBase64 end');\n }\n\n return {\n tree: {\n node: null,\n children: [],\n },\n size,\n screenshotBase64: screenshotBase64!,\n };\n}\n\nexport function getReportFileName(tag = 'web') {\n const reportTagName = globalConfigManager.getEnvConfigValue(\n MIDSCENE_REPORT_TAG_NAME,\n );\n const dateTimeInFileName = dayjs().format('YYYY-MM-DD_HH-mm-ss');\n // ensure uniqueness at the same time\n const uniqueId = uuid().substring(0, 8);\n return `${reportTagName || tag}-${dateTimeInFileName}-${uniqueId}`;\n}\n\nexport function printReportMsg(filepath: string) {\n logMsg(`Midscene - report file updated: ${filepath}`);\n}\n\n/**\n * Get the current execution file name\n * @returns The name of the current execution file\n */\nexport function getCurrentExecutionFile(trace?: string): string | false {\n const error = new Error();\n const stackTrace = trace || error.stack;\n const pkgDir = process.cwd() || '';\n if (stackTrace) {\n const stackLines = stackTrace.split('\\n');\n for (const line of stackLines) {\n if (\n line.includes('.spec.') ||\n line.includes('.test.') ||\n line.includes('.ts') ||\n line.includes('.js')\n ) {\n const match = line.match(/(?:at\\s+)?(.*?\\.(?:spec|test)\\.[jt]s)/);\n if (match?.[1]) {\n const targetFileName = match[1]\n .replace(pkgDir, '')\n .trim()\n .replace('at ', '');\n return targetFileName;\n }\n }\n }\n }\n return false;\n}\n\nconst testFileIndex = new Map<string, number>();\n\nexport function generateCacheId(fileName?: string): string {\n let taskFile = fileName || getCurrentExecutionFile();\n if (!taskFile) {\n taskFile = uuid();\n console.warn(\n 'Midscene - using random UUID for cache id. Cache may be invalid.',\n );\n }\n\n if (testFileIndex.has(taskFile)) {\n const currentIndex = testFileIndex.get(taskFile);\n if (currentIndex !== undefined) {\n testFileIndex.set(taskFile, currentIndex + 1);\n }\n } else {\n testFileIndex.set(taskFile, 1);\n }\n return `${taskFile}-${testFileIndex.get(taskFile)}`;\n}\n\nexport function matchElementFromPlan(\n planLocateParam: PlanningLocateParam,\n tree: ElementTreeNode<BaseElement>,\n) {\n if (!planLocateParam) {\n return undefined;\n }\n if (planLocateParam.id) {\n return getNodeFromCacheList(planLocateParam.id);\n }\n\n if (planLocateParam.bbox) {\n const centerPosition = {\n x: Math.floor((planLocateParam.bbox[0] + planLocateParam.bbox[2]) / 2),\n y: Math.floor((planLocateParam.bbox[1] + planLocateParam.bbox[3]) / 2),\n };\n let element = elementByPositionWithElementInfo(tree, centerPosition);\n\n if (!element) {\n element = generateElementByPosition(centerPosition) as BaseElement;\n }\n\n return element;\n }\n\n return undefined;\n}\n\nexport async function matchElementFromCache(\n taskExecutor: TaskExecutor,\n xpaths: string[] | undefined,\n cachePrompt: TUserPrompt,\n cacheable: boolean | undefined,\n) {\n try {\n if (\n xpaths?.length &&\n taskExecutor.taskCache?.isCacheResultUsed &&\n cacheable !== false &&\n (taskExecutor.interface as any).getElementInfoByXpath\n ) {\n // hit cache, use new id\n for (let i = 0; i < xpaths.length; i++) {\n const element = await (\n taskExecutor.interface as any\n ).getElementInfoByXpath(xpaths[i]);\n\n if (element?.id) {\n cacheDebug('cache hit, prompt: %s', cachePrompt);\n cacheDebug(\n 'found a new element with same xpath, xpath: %s, id: %s',\n xpaths[i],\n element?.id,\n );\n return element;\n }\n }\n }\n } catch (error) {\n cacheDebug('get element info by xpath error: ', error);\n }\n}\n\nexport function trimContextByViewport(execution: ExecutionDump) {\n function filterVisibleTree(\n node: ElementTreeNode<BaseElement>,\n ): ElementTreeNode<BaseElement> | null {\n if (!node) return null;\n\n // recursively process all children\n const filteredChildren = Array.isArray(node.children)\n ? (node.children\n .map(filterVisibleTree)\n .filter((child) => child !== null) as ElementTreeNode<BaseElement>[])\n : [];\n\n // if the current node is visible, keep it and the filtered children\n if (node.node && node.node.isVisible === true) {\n return {\n ...node,\n children: filteredChildren,\n };\n }\n\n // if the current node is invisible, but has visible children, create an empty node to include these children\n if (filteredChildren.length > 0) {\n return {\n node: null,\n children: filteredChildren,\n };\n }\n\n // if the current node is invisible and has no visible children, return null\n return null;\n }\n\n return {\n ...execution,\n tasks: Array.isArray(execution.tasks)\n ? execution.tasks.map((task: ExecutionTask) => {\n const newTask = { ...task };\n if (task.uiContext?.tree) {\n newTask.uiContext = {\n ...task.uiContext,\n tree: filterVisibleTree(task.uiContext.tree) || {\n node: null,\n children: [],\n },\n };\n }\n return newTask;\n })\n : execution.tasks,\n };\n}\n\ndeclare const __VERSION__: string | undefined;\n\nexport const getMidsceneVersion = (): string => {\n if (typeof __VERSION__ !== 'undefined') {\n return __VERSION__;\n } else if (\n process.env.__VERSION__ &&\n process.env.__VERSION__ !== 'undefined'\n ) {\n return process.env.__VERSION__;\n }\n throw new Error('__VERSION__ inject failed during build');\n};\n\nexport const parsePrompt = (\n prompt: TUserPrompt,\n): {\n textPrompt: string;\n multimodalPrompt?: TMultimodalPrompt;\n} => {\n if (typeof prompt === 'string') {\n return {\n textPrompt: prompt,\n multimodalPrompt: undefined,\n };\n }\n return {\n textPrompt: prompt.prompt,\n multimodalPrompt: prompt.images\n ? {\n images: prompt.images,\n convertHttpImage2Base64: !!prompt.convertHttpImage2Base64,\n }\n : undefined,\n };\n};\n"],"names":["debugProfile","getDebug","commonContextParser","interfaceInstance","_opt","assert","description","uploadTestInfoToServer","screenshotBase64","size","resizeImgBase64","getReportFileName","tag","reportTagName","globalConfigManager","MIDSCENE_REPORT_TAG_NAME","dateTimeInFileName","dayjs","uniqueId","uuid","printReportMsg","filepath","logMsg","getCurrentExecutionFile","trace","error","Error","stackTrace","pkgDir","process","stackLines","line","match","targetFileName","testFileIndex","Map","generateCacheId","fileName","taskFile","console","currentIndex","undefined","matchElementFromPlan","planLocateParam","tree","getNodeFromCacheList","centerPosition","Math","element","elementByPositionWithElementInfo","generateElementByPosition","matchElementFromCache","taskExecutor","xpaths","cachePrompt","cacheable","_taskExecutor_taskCache","i","cacheDebug","trimContextByViewport","execution","filterVisibleTree","node","filteredChildren","Array","child","task","_task_uiContext","newTask","getMidsceneVersion","__VERSION__","parsePrompt","prompt"],"mappings":";;;;;;;;;AA8BA,MAAMA,eAAeC,SAAS;AAEvB,eAAeC,oBACpBC,iBAAoC,EACpCC,IAAkC;QAKdD;IAHpBE,OAAOF,mBAAmB;IAE1BH,aAAa;IACb,MAAMM,cAAcH,AAAAA,SAAAA,CAAAA,8BAAAA,kBAAkB,QAAQ,AAAD,IAAzBA,KAAAA,IAAAA,4BAAAA,IAAAA,CAAAA,kBAAiB,KAAiB;IACtDH,aAAa;IAEbA,aAAa;IACbO,uBAAuB;QACrB,SAASD;QACT,WAAWF,KAAK,eAAe;IACjC;IACAJ,aAAa;IAEb,IAAIQ,mBAAmB,MAAML,kBAAkB,gBAAgB;IAC/DE,OAAOG,kBAAmB;IAE1B,MAAMC,OAAO,MAAMN,kBAAkB,IAAI;IACzCH,aAAa,CAAC,MAAM,EAAES,KAAK,KAAK,CAAC,CAAC,EAAEA,KAAK,MAAM,CAAC,MAAM,EAAEA,KAAK,GAAG,EAAE;IAElE,IAAIA,KAAK,GAAG,IAAIA,KAAK,GAAG,GAAG,GAAG;QAC5BT,aAAa;QACbQ,mBAAmB,MAAME,gBAAgBF,kBAAkB;YACzD,OAAOC,KAAK,KAAK;YACjB,QAAQA,KAAK,MAAM;QACrB;QACAT,aAAa;IACf;IAEA,OAAO;QACL,MAAM;YACJ,MAAM;YACN,UAAU,EAAE;QACd;QACAS;QACA,kBAAkBD;IACpB;AACF;AAEO,SAASG,kBAAkBC,MAAM,KAAK;IAC3C,MAAMC,gBAAgBC,oBAAoB,iBAAiB,CACzDC;IAEF,MAAMC,qBAAqBC,QAAQ,MAAM,CAAC;IAE1C,MAAMC,WAAWC,OAAO,SAAS,CAAC,GAAG;IACrC,OAAO,GAAGN,iBAAiBD,IAAI,CAAC,EAAEI,mBAAmB,CAAC,EAAEE,UAAU;AACpE;AAEO,SAASE,eAAeC,QAAgB;IAC7CC,OAAO,CAAC,gCAAgC,EAAED,UAAU;AACtD;AAMO,SAASE,wBAAwBC,KAAc;IACpD,MAAMC,QAAQ,IAAIC;IAClB,MAAMC,aAAaH,SAASC,MAAM,KAAK;IACvC,MAAMG,SAASC,QAAQ,GAAG,MAAM;IAChC,IAAIF,YAAY;QACd,MAAMG,aAAaH,WAAW,KAAK,CAAC;QACpC,KAAK,MAAMI,QAAQD,WACjB,IACEC,KAAK,QAAQ,CAAC,aACdA,KAAK,QAAQ,CAAC,aACdA,KAAK,QAAQ,CAAC,UACdA,KAAK,QAAQ,CAAC,QACd;YACA,MAAMC,QAAQD,KAAK,KAAK,CAAC;YACzB,IAAIC,QAAAA,QAAAA,KAAAA,IAAAA,KAAO,CAAC,EAAE,EAAE;gBACd,MAAMC,iBAAiBD,KAAK,CAAC,EAAE,CAC5B,OAAO,CAACJ,QAAQ,IAChB,IAAI,GACJ,OAAO,CAAC,OAAO;gBAClB,OAAOK;YACT;QACF;IAEJ;IACA,OAAO;AACT;AAEA,MAAMC,gBAAgB,IAAIC;AAEnB,SAASC,gBAAgBC,QAAiB;IAC/C,IAAIC,WAAWD,YAAYd;IAC3B,IAAI,CAACe,UAAU;QACbA,WAAWnB;QACXoB,QAAQ,IAAI,CACV;IAEJ;IAEA,IAAIL,cAAc,GAAG,CAACI,WAAW;QAC/B,MAAME,eAAeN,cAAc,GAAG,CAACI;QACvC,IAAIE,AAAiBC,WAAjBD,cACFN,cAAc,GAAG,CAACI,UAAUE,eAAe;IAE/C,OACEN,cAAc,GAAG,CAACI,UAAU;IAE9B,OAAO,GAAGA,SAAS,CAAC,EAAEJ,cAAc,GAAG,CAACI,WAAW;AACrD;AAEO,SAASI,qBACdC,eAAoC,EACpCC,IAAkC;IAElC,IAAI,CAACD,iBACH;IAEF,IAAIA,gBAAgB,EAAE,EACpB,OAAOE,qBAAqBF,gBAAgB,EAAE;IAGhD,IAAIA,gBAAgB,IAAI,EAAE;QACxB,MAAMG,iBAAiB;YACrB,GAAGC,KAAK,KAAK,CAAEJ,AAAAA,CAAAA,gBAAgB,IAAI,CAAC,EAAE,GAAGA,gBAAgB,IAAI,CAAC,EAAC,IAAK;YACpE,GAAGI,KAAK,KAAK,CAAEJ,AAAAA,CAAAA,gBAAgB,IAAI,CAAC,EAAE,GAAGA,gBAAgB,IAAI,CAAC,EAAC,IAAK;QACtE;QACA,IAAIK,UAAUC,iCAAiCL,MAAME;QAErD,IAAI,CAACE,SACHA,UAAUE,0BAA0BJ;QAGtC,OAAOE;IACT;AAGF;AAEO,eAAeG,sBACpBC,YAA0B,EAC1BC,MAA4B,EAC5BC,WAAwB,EACxBC,SAA8B;IAE9B,IAAI;YAGAC;QAFF,IACEH,AAAAA,CAAAA,QAAAA,SAAAA,KAAAA,IAAAA,OAAQ,MAAM,AAAD,KAAC,SACdG,CAAAA,0BAAAA,aAAa,SAAS,AAAD,IAArBA,KAAAA,IAAAA,wBAAwB,iBAAiB,AAAD,KACxCD,AAAc,UAAdA,aACCH,aAAa,SAAS,CAAS,qBAAqB,EAGrD,IAAK,IAAIK,IAAI,GAAGA,IAAIJ,OAAO,MAAM,EAAEI,IAAK;YACtC,MAAMT,UAAU,MACdI,aAAa,SAAS,CACtB,qBAAqB,CAACC,MAAM,CAACI,EAAE;YAEjC,IAAIT,QAAAA,UAAAA,KAAAA,IAAAA,QAAS,EAAE,EAAE;gBACfU,MAAW,yBAAyBJ;gBACpCI,MACE,0DACAL,MAAM,CAACI,EAAE,EACTT,QAAAA,UAAAA,KAAAA,IAAAA,QAAS,EAAE;gBAEb,OAAOA;YACT;QACF;IAEJ,EAAE,OAAOvB,OAAO;QACdiC,MAAW,qCAAqCjC;IAClD;AACF;AAEO,SAASkC,sBAAsBC,SAAwB;IAC5D,SAASC,kBACPC,IAAkC;QAElC,IAAI,CAACA,MAAM,OAAO;QAGlB,MAAMC,mBAAmBC,MAAM,OAAO,CAACF,KAAK,QAAQ,IAC/CA,KAAK,QAAQ,CACX,GAAG,CAACD,mBACJ,MAAM,CAAC,CAACI,QAAUA,AAAU,SAAVA,SACrB,EAAE;QAGN,IAAIH,KAAK,IAAI,IAAIA,AAAwB,SAAxBA,KAAK,IAAI,CAAC,SAAS,EAClC,OAAO;YACL,GAAGA,IAAI;YACP,UAAUC;QACZ;QAIF,IAAIA,iBAAiB,MAAM,GAAG,GAC5B,OAAO;YACL,MAAM;YACN,UAAUA;QACZ;QAIF,OAAO;IACT;IAEA,OAAO;QACL,GAAGH,SAAS;QACZ,OAAOI,MAAM,OAAO,CAACJ,UAAU,KAAK,IAChCA,UAAU,KAAK,CAAC,GAAG,CAAC,CAACM;gBAEfC;YADJ,MAAMC,UAAU;gBAAE,GAAGF,IAAI;YAAC;YAC1B,IAAI,QAAAC,CAAAA,kBAAAA,KAAK,SAAS,AAAD,IAAbA,KAAAA,IAAAA,gBAAgB,IAAI,EACtBC,QAAQ,SAAS,GAAG;gBAClB,GAAGF,KAAK,SAAS;gBACjB,MAAML,kBAAkBK,KAAK,SAAS,CAAC,IAAI,KAAK;oBAC9C,MAAM;oBACN,UAAU,EAAE;gBACd;YACF;YAEF,OAAOE;QACT,KACAR,UAAU,KAAK;IACrB;AACF;AAIO,MAAMS,qBAAqB,IAEvBC;AAUJ,MAAMC,cAAc,CACzBC;IAKA,IAAI,AAAkB,YAAlB,OAAOA,QACT,OAAO;QACL,YAAYA;QACZ,kBAAkB/B;IACpB;IAEF,OAAO;QACL,YAAY+B,OAAO,MAAM;QACzB,kBAAkBA,OAAO,MAAM,GAC3B;YACE,QAAQA,OAAO,MAAM;YACrB,yBAAyB,CAAC,CAACA,OAAO,uBAAuB;QAC3D,IACA/B;IACN;AACF"}
|
|
1
|
+
{"version":3,"file":"agent/utils.mjs","sources":["webpack://@midscene/core/./src/agent/utils.ts"],"sourcesContent":["import { elementByPositionWithElementInfo } from '@/ai-model';\nimport type { AbstractInterface } from '@/device';\nimport type {\n BaseElement,\n ElementTreeNode,\n ExecutionDump,\n ExecutionTask,\n ExecutorContext,\n PlanningLocateParam,\n TMultimodalPrompt,\n TUserPrompt,\n UIContext,\n} from '@/index';\nimport { uploadTestInfoToServer } from '@/utils';\nimport {\n MIDSCENE_REPORT_TAG_NAME,\n globalConfigManager,\n} from '@midscene/shared/env';\nimport {\n generateElementByPosition,\n getNodeFromCacheList,\n} from '@midscene/shared/extractor';\nimport { resizeImgBase64 } from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { _keyDefinitions } from '@midscene/shared/us-keyboard-layout';\nimport { assert, logMsg, uuid } from '@midscene/shared/utils';\nimport dayjs from 'dayjs';\nimport { debug as cacheDebug } from './task-cache';\nimport type { TaskExecutor } from './tasks';\n\nconst debugProfile = getDebug('web:tool:profile');\n\nexport async function commonContextParser(\n interfaceInstance: AbstractInterface,\n): Promise<UIContext> {\n assert(interfaceInstance, 'interfaceInstance is required');\n\n debugProfile('Getting interface description');\n const description = interfaceInstance.describe?.() || '';\n debugProfile('Interface description end');\n\n debugProfile('Uploading test info to server');\n uploadTestInfoToServer({ testUrl: description });\n debugProfile('UploadTestInfoToServer end');\n\n let screenshotBase64 = await interfaceInstance.screenshotBase64();\n assert(screenshotBase64!, 'screenshotBase64 is required');\n\n const size = await interfaceInstance.size();\n debugProfile(`size: ${size.width}x${size.height} dpr: ${size.dpr}`);\n\n if (size.dpr && size.dpr > 1) {\n debugProfile('Resizing screenshot for high DPR display');\n screenshotBase64 = await resizeImgBase64(screenshotBase64, {\n width: size.width,\n height: size.height,\n });\n debugProfile('ResizeImgBase64 end');\n }\n\n return {\n tree: {\n node: null,\n children: [],\n },\n size,\n screenshotBase64: screenshotBase64!,\n };\n}\n\nexport function getReportFileName(tag = 'web') {\n const reportTagName = globalConfigManager.getEnvConfigValue(\n MIDSCENE_REPORT_TAG_NAME,\n );\n const dateTimeInFileName = dayjs().format('YYYY-MM-DD_HH-mm-ss');\n // ensure uniqueness at the same time\n const uniqueId = uuid().substring(0, 8);\n return `${reportTagName || tag}-${dateTimeInFileName}-${uniqueId}`;\n}\n\nexport function printReportMsg(filepath: string) {\n logMsg(`Midscene - report file updated: ${filepath}`);\n}\n\n/**\n * Get the current execution file name\n * @returns The name of the current execution file\n */\nexport function getCurrentExecutionFile(trace?: string): string | false {\n const error = new Error();\n const stackTrace = trace || error.stack;\n const pkgDir = process.cwd() || '';\n if (stackTrace) {\n const stackLines = stackTrace.split('\\n');\n for (const line of stackLines) {\n if (\n line.includes('.spec.') ||\n line.includes('.test.') ||\n line.includes('.ts') ||\n line.includes('.js')\n ) {\n const match = line.match(/(?:at\\s+)?(.*?\\.(?:spec|test)\\.[jt]s)/);\n if (match?.[1]) {\n const targetFileName = match[1]\n .replace(pkgDir, '')\n .trim()\n .replace('at ', '');\n return targetFileName;\n }\n }\n }\n }\n return false;\n}\n\nconst testFileIndex = new Map<string, number>();\n\nexport function generateCacheId(fileName?: string): string {\n let taskFile = fileName || getCurrentExecutionFile();\n if (!taskFile) {\n taskFile = uuid();\n console.warn(\n 'Midscene - using random UUID for cache id. Cache may be invalid.',\n );\n }\n\n if (testFileIndex.has(taskFile)) {\n const currentIndex = testFileIndex.get(taskFile);\n if (currentIndex !== undefined) {\n testFileIndex.set(taskFile, currentIndex + 1);\n }\n } else {\n testFileIndex.set(taskFile, 1);\n }\n return `${taskFile}-${testFileIndex.get(taskFile)}`;\n}\n\nexport function matchElementFromPlan(\n planLocateParam: PlanningLocateParam,\n tree: ElementTreeNode<BaseElement>,\n) {\n if (!planLocateParam) {\n return undefined;\n }\n if (planLocateParam.id) {\n return getNodeFromCacheList(planLocateParam.id);\n }\n\n if (planLocateParam.bbox) {\n const centerPosition = {\n x: Math.floor((planLocateParam.bbox[0] + planLocateParam.bbox[2]) / 2),\n y: Math.floor((planLocateParam.bbox[1] + planLocateParam.bbox[3]) / 2),\n };\n let element = elementByPositionWithElementInfo(tree, centerPosition);\n\n if (!element) {\n element = generateElementByPosition(centerPosition) as BaseElement;\n }\n\n return element;\n }\n\n return undefined;\n}\n\nexport async function matchElementFromCache(\n taskExecutor: TaskExecutor,\n xpaths: string[] | undefined,\n cachePrompt: TUserPrompt,\n cacheable: boolean | undefined,\n) {\n try {\n if (\n xpaths?.length &&\n taskExecutor.taskCache?.isCacheResultUsed &&\n cacheable !== false &&\n (taskExecutor.interface as any).getElementInfoByXpath\n ) {\n // hit cache, use new id\n for (let i = 0; i < xpaths.length; i++) {\n const element = await (\n taskExecutor.interface as any\n ).getElementInfoByXpath(xpaths[i]);\n\n if (element?.id) {\n cacheDebug('cache hit, prompt: %s', cachePrompt);\n cacheDebug(\n 'found a new element with same xpath, xpath: %s, id: %s',\n xpaths[i],\n element?.id,\n );\n return element;\n }\n }\n }\n } catch (error) {\n cacheDebug('get element info by xpath error: ', error);\n }\n}\n\nexport function trimContextByViewport(execution: ExecutionDump) {\n function filterVisibleTree(\n node: ElementTreeNode<BaseElement>,\n ): ElementTreeNode<BaseElement> | null {\n if (!node) return null;\n\n // recursively process all children\n const filteredChildren = Array.isArray(node.children)\n ? (node.children\n .map(filterVisibleTree)\n .filter((child) => child !== null) as ElementTreeNode<BaseElement>[])\n : [];\n\n // if the current node is visible, keep it and the filtered children\n if (node.node && node.node.isVisible === true) {\n return {\n ...node,\n children: filteredChildren,\n };\n }\n\n // if the current node is invisible, but has visible children, create an empty node to include these children\n if (filteredChildren.length > 0) {\n return {\n node: null,\n children: filteredChildren,\n };\n }\n\n // if the current node is invisible and has no visible children, return null\n return null;\n }\n\n return {\n ...execution,\n tasks: Array.isArray(execution.tasks)\n ? execution.tasks.map((task: ExecutionTask) => {\n const newTask = { ...task };\n if (task.uiContext?.tree) {\n newTask.uiContext = {\n ...task.uiContext,\n tree: filterVisibleTree(task.uiContext.tree) || {\n node: null,\n children: [],\n },\n };\n }\n return newTask;\n })\n : execution.tasks,\n };\n}\n\ndeclare const __VERSION__: string | undefined;\n\nexport const getMidsceneVersion = (): string => {\n if (typeof __VERSION__ !== 'undefined') {\n return __VERSION__;\n } else if (\n process.env.__VERSION__ &&\n process.env.__VERSION__ !== 'undefined'\n ) {\n return process.env.__VERSION__;\n }\n throw new Error('__VERSION__ inject failed during build');\n};\n\nexport const parsePrompt = (\n prompt: TUserPrompt,\n): {\n textPrompt: string;\n multimodalPrompt?: TMultimodalPrompt;\n} => {\n if (typeof prompt === 'string') {\n return {\n textPrompt: prompt,\n multimodalPrompt: undefined,\n };\n }\n return {\n textPrompt: prompt.prompt,\n multimodalPrompt: prompt.images\n ? {\n images: prompt.images,\n convertHttpImage2Base64: !!prompt.convertHttpImage2Base64,\n }\n : undefined,\n };\n};\n"],"names":["debugProfile","getDebug","commonContextParser","interfaceInstance","assert","description","uploadTestInfoToServer","screenshotBase64","size","resizeImgBase64","getReportFileName","tag","reportTagName","globalConfigManager","MIDSCENE_REPORT_TAG_NAME","dateTimeInFileName","dayjs","uniqueId","uuid","printReportMsg","filepath","logMsg","getCurrentExecutionFile","trace","error","Error","stackTrace","pkgDir","process","stackLines","line","match","targetFileName","testFileIndex","Map","generateCacheId","fileName","taskFile","console","currentIndex","undefined","matchElementFromPlan","planLocateParam","tree","getNodeFromCacheList","centerPosition","Math","element","elementByPositionWithElementInfo","generateElementByPosition","matchElementFromCache","taskExecutor","xpaths","cachePrompt","cacheable","_taskExecutor_taskCache","i","cacheDebug","trimContextByViewport","execution","filterVisibleTree","node","filteredChildren","Array","child","task","_task_uiContext","newTask","getMidsceneVersion","__VERSION__","parsePrompt","prompt"],"mappings":";;;;;;;;;AA8BA,MAAMA,eAAeC,SAAS;AAEvB,eAAeC,oBACpBC,iBAAoC;QAKhBA;IAHpBC,OAAOD,mBAAmB;IAE1BH,aAAa;IACb,MAAMK,cAAcF,AAAAA,SAAAA,CAAAA,8BAAAA,kBAAkB,QAAQ,AAAD,IAAzBA,KAAAA,IAAAA,4BAAAA,IAAAA,CAAAA,kBAAiB,KAAiB;IACtDH,aAAa;IAEbA,aAAa;IACbM,uBAAuB;QAAE,SAASD;IAAY;IAC9CL,aAAa;IAEb,IAAIO,mBAAmB,MAAMJ,kBAAkB,gBAAgB;IAC/DC,OAAOG,kBAAmB;IAE1B,MAAMC,OAAO,MAAML,kBAAkB,IAAI;IACzCH,aAAa,CAAC,MAAM,EAAEQ,KAAK,KAAK,CAAC,CAAC,EAAEA,KAAK,MAAM,CAAC,MAAM,EAAEA,KAAK,GAAG,EAAE;IAElE,IAAIA,KAAK,GAAG,IAAIA,KAAK,GAAG,GAAG,GAAG;QAC5BR,aAAa;QACbO,mBAAmB,MAAME,gBAAgBF,kBAAkB;YACzD,OAAOC,KAAK,KAAK;YACjB,QAAQA,KAAK,MAAM;QACrB;QACAR,aAAa;IACf;IAEA,OAAO;QACL,MAAM;YACJ,MAAM;YACN,UAAU,EAAE;QACd;QACAQ;QACA,kBAAkBD;IACpB;AACF;AAEO,SAASG,kBAAkBC,MAAM,KAAK;IAC3C,MAAMC,gBAAgBC,oBAAoB,iBAAiB,CACzDC;IAEF,MAAMC,qBAAqBC,QAAQ,MAAM,CAAC;IAE1C,MAAMC,WAAWC,OAAO,SAAS,CAAC,GAAG;IACrC,OAAO,GAAGN,iBAAiBD,IAAI,CAAC,EAAEI,mBAAmB,CAAC,EAAEE,UAAU;AACpE;AAEO,SAASE,eAAeC,QAAgB;IAC7CC,OAAO,CAAC,gCAAgC,EAAED,UAAU;AACtD;AAMO,SAASE,wBAAwBC,KAAc;IACpD,MAAMC,QAAQ,IAAIC;IAClB,MAAMC,aAAaH,SAASC,MAAM,KAAK;IACvC,MAAMG,SAASC,QAAQ,GAAG,MAAM;IAChC,IAAIF,YAAY;QACd,MAAMG,aAAaH,WAAW,KAAK,CAAC;QACpC,KAAK,MAAMI,QAAQD,WACjB,IACEC,KAAK,QAAQ,CAAC,aACdA,KAAK,QAAQ,CAAC,aACdA,KAAK,QAAQ,CAAC,UACdA,KAAK,QAAQ,CAAC,QACd;YACA,MAAMC,QAAQD,KAAK,KAAK,CAAC;YACzB,IAAIC,QAAAA,QAAAA,KAAAA,IAAAA,KAAO,CAAC,EAAE,EAAE;gBACd,MAAMC,iBAAiBD,KAAK,CAAC,EAAE,CAC5B,OAAO,CAACJ,QAAQ,IAChB,IAAI,GACJ,OAAO,CAAC,OAAO;gBAClB,OAAOK;YACT;QACF;IAEJ;IACA,OAAO;AACT;AAEA,MAAMC,gBAAgB,IAAIC;AAEnB,SAASC,gBAAgBC,QAAiB;IAC/C,IAAIC,WAAWD,YAAYd;IAC3B,IAAI,CAACe,UAAU;QACbA,WAAWnB;QACXoB,QAAQ,IAAI,CACV;IAEJ;IAEA,IAAIL,cAAc,GAAG,CAACI,WAAW;QAC/B,MAAME,eAAeN,cAAc,GAAG,CAACI;QACvC,IAAIE,AAAiBC,WAAjBD,cACFN,cAAc,GAAG,CAACI,UAAUE,eAAe;IAE/C,OACEN,cAAc,GAAG,CAACI,UAAU;IAE9B,OAAO,GAAGA,SAAS,CAAC,EAAEJ,cAAc,GAAG,CAACI,WAAW;AACrD;AAEO,SAASI,qBACdC,eAAoC,EACpCC,IAAkC;IAElC,IAAI,CAACD,iBACH;IAEF,IAAIA,gBAAgB,EAAE,EACpB,OAAOE,qBAAqBF,gBAAgB,EAAE;IAGhD,IAAIA,gBAAgB,IAAI,EAAE;QACxB,MAAMG,iBAAiB;YACrB,GAAGC,KAAK,KAAK,CAAEJ,AAAAA,CAAAA,gBAAgB,IAAI,CAAC,EAAE,GAAGA,gBAAgB,IAAI,CAAC,EAAC,IAAK;YACpE,GAAGI,KAAK,KAAK,CAAEJ,AAAAA,CAAAA,gBAAgB,IAAI,CAAC,EAAE,GAAGA,gBAAgB,IAAI,CAAC,EAAC,IAAK;QACtE;QACA,IAAIK,UAAUC,iCAAiCL,MAAME;QAErD,IAAI,CAACE,SACHA,UAAUE,0BAA0BJ;QAGtC,OAAOE;IACT;AAGF;AAEO,eAAeG,sBACpBC,YAA0B,EAC1BC,MAA4B,EAC5BC,WAAwB,EACxBC,SAA8B;IAE9B,IAAI;YAGAC;QAFF,IACEH,AAAAA,CAAAA,QAAAA,SAAAA,KAAAA,IAAAA,OAAQ,MAAM,AAAD,KAAC,SACdG,CAAAA,0BAAAA,aAAa,SAAS,AAAD,IAArBA,KAAAA,IAAAA,wBAAwB,iBAAiB,AAAD,KACxCD,AAAc,UAAdA,aACCH,aAAa,SAAS,CAAS,qBAAqB,EAGrD,IAAK,IAAIK,IAAI,GAAGA,IAAIJ,OAAO,MAAM,EAAEI,IAAK;YACtC,MAAMT,UAAU,MACdI,aAAa,SAAS,CACtB,qBAAqB,CAACC,MAAM,CAACI,EAAE;YAEjC,IAAIT,QAAAA,UAAAA,KAAAA,IAAAA,QAAS,EAAE,EAAE;gBACfU,MAAW,yBAAyBJ;gBACpCI,MACE,0DACAL,MAAM,CAACI,EAAE,EACTT,QAAAA,UAAAA,KAAAA,IAAAA,QAAS,EAAE;gBAEb,OAAOA;YACT;QACF;IAEJ,EAAE,OAAOvB,OAAO;QACdiC,MAAW,qCAAqCjC;IAClD;AACF;AAEO,SAASkC,sBAAsBC,SAAwB;IAC5D,SAASC,kBACPC,IAAkC;QAElC,IAAI,CAACA,MAAM,OAAO;QAGlB,MAAMC,mBAAmBC,MAAM,OAAO,CAACF,KAAK,QAAQ,IAC/CA,KAAK,QAAQ,CACX,GAAG,CAACD,mBACJ,MAAM,CAAC,CAACI,QAAUA,AAAU,SAAVA,SACrB,EAAE;QAGN,IAAIH,KAAK,IAAI,IAAIA,AAAwB,SAAxBA,KAAK,IAAI,CAAC,SAAS,EAClC,OAAO;YACL,GAAGA,IAAI;YACP,UAAUC;QACZ;QAIF,IAAIA,iBAAiB,MAAM,GAAG,GAC5B,OAAO;YACL,MAAM;YACN,UAAUA;QACZ;QAIF,OAAO;IACT;IAEA,OAAO;QACL,GAAGH,SAAS;QACZ,OAAOI,MAAM,OAAO,CAACJ,UAAU,KAAK,IAChCA,UAAU,KAAK,CAAC,GAAG,CAAC,CAACM;gBAEfC;YADJ,MAAMC,UAAU;gBAAE,GAAGF,IAAI;YAAC;YAC1B,IAAI,QAAAC,CAAAA,kBAAAA,KAAK,SAAS,AAAD,IAAbA,KAAAA,IAAAA,gBAAgB,IAAI,EACtBC,QAAQ,SAAS,GAAG;gBAClB,GAAGF,KAAK,SAAS;gBACjB,MAAML,kBAAkBK,KAAK,SAAS,CAAC,IAAI,KAAK;oBAC9C,MAAM;oBACN,UAAU,EAAE;gBACd;YACF;YAEF,OAAOE;QACT,KACAR,UAAU,KAAK;IACrB;AACF;AAIO,MAAMS,qBAAqB,IAEvBC;AAUJ,MAAMC,cAAc,CACzBC;IAKA,IAAI,AAAkB,YAAlB,OAAOA,QACT,OAAO;QACL,YAAYA;QACZ,kBAAkB/B;IACpB;IAEF,OAAO;QACL,YAAY+B,OAAO,MAAM;QACzB,kBAAkBA,OAAO,MAAM,GAC3B;YACE,QAAQA,OAAO,MAAM;YACrB,yBAAyB,CAAC,CAACA,OAAO,uBAAuB;QAC3D,IACA/B;IACN;AACF"}
|
|
@@ -1,5 +1,7 @@
|
|
|
1
1
|
import { assert } from "@midscene/shared/utils";
|
|
2
|
+
import { callToGetJSONObject } from "./service-caller/index.mjs";
|
|
2
3
|
import { NodeType } from "@midscene/shared/constants";
|
|
4
|
+
import { getModelName, vlLocateMode } from "@midscene/shared/env";
|
|
3
5
|
import { treeToList } from "@midscene/shared/extractor";
|
|
4
6
|
import { compositeElementInfoImg } from "@midscene/shared/img";
|
|
5
7
|
import { getDebug } from "@midscene/shared/logger";
|
|
@@ -12,14 +14,21 @@ var common_AIActionType = /*#__PURE__*/ function(AIActionType) {
|
|
|
12
14
|
AIActionType[AIActionType["DESCRIBE_ELEMENT"] = 4] = "DESCRIBE_ELEMENT";
|
|
13
15
|
return AIActionType;
|
|
14
16
|
}({});
|
|
17
|
+
async function callAiFn(msgs, AIActionTypeValue, modelPreferences) {
|
|
18
|
+
const jsonObject = await callToGetJSONObject(msgs, AIActionTypeValue, modelPreferences);
|
|
19
|
+
return {
|
|
20
|
+
content: jsonObject.content,
|
|
21
|
+
usage: jsonObject.usage
|
|
22
|
+
};
|
|
23
|
+
}
|
|
15
24
|
const defaultBboxSize = 20;
|
|
16
25
|
const debugInspectUtils = getDebug('ai:common');
|
|
17
|
-
function fillBboxParam(locate, width, height,
|
|
26
|
+
function fillBboxParam(locate, width, height, modelPreferences) {
|
|
18
27
|
if (locate.bbox_2d && !(null == locate ? void 0 : locate.bbox)) {
|
|
19
28
|
locate.bbox = locate.bbox_2d;
|
|
20
29
|
delete locate.bbox_2d;
|
|
21
30
|
}
|
|
22
|
-
if (null == locate ? void 0 : locate.bbox) locate.bbox = adaptBbox(locate.bbox, width, height,
|
|
31
|
+
if (null == locate ? void 0 : locate.bbox) locate.bbox = adaptBbox(locate.bbox, width, height, modelPreferences);
|
|
23
32
|
return locate;
|
|
24
33
|
}
|
|
25
34
|
function adaptQwenBbox(bbox) {
|
|
@@ -81,9 +90,9 @@ function adaptDoubaoBbox(bbox, width, height) {
|
|
|
81
90
|
const msg = `invalid bbox data for doubao-vision mode: ${JSON.stringify(bbox)} `;
|
|
82
91
|
throw new Error(msg);
|
|
83
92
|
}
|
|
84
|
-
function adaptBbox(bbox, width, height,
|
|
85
|
-
if ('doubao-vision' ===
|
|
86
|
-
if ('gemini' ===
|
|
93
|
+
function adaptBbox(bbox, width, height, modelPreferences) {
|
|
94
|
+
if ('doubao-vision' === vlLocateMode(modelPreferences) || 'vlm-ui-tars' === vlLocateMode(modelPreferences)) return adaptDoubaoBbox(bbox, width, height);
|
|
95
|
+
if ('gemini' === vlLocateMode(modelPreferences)) return adaptGeminiBbox(bbox, width, height);
|
|
87
96
|
return adaptQwenBbox(bbox);
|
|
88
97
|
}
|
|
89
98
|
function adaptGeminiBbox(bbox, width, height) {
|
|
@@ -98,11 +107,11 @@ function adaptGeminiBbox(bbox, width, height) {
|
|
|
98
107
|
bottom
|
|
99
108
|
];
|
|
100
109
|
}
|
|
101
|
-
function adaptBboxToRect(bbox, width, height, offsetX = 0, offsetY = 0
|
|
110
|
+
function adaptBboxToRect(bbox, width, height, modelPreferences, offsetX = 0, offsetY = 0) {
|
|
102
111
|
debugInspectUtils('adaptBboxToRect', bbox, width, height, offsetX, offsetY);
|
|
103
|
-
const [left, top, right, bottom] = adaptBbox(bbox, width, height,
|
|
104
|
-
const rectLeft = left
|
|
105
|
-
const rectTop = top
|
|
112
|
+
const [left, top, right, bottom] = adaptBbox(bbox, width, height, modelPreferences);
|
|
113
|
+
const rectLeft = left;
|
|
114
|
+
const rectTop = top;
|
|
106
115
|
let rectWidth = right - left;
|
|
107
116
|
let rectHeight = bottom - top;
|
|
108
117
|
if (rectLeft + rectWidth > width) rectWidth = width - rectLeft;
|
|
@@ -110,8 +119,8 @@ function adaptBboxToRect(bbox, width, height, offsetX = 0, offsetY = 0, vlMode)
|
|
|
110
119
|
rectWidth = Math.max(1, rectWidth);
|
|
111
120
|
rectHeight = Math.max(1, rectHeight);
|
|
112
121
|
const rect = {
|
|
113
|
-
left: rectLeft,
|
|
114
|
-
top: rectTop,
|
|
122
|
+
left: rectLeft + offsetX,
|
|
123
|
+
top: rectTop + offsetY,
|
|
115
124
|
width: rectWidth,
|
|
116
125
|
height: rectHeight
|
|
117
126
|
};
|
|
@@ -119,9 +128,10 @@ function adaptBboxToRect(bbox, width, height, offsetX = 0, offsetY = 0, vlMode)
|
|
|
119
128
|
return rect;
|
|
120
129
|
}
|
|
121
130
|
let warned = false;
|
|
122
|
-
function warnGPT4oSizeLimit(size,
|
|
131
|
+
function warnGPT4oSizeLimit(size, modelPreferences) {
|
|
132
|
+
var _getModelName;
|
|
123
133
|
if (warned) return;
|
|
124
|
-
if (
|
|
134
|
+
if (null == (_getModelName = getModelName(modelPreferences)) ? void 0 : _getModelName.toLowerCase().includes('gpt-4o')) {
|
|
125
135
|
const warningMsg = `GPT-4o has a maximum image input size of 2000x768 or 768x2000, but got ${size.width}x${size.height}. Please set your interface to a smaller resolution. Otherwise, the result may be inaccurate.`;
|
|
126
136
|
if (Math.max(size.width, size.height) > 2000 || Math.min(size.width, size.height) > 768) {
|
|
127
137
|
console.warn(warningMsg);
|
|
@@ -144,8 +154,8 @@ function mergeRects(rects) {
|
|
|
144
154
|
height: maxBottom - minTop
|
|
145
155
|
};
|
|
146
156
|
}
|
|
147
|
-
function expandSearchArea(rect, screenSize,
|
|
148
|
-
const minEdgeSize = 'doubao-vision' ===
|
|
157
|
+
function expandSearchArea(rect, screenSize, modelPreferences) {
|
|
158
|
+
const minEdgeSize = 'doubao-vision' === vlLocateMode(modelPreferences) ? 500 : 300;
|
|
149
159
|
const defaultPadding = 160;
|
|
150
160
|
const paddingSizeHorizontal = rect.width < minEdgeSize ? Math.ceil((minEdgeSize - rect.width) / 2) : defaultPadding;
|
|
151
161
|
const paddingSizeVertical = rect.height < minEdgeSize ? Math.ceil((minEdgeSize - rect.height) / 2) : defaultPadding;
|
|
@@ -299,6 +309,6 @@ const loadActionParam = (jsonObject, zodSchema)=>{
|
|
|
299
309
|
}
|
|
300
310
|
return result;
|
|
301
311
|
};
|
|
302
|
-
export { common_AIActionType as AIActionType, PointSchema, RectSchema, SizeSchema, TMultimodalPromptSchema, TUserPromptSchema, adaptBbox, adaptBboxToRect, adaptDoubaoBbox, adaptGeminiBbox, adaptQwenBbox, buildYamlFlowFromPlans, dumpActionParam, dumpMidsceneLocatorField, expandSearchArea, fillBboxParam, findAllMidsceneLocatorField, getMidsceneLocationSchema, ifMidsceneLocatorField, loadActionParam, markupImageForLLM, mergeRects, warnGPT4oSizeLimit };
|
|
312
|
+
export { common_AIActionType as AIActionType, PointSchema, RectSchema, SizeSchema, TMultimodalPromptSchema, TUserPromptSchema, adaptBbox, adaptBboxToRect, adaptDoubaoBbox, adaptGeminiBbox, adaptQwenBbox, buildYamlFlowFromPlans, callAiFn, dumpActionParam, dumpMidsceneLocatorField, expandSearchArea, fillBboxParam, findAllMidsceneLocatorField, getMidsceneLocationSchema, ifMidsceneLocatorField, loadActionParam, markupImageForLLM, mergeRects, warnGPT4oSizeLimit };
|
|
303
313
|
|
|
304
314
|
//# sourceMappingURL=common.mjs.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"ai-model/common.mjs","sources":["webpack://@midscene/core/./src/ai-model/common.ts"],"sourcesContent":["import type {\n AIUsageInfo,\n BaseElement,\n DeviceAction,\n ElementTreeNode,\n MidsceneYamlFlowItem,\n PlanningAction,\n Rect,\n Size,\n} from '@/types';\nimport { assert } from '@midscene/shared/utils';\n\nimport type {\n ChatCompletionSystemMessageParam,\n ChatCompletionUserMessageParam,\n} from 'openai/resources/index';\n\nimport type { PlanningLocateParam } from '@/types';\nimport { NodeType } from '@midscene/shared/constants';\nimport type { TVlModeTypes } from '@midscene/shared/env';\nimport { treeToList } from '@midscene/shared/extractor';\nimport { compositeElementInfoImg } from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { z } from 'zod';\n\nexport type AIArgs = [\n ChatCompletionSystemMessageParam,\n ...ChatCompletionUserMessageParam[],\n];\n\nexport enum AIActionType {\n ASSERT = 0,\n INSPECT_ELEMENT = 1,\n EXTRACT_DATA = 2,\n PLAN = 3,\n DESCRIBE_ELEMENT = 4,\n}\n\nconst defaultBboxSize = 20; // must be even number\nconst debugInspectUtils = getDebug('ai:common');\n\n// transform the param of locate from qwen mode\nexport function fillBboxParam(\n locate: PlanningLocateParam,\n width: number,\n height: number,\n vlMode: TVlModeTypes | undefined,\n) {\n // The Qwen model might have hallucinations of naming bbox as bbox_2d.\n if ((locate as any).bbox_2d && !locate?.bbox) {\n locate.bbox = (locate as any).bbox_2d;\n // biome-ignore lint/performance/noDelete: <explanation>\n delete (locate as any).bbox_2d;\n }\n\n if (locate?.bbox) {\n locate.bbox = adaptBbox(locate.bbox, width, height, vlMode);\n }\n\n return locate;\n}\n\nexport function adaptQwenBbox(\n bbox: number[],\n): [number, number, number, number] {\n if (bbox.length < 2) {\n const msg = `invalid bbox data for qwen-vl mode: ${JSON.stringify(bbox)} `;\n throw new Error(msg);\n }\n\n const result: [number, number, number, number] = [\n Math.round(bbox[0]),\n Math.round(bbox[1]),\n typeof bbox[2] === 'number'\n ? Math.round(bbox[2])\n : Math.round(bbox[0] + defaultBboxSize),\n typeof bbox[3] === 'number'\n ? Math.round(bbox[3])\n : Math.round(bbox[1] + defaultBboxSize),\n ];\n return result;\n}\n\nexport function adaptDoubaoBbox(\n bbox: string[] | number[] | string,\n width: number,\n height: number,\n): [number, number, number, number] {\n assert(\n width > 0 && height > 0,\n 'width and height must be greater than 0 in doubao mode',\n );\n\n if (typeof bbox === 'string') {\n assert(\n /^(\\d+)\\s(\\d+)\\s(\\d+)\\s(\\d+)$/.test(bbox.trim()),\n `invalid bbox data string for doubao-vision mode: ${bbox}`,\n );\n const splitted = bbox.split(' ');\n if (splitted.length === 4) {\n return [\n Math.round((Number(splitted[0]) * width) / 1000),\n Math.round((Number(splitted[1]) * height) / 1000),\n Math.round((Number(splitted[2]) * width) / 1000),\n Math.round((Number(splitted[3]) * height) / 1000),\n ];\n }\n throw new Error(`invalid bbox data string for doubao-vision mode: ${bbox}`);\n }\n\n if (Array.isArray(bbox) && Array.isArray(bbox[0])) {\n bbox = bbox[0];\n }\n\n let bboxList: number[] = [];\n if (Array.isArray(bbox) && typeof bbox[0] === 'string') {\n bbox.forEach((item) => {\n if (typeof item === 'string' && item.includes(',')) {\n const [x, y] = item.split(',');\n bboxList.push(Number(x.trim()), Number(y.trim()));\n } else if (typeof item === 'string' && item.includes(' ')) {\n const [x, y] = item.split(' ');\n bboxList.push(Number(x.trim()), Number(y.trim()));\n } else {\n bboxList.push(Number(item));\n }\n });\n } else {\n bboxList = bbox as any;\n }\n\n if (bboxList.length === 4 || bboxList.length === 5) {\n return [\n Math.round((bboxList[0] * width) / 1000),\n Math.round((bboxList[1] * height) / 1000),\n Math.round((bboxList[2] * width) / 1000),\n Math.round((bboxList[3] * height) / 1000),\n ];\n }\n\n // treat the bbox as a center point\n if (\n bboxList.length === 6 ||\n bboxList.length === 2 ||\n bboxList.length === 3 ||\n bboxList.length === 7\n ) {\n return [\n Math.max(\n 0,\n Math.round((bboxList[0] * width) / 1000) - defaultBboxSize / 2,\n ),\n Math.max(\n 0,\n Math.round((bboxList[1] * height) / 1000) - defaultBboxSize / 2,\n ),\n Math.min(\n width,\n Math.round((bboxList[0] * width) / 1000) + defaultBboxSize / 2,\n ),\n Math.min(\n height,\n Math.round((bboxList[1] * height) / 1000) + defaultBboxSize / 2,\n ),\n ];\n }\n\n if (bbox.length === 8) {\n return [\n Math.round((bboxList[0] * width) / 1000),\n Math.round((bboxList[1] * height) / 1000),\n Math.round((bboxList[4] * width) / 1000),\n Math.round((bboxList[5] * height) / 1000),\n ];\n }\n\n const msg = `invalid bbox data for doubao-vision mode: ${JSON.stringify(bbox)} `;\n throw new Error(msg);\n}\n\nexport function adaptBbox(\n bbox: number[],\n width: number,\n height: number,\n vlMode: TVlModeTypes | undefined,\n): [number, number, number, number] {\n if (vlMode === 'doubao-vision' || vlMode === 'vlm-ui-tars') {\n return adaptDoubaoBbox(bbox, width, height);\n }\n\n if (vlMode === 'gemini') {\n return adaptGeminiBbox(bbox, width, height);\n }\n\n return adaptQwenBbox(bbox);\n}\n\nexport function adaptGeminiBbox(\n bbox: number[],\n width: number,\n height: number,\n): [number, number, number, number] {\n const left = Math.round((bbox[1] * width) / 1000);\n const top = Math.round((bbox[0] * height) / 1000);\n const right = Math.round((bbox[3] * width) / 1000);\n const bottom = Math.round((bbox[2] * height) / 1000);\n return [left, top, right, bottom];\n}\n\nexport function adaptBboxToRect(\n bbox: number[],\n width: number,\n height: number,\n offsetX = 0,\n offsetY = 0,\n vlMode?: TVlModeTypes | undefined,\n): Rect {\n debugInspectUtils('adaptBboxToRect', bbox, width, height, offsetX, offsetY);\n const [left, top, right, bottom] = adaptBbox(bbox, width, height, vlMode);\n\n // Calculate initial rect dimensions\n const rectLeft = left + offsetX;\n const rectTop = top + offsetY;\n let rectWidth = right - left;\n let rectHeight = bottom - top;\n\n // Ensure the rect doesn't exceed image boundaries\n // If right edge exceeds width, adjust the width\n if (rectLeft + rectWidth > width) {\n rectWidth = width - rectLeft;\n }\n\n // If bottom edge exceeds height, adjust the height\n if (rectTop + rectHeight > height) {\n rectHeight = height - rectTop;\n }\n\n // Ensure minimum dimensions (width and height should be at least 1)\n rectWidth = Math.max(1, rectWidth);\n rectHeight = Math.max(1, rectHeight);\n\n const rect = {\n left: rectLeft,\n top: rectTop,\n width: rectWidth,\n height: rectHeight,\n };\n debugInspectUtils('adaptBboxToRect, result=', rect);\n return rect;\n}\n\nlet warned = false;\nexport function warnGPT4oSizeLimit(size: Size, modelName: string) {\n if (warned) return;\n if (modelName.toLowerCase().includes('gpt-4o')) {\n const warningMsg = `GPT-4o has a maximum image input size of 2000x768 or 768x2000, but got ${size.width}x${size.height}. Please set your interface to a smaller resolution. Otherwise, the result may be inaccurate.`;\n\n if (\n Math.max(size.width, size.height) > 2000 ||\n Math.min(size.width, size.height) > 768\n ) {\n console.warn(warningMsg);\n warned = true;\n }\n } else if (size.width > 1800 || size.height > 1800) {\n console.warn(\n `The image size seems too large (${size.width}x${size.height}). It may lead to more token usage, slower response, and inaccurate result.`,\n );\n warned = true;\n }\n}\n\nexport function mergeRects(rects: Rect[]) {\n const minLeft = Math.min(...rects.map((r) => r.left));\n const minTop = Math.min(...rects.map((r) => r.top));\n const maxRight = Math.max(...rects.map((r) => r.left + r.width));\n const maxBottom = Math.max(...rects.map((r) => r.top + r.height));\n return {\n left: minLeft,\n top: minTop,\n width: maxRight - minLeft,\n height: maxBottom - minTop,\n };\n}\n\n// expand the search area to at least 300 x 300, or add a default padding\nexport function expandSearchArea(\n rect: Rect,\n screenSize: Size,\n vlMode: TVlModeTypes | undefined,\n) {\n const minEdgeSize = vlMode === 'doubao-vision' ? 500 : 300;\n const defaultPadding = 160;\n\n const paddingSizeHorizontal =\n rect.width < minEdgeSize\n ? Math.ceil((minEdgeSize - rect.width) / 2)\n : defaultPadding;\n const paddingSizeVertical =\n rect.height < minEdgeSize\n ? Math.ceil((minEdgeSize - rect.height) / 2)\n : defaultPadding;\n rect.left = Math.max(0, rect.left - paddingSizeHorizontal);\n rect.width = Math.min(\n rect.width + paddingSizeHorizontal * 2,\n screenSize.width - rect.left,\n );\n rect.top = Math.max(0, rect.top - paddingSizeVertical);\n rect.height = Math.min(\n rect.height + paddingSizeVertical * 2,\n screenSize.height - rect.top,\n );\n return rect;\n}\n\nexport async function markupImageForLLM(\n screenshotBase64: string,\n tree: ElementTreeNode<BaseElement>,\n size: Size,\n) {\n const elementsInfo = treeToList(tree);\n const elementsPositionInfoWithoutText = elementsInfo!.filter(\n (elementInfo) => {\n if (elementInfo.attributes.nodeType === NodeType.TEXT) {\n return false;\n }\n return true;\n },\n );\n\n const imagePayload = await compositeElementInfoImg({\n inputImgBase64: screenshotBase64,\n elementsPositionInfo: elementsPositionInfoWithoutText,\n size,\n });\n return imagePayload;\n}\n\nexport function buildYamlFlowFromPlans(\n plans: PlanningAction[],\n actionSpace: DeviceAction<any>[],\n sleep?: number,\n): MidsceneYamlFlowItem[] {\n const flow: MidsceneYamlFlowItem[] = [];\n\n for (const plan of plans) {\n const verb = plan.type;\n\n const action = actionSpace.find((action) => action.name === verb);\n if (!action) {\n console.warn(\n `Cannot convert action ${verb} to yaml flow. Will ignore it.`,\n );\n continue;\n }\n\n const flowKey = action.interfaceAlias || verb;\n const flowParam = action.paramSchema\n ? dumpActionParam(plan.param || {}, action.paramSchema)\n : {};\n\n const flowItem: MidsceneYamlFlowItem = {\n [flowKey]: '',\n ...flowParam,\n };\n\n flow.push(flowItem);\n }\n\n if (sleep) {\n flow.push({\n sleep,\n });\n }\n\n return flow;\n}\n\n// Zod schemas for shared types\nexport const PointSchema = z.object({\n left: z.number(),\n top: z.number(),\n});\n\nexport const SizeSchema = z.object({\n width: z.number(),\n height: z.number(),\n dpr: z.number().optional(),\n});\n\nexport const RectSchema = PointSchema.and(SizeSchema).and(\n z.object({\n zoom: z.number().optional(),\n }),\n);\n\n// Zod schema for TMultimodalPrompt\nexport const TMultimodalPromptSchema = z.object({\n images: z\n .array(\n z.object({\n name: z.string(),\n url: z.string(),\n }),\n )\n .optional(),\n convertHttpImage2Base64: z.boolean().optional(),\n});\n\n// Zod schema for TUserPrompt\nexport const TUserPromptSchema = z.union([\n z.string(),\n z\n .object({\n prompt: z.string(),\n })\n .and(TMultimodalPromptSchema.partial()),\n]);\n\n// Generate TypeScript types from Zod schemas\nexport type TMultimodalPrompt = z.infer<typeof TMultimodalPromptSchema>;\nexport type TUserPrompt = z.infer<typeof TUserPromptSchema>;\n\nconst locateFieldFlagName = 'midscene_location_field_flag';\n\nconst MidsceneLocationResult = z\n .object({\n [locateFieldFlagName]: z.literal(true),\n prompt: TUserPromptSchema,\n\n // optional fields\n deepThink: z.boolean().optional(), // only available in vl model\n cacheable: z.boolean().optional(),\n xpath: z.boolean().optional(), // preset result for xpath\n\n // these two fields will only appear in the result\n center: z.tuple([z.number(), z.number()]),\n rect: RectSchema,\n })\n .passthrough();\n\nexport type MidsceneLocationResultType = z.infer<typeof MidsceneLocationResult>;\nexport const getMidsceneLocationSchema = () => {\n return MidsceneLocationResult;\n};\n\nexport const ifMidsceneLocatorField = (field: any): boolean => {\n // Handle optional fields by getting the inner type\n let actualField = field;\n if (actualField._def?.typeName === 'ZodOptional') {\n actualField = actualField._def.innerType;\n }\n\n // Check if this is a ZodUnion (the new MidsceneLocation structure)\n if (actualField._def?.typeName === 'ZodObject') {\n const shape = actualField._def.shape();\n return locateFieldFlagName in shape;\n }\n\n return false;\n};\n\nexport const dumpMidsceneLocatorField = (field: any): string => {\n assert(\n ifMidsceneLocatorField(field),\n 'field is not a midscene locator field',\n );\n\n // If field is a string, return it directly\n if (typeof field === 'string') {\n return field;\n }\n\n // If field is an object with prompt property\n if (field && typeof field === 'object' && field.prompt) {\n // If prompt is a string, return it directly\n if (typeof field.prompt === 'string') {\n return field.prompt;\n }\n // If prompt is a TUserPrompt object, extract the prompt string\n if (typeof field.prompt === 'object' && field.prompt.prompt) {\n return field.prompt.prompt; // TODO: dump images if necessary\n }\n }\n\n // Fallback: try to convert to string\n return String(field);\n};\n\nexport const findAllMidsceneLocatorField = (\n zodType?: z.ZodType<any>,\n requiredOnly?: boolean,\n): string[] => {\n if (!zodType) {\n return [];\n }\n\n // Check if this is a ZodObject by checking if it has a shape property\n const zodObject = zodType as any;\n if (zodObject._def?.typeName === 'ZodObject' && zodObject.shape) {\n const keys = Object.keys(zodObject.shape);\n return keys.filter((key) => {\n const field = zodObject.shape[key];\n if (!ifMidsceneLocatorField(field)) {\n return false;\n }\n\n // If requiredOnly is true, filter out optional fields\n if (requiredOnly) {\n return field._def?.typeName !== 'ZodOptional';\n }\n\n return true;\n });\n }\n\n // For other ZodType instances, we can't extract field names\n return [];\n};\n\nexport const dumpActionParam = (\n jsonObject: Record<string, any>,\n zodSchema: z.ZodType<any>,\n): Record<string, any> => {\n const locatorFields = findAllMidsceneLocatorField(zodSchema);\n const result = { ...jsonObject };\n\n for (const fieldName of locatorFields) {\n const fieldValue = result[fieldName];\n if (fieldValue) {\n // If it's already a string, keep it as is\n if (typeof fieldValue === 'string') {\n result[fieldName] = fieldValue;\n } else if (typeof fieldValue === 'object') {\n // Check if this field is actually a MidsceneLocationType object\n if (fieldValue.prompt) {\n // If prompt is a string, use it directly\n if (typeof fieldValue.prompt === 'string') {\n result[fieldName] = fieldValue.prompt;\n } else if (\n typeof fieldValue.prompt === 'object' &&\n fieldValue.prompt.prompt\n ) {\n // If prompt is a TUserPrompt object, extract the prompt string\n result[fieldName] = fieldValue.prompt.prompt;\n }\n }\n }\n }\n }\n\n return result;\n};\n\nexport const loadActionParam = (\n jsonObject: Record<string, any>,\n zodSchema: z.ZodType<any>,\n): Record<string, any> => {\n const locatorFields = findAllMidsceneLocatorField(zodSchema);\n const result = { ...jsonObject };\n\n for (const fieldName of locatorFields) {\n const fieldValue = result[fieldName];\n if (fieldValue && typeof fieldValue === 'string') {\n result[fieldName] = {\n [locateFieldFlagName]: true,\n prompt: fieldValue,\n };\n }\n }\n\n return result;\n};\n"],"names":["AIActionType","defaultBboxSize","debugInspectUtils","getDebug","fillBboxParam","locate","width","height","vlMode","adaptBbox","adaptQwenBbox","bbox","msg","JSON","Error","result","Math","adaptDoubaoBbox","assert","splitted","Number","Array","bboxList","item","x","y","adaptGeminiBbox","left","top","right","bottom","adaptBboxToRect","offsetX","offsetY","rectLeft","rectTop","rectWidth","rectHeight","rect","warned","warnGPT4oSizeLimit","size","modelName","warningMsg","console","mergeRects","rects","minLeft","r","minTop","maxRight","maxBottom","expandSearchArea","screenSize","minEdgeSize","defaultPadding","paddingSizeHorizontal","paddingSizeVertical","markupImageForLLM","screenshotBase64","tree","elementsInfo","treeToList","elementsPositionInfoWithoutText","elementInfo","NodeType","imagePayload","compositeElementInfoImg","buildYamlFlowFromPlans","plans","actionSpace","sleep","flow","plan","verb","action","flowKey","flowParam","dumpActionParam","flowItem","PointSchema","z","SizeSchema","RectSchema","TMultimodalPromptSchema","TUserPromptSchema","locateFieldFlagName","MidsceneLocationResult","getMidsceneLocationSchema","ifMidsceneLocatorField","field","_actualField__def","_actualField__def1","actualField","shape","dumpMidsceneLocatorField","String","findAllMidsceneLocatorField","zodType","requiredOnly","_zodObject__def","zodObject","keys","Object","key","_field__def","jsonObject","zodSchema","locatorFields","fieldName","fieldValue","loadActionParam"],"mappings":";;;;;;AA8BO,IAAKA,sBAAYA,WAAAA,GAAAA,SAAZA,YAAY;;;;;;WAAZA;;AAQZ,MAAMC,kBAAkB;AACxB,MAAMC,oBAAoBC,SAAS;AAG5B,SAASC,cACdC,MAA2B,EAC3BC,KAAa,EACbC,MAAc,EACdC,MAAgC;IAGhC,IAAKH,OAAe,OAAO,IAAI,CAACA,CAAAA,QAAAA,SAAAA,KAAAA,IAAAA,OAAQ,IAAI,AAAD,GAAG;QAC5CA,OAAO,IAAI,GAAIA,OAAe,OAAO;QAErC,OAAQA,OAAe,OAAO;IAChC;IAEA,IAAIA,QAAAA,SAAAA,KAAAA,IAAAA,OAAQ,IAAI,EACdA,OAAO,IAAI,GAAGI,UAAUJ,OAAO,IAAI,EAAEC,OAAOC,QAAQC;IAGtD,OAAOH;AACT;AAEO,SAASK,cACdC,IAAc;IAEd,IAAIA,KAAK,MAAM,GAAG,GAAG;QACnB,MAAMC,MAAM,CAAC,oCAAoC,EAAEC,KAAK,SAAS,CAACF,MAAM,CAAC,CAAC;QAC1E,MAAM,IAAIG,MAAMF;IAClB;IAEA,MAAMG,SAA2C;QAC/CC,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE;QAClBK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE;QACC,YAAnB,OAAOA,IAAI,CAAC,EAAE,GACVK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,IAClBK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,GAAGV;QACN,YAAnB,OAAOU,IAAI,CAAC,EAAE,GACVK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,IAClBK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,GAAGV;KAC1B;IACD,OAAOc;AACT;AAEO,SAASE,gBACdN,IAAkC,EAClCL,KAAa,EACbC,MAAc;IAEdW,OACEZ,QAAQ,KAAKC,SAAS,GACtB;IAGF,IAAI,AAAgB,YAAhB,OAAOI,MAAmB;QAC5BO,OACE,+BAA+B,IAAI,CAACP,KAAK,IAAI,KAC7C,CAAC,iDAAiD,EAAEA,MAAM;QAE5D,MAAMQ,WAAWR,KAAK,KAAK,CAAC;QAC5B,IAAIQ,AAAoB,MAApBA,SAAS,MAAM,EACjB,OAAO;YACLH,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAIb,QAAS;YAC3CU,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAIZ,SAAU;YAC5CS,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAIb,QAAS;YAC3CU,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAIZ,SAAU;SAC7C;QAEH,MAAM,IAAIO,MAAM,CAAC,iDAAiD,EAAEH,MAAM;IAC5E;IAEA,IAAIU,MAAM,OAAO,CAACV,SAASU,MAAM,OAAO,CAACV,IAAI,CAAC,EAAE,GAC9CA,OAAOA,IAAI,CAAC,EAAE;IAGhB,IAAIW,WAAqB,EAAE;IAC3B,IAAID,MAAM,OAAO,CAACV,SAAS,AAAmB,YAAnB,OAAOA,IAAI,CAAC,EAAE,EACvCA,KAAK,OAAO,CAAC,CAACY;QACZ,IAAI,AAAgB,YAAhB,OAAOA,QAAqBA,KAAK,QAAQ,CAAC,MAAM;YAClD,MAAM,CAACC,GAAGC,EAAE,GAAGF,KAAK,KAAK,CAAC;YAC1BD,SAAS,IAAI,CAACF,OAAOI,EAAE,IAAI,KAAKJ,OAAOK,EAAE,IAAI;QAC/C,OAAO,IAAI,AAAgB,YAAhB,OAAOF,QAAqBA,KAAK,QAAQ,CAAC,MAAM;YACzD,MAAM,CAACC,GAAGC,EAAE,GAAGF,KAAK,KAAK,CAAC;YAC1BD,SAAS,IAAI,CAACF,OAAOI,EAAE,IAAI,KAAKJ,OAAOK,EAAE,IAAI;QAC/C,OACEH,SAAS,IAAI,CAACF,OAAOG;IAEzB;SAEAD,WAAWX;IAGb,IAAIW,AAAoB,MAApBA,SAAS,MAAM,IAAUA,AAAoB,MAApBA,SAAS,MAAM,EAC1C,OAAO;QACLN,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGhB,QAAS;QACnCU,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,SAAU;QACpCS,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGhB,QAAS;QACnCU,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,SAAU;KACrC;IAIH,IACEe,AAAoB,MAApBA,SAAS,MAAM,IACfA,AAAoB,MAApBA,SAAS,MAAM,IACfA,AAAoB,MAApBA,SAAS,MAAM,IACfA,AAAoB,MAApBA,SAAS,MAAM,EAEf,OAAO;QACLN,KAAK,GAAG,CACN,GACAA,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGhB,QAAS,QAAQL,kBAAkB;QAE/De,KAAK,GAAG,CACN,GACAA,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,SAAU,QAAQN,kBAAkB;QAEhEe,KAAK,GAAG,CACNV,OACAU,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGhB,QAAS,QAAQL,kBAAkB;QAE/De,KAAK,GAAG,CACNT,QACAS,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,SAAU,QAAQN,kBAAkB;KAEjE;IAGH,IAAIU,AAAgB,MAAhBA,KAAK,MAAM,EACb,OAAO;QACLK,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGhB,QAAS;QACnCU,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,SAAU;QACpCS,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGhB,QAAS;QACnCU,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,SAAU;KACrC;IAGH,MAAMK,MAAM,CAAC,0CAA0C,EAAEC,KAAK,SAAS,CAACF,MAAM,CAAC,CAAC;IAChF,MAAM,IAAIG,MAAMF;AAClB;AAEO,SAASH,UACdE,IAAc,EACdL,KAAa,EACbC,MAAc,EACdC,MAAgC;IAEhC,IAAIA,AAAW,oBAAXA,UAA8BA,AAAW,kBAAXA,QAChC,OAAOS,gBAAgBN,MAAML,OAAOC;IAGtC,IAAIC,AAAW,aAAXA,QACF,OAAOkB,gBAAgBf,MAAML,OAAOC;IAGtC,OAAOG,cAAcC;AACvB;AAEO,SAASe,gBACdf,IAAc,EACdL,KAAa,EACbC,MAAc;IAEd,MAAMoB,OAAOX,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGL,QAAS;IAC5C,MAAMsB,MAAMZ,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGJ,SAAU;IAC5C,MAAMsB,QAAQb,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGL,QAAS;IAC7C,MAAMwB,SAASd,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGJ,SAAU;IAC/C,OAAO;QAACoB;QAAMC;QAAKC;QAAOC;KAAO;AACnC;AAEO,SAASC,gBACdpB,IAAc,EACdL,KAAa,EACbC,MAAc,EACdyB,UAAU,CAAC,EACXC,UAAU,CAAC,EACXzB,MAAiC;IAEjCN,kBAAkB,mBAAmBS,MAAML,OAAOC,QAAQyB,SAASC;IACnE,MAAM,CAACN,MAAMC,KAAKC,OAAOC,OAAO,GAAGrB,UAAUE,MAAML,OAAOC,QAAQC;IAGlE,MAAM0B,WAAWP,OAAOK;IACxB,MAAMG,UAAUP,MAAMK;IACtB,IAAIG,YAAYP,QAAQF;IACxB,IAAIU,aAAaP,SAASF;IAI1B,IAAIM,WAAWE,YAAY9B,OACzB8B,YAAY9B,QAAQ4B;IAItB,IAAIC,UAAUE,aAAa9B,QACzB8B,aAAa9B,SAAS4B;IAIxBC,YAAYpB,KAAK,GAAG,CAAC,GAAGoB;IACxBC,aAAarB,KAAK,GAAG,CAAC,GAAGqB;IAEzB,MAAMC,OAAO;QACX,MAAMJ;QACN,KAAKC;QACL,OAAOC;QACP,QAAQC;IACV;IACAnC,kBAAkB,4BAA4BoC;IAC9C,OAAOA;AACT;AAEA,IAAIC,SAAS;AACN,SAASC,mBAAmBC,IAAU,EAAEC,SAAiB;IAC9D,IAAIH,QAAQ;IACZ,IAAIG,UAAU,WAAW,GAAG,QAAQ,CAAC,WAAW;QAC9C,MAAMC,aAAa,CAAC,uEAAuE,EAAEF,KAAK,KAAK,CAAC,CAAC,EAAEA,KAAK,MAAM,CAAC,6FAA6F,CAAC;QAErN,IACEzB,KAAK,GAAG,CAACyB,KAAK,KAAK,EAAEA,KAAK,MAAM,IAAI,QACpCzB,KAAK,GAAG,CAACyB,KAAK,KAAK,EAAEA,KAAK,MAAM,IAAI,KACpC;YACAG,QAAQ,IAAI,CAACD;YACbJ,SAAS;QACX;IACF,OAAO,IAAIE,KAAK,KAAK,GAAG,QAAQA,KAAK,MAAM,GAAG,MAAM;QAClDG,QAAQ,IAAI,CACV,CAAC,gCAAgC,EAAEH,KAAK,KAAK,CAAC,CAAC,EAAEA,KAAK,MAAM,CAAC,2EAA2E,CAAC;QAE3IF,SAAS;IACX;AACF;AAEO,SAASM,WAAWC,KAAa;IACtC,MAAMC,UAAU/B,KAAK,GAAG,IAAI8B,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,IAAI;IACnD,MAAMC,SAASjC,KAAK,GAAG,IAAI8B,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,GAAG;IACjD,MAAME,WAAWlC,KAAK,GAAG,IAAI8B,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,IAAI,GAAGA,EAAE,KAAK;IAC9D,MAAMG,YAAYnC,KAAK,GAAG,IAAI8B,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,GAAG,GAAGA,EAAE,MAAM;IAC/D,OAAO;QACL,MAAMD;QACN,KAAKE;QACL,OAAOC,WAAWH;QAClB,QAAQI,YAAYF;IACtB;AACF;AAGO,SAASG,iBACdd,IAAU,EACVe,UAAgB,EAChB7C,MAAgC;IAEhC,MAAM8C,cAAc9C,AAAW,oBAAXA,SAA6B,MAAM;IACvD,MAAM+C,iBAAiB;IAEvB,MAAMC,wBACJlB,KAAK,KAAK,GAAGgB,cACTtC,KAAK,IAAI,CAAEsC,AAAAA,CAAAA,cAAchB,KAAK,KAAI,IAAK,KACvCiB;IACN,MAAME,sBACJnB,KAAK,MAAM,GAAGgB,cACVtC,KAAK,IAAI,CAAEsC,AAAAA,CAAAA,cAAchB,KAAK,MAAK,IAAK,KACxCiB;IACNjB,KAAK,IAAI,GAAGtB,KAAK,GAAG,CAAC,GAAGsB,KAAK,IAAI,GAAGkB;IACpClB,KAAK,KAAK,GAAGtB,KAAK,GAAG,CACnBsB,KAAK,KAAK,GAAGkB,AAAwB,IAAxBA,uBACbH,WAAW,KAAK,GAAGf,KAAK,IAAI;IAE9BA,KAAK,GAAG,GAAGtB,KAAK,GAAG,CAAC,GAAGsB,KAAK,GAAG,GAAGmB;IAClCnB,KAAK,MAAM,GAAGtB,KAAK,GAAG,CACpBsB,KAAK,MAAM,GAAGmB,AAAsB,IAAtBA,qBACdJ,WAAW,MAAM,GAAGf,KAAK,GAAG;IAE9B,OAAOA;AACT;AAEO,eAAeoB,kBACpBC,gBAAwB,EACxBC,IAAkC,EAClCnB,IAAU;IAEV,MAAMoB,eAAeC,WAAWF;IAChC,MAAMG,kCAAkCF,aAAc,MAAM,CAC1D,CAACG;QACC,IAAIA,YAAY,UAAU,CAAC,QAAQ,KAAKC,SAAS,IAAI,EACnD,OAAO;QAET,OAAO;IACT;IAGF,MAAMC,eAAe,MAAMC,wBAAwB;QACjD,gBAAgBR;QAChB,sBAAsBI;QACtBtB;IACF;IACA,OAAOyB;AACT;AAEO,SAASE,uBACdC,KAAuB,EACvBC,WAAgC,EAChCC,KAAc;IAEd,MAAMC,OAA+B,EAAE;IAEvC,KAAK,MAAMC,QAAQJ,MAAO;QACxB,MAAMK,OAAOD,KAAK,IAAI;QAEtB,MAAME,SAASL,YAAY,IAAI,CAAC,CAACK,SAAWA,OAAO,IAAI,KAAKD;QAC5D,IAAI,CAACC,QAAQ;YACX/B,QAAQ,IAAI,CACV,CAAC,sBAAsB,EAAE8B,KAAK,8BAA8B,CAAC;YAE/D;QACF;QAEA,MAAME,UAAUD,OAAO,cAAc,IAAID;QACzC,MAAMG,YAAYF,OAAO,WAAW,GAChCG,gBAAgBL,KAAK,KAAK,IAAI,CAAC,GAAGE,OAAO,WAAW,IACpD,CAAC;QAEL,MAAMI,WAAiC;YACrC,CAACH,QAAQ,EAAE;YACX,GAAGC,SAAS;QACd;QAEAL,KAAK,IAAI,CAACO;IACZ;IAEA,IAAIR,OACFC,KAAK,IAAI,CAAC;QACRD;IACF;IAGF,OAAOC;AACT;AAGO,MAAMQ,cAAcC,EAAE,MAAM,CAAC;IAClC,MAAMA,EAAE,MAAM;IACd,KAAKA,EAAE,MAAM;AACf;AAEO,MAAMC,aAAaD,EAAE,MAAM,CAAC;IACjC,OAAOA,EAAE,MAAM;IACf,QAAQA,EAAE,MAAM;IAChB,KAAKA,EAAE,MAAM,GAAG,QAAQ;AAC1B;AAEO,MAAME,aAAaH,YAAY,GAAG,CAACE,YAAY,GAAG,CACvDD,EAAE,MAAM,CAAC;IACP,MAAMA,EAAE,MAAM,GAAG,QAAQ;AAC3B;AAIK,MAAMG,0BAA0BH,EAAE,MAAM,CAAC;IAC9C,QAAQA,EAAAA,KACA,CACJA,EAAE,MAAM,CAAC;QACP,MAAMA,EAAE,MAAM;QACd,KAAKA,EAAE,MAAM;IACf,IAED,QAAQ;IACX,yBAAyBA,EAAE,OAAO,GAAG,QAAQ;AAC/C;AAGO,MAAMI,oBAAoBJ,EAAE,KAAK,CAAC;IACvCA,EAAE,MAAM;IACRA,EAAAA,MACS,CAAC;QACN,QAAQA,EAAE,MAAM;IAClB,GACC,GAAG,CAACG,wBAAwB,OAAO;CACvC;AAMD,MAAME,sBAAsB;AAE5B,MAAMC,yBAAyBN,EAAAA,MACtB,CAAC;IACN,CAACK,oBAAoB,EAAEL,EAAE,OAAO,CAAC;IACjC,QAAQI;IAGR,WAAWJ,EAAE,OAAO,GAAG,QAAQ;IAC/B,WAAWA,EAAE,OAAO,GAAG,QAAQ;IAC/B,OAAOA,EAAE,OAAO,GAAG,QAAQ;IAG3B,QAAQA,EAAE,KAAK,CAAC;QAACA,EAAE,MAAM;QAAIA,EAAE,MAAM;KAAG;IACxC,MAAME;AACR,GACC,WAAW;AAGP,MAAMK,4BAA4B,IAChCD;AAGF,MAAME,yBAAyB,CAACC;QAGjCC,mBAKAC;IANJ,IAAIC,cAAcH;IAClB,IAAIC,AAAAA,SAAAA,CAAAA,oBAAAA,YAAY,IAAI,AAAD,IAAfA,KAAAA,IAAAA,kBAAkB,QAAQ,AAAD,MAAM,eACjCE,cAAcA,YAAY,IAAI,CAAC,SAAS;IAI1C,IAAID,AAAAA,SAAAA,CAAAA,qBAAAA,YAAY,IAAI,AAAD,IAAfA,KAAAA,IAAAA,mBAAkB,QAAQ,AAAD,MAAM,aAAa;QAC9C,MAAME,QAAQD,YAAY,IAAI,CAAC,KAAK;QACpC,OAAOP,uBAAuBQ;IAChC;IAEA,OAAO;AACT;AAEO,MAAMC,2BAA2B,CAACL;IACvCxE,OACEuE,uBAAuBC,QACvB;IAIF,IAAI,AAAiB,YAAjB,OAAOA,OACT,OAAOA;IAIT,IAAIA,SAAS,AAAiB,YAAjB,OAAOA,SAAsBA,MAAM,MAAM,EAAE;QAEtD,IAAI,AAAwB,YAAxB,OAAOA,MAAM,MAAM,EACrB,OAAOA,MAAM,MAAM;QAGrB,IAAI,AAAwB,YAAxB,OAAOA,MAAM,MAAM,IAAiBA,MAAM,MAAM,CAAC,MAAM,EACzD,OAAOA,MAAM,MAAM,CAAC,MAAM;IAE9B;IAGA,OAAOM,OAAON;AAChB;AAEO,MAAMO,8BAA8B,CACzCC,SACAC;QAQIC;IANJ,IAAI,CAACF,SACH,OAAO,EAAE;IAIX,MAAMG,YAAYH;IAClB,IAAIE,AAAAA,SAAAA,CAAAA,kBAAAA,UAAU,IAAI,AAAD,IAAbA,KAAAA,IAAAA,gBAAgB,QAAQ,AAAD,MAAM,eAAeC,UAAU,KAAK,EAAE;QAC/D,MAAMC,OAAOC,OAAO,IAAI,CAACF,UAAU,KAAK;QACxC,OAAOC,KAAK,MAAM,CAAC,CAACE;YAClB,MAAMd,QAAQW,UAAU,KAAK,CAACG,IAAI;YAClC,IAAI,CAACf,uBAAuBC,QAC1B,OAAO;YAIT,IAAIS,cAAc;oBACTM;gBAAP,OAAOA,AAAAA,SAAAA,CAAAA,cAAAA,MAAM,IAAI,AAAD,IAATA,KAAAA,IAAAA,YAAY,QAAQ,AAAD,MAAM;YAClC;YAEA,OAAO;QACT;IACF;IAGA,OAAO,EAAE;AACX;AAEO,MAAM3B,kBAAkB,CAC7B4B,YACAC;IAEA,MAAMC,gBAAgBX,4BAA4BU;IAClD,MAAM5F,SAAS;QAAE,GAAG2F,UAAU;IAAC;IAE/B,KAAK,MAAMG,aAAaD,cAAe;QACrC,MAAME,aAAa/F,MAAM,CAAC8F,UAAU;QACpC,IAAIC,YAEF;YAAA,IAAI,AAAsB,YAAtB,OAAOA,YACT/F,MAAM,CAAC8F,UAAU,GAAGC;iBACf,IAAI,AAAsB,YAAtB,OAAOA,YAEhB;gBAAA,IAAIA,WAAW,MAAM,EAEnB;oBAAA,IAAI,AAA6B,YAA7B,OAAOA,WAAW,MAAM,EAC1B/F,MAAM,CAAC8F,UAAU,GAAGC,WAAW,MAAM;yBAChC,IACL,AAA6B,YAA7B,OAAOA,WAAW,MAAM,IACxBA,WAAW,MAAM,CAAC,MAAM,EAGxB/F,MAAM,CAAC8F,UAAU,GAAGC,WAAW,MAAM,CAAC,MAAM;gBAC9C;YACF;QACF;IAEJ;IAEA,OAAO/F;AACT;AAEO,MAAMgG,kBAAkB,CAC7BL,YACAC;IAEA,MAAMC,gBAAgBX,4BAA4BU;IAClD,MAAM5F,SAAS;QAAE,GAAG2F,UAAU;IAAC;IAE/B,KAAK,MAAMG,aAAaD,cAAe;QACrC,MAAME,aAAa/F,MAAM,CAAC8F,UAAU;QACpC,IAAIC,cAAc,AAAsB,YAAtB,OAAOA,YACvB/F,MAAM,CAAC8F,UAAU,GAAG;YAClB,CAACvB,oBAAoB,EAAE;YACvB,QAAQwB;QACV;IAEJ;IAEA,OAAO/F;AACT"}
|
|
1
|
+
{"version":3,"file":"ai-model/common.mjs","sources":["webpack://@midscene/core/./src/ai-model/common.ts"],"sourcesContent":["import type {\n AIUsageInfo,\n BaseElement,\n DeviceAction,\n ElementTreeNode,\n MidsceneYamlFlowItem,\n PlanningAction,\n Rect,\n Size,\n} from '@/types';\nimport { assert } from '@midscene/shared/utils';\n\nimport type {\n ChatCompletionSystemMessageParam,\n ChatCompletionUserMessageParam,\n} from 'openai/resources/index';\nimport { callToGetJSONObject } from './service-caller/index';\n\nimport type { PlanningLocateParam } from '@/types';\nimport { NodeType } from '@midscene/shared/constants';\nimport {\n type IModelPreferences,\n getModelName,\n vlLocateMode,\n} from '@midscene/shared/env';\nimport { treeToList } from '@midscene/shared/extractor';\nimport { compositeElementInfoImg } from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { z } from 'zod';\n\nexport type AIArgs = [\n ChatCompletionSystemMessageParam,\n ...ChatCompletionUserMessageParam[],\n];\n\nexport enum AIActionType {\n ASSERT = 0,\n INSPECT_ELEMENT = 1,\n EXTRACT_DATA = 2,\n PLAN = 3,\n DESCRIBE_ELEMENT = 4,\n}\n\nexport async function callAiFn<T>(\n msgs: AIArgs,\n AIActionTypeValue: AIActionType,\n modelPreferences: IModelPreferences,\n): Promise<{ content: T; usage?: AIUsageInfo }> {\n const jsonObject = await callToGetJSONObject<T>(\n msgs,\n AIActionTypeValue,\n modelPreferences,\n );\n\n return {\n content: jsonObject.content,\n usage: jsonObject.usage,\n };\n}\n\nconst defaultBboxSize = 20; // must be even number\nconst debugInspectUtils = getDebug('ai:common');\n\n// transform the param of locate from qwen mode\nexport function fillBboxParam(\n locate: PlanningLocateParam,\n width: number,\n height: number,\n modelPreferences: IModelPreferences,\n) {\n // The Qwen model might have hallucinations of naming bbox as bbox_2d.\n if ((locate as any).bbox_2d && !locate?.bbox) {\n locate.bbox = (locate as any).bbox_2d;\n // biome-ignore lint/performance/noDelete: <explanation>\n delete (locate as any).bbox_2d;\n }\n\n if (locate?.bbox) {\n locate.bbox = adaptBbox(locate.bbox, width, height, modelPreferences);\n }\n\n return locate;\n}\n\nexport function adaptQwenBbox(\n bbox: number[],\n): [number, number, number, number] {\n if (bbox.length < 2) {\n const msg = `invalid bbox data for qwen-vl mode: ${JSON.stringify(bbox)} `;\n throw new Error(msg);\n }\n\n const result: [number, number, number, number] = [\n Math.round(bbox[0]),\n Math.round(bbox[1]),\n typeof bbox[2] === 'number'\n ? Math.round(bbox[2])\n : Math.round(bbox[0] + defaultBboxSize),\n typeof bbox[3] === 'number'\n ? Math.round(bbox[3])\n : Math.round(bbox[1] + defaultBboxSize),\n ];\n return result;\n}\n\nexport function adaptDoubaoBbox(\n bbox: string[] | number[] | string,\n width: number,\n height: number,\n): [number, number, number, number] {\n assert(\n width > 0 && height > 0,\n 'width and height must be greater than 0 in doubao mode',\n );\n\n if (typeof bbox === 'string') {\n assert(\n /^(\\d+)\\s(\\d+)\\s(\\d+)\\s(\\d+)$/.test(bbox.trim()),\n `invalid bbox data string for doubao-vision mode: ${bbox}`,\n );\n const splitted = bbox.split(' ');\n if (splitted.length === 4) {\n return [\n Math.round((Number(splitted[0]) * width) / 1000),\n Math.round((Number(splitted[1]) * height) / 1000),\n Math.round((Number(splitted[2]) * width) / 1000),\n Math.round((Number(splitted[3]) * height) / 1000),\n ];\n }\n throw new Error(`invalid bbox data string for doubao-vision mode: ${bbox}`);\n }\n\n if (Array.isArray(bbox) && Array.isArray(bbox[0])) {\n bbox = bbox[0];\n }\n\n let bboxList: number[] = [];\n if (Array.isArray(bbox) && typeof bbox[0] === 'string') {\n bbox.forEach((item) => {\n if (typeof item === 'string' && item.includes(',')) {\n const [x, y] = item.split(',');\n bboxList.push(Number(x.trim()), Number(y.trim()));\n } else if (typeof item === 'string' && item.includes(' ')) {\n const [x, y] = item.split(' ');\n bboxList.push(Number(x.trim()), Number(y.trim()));\n } else {\n bboxList.push(Number(item));\n }\n });\n } else {\n bboxList = bbox as any;\n }\n\n if (bboxList.length === 4 || bboxList.length === 5) {\n return [\n Math.round((bboxList[0] * width) / 1000),\n Math.round((bboxList[1] * height) / 1000),\n Math.round((bboxList[2] * width) / 1000),\n Math.round((bboxList[3] * height) / 1000),\n ];\n }\n\n // treat the bbox as a center point\n if (\n bboxList.length === 6 ||\n bboxList.length === 2 ||\n bboxList.length === 3 ||\n bboxList.length === 7\n ) {\n return [\n Math.max(\n 0,\n Math.round((bboxList[0] * width) / 1000) - defaultBboxSize / 2,\n ),\n Math.max(\n 0,\n Math.round((bboxList[1] * height) / 1000) - defaultBboxSize / 2,\n ),\n Math.min(\n width,\n Math.round((bboxList[0] * width) / 1000) + defaultBboxSize / 2,\n ),\n Math.min(\n height,\n Math.round((bboxList[1] * height) / 1000) + defaultBboxSize / 2,\n ),\n ];\n }\n\n if (bbox.length === 8) {\n return [\n Math.round((bboxList[0] * width) / 1000),\n Math.round((bboxList[1] * height) / 1000),\n Math.round((bboxList[4] * width) / 1000),\n Math.round((bboxList[5] * height) / 1000),\n ];\n }\n\n const msg = `invalid bbox data for doubao-vision mode: ${JSON.stringify(bbox)} `;\n throw new Error(msg);\n}\n\nexport function adaptBbox(\n bbox: number[],\n width: number,\n height: number,\n modelPreferences: IModelPreferences,\n): [number, number, number, number] {\n if (\n vlLocateMode(modelPreferences) === 'doubao-vision' ||\n vlLocateMode(modelPreferences) === 'vlm-ui-tars'\n ) {\n return adaptDoubaoBbox(bbox, width, height);\n }\n\n if (vlLocateMode(modelPreferences) === 'gemini') {\n return adaptGeminiBbox(bbox, width, height);\n }\n\n return adaptQwenBbox(bbox);\n}\n\nexport function adaptGeminiBbox(\n bbox: number[],\n width: number,\n height: number,\n): [number, number, number, number] {\n const left = Math.round((bbox[1] * width) / 1000);\n const top = Math.round((bbox[0] * height) / 1000);\n const right = Math.round((bbox[3] * width) / 1000);\n const bottom = Math.round((bbox[2] * height) / 1000);\n return [left, top, right, bottom];\n}\n\nexport function adaptBboxToRect(\n bbox: number[],\n width: number,\n height: number,\n modelPreferences: IModelPreferences,\n offsetX = 0,\n offsetY = 0,\n): Rect {\n debugInspectUtils('adaptBboxToRect', bbox, width, height, offsetX, offsetY);\n const [left, top, right, bottom] = adaptBbox(\n bbox,\n width,\n height,\n modelPreferences,\n );\n\n // Calculate initial rect dimensions\n const rectLeft = left;\n const rectTop = top;\n let rectWidth = right - left;\n let rectHeight = bottom - top;\n\n // Ensure the rect doesn't exceed image boundaries\n // If right edge exceeds width, adjust the width\n if (rectLeft + rectWidth > width) {\n rectWidth = width - rectLeft;\n }\n\n // If bottom edge exceeds height, adjust the height\n if (rectTop + rectHeight > height) {\n rectHeight = height - rectTop;\n }\n\n // Ensure minimum dimensions (width and height should be at least 1)\n rectWidth = Math.max(1, rectWidth);\n rectHeight = Math.max(1, rectHeight);\n\n const rect = {\n left: rectLeft + offsetX,\n top: rectTop + offsetY,\n width: rectWidth,\n height: rectHeight,\n };\n debugInspectUtils('adaptBboxToRect, result=', rect);\n return rect;\n}\n\nlet warned = false;\nexport function warnGPT4oSizeLimit(\n size: Size,\n modelPreferences: IModelPreferences,\n) {\n if (warned) return;\n if (getModelName(modelPreferences)?.toLowerCase().includes('gpt-4o')) {\n const warningMsg = `GPT-4o has a maximum image input size of 2000x768 or 768x2000, but got ${size.width}x${size.height}. Please set your interface to a smaller resolution. Otherwise, the result may be inaccurate.`;\n\n if (\n Math.max(size.width, size.height) > 2000 ||\n Math.min(size.width, size.height) > 768\n ) {\n console.warn(warningMsg);\n warned = true;\n }\n } else if (size.width > 1800 || size.height > 1800) {\n console.warn(\n `The image size seems too large (${size.width}x${size.height}). It may lead to more token usage, slower response, and inaccurate result.`,\n );\n warned = true;\n }\n}\n\nexport function mergeRects(rects: Rect[]) {\n const minLeft = Math.min(...rects.map((r) => r.left));\n const minTop = Math.min(...rects.map((r) => r.top));\n const maxRight = Math.max(...rects.map((r) => r.left + r.width));\n const maxBottom = Math.max(...rects.map((r) => r.top + r.height));\n return {\n left: minLeft,\n top: minTop,\n width: maxRight - minLeft,\n height: maxBottom - minTop,\n };\n}\n\n// expand the search area to at least 300 x 300, or add a default padding\nexport function expandSearchArea(\n rect: Rect,\n screenSize: Size,\n modelPreferences: IModelPreferences,\n) {\n const minEdgeSize =\n vlLocateMode(modelPreferences) === 'doubao-vision' ? 500 : 300;\n const defaultPadding = 160;\n\n const paddingSizeHorizontal =\n rect.width < minEdgeSize\n ? Math.ceil((minEdgeSize - rect.width) / 2)\n : defaultPadding;\n const paddingSizeVertical =\n rect.height < minEdgeSize\n ? Math.ceil((minEdgeSize - rect.height) / 2)\n : defaultPadding;\n rect.left = Math.max(0, rect.left - paddingSizeHorizontal);\n rect.width = Math.min(\n rect.width + paddingSizeHorizontal * 2,\n screenSize.width - rect.left,\n );\n rect.top = Math.max(0, rect.top - paddingSizeVertical);\n rect.height = Math.min(\n rect.height + paddingSizeVertical * 2,\n screenSize.height - rect.top,\n );\n return rect;\n}\n\nexport async function markupImageForLLM(\n screenshotBase64: string,\n tree: ElementTreeNode<BaseElement>,\n size: Size,\n) {\n const elementsInfo = treeToList(tree);\n const elementsPositionInfoWithoutText = elementsInfo!.filter(\n (elementInfo) => {\n if (elementInfo.attributes.nodeType === NodeType.TEXT) {\n return false;\n }\n return true;\n },\n );\n\n const imagePayload = await compositeElementInfoImg({\n inputImgBase64: screenshotBase64,\n elementsPositionInfo: elementsPositionInfoWithoutText,\n size,\n });\n return imagePayload;\n}\n\nexport function buildYamlFlowFromPlans(\n plans: PlanningAction[],\n actionSpace: DeviceAction<any>[],\n sleep?: number,\n): MidsceneYamlFlowItem[] {\n const flow: MidsceneYamlFlowItem[] = [];\n\n for (const plan of plans) {\n const verb = plan.type;\n\n const action = actionSpace.find((action) => action.name === verb);\n if (!action) {\n console.warn(\n `Cannot convert action ${verb} to yaml flow. Will ignore it.`,\n );\n continue;\n }\n\n const flowKey = action.interfaceAlias || verb;\n const flowParam = action.paramSchema\n ? dumpActionParam(plan.param || {}, action.paramSchema)\n : {};\n\n const flowItem: MidsceneYamlFlowItem = {\n [flowKey]: '',\n ...flowParam,\n };\n\n flow.push(flowItem);\n }\n\n if (sleep) {\n flow.push({\n sleep,\n });\n }\n\n return flow;\n}\n\n// Zod schemas for shared types\nexport const PointSchema = z.object({\n left: z.number(),\n top: z.number(),\n});\n\nexport const SizeSchema = z.object({\n width: z.number(),\n height: z.number(),\n dpr: z.number().optional(),\n});\n\nexport const RectSchema = PointSchema.and(SizeSchema).and(\n z.object({\n zoom: z.number().optional(),\n }),\n);\n\n// Zod schema for TMultimodalPrompt\nexport const TMultimodalPromptSchema = z.object({\n images: z\n .array(\n z.object({\n name: z.string(),\n url: z.string(),\n }),\n )\n .optional(),\n convertHttpImage2Base64: z.boolean().optional(),\n});\n\n// Zod schema for TUserPrompt\nexport const TUserPromptSchema = z.union([\n z.string(),\n z\n .object({\n prompt: z.string(),\n })\n .and(TMultimodalPromptSchema.partial()),\n]);\n\n// Generate TypeScript types from Zod schemas\nexport type TMultimodalPrompt = z.infer<typeof TMultimodalPromptSchema>;\nexport type TUserPrompt = z.infer<typeof TUserPromptSchema>;\n\nconst locateFieldFlagName = 'midscene_location_field_flag';\n\nconst MidsceneLocationResult = z\n .object({\n [locateFieldFlagName]: z.literal(true),\n prompt: TUserPromptSchema,\n\n // optional fields\n deepThink: z.boolean().optional(), // only available in vl model\n cacheable: z.boolean().optional(),\n xpath: z.boolean().optional(), // preset result for xpath\n\n // these two fields will only appear in the result\n center: z.tuple([z.number(), z.number()]),\n rect: RectSchema,\n })\n .passthrough();\n\nexport type MidsceneLocationResultType = z.infer<typeof MidsceneLocationResult>;\nexport const getMidsceneLocationSchema = () => {\n return MidsceneLocationResult;\n};\n\nexport const ifMidsceneLocatorField = (field: any): boolean => {\n // Handle optional fields by getting the inner type\n let actualField = field;\n if (actualField._def?.typeName === 'ZodOptional') {\n actualField = actualField._def.innerType;\n }\n\n // Check if this is a ZodUnion (the new MidsceneLocation structure)\n if (actualField._def?.typeName === 'ZodObject') {\n const shape = actualField._def.shape();\n return locateFieldFlagName in shape;\n }\n\n return false;\n};\n\nexport const dumpMidsceneLocatorField = (field: any): string => {\n assert(\n ifMidsceneLocatorField(field),\n 'field is not a midscene locator field',\n );\n\n // If field is a string, return it directly\n if (typeof field === 'string') {\n return field;\n }\n\n // If field is an object with prompt property\n if (field && typeof field === 'object' && field.prompt) {\n // If prompt is a string, return it directly\n if (typeof field.prompt === 'string') {\n return field.prompt;\n }\n // If prompt is a TUserPrompt object, extract the prompt string\n if (typeof field.prompt === 'object' && field.prompt.prompt) {\n return field.prompt.prompt; // TODO: dump images if necessary\n }\n }\n\n // Fallback: try to convert to string\n return String(field);\n};\n\nexport const findAllMidsceneLocatorField = (\n zodType?: z.ZodType<any>,\n requiredOnly?: boolean,\n): string[] => {\n if (!zodType) {\n return [];\n }\n\n // Check if this is a ZodObject by checking if it has a shape property\n const zodObject = zodType as any;\n if (zodObject._def?.typeName === 'ZodObject' && zodObject.shape) {\n const keys = Object.keys(zodObject.shape);\n return keys.filter((key) => {\n const field = zodObject.shape[key];\n if (!ifMidsceneLocatorField(field)) {\n return false;\n }\n\n // If requiredOnly is true, filter out optional fields\n if (requiredOnly) {\n return field._def?.typeName !== 'ZodOptional';\n }\n\n return true;\n });\n }\n\n // For other ZodType instances, we can't extract field names\n return [];\n};\n\nexport const dumpActionParam = (\n jsonObject: Record<string, any>,\n zodSchema: z.ZodType<any>,\n): Record<string, any> => {\n const locatorFields = findAllMidsceneLocatorField(zodSchema);\n const result = { ...jsonObject };\n\n for (const fieldName of locatorFields) {\n const fieldValue = result[fieldName];\n if (fieldValue) {\n // If it's already a string, keep it as is\n if (typeof fieldValue === 'string') {\n result[fieldName] = fieldValue;\n } else if (typeof fieldValue === 'object') {\n // Check if this field is actually a MidsceneLocationType object\n if (fieldValue.prompt) {\n // If prompt is a string, use it directly\n if (typeof fieldValue.prompt === 'string') {\n result[fieldName] = fieldValue.prompt;\n } else if (\n typeof fieldValue.prompt === 'object' &&\n fieldValue.prompt.prompt\n ) {\n // If prompt is a TUserPrompt object, extract the prompt string\n result[fieldName] = fieldValue.prompt.prompt;\n }\n }\n }\n }\n }\n\n return result;\n};\n\nexport const loadActionParam = (\n jsonObject: Record<string, any>,\n zodSchema: z.ZodType<any>,\n): Record<string, any> => {\n const locatorFields = findAllMidsceneLocatorField(zodSchema);\n const result = { ...jsonObject };\n\n for (const fieldName of locatorFields) {\n const fieldValue = result[fieldName];\n if (fieldValue && typeof fieldValue === 'string') {\n result[fieldName] = {\n [locateFieldFlagName]: true,\n prompt: fieldValue,\n };\n }\n }\n\n return result;\n};\n"],"names":["AIActionType","callAiFn","msgs","AIActionTypeValue","modelPreferences","jsonObject","callToGetJSONObject","defaultBboxSize","debugInspectUtils","getDebug","fillBboxParam","locate","width","height","adaptBbox","adaptQwenBbox","bbox","msg","JSON","Error","result","Math","adaptDoubaoBbox","assert","splitted","Number","Array","bboxList","item","x","y","vlLocateMode","adaptGeminiBbox","left","top","right","bottom","adaptBboxToRect","offsetX","offsetY","rectLeft","rectTop","rectWidth","rectHeight","rect","warned","warnGPT4oSizeLimit","size","_getModelName","warningMsg","console","mergeRects","rects","minLeft","r","minTop","maxRight","maxBottom","expandSearchArea","screenSize","minEdgeSize","defaultPadding","paddingSizeHorizontal","paddingSizeVertical","markupImageForLLM","screenshotBase64","tree","elementsInfo","treeToList","elementsPositionInfoWithoutText","elementInfo","NodeType","imagePayload","compositeElementInfoImg","buildYamlFlowFromPlans","plans","actionSpace","sleep","flow","plan","verb","action","flowKey","flowParam","dumpActionParam","flowItem","PointSchema","z","SizeSchema","RectSchema","TMultimodalPromptSchema","TUserPromptSchema","locateFieldFlagName","MidsceneLocationResult","getMidsceneLocationSchema","ifMidsceneLocatorField","field","_actualField__def","_actualField__def1","actualField","shape","dumpMidsceneLocatorField","String","findAllMidsceneLocatorField","zodType","requiredOnly","_zodObject__def","zodObject","keys","Object","key","_field__def","zodSchema","locatorFields","fieldName","fieldValue","loadActionParam"],"mappings":";;;;;;;;AAmCO,IAAKA,sBAAYA,WAAAA,GAAAA,SAAZA,YAAY;;;;;;WAAZA;;AAQL,eAAeC,SACpBC,IAAY,EACZC,iBAA+B,EAC/BC,gBAAmC;IAEnC,MAAMC,aAAa,MAAMC,oBACvBJ,MACAC,mBACAC;IAGF,OAAO;QACL,SAASC,WAAW,OAAO;QAC3B,OAAOA,WAAW,KAAK;IACzB;AACF;AAEA,MAAME,kBAAkB;AACxB,MAAMC,oBAAoBC,SAAS;AAG5B,SAASC,cACdC,MAA2B,EAC3BC,KAAa,EACbC,MAAc,EACdT,gBAAmC;IAGnC,IAAKO,OAAe,OAAO,IAAI,CAACA,CAAAA,QAAAA,SAAAA,KAAAA,IAAAA,OAAQ,IAAI,AAAD,GAAG;QAC5CA,OAAO,IAAI,GAAIA,OAAe,OAAO;QAErC,OAAQA,OAAe,OAAO;IAChC;IAEA,IAAIA,QAAAA,SAAAA,KAAAA,IAAAA,OAAQ,IAAI,EACdA,OAAO,IAAI,GAAGG,UAAUH,OAAO,IAAI,EAAEC,OAAOC,QAAQT;IAGtD,OAAOO;AACT;AAEO,SAASI,cACdC,IAAc;IAEd,IAAIA,KAAK,MAAM,GAAG,GAAG;QACnB,MAAMC,MAAM,CAAC,oCAAoC,EAAEC,KAAK,SAAS,CAACF,MAAM,CAAC,CAAC;QAC1E,MAAM,IAAIG,MAAMF;IAClB;IAEA,MAAMG,SAA2C;QAC/CC,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE;QAClBK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE;QACC,YAAnB,OAAOA,IAAI,CAAC,EAAE,GACVK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,IAClBK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,GAAGT;QACN,YAAnB,OAAOS,IAAI,CAAC,EAAE,GACVK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,IAClBK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,GAAGT;KAC1B;IACD,OAAOa;AACT;AAEO,SAASE,gBACdN,IAAkC,EAClCJ,KAAa,EACbC,MAAc;IAEdU,OACEX,QAAQ,KAAKC,SAAS,GACtB;IAGF,IAAI,AAAgB,YAAhB,OAAOG,MAAmB;QAC5BO,OACE,+BAA+B,IAAI,CAACP,KAAK,IAAI,KAC7C,CAAC,iDAAiD,EAAEA,MAAM;QAE5D,MAAMQ,WAAWR,KAAK,KAAK,CAAC;QAC5B,IAAIQ,AAAoB,MAApBA,SAAS,MAAM,EACjB,OAAO;YACLH,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAIZ,QAAS;YAC3CS,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAIX,SAAU;YAC5CQ,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAIZ,QAAS;YAC3CS,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAIX,SAAU;SAC7C;QAEH,MAAM,IAAIM,MAAM,CAAC,iDAAiD,EAAEH,MAAM;IAC5E;IAEA,IAAIU,MAAM,OAAO,CAACV,SAASU,MAAM,OAAO,CAACV,IAAI,CAAC,EAAE,GAC9CA,OAAOA,IAAI,CAAC,EAAE;IAGhB,IAAIW,WAAqB,EAAE;IAC3B,IAAID,MAAM,OAAO,CAACV,SAAS,AAAmB,YAAnB,OAAOA,IAAI,CAAC,EAAE,EACvCA,KAAK,OAAO,CAAC,CAACY;QACZ,IAAI,AAAgB,YAAhB,OAAOA,QAAqBA,KAAK,QAAQ,CAAC,MAAM;YAClD,MAAM,CAACC,GAAGC,EAAE,GAAGF,KAAK,KAAK,CAAC;YAC1BD,SAAS,IAAI,CAACF,OAAOI,EAAE,IAAI,KAAKJ,OAAOK,EAAE,IAAI;QAC/C,OAAO,IAAI,AAAgB,YAAhB,OAAOF,QAAqBA,KAAK,QAAQ,CAAC,MAAM;YACzD,MAAM,CAACC,GAAGC,EAAE,GAAGF,KAAK,KAAK,CAAC;YAC1BD,SAAS,IAAI,CAACF,OAAOI,EAAE,IAAI,KAAKJ,OAAOK,EAAE,IAAI;QAC/C,OACEH,SAAS,IAAI,CAACF,OAAOG;IAEzB;SAEAD,WAAWX;IAGb,IAAIW,AAAoB,MAApBA,SAAS,MAAM,IAAUA,AAAoB,MAApBA,SAAS,MAAM,EAC1C,OAAO;QACLN,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,QAAS;QACnCS,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGd,SAAU;QACpCQ,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,QAAS;QACnCS,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGd,SAAU;KACrC;IAIH,IACEc,AAAoB,MAApBA,SAAS,MAAM,IACfA,AAAoB,MAApBA,SAAS,MAAM,IACfA,AAAoB,MAApBA,SAAS,MAAM,IACfA,AAAoB,MAApBA,SAAS,MAAM,EAEf,OAAO;QACLN,KAAK,GAAG,CACN,GACAA,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,QAAS,QAAQL,kBAAkB;QAE/Dc,KAAK,GAAG,CACN,GACAA,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGd,SAAU,QAAQN,kBAAkB;QAEhEc,KAAK,GAAG,CACNT,OACAS,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,QAAS,QAAQL,kBAAkB;QAE/Dc,KAAK,GAAG,CACNR,QACAQ,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGd,SAAU,QAAQN,kBAAkB;KAEjE;IAGH,IAAIS,AAAgB,MAAhBA,KAAK,MAAM,EACb,OAAO;QACLK,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,QAAS;QACnCS,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGd,SAAU;QACpCQ,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,QAAS;QACnCS,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGd,SAAU;KACrC;IAGH,MAAMI,MAAM,CAAC,0CAA0C,EAAEC,KAAK,SAAS,CAACF,MAAM,CAAC,CAAC;IAChF,MAAM,IAAIG,MAAMF;AAClB;AAEO,SAASH,UACdE,IAAc,EACdJ,KAAa,EACbC,MAAc,EACdT,gBAAmC;IAEnC,IACE2B,AAAmC,oBAAnCA,aAAa3B,qBACb2B,AAAmC,kBAAnCA,aAAa3B,mBAEb,OAAOkB,gBAAgBN,MAAMJ,OAAOC;IAGtC,IAAIkB,AAAmC,aAAnCA,aAAa3B,mBACf,OAAO4B,gBAAgBhB,MAAMJ,OAAOC;IAGtC,OAAOE,cAAcC;AACvB;AAEO,SAASgB,gBACdhB,IAAc,EACdJ,KAAa,EACbC,MAAc;IAEd,MAAMoB,OAAOZ,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGJ,QAAS;IAC5C,MAAMsB,MAAMb,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGH,SAAU;IAC5C,MAAMsB,QAAQd,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGJ,QAAS;IAC7C,MAAMwB,SAASf,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGH,SAAU;IAC/C,OAAO;QAACoB;QAAMC;QAAKC;QAAOC;KAAO;AACnC;AAEO,SAASC,gBACdrB,IAAc,EACdJ,KAAa,EACbC,MAAc,EACdT,gBAAmC,EACnCkC,UAAU,CAAC,EACXC,UAAU,CAAC;IAEX/B,kBAAkB,mBAAmBQ,MAAMJ,OAAOC,QAAQyB,SAASC;IACnE,MAAM,CAACN,MAAMC,KAAKC,OAAOC,OAAO,GAAGtB,UACjCE,MACAJ,OACAC,QACAT;IAIF,MAAMoC,WAAWP;IACjB,MAAMQ,UAAUP;IAChB,IAAIQ,YAAYP,QAAQF;IACxB,IAAIU,aAAaP,SAASF;IAI1B,IAAIM,WAAWE,YAAY9B,OACzB8B,YAAY9B,QAAQ4B;IAItB,IAAIC,UAAUE,aAAa9B,QACzB8B,aAAa9B,SAAS4B;IAIxBC,YAAYrB,KAAK,GAAG,CAAC,GAAGqB;IACxBC,aAAatB,KAAK,GAAG,CAAC,GAAGsB;IAEzB,MAAMC,OAAO;QACX,MAAMJ,WAAWF;QACjB,KAAKG,UAAUF;QACf,OAAOG;QACP,QAAQC;IACV;IACAnC,kBAAkB,4BAA4BoC;IAC9C,OAAOA;AACT;AAEA,IAAIC,SAAS;AACN,SAASC,mBACdC,IAAU,EACV3C,gBAAmC;QAG/B4C;IADJ,IAAIH,QAAQ;IACZ,IAAI,QAAAG,CAAAA,gBAAAA,aAAa5C,iBAAgB,IAA7B4C,KAAAA,IAAAA,cAAgC,WAAW,GAAG,QAAQ,CAAC,WAAW;QACpE,MAAMC,aAAa,CAAC,uEAAuE,EAAEF,KAAK,KAAK,CAAC,CAAC,EAAEA,KAAK,MAAM,CAAC,6FAA6F,CAAC;QAErN,IACE1B,KAAK,GAAG,CAAC0B,KAAK,KAAK,EAAEA,KAAK,MAAM,IAAI,QACpC1B,KAAK,GAAG,CAAC0B,KAAK,KAAK,EAAEA,KAAK,MAAM,IAAI,KACpC;YACAG,QAAQ,IAAI,CAACD;YACbJ,SAAS;QACX;IACF,OAAO,IAAIE,KAAK,KAAK,GAAG,QAAQA,KAAK,MAAM,GAAG,MAAM;QAClDG,QAAQ,IAAI,CACV,CAAC,gCAAgC,EAAEH,KAAK,KAAK,CAAC,CAAC,EAAEA,KAAK,MAAM,CAAC,2EAA2E,CAAC;QAE3IF,SAAS;IACX;AACF;AAEO,SAASM,WAAWC,KAAa;IACtC,MAAMC,UAAUhC,KAAK,GAAG,IAAI+B,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,IAAI;IACnD,MAAMC,SAASlC,KAAK,GAAG,IAAI+B,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,GAAG;IACjD,MAAME,WAAWnC,KAAK,GAAG,IAAI+B,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,IAAI,GAAGA,EAAE,KAAK;IAC9D,MAAMG,YAAYpC,KAAK,GAAG,IAAI+B,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,GAAG,GAAGA,EAAE,MAAM;IAC/D,OAAO;QACL,MAAMD;QACN,KAAKE;QACL,OAAOC,WAAWH;QAClB,QAAQI,YAAYF;IACtB;AACF;AAGO,SAASG,iBACdd,IAAU,EACVe,UAAgB,EAChBvD,gBAAmC;IAEnC,MAAMwD,cACJ7B,AAAmC,oBAAnCA,aAAa3B,oBAAwC,MAAM;IAC7D,MAAMyD,iBAAiB;IAEvB,MAAMC,wBACJlB,KAAK,KAAK,GAAGgB,cACTvC,KAAK,IAAI,CAAEuC,AAAAA,CAAAA,cAAchB,KAAK,KAAI,IAAK,KACvCiB;IACN,MAAME,sBACJnB,KAAK,MAAM,GAAGgB,cACVvC,KAAK,IAAI,CAAEuC,AAAAA,CAAAA,cAAchB,KAAK,MAAK,IAAK,KACxCiB;IACNjB,KAAK,IAAI,GAAGvB,KAAK,GAAG,CAAC,GAAGuB,KAAK,IAAI,GAAGkB;IACpClB,KAAK,KAAK,GAAGvB,KAAK,GAAG,CACnBuB,KAAK,KAAK,GAAGkB,AAAwB,IAAxBA,uBACbH,WAAW,KAAK,GAAGf,KAAK,IAAI;IAE9BA,KAAK,GAAG,GAAGvB,KAAK,GAAG,CAAC,GAAGuB,KAAK,GAAG,GAAGmB;IAClCnB,KAAK,MAAM,GAAGvB,KAAK,GAAG,CACpBuB,KAAK,MAAM,GAAGmB,AAAsB,IAAtBA,qBACdJ,WAAW,MAAM,GAAGf,KAAK,GAAG;IAE9B,OAAOA;AACT;AAEO,eAAeoB,kBACpBC,gBAAwB,EACxBC,IAAkC,EAClCnB,IAAU;IAEV,MAAMoB,eAAeC,WAAWF;IAChC,MAAMG,kCAAkCF,aAAc,MAAM,CAC1D,CAACG;QACC,IAAIA,YAAY,UAAU,CAAC,QAAQ,KAAKC,SAAS,IAAI,EACnD,OAAO;QAET,OAAO;IACT;IAGF,MAAMC,eAAe,MAAMC,wBAAwB;QACjD,gBAAgBR;QAChB,sBAAsBI;QACtBtB;IACF;IACA,OAAOyB;AACT;AAEO,SAASE,uBACdC,KAAuB,EACvBC,WAAgC,EAChCC,KAAc;IAEd,MAAMC,OAA+B,EAAE;IAEvC,KAAK,MAAMC,QAAQJ,MAAO;QACxB,MAAMK,OAAOD,KAAK,IAAI;QAEtB,MAAME,SAASL,YAAY,IAAI,CAAC,CAACK,SAAWA,OAAO,IAAI,KAAKD;QAC5D,IAAI,CAACC,QAAQ;YACX/B,QAAQ,IAAI,CACV,CAAC,sBAAsB,EAAE8B,KAAK,8BAA8B,CAAC;YAE/D;QACF;QAEA,MAAME,UAAUD,OAAO,cAAc,IAAID;QACzC,MAAMG,YAAYF,OAAO,WAAW,GAChCG,gBAAgBL,KAAK,KAAK,IAAI,CAAC,GAAGE,OAAO,WAAW,IACpD,CAAC;QAEL,MAAMI,WAAiC;YACrC,CAACH,QAAQ,EAAE;YACX,GAAGC,SAAS;QACd;QAEAL,KAAK,IAAI,CAACO;IACZ;IAEA,IAAIR,OACFC,KAAK,IAAI,CAAC;QACRD;IACF;IAGF,OAAOC;AACT;AAGO,MAAMQ,cAAcC,EAAE,MAAM,CAAC;IAClC,MAAMA,EAAE,MAAM;IACd,KAAKA,EAAE,MAAM;AACf;AAEO,MAAMC,aAAaD,EAAE,MAAM,CAAC;IACjC,OAAOA,EAAE,MAAM;IACf,QAAQA,EAAE,MAAM;IAChB,KAAKA,EAAE,MAAM,GAAG,QAAQ;AAC1B;AAEO,MAAME,aAAaH,YAAY,GAAG,CAACE,YAAY,GAAG,CACvDD,EAAE,MAAM,CAAC;IACP,MAAMA,EAAE,MAAM,GAAG,QAAQ;AAC3B;AAIK,MAAMG,0BAA0BH,EAAE,MAAM,CAAC;IAC9C,QAAQA,EAAAA,KACA,CACJA,EAAE,MAAM,CAAC;QACP,MAAMA,EAAE,MAAM;QACd,KAAKA,EAAE,MAAM;IACf,IAED,QAAQ;IACX,yBAAyBA,EAAE,OAAO,GAAG,QAAQ;AAC/C;AAGO,MAAMI,oBAAoBJ,EAAE,KAAK,CAAC;IACvCA,EAAE,MAAM;IACRA,EAAAA,MACS,CAAC;QACN,QAAQA,EAAE,MAAM;IAClB,GACC,GAAG,CAACG,wBAAwB,OAAO;CACvC;AAMD,MAAME,sBAAsB;AAE5B,MAAMC,yBAAyBN,EAAAA,MACtB,CAAC;IACN,CAACK,oBAAoB,EAAEL,EAAE,OAAO,CAAC;IACjC,QAAQI;IAGR,WAAWJ,EAAE,OAAO,GAAG,QAAQ;IAC/B,WAAWA,EAAE,OAAO,GAAG,QAAQ;IAC/B,OAAOA,EAAE,OAAO,GAAG,QAAQ;IAG3B,QAAQA,EAAE,KAAK,CAAC;QAACA,EAAE,MAAM;QAAIA,EAAE,MAAM;KAAG;IACxC,MAAME;AACR,GACC,WAAW;AAGP,MAAMK,4BAA4B,IAChCD;AAGF,MAAME,yBAAyB,CAACC;QAGjCC,mBAKAC;IANJ,IAAIC,cAAcH;IAClB,IAAIC,AAAAA,SAAAA,CAAAA,oBAAAA,YAAY,IAAI,AAAD,IAAfA,KAAAA,IAAAA,kBAAkB,QAAQ,AAAD,MAAM,eACjCE,cAAcA,YAAY,IAAI,CAAC,SAAS;IAI1C,IAAID,AAAAA,SAAAA,CAAAA,qBAAAA,YAAY,IAAI,AAAD,IAAfA,KAAAA,IAAAA,mBAAkB,QAAQ,AAAD,MAAM,aAAa;QAC9C,MAAME,QAAQD,YAAY,IAAI,CAAC,KAAK;QACpC,OAAOP,uBAAuBQ;IAChC;IAEA,OAAO;AACT;AAEO,MAAMC,2BAA2B,CAACL;IACvCzE,OACEwE,uBAAuBC,QACvB;IAIF,IAAI,AAAiB,YAAjB,OAAOA,OACT,OAAOA;IAIT,IAAIA,SAAS,AAAiB,YAAjB,OAAOA,SAAsBA,MAAM,MAAM,EAAE;QAEtD,IAAI,AAAwB,YAAxB,OAAOA,MAAM,MAAM,EACrB,OAAOA,MAAM,MAAM;QAGrB,IAAI,AAAwB,YAAxB,OAAOA,MAAM,MAAM,IAAiBA,MAAM,MAAM,CAAC,MAAM,EACzD,OAAOA,MAAM,MAAM,CAAC,MAAM;IAE9B;IAGA,OAAOM,OAAON;AAChB;AAEO,MAAMO,8BAA8B,CACzCC,SACAC;QAQIC;IANJ,IAAI,CAACF,SACH,OAAO,EAAE;IAIX,MAAMG,YAAYH;IAClB,IAAIE,AAAAA,SAAAA,CAAAA,kBAAAA,UAAU,IAAI,AAAD,IAAbA,KAAAA,IAAAA,gBAAgB,QAAQ,AAAD,MAAM,eAAeC,UAAU,KAAK,EAAE;QAC/D,MAAMC,OAAOC,OAAO,IAAI,CAACF,UAAU,KAAK;QACxC,OAAOC,KAAK,MAAM,CAAC,CAACE;YAClB,MAAMd,QAAQW,UAAU,KAAK,CAACG,IAAI;YAClC,IAAI,CAACf,uBAAuBC,QAC1B,OAAO;YAIT,IAAIS,cAAc;oBACTM;gBAAP,OAAOA,AAAAA,SAAAA,CAAAA,cAAAA,MAAM,IAAI,AAAD,IAATA,KAAAA,IAAAA,YAAY,QAAQ,AAAD,MAAM;YAClC;YAEA,OAAO;QACT;IACF;IAGA,OAAO,EAAE;AACX;AAEO,MAAM3B,kBAAkB,CAC7B/E,YACA2G;IAEA,MAAMC,gBAAgBV,4BAA4BS;IAClD,MAAM5F,SAAS;QAAE,GAAGf,UAAU;IAAC;IAE/B,KAAK,MAAM6G,aAAaD,cAAe;QACrC,MAAME,aAAa/F,MAAM,CAAC8F,UAAU;QACpC,IAAIC,YAEF;YAAA,IAAI,AAAsB,YAAtB,OAAOA,YACT/F,MAAM,CAAC8F,UAAU,GAAGC;iBACf,IAAI,AAAsB,YAAtB,OAAOA,YAEhB;gBAAA,IAAIA,WAAW,MAAM,EAEnB;oBAAA,IAAI,AAA6B,YAA7B,OAAOA,WAAW,MAAM,EAC1B/F,MAAM,CAAC8F,UAAU,GAAGC,WAAW,MAAM;yBAChC,IACL,AAA6B,YAA7B,OAAOA,WAAW,MAAM,IACxBA,WAAW,MAAM,CAAC,MAAM,EAGxB/F,MAAM,CAAC8F,UAAU,GAAGC,WAAW,MAAM,CAAC,MAAM;gBAC9C;YACF;QACF;IAEJ;IAEA,OAAO/F;AACT;AAEO,MAAMgG,kBAAkB,CAC7B/G,YACA2G;IAEA,MAAMC,gBAAgBV,4BAA4BS;IAClD,MAAM5F,SAAS;QAAE,GAAGf,UAAU;IAAC;IAE/B,KAAK,MAAM6G,aAAaD,cAAe;QACrC,MAAME,aAAa/F,MAAM,CAAC8F,UAAU;QACpC,IAAIC,cAAc,AAAsB,YAAtB,OAAOA,YACvB/F,MAAM,CAAC8F,UAAU,GAAG;YAClB,CAACtB,oBAAoB,EAAE;YACvB,QAAQuB;QACV;IAEJ;IAEA,OAAO/F;AACT"}
|
|
@@ -1,10 +1,10 @@
|
|
|
1
|
-
import {
|
|
1
|
+
import { call, callAiFnWithStringResponse, callToGetJSONObject } from "./service-caller/index.mjs";
|
|
2
2
|
import { systemPromptToLocateElement } from "./prompt/llm-locator.mjs";
|
|
3
3
|
import { describeUserPage, elementByPositionWithElementInfo } from "./prompt/util.mjs";
|
|
4
4
|
import { generatePlaywrightTest, generatePlaywrightTestStream } from "./prompt/playwright-generator.mjs";
|
|
5
5
|
import { generateYamlTest, generateYamlTestStream } from "./prompt/yaml-generator.mjs";
|
|
6
6
|
import { AiExtractElementInfo, AiLocateElement, AiLocateSection } from "./inspect.mjs";
|
|
7
7
|
import { plan } from "./llm-planning.mjs";
|
|
8
|
-
import { AIActionType, PointSchema, RectSchema, SizeSchema, TMultimodalPromptSchema, TUserPromptSchema, adaptBboxToRect, dumpActionParam, findAllMidsceneLocatorField, getMidsceneLocationSchema, loadActionParam } from "./common.mjs";
|
|
8
|
+
import { AIActionType, PointSchema, RectSchema, SizeSchema, TMultimodalPromptSchema, TUserPromptSchema, adaptBboxToRect, callAiFn, dumpActionParam, findAllMidsceneLocatorField, getMidsceneLocationSchema, loadActionParam } from "./common.mjs";
|
|
9
9
|
import { resizeImageForUiTars, vlmPlanning } from "./ui-tars-planning.mjs";
|
|
10
|
-
export { AIActionType, AiExtractElementInfo, AiLocateElement, AiLocateSection, PointSchema, RectSchema, SizeSchema, TMultimodalPromptSchema, TUserPromptSchema, adaptBboxToRect,
|
|
10
|
+
export { AIActionType, AiExtractElementInfo, AiLocateElement, AiLocateSection, PointSchema, RectSchema, SizeSchema, TMultimodalPromptSchema, TUserPromptSchema, adaptBboxToRect, call as callAi, callAiFn, callAiFnWithStringResponse, callToGetJSONObject, describeUserPage, dumpActionParam, elementByPositionWithElementInfo, findAllMidsceneLocatorField, generatePlaywrightTest, generatePlaywrightTestStream, generateYamlTest, generateYamlTestStream, getMidsceneLocationSchema, loadActionParam, plan, resizeImageForUiTars, systemPromptToLocateElement, vlmPlanning };
|
|
@@ -1,12 +1,13 @@
|
|
|
1
|
+
import { getIsUseQwenVl, vlLocateMode } from "@midscene/shared/env";
|
|
1
2
|
import { cropByRect, paddingToMatchBlockByBase64, preProcessImageUrl } from "@midscene/shared/img";
|
|
2
3
|
import { getDebug } from "@midscene/shared/logger";
|
|
3
4
|
import { assert } from "@midscene/shared/utils";
|
|
4
|
-
import { AIActionType, adaptBboxToRect, expandSearchArea, markupImageForLLM, mergeRects } from "./common.mjs";
|
|
5
|
+
import { AIActionType, adaptBboxToRect, callAiFn, expandSearchArea, markupImageForLLM, mergeRects } from "./common.mjs";
|
|
5
6
|
import { extractDataQueryPrompt, systemPromptToExtract } from "./prompt/extraction.mjs";
|
|
6
7
|
import { findElementPrompt, systemPromptToLocateElement } from "./prompt/llm-locator.mjs";
|
|
7
8
|
import { sectionLocatorInstruction, systemPromptToLocateSection } from "./prompt/llm-section-locator.mjs";
|
|
8
9
|
import { describeUserPage, distance, distanceThreshold, elementByPositionWithElementInfo } from "./prompt/util.mjs";
|
|
9
|
-
import {
|
|
10
|
+
import { callToGetJSONObject } from "./service-caller/index.mjs";
|
|
10
11
|
const debugInspect = getDebug('ai:inspect');
|
|
11
12
|
const debugSection = getDebug('ai:section');
|
|
12
13
|
const extraTextFromUserPrompt = (prompt)=>{
|
|
@@ -54,25 +55,25 @@ const promptsToChatParam = async (multimodalPrompt)=>{
|
|
|
54
55
|
return msgs;
|
|
55
56
|
};
|
|
56
57
|
async function AiLocateElement(options) {
|
|
57
|
-
const { context, targetElementDescription,
|
|
58
|
-
const { vlMode } = modelConfig;
|
|
58
|
+
const { context, targetElementDescription, callAI } = options;
|
|
59
59
|
const { screenshotBase64 } = context;
|
|
60
|
-
const
|
|
61
|
-
|
|
62
|
-
}
|
|
60
|
+
const modelPreferences = {
|
|
61
|
+
intent: 'grounding'
|
|
62
|
+
};
|
|
63
|
+
const { description, elementById, insertElementByPosition } = await describeUserPage(context, modelPreferences);
|
|
63
64
|
assert(targetElementDescription, "cannot find the target element description");
|
|
64
65
|
const userInstructionPrompt = await findElementPrompt.format({
|
|
65
66
|
pageDescription: description,
|
|
66
67
|
targetElementDescription: extraTextFromUserPrompt(targetElementDescription)
|
|
67
68
|
});
|
|
68
|
-
const systemPrompt = systemPromptToLocateElement(
|
|
69
|
+
const systemPrompt = systemPromptToLocateElement(vlLocateMode(modelPreferences));
|
|
69
70
|
let imagePayload = screenshotBase64;
|
|
70
71
|
if (options.searchConfig) {
|
|
71
72
|
assert(options.searchConfig.rect, 'searchArea is provided but its rect cannot be found. Failed to locate element');
|
|
72
73
|
assert(options.searchConfig.imageBase64, 'searchArea is provided but its imageBase64 cannot be found. Failed to locate element');
|
|
73
74
|
imagePayload = options.searchConfig.imageBase64;
|
|
74
|
-
} else if ('qwen-vl' ===
|
|
75
|
-
else if (!
|
|
75
|
+
} else if ('qwen-vl' === vlLocateMode(modelPreferences)) imagePayload = await paddingToMatchBlockByBase64(imagePayload);
|
|
76
|
+
else if (!vlLocateMode(modelPreferences)) imagePayload = await markupImageForLLM(screenshotBase64, context.tree, context.size);
|
|
76
77
|
const msgs = [
|
|
77
78
|
{
|
|
78
79
|
role: 'system',
|
|
@@ -102,7 +103,10 @@ async function AiLocateElement(options) {
|
|
|
102
103
|
});
|
|
103
104
|
msgs.push(...addOns);
|
|
104
105
|
}
|
|
105
|
-
const
|
|
106
|
+
const callAIFn = callAI || callToGetJSONObject;
|
|
107
|
+
const res = await callAIFn(msgs, AIActionType.INSPECT_ELEMENT, {
|
|
108
|
+
intent: 'grounding'
|
|
109
|
+
});
|
|
106
110
|
const rawResponse = JSON.stringify(res.content);
|
|
107
111
|
let resRect;
|
|
108
112
|
let matchedElements = 'elements' in res.content ? res.content.elements : [];
|
|
@@ -110,7 +114,7 @@ async function AiLocateElement(options) {
|
|
|
110
114
|
try {
|
|
111
115
|
if ('bbox' in res.content && Array.isArray(res.content.bbox)) {
|
|
112
116
|
var _options_searchConfig_rect, _options_searchConfig, _options_searchConfig_rect1, _options_searchConfig1, _options_searchConfig_rect2, _options_searchConfig2, _options_searchConfig_rect3, _options_searchConfig3;
|
|
113
|
-
resRect = adaptBboxToRect(res.content.bbox, (null == (_options_searchConfig = options.searchConfig) ? void 0 : null == (_options_searchConfig_rect = _options_searchConfig.rect) ? void 0 : _options_searchConfig_rect.width) || context.size.width, (null == (_options_searchConfig1 = options.searchConfig) ? void 0 : null == (_options_searchConfig_rect1 = _options_searchConfig1.rect) ? void 0 : _options_searchConfig_rect1.height) || context.size.height, null == (_options_searchConfig2 = options.searchConfig) ? void 0 : null == (_options_searchConfig_rect2 = _options_searchConfig2.rect) ? void 0 : _options_searchConfig_rect2.left, null == (_options_searchConfig3 = options.searchConfig) ? void 0 : null == (_options_searchConfig_rect3 = _options_searchConfig3.rect) ? void 0 : _options_searchConfig_rect3.top
|
|
117
|
+
resRect = adaptBboxToRect(res.content.bbox, (null == (_options_searchConfig = options.searchConfig) ? void 0 : null == (_options_searchConfig_rect = _options_searchConfig.rect) ? void 0 : _options_searchConfig_rect.width) || context.size.width, (null == (_options_searchConfig1 = options.searchConfig) ? void 0 : null == (_options_searchConfig_rect1 = _options_searchConfig1.rect) ? void 0 : _options_searchConfig_rect1.height) || context.size.height, modelPreferences, null == (_options_searchConfig2 = options.searchConfig) ? void 0 : null == (_options_searchConfig_rect2 = _options_searchConfig2.rect) ? void 0 : _options_searchConfig_rect2.left, null == (_options_searchConfig3 = options.searchConfig) ? void 0 : null == (_options_searchConfig_rect3 = _options_searchConfig3.rect) ? void 0 : _options_searchConfig_rect3.top);
|
|
114
118
|
debugInspect('resRect', resRect);
|
|
115
119
|
const rectCenter = {
|
|
116
120
|
x: resRect.left + resRect.width / 2,
|
|
@@ -149,10 +153,12 @@ async function AiLocateElement(options) {
|
|
|
149
153
|
};
|
|
150
154
|
}
|
|
151
155
|
async function AiLocateSection(options) {
|
|
152
|
-
const { context, sectionDescription
|
|
153
|
-
const { vlMode } = modelConfig;
|
|
156
|
+
const { context, sectionDescription } = options;
|
|
154
157
|
const { screenshotBase64 } = context;
|
|
155
|
-
const
|
|
158
|
+
const modelPreferences = {
|
|
159
|
+
intent: 'grounding'
|
|
160
|
+
};
|
|
161
|
+
const systemPrompt = systemPromptToLocateSection(vlLocateMode(modelPreferences));
|
|
156
162
|
const sectionLocatorInstructionText = await sectionLocatorInstruction.format({
|
|
157
163
|
sectionDescription: extraTextFromUserPrompt(sectionDescription)
|
|
158
164
|
});
|
|
@@ -185,26 +191,30 @@ async function AiLocateSection(options) {
|
|
|
185
191
|
});
|
|
186
192
|
msgs.push(...addOns);
|
|
187
193
|
}
|
|
188
|
-
const result = await
|
|
194
|
+
const result = await callAiFn(msgs, AIActionType.EXTRACT_DATA, {
|
|
195
|
+
intent: 'grounding'
|
|
196
|
+
});
|
|
189
197
|
let sectionRect;
|
|
190
198
|
const sectionBbox = result.content.bbox;
|
|
191
199
|
if (sectionBbox) {
|
|
192
|
-
const targetRect = adaptBboxToRect(sectionBbox, context.size.width, context.size.height,
|
|
200
|
+
const targetRect = adaptBboxToRect(sectionBbox, context.size.width, context.size.height, modelPreferences);
|
|
193
201
|
debugSection('original targetRect %j', targetRect);
|
|
194
202
|
const referenceBboxList = result.content.references_bbox || [];
|
|
195
203
|
debugSection('referenceBboxList %j', referenceBboxList);
|
|
196
|
-
const referenceRects = referenceBboxList.filter((bbox)=>Array.isArray(bbox)).map((bbox)=>adaptBboxToRect(bbox, context.size.width, context.size.height,
|
|
204
|
+
const referenceRects = referenceBboxList.filter((bbox)=>Array.isArray(bbox)).map((bbox)=>adaptBboxToRect(bbox, context.size.width, context.size.height, modelPreferences));
|
|
197
205
|
debugSection('referenceRects %j', referenceRects);
|
|
198
206
|
const mergedRect = mergeRects([
|
|
199
207
|
targetRect,
|
|
200
208
|
...referenceRects
|
|
201
209
|
]);
|
|
202
210
|
debugSection('mergedRect %j', mergedRect);
|
|
203
|
-
sectionRect = expandSearchArea(mergedRect, context.size,
|
|
211
|
+
sectionRect = expandSearchArea(mergedRect, context.size, modelPreferences);
|
|
204
212
|
debugSection('expanded sectionRect %j', sectionRect);
|
|
205
213
|
}
|
|
206
214
|
let imageBase64 = screenshotBase64;
|
|
207
|
-
if (sectionRect) imageBase64 = await cropByRect(screenshotBase64, sectionRect,
|
|
215
|
+
if (sectionRect) imageBase64 = await cropByRect(screenshotBase64, sectionRect, getIsUseQwenVl({
|
|
216
|
+
intent: 'grounding'
|
|
217
|
+
}));
|
|
208
218
|
return {
|
|
209
219
|
rect: sectionRect,
|
|
210
220
|
imageBase64,
|
|
@@ -215,16 +225,14 @@ async function AiLocateSection(options) {
|
|
|
215
225
|
}
|
|
216
226
|
async function AiExtractElementInfo(options) {
|
|
217
227
|
var _options_extractOption;
|
|
218
|
-
const { dataQuery, context, extractOption, multimodalPrompt,
|
|
219
|
-
const { vlMode } = modelConfig;
|
|
228
|
+
const { dataQuery, context, extractOption, multimodalPrompt, modelPreferences } = options;
|
|
220
229
|
const systemPrompt = systemPromptToExtract();
|
|
221
230
|
const { screenshotBase64 } = context;
|
|
222
|
-
const { description, elementById } = await describeUserPage(context, {
|
|
231
|
+
const { description, elementById } = await describeUserPage(context, modelPreferences, {
|
|
223
232
|
truncateTextLength: 200,
|
|
224
233
|
filterNonTextContent: false,
|
|
225
234
|
visibleOnly: false,
|
|
226
|
-
domIncluded: null == extractOption ? void 0 : extractOption.domIncluded
|
|
227
|
-
vlMode
|
|
235
|
+
domIncluded: null == extractOption ? void 0 : extractOption.domIncluded
|
|
228
236
|
});
|
|
229
237
|
const extractDataPromptText = await extractDataQueryPrompt(description, dataQuery);
|
|
230
238
|
const userContent = [];
|
|
@@ -260,7 +268,7 @@ async function AiExtractElementInfo(options) {
|
|
|
260
268
|
});
|
|
261
269
|
msgs.push(...addOns);
|
|
262
270
|
}
|
|
263
|
-
const result = await
|
|
271
|
+
const result = await callAiFn(msgs, AIActionType.EXTRACT_DATA, modelPreferences);
|
|
264
272
|
return {
|
|
265
273
|
parseResult: result.content,
|
|
266
274
|
elementById,
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"ai-model/inspect.mjs","sources":["webpack://@midscene/core/./src/ai-model/inspect.ts"],"sourcesContent":["import type {\n AIDataExtractionResponse,\n AIElementLocatorResponse,\n AIElementResponse,\n AISectionLocatorResponse,\n AIUsageInfo,\n BaseElement,\n ElementById,\n InsightExtractOption,\n Rect,\n ReferenceImage,\n UIContext,\n} from '@/types';\nimport type { IModelConfig } from '@midscene/shared/env';\nimport {\n cropByRect,\n paddingToMatchBlockByBase64,\n preProcessImageUrl,\n} from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type {\n ChatCompletionSystemMessageParam,\n ChatCompletionUserMessageParam,\n} from 'openai/resources/index';\nimport type { TMultimodalPrompt, TUserPrompt } from './common';\nimport {\n AIActionType,\n adaptBboxToRect,\n expandSearchArea,\n markupImageForLLM,\n mergeRects,\n} from './common';\nimport {\n extractDataQueryPrompt,\n systemPromptToExtract,\n} from './prompt/extraction';\nimport {\n findElementPrompt,\n systemPromptToLocateElement,\n} from './prompt/llm-locator';\nimport {\n sectionLocatorInstruction,\n systemPromptToLocateSection,\n} from './prompt/llm-section-locator';\nimport {\n describeUserPage,\n distance,\n distanceThreshold,\n elementByPositionWithElementInfo,\n} from './prompt/util';\nimport { callAIWithObjectResponse } from './service-caller/index';\n\nexport type AIArgs = [\n ChatCompletionSystemMessageParam,\n ...ChatCompletionUserMessageParam[],\n];\n\nconst debugInspect = getDebug('ai:inspect');\nconst debugSection = getDebug('ai:section');\n\nconst extraTextFromUserPrompt = (prompt: TUserPrompt): string => {\n if (typeof prompt === 'string') {\n return prompt;\n } else {\n return prompt.prompt;\n }\n};\n\nconst promptsToChatParam = async (\n multimodalPrompt: TMultimodalPrompt,\n): Promise<ChatCompletionUserMessageParam[]> => {\n const msgs: ChatCompletionUserMessageParam[] = [];\n if (multimodalPrompt?.images?.length) {\n msgs.push({\n role: 'user',\n content: [\n {\n type: 'text',\n text: 'Next, I will provide all the reference images.',\n },\n ],\n });\n\n for (const item of multimodalPrompt.images) {\n const base64 = await preProcessImageUrl(\n item.url,\n !!multimodalPrompt.convertHttpImage2Base64,\n );\n\n msgs.push({\n role: 'user',\n content: [\n {\n type: 'text',\n text: `reference image ${item.name}:`,\n },\n ],\n });\n\n msgs.push({\n role: 'user',\n content: [\n {\n type: 'image_url',\n image_url: {\n url: base64,\n detail: 'high',\n },\n },\n ],\n });\n }\n }\n return msgs;\n};\n\nexport async function AiLocateElement<\n ElementType extends BaseElement = BaseElement,\n>(options: {\n context: UIContext<ElementType>;\n targetElementDescription: TUserPrompt;\n referenceImage?: ReferenceImage;\n callAIFn: typeof callAIWithObjectResponse<\n AIElementResponse | [number, number]\n >;\n searchConfig?: Awaited<ReturnType<typeof AiLocateSection>>;\n modelConfig: IModelConfig;\n}): Promise<{\n parseResult: AIElementLocatorResponse;\n rect?: Rect;\n rawResponse: string;\n elementById: ElementById;\n usage?: AIUsageInfo;\n isOrderSensitive?: boolean;\n}> {\n const { context, targetElementDescription, callAIFn, modelConfig } = options;\n const { vlMode } = modelConfig;\n const { screenshotBase64 } = context;\n\n const { description, elementById, insertElementByPosition } =\n await describeUserPage(context, { vlMode });\n\n assert(\n targetElementDescription,\n 'cannot find the target element description',\n );\n const userInstructionPrompt = await findElementPrompt.format({\n pageDescription: description,\n targetElementDescription: extraTextFromUserPrompt(targetElementDescription),\n });\n const systemPrompt = systemPromptToLocateElement(vlMode);\n\n let imagePayload = screenshotBase64;\n\n if (options.searchConfig) {\n assert(\n options.searchConfig.rect,\n 'searchArea is provided but its rect cannot be found. Failed to locate element',\n );\n assert(\n options.searchConfig.imageBase64,\n 'searchArea is provided but its imageBase64 cannot be found. Failed to locate element',\n );\n\n imagePayload = options.searchConfig.imageBase64;\n } else if (vlMode === 'qwen-vl') {\n imagePayload = await paddingToMatchBlockByBase64(imagePayload);\n } else if (!vlMode) {\n imagePayload = await markupImageForLLM(\n screenshotBase64,\n context.tree,\n context.size,\n );\n }\n\n const msgs: AIArgs = [\n { role: 'system', content: systemPrompt },\n {\n role: 'user',\n content: [\n {\n type: 'image_url',\n image_url: {\n url: imagePayload,\n detail: 'high',\n },\n },\n {\n type: 'text',\n text: userInstructionPrompt,\n },\n ],\n },\n ];\n\n if (typeof targetElementDescription !== 'string') {\n const addOns = await promptsToChatParam({\n images: targetElementDescription.images,\n convertHttpImage2Base64: targetElementDescription.convertHttpImage2Base64,\n });\n msgs.push(...addOns);\n }\n\n const res = await callAIFn(msgs, AIActionType.INSPECT_ELEMENT, modelConfig);\n\n const rawResponse = JSON.stringify(res.content);\n\n let resRect: Rect | undefined;\n let matchedElements: AIElementLocatorResponse['elements'] =\n 'elements' in res.content ? res.content.elements : [];\n let errors: AIElementLocatorResponse['errors'] | undefined =\n 'errors' in res.content ? res.content.errors : [];\n try {\n if ('bbox' in res.content && Array.isArray(res.content.bbox)) {\n resRect = adaptBboxToRect(\n res.content.bbox,\n options.searchConfig?.rect?.width || context.size.width,\n options.searchConfig?.rect?.height || context.size.height,\n options.searchConfig?.rect?.left,\n options.searchConfig?.rect?.top,\n vlMode,\n );\n debugInspect('resRect', resRect);\n\n const rectCenter = {\n x: resRect.left + resRect.width / 2,\n y: resRect.top + resRect.height / 2,\n };\n let element = elementByPositionWithElementInfo(context.tree, rectCenter);\n\n const distanceToCenter = element\n ? distance({ x: element.center[0], y: element.center[1] }, rectCenter)\n : 0;\n\n if (!element || distanceToCenter > distanceThreshold) {\n element = insertElementByPosition(rectCenter);\n }\n\n if (element) {\n matchedElements = [element];\n errors = [];\n }\n }\n } catch (e) {\n const msg =\n e instanceof Error\n ? `Failed to parse bbox: ${e.message}`\n : 'unknown error in locate';\n if (!errors || errors?.length === 0) {\n errors = [msg];\n } else {\n errors.push(`(${msg})`);\n }\n }\n\n return {\n rect: resRect,\n parseResult: {\n elements: matchedElements,\n errors,\n },\n rawResponse,\n elementById,\n usage: res.usage,\n isOrderSensitive:\n typeof res.content === 'object' &&\n res.content !== null &&\n 'isOrderSensitive' in res.content\n ? (res.content as any).isOrderSensitive\n : undefined,\n };\n}\n\nexport async function AiLocateSection(options: {\n context: UIContext<BaseElement>;\n sectionDescription: TUserPrompt;\n modelConfig: IModelConfig;\n}): Promise<{\n rect?: Rect;\n imageBase64?: string;\n error?: string;\n rawResponse: string;\n usage?: AIUsageInfo;\n}> {\n const { context, sectionDescription, modelConfig } = options;\n const { vlMode } = modelConfig;\n const { screenshotBase64 } = context;\n\n const systemPrompt = systemPromptToLocateSection(vlMode);\n const sectionLocatorInstructionText = await sectionLocatorInstruction.format({\n sectionDescription: extraTextFromUserPrompt(sectionDescription),\n });\n const msgs: AIArgs = [\n { role: 'system', content: systemPrompt },\n {\n role: 'user',\n content: [\n {\n type: 'image_url',\n image_url: {\n url: screenshotBase64,\n detail: 'high',\n },\n },\n {\n type: 'text',\n text: sectionLocatorInstructionText,\n },\n ],\n },\n ];\n\n if (typeof sectionDescription !== 'string') {\n const addOns = await promptsToChatParam({\n images: sectionDescription.images,\n convertHttpImage2Base64: sectionDescription.convertHttpImage2Base64,\n });\n msgs.push(...addOns);\n }\n\n const result = await callAIWithObjectResponse<AISectionLocatorResponse>(\n msgs,\n AIActionType.EXTRACT_DATA,\n modelConfig,\n );\n\n let sectionRect: Rect | undefined;\n const sectionBbox = result.content.bbox;\n if (sectionBbox) {\n const targetRect = adaptBboxToRect(\n sectionBbox,\n context.size.width,\n context.size.height,\n 0,\n 0,\n vlMode,\n );\n debugSection('original targetRect %j', targetRect);\n\n const referenceBboxList = result.content.references_bbox || [];\n debugSection('referenceBboxList %j', referenceBboxList);\n\n const referenceRects = referenceBboxList\n .filter((bbox) => Array.isArray(bbox))\n .map((bbox) => {\n return adaptBboxToRect(\n bbox,\n context.size.width,\n context.size.height,\n 0,\n 0,\n vlMode,\n );\n });\n debugSection('referenceRects %j', referenceRects);\n\n // merge the sectionRect and referenceRects\n const mergedRect = mergeRects([targetRect, ...referenceRects]);\n debugSection('mergedRect %j', mergedRect);\n\n // expand search area to at least 200 x 200\n sectionRect = expandSearchArea(mergedRect, context.size, vlMode);\n debugSection('expanded sectionRect %j', sectionRect);\n }\n\n let imageBase64 = screenshotBase64;\n if (sectionRect) {\n imageBase64 = await cropByRect(\n screenshotBase64,\n sectionRect,\n vlMode === 'qwen-vl',\n );\n }\n\n return {\n rect: sectionRect,\n imageBase64,\n error: result.content.error,\n rawResponse: JSON.stringify(result.content),\n usage: result.usage,\n };\n}\n\nexport async function AiExtractElementInfo<\n T,\n ElementType extends BaseElement = BaseElement,\n>(options: {\n dataQuery: string | Record<string, string>;\n multimodalPrompt?: TMultimodalPrompt;\n context: UIContext<ElementType>;\n extractOption?: InsightExtractOption;\n modelConfig: IModelConfig;\n}) {\n const { dataQuery, context, extractOption, multimodalPrompt, modelConfig } =\n options;\n const { vlMode } = modelConfig;\n const systemPrompt = systemPromptToExtract();\n\n const { screenshotBase64 } = context;\n\n const { description, elementById } = await describeUserPage(context, {\n truncateTextLength: 200,\n filterNonTextContent: false,\n visibleOnly: false,\n domIncluded: extractOption?.domIncluded,\n vlMode,\n });\n\n const extractDataPromptText = await extractDataQueryPrompt(\n description,\n dataQuery,\n );\n\n const userContent: ChatCompletionUserMessageParam['content'] = [];\n\n if (extractOption?.screenshotIncluded !== false) {\n userContent.push({\n type: 'image_url',\n image_url: {\n url: screenshotBase64,\n detail: 'high',\n },\n });\n }\n\n userContent.push({\n type: 'text',\n text: extractDataPromptText,\n });\n\n const msgs: AIArgs = [\n { role: 'system', content: systemPrompt },\n {\n role: 'user',\n content: userContent,\n },\n ];\n\n if (options.extractOption?.returnThought) {\n msgs.push({\n role: 'user',\n content: 'Please provide reasons.',\n });\n }\n\n if (multimodalPrompt) {\n const addOns = await promptsToChatParam({\n images: multimodalPrompt.images,\n convertHttpImage2Base64: multimodalPrompt.convertHttpImage2Base64,\n });\n msgs.push(...addOns);\n }\n\n const result = await callAIWithObjectResponse<AIDataExtractionResponse<T>>(\n msgs,\n AIActionType.EXTRACT_DATA,\n modelConfig,\n );\n return {\n parseResult: result.content,\n elementById,\n usage: result.usage,\n };\n}\n"],"names":["debugInspect","getDebug","debugSection","extraTextFromUserPrompt","prompt","promptsToChatParam","multimodalPrompt","_multimodalPrompt_images","msgs","item","base64","preProcessImageUrl","AiLocateElement","options","context","targetElementDescription","callAIFn","modelConfig","vlMode","screenshotBase64","description","elementById","insertElementByPosition","describeUserPage","assert","userInstructionPrompt","findElementPrompt","systemPrompt","systemPromptToLocateElement","imagePayload","paddingToMatchBlockByBase64","markupImageForLLM","addOns","res","AIActionType","rawResponse","JSON","resRect","matchedElements","errors","Array","_options_searchConfig_rect","_options_searchConfig_rect1","_options_searchConfig_rect2","_options_searchConfig_rect3","adaptBboxToRect","rectCenter","element","elementByPositionWithElementInfo","distanceToCenter","distance","distanceThreshold","e","msg","Error","undefined","AiLocateSection","sectionDescription","systemPromptToLocateSection","sectionLocatorInstructionText","sectionLocatorInstruction","result","callAIWithObjectResponse","sectionRect","sectionBbox","targetRect","referenceBboxList","referenceRects","bbox","mergedRect","mergeRects","expandSearchArea","imageBase64","cropByRect","AiExtractElementInfo","_options_extractOption","dataQuery","extractOption","systemPromptToExtract","extractDataPromptText","extractDataQueryPrompt","userContent"],"mappings":";;;;;;;;;AA0DA,MAAMA,eAAeC,SAAS;AAC9B,MAAMC,eAAeD,SAAS;AAE9B,MAAME,0BAA0B,CAACC;IAC/B,IAAI,AAAkB,YAAlB,OAAOA,QACT,OAAOA;IAEP,OAAOA,OAAO,MAAM;AAExB;AAEA,MAAMC,qBAAqB,OACzBC;QAGIC;IADJ,MAAMC,OAAyC,EAAE;IACjD,IAAID,QAAAA,mBAAAA,KAAAA,IAAAA,QAAAA,CAAAA,2BAAAA,iBAAkB,MAAM,AAAD,IAAvBA,KAAAA,IAAAA,yBAA0B,MAAM,EAAE;QACpCC,KAAK,IAAI,CAAC;YACR,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,MAAM;gBACR;aACD;QACH;QAEA,KAAK,MAAMC,QAAQH,iBAAiB,MAAM,CAAE;YAC1C,MAAMI,SAAS,MAAMC,mBACnBF,KAAK,GAAG,EACR,CAAC,CAACH,iBAAiB,uBAAuB;YAG5CE,KAAK,IAAI,CAAC;gBACR,MAAM;gBACN,SAAS;oBACP;wBACE,MAAM;wBACN,MAAM,CAAC,gBAAgB,EAAEC,KAAK,IAAI,CAAC,CAAC,CAAC;oBACvC;iBACD;YACH;YAEAD,KAAK,IAAI,CAAC;gBACR,MAAM;gBACN,SAAS;oBACP;wBACE,MAAM;wBACN,WAAW;4BACT,KAAKE;4BACL,QAAQ;wBACV;oBACF;iBACD;YACH;QACF;IACF;IACA,OAAOF;AACT;AAEO,eAAeI,gBAEpBC,OASD;IAQC,MAAM,EAAEC,OAAO,EAAEC,wBAAwB,EAAEC,QAAQ,EAAEC,WAAW,EAAE,GAAGJ;IACrE,MAAM,EAAEK,MAAM,EAAE,GAAGD;IACnB,MAAM,EAAEE,gBAAgB,EAAE,GAAGL;IAE7B,MAAM,EAAEM,WAAW,EAAEC,WAAW,EAAEC,uBAAuB,EAAE,GACzD,MAAMC,iBAAiBT,SAAS;QAAEI;IAAO;IAE3CM,OACET,0BACA;IAEF,MAAMU,wBAAwB,MAAMC,kBAAkB,MAAM,CAAC;QAC3D,iBAAiBN;QACjB,0BAA0BjB,wBAAwBY;IACpD;IACA,MAAMY,eAAeC,4BAA4BV;IAEjD,IAAIW,eAAeV;IAEnB,IAAIN,QAAQ,YAAY,EAAE;QACxBW,OACEX,QAAQ,YAAY,CAAC,IAAI,EACzB;QAEFW,OACEX,QAAQ,YAAY,CAAC,WAAW,EAChC;QAGFgB,eAAehB,QAAQ,YAAY,CAAC,WAAW;IACjD,OAAO,IAAIK,AAAW,cAAXA,QACTW,eAAe,MAAMC,4BAA4BD;SAC5C,IAAI,CAACX,QACVW,eAAe,MAAME,kBACnBZ,kBACAL,QAAQ,IAAI,EACZA,QAAQ,IAAI;IAIhB,MAAMN,OAAe;QACnB;YAAE,MAAM;YAAU,SAASmB;QAAa;QACxC;YACE,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,WAAW;wBACT,KAAKE;wBACL,QAAQ;oBACV;gBACF;gBACA;oBACE,MAAM;oBACN,MAAMJ;gBACR;aACD;QACH;KACD;IAED,IAAI,AAAoC,YAApC,OAAOV,0BAAuC;QAChD,MAAMiB,SAAS,MAAM3B,mBAAmB;YACtC,QAAQU,yBAAyB,MAAM;YACvC,yBAAyBA,yBAAyB,uBAAuB;QAC3E;QACAP,KAAK,IAAI,IAAIwB;IACf;IAEA,MAAMC,MAAM,MAAMjB,SAASR,MAAM0B,aAAa,eAAe,EAAEjB;IAE/D,MAAMkB,cAAcC,KAAK,SAAS,CAACH,IAAI,OAAO;IAE9C,IAAII;IACJ,IAAIC,kBACF,cAAcL,IAAI,OAAO,GAAGA,IAAI,OAAO,CAAC,QAAQ,GAAG,EAAE;IACvD,IAAIM,SACF,YAAYN,IAAI,OAAO,GAAGA,IAAI,OAAO,CAAC,MAAM,GAAG,EAAE;IACnD,IAAI;QACF,IAAI,UAAUA,IAAI,OAAO,IAAIO,MAAM,OAAO,CAACP,IAAI,OAAO,CAAC,IAAI,GAAG;gBAG1DQ,4BAAAA,uBACAC,6BAAAA,wBACAC,6BAAAA,wBACAC,6BAAAA;YALFP,UAAUQ,gBACRZ,IAAI,OAAO,CAAC,IAAI,EAChBQ,AAAAA,SAAAA,CAAAA,wBAAAA,QAAQ,YAAY,AAAD,IAAnBA,KAAAA,IAAAA,QAAAA,CAAAA,6BAAAA,sBAAsB,IAAI,AAAD,IAAzBA,KAAAA,IAAAA,2BAA4B,KAAK,AAAD,KAAK3B,QAAQ,IAAI,CAAC,KAAK,EACvD4B,AAAAA,SAAAA,CAAAA,yBAAAA,QAAQ,YAAY,AAAD,IAAnBA,KAAAA,IAAAA,QAAAA,CAAAA,8BAAAA,uBAAsB,IAAI,AAAD,IAAzBA,KAAAA,IAAAA,4BAA4B,MAAM,AAAD,KAAK5B,QAAQ,IAAI,CAAC,MAAM,UACzD6B,CAAAA,yBAAAA,QAAQ,YAAY,AAAD,IAAnBA,KAAAA,IAAAA,QAAAA,CAAAA,8BAAAA,uBAAsB,IAAI,AAAD,IAAzBA,KAAAA,IAAAA,4BAA4B,IAAI,UAChCC,CAAAA,yBAAAA,QAAQ,YAAY,AAAD,IAAnBA,KAAAA,IAAAA,QAAAA,CAAAA,8BAAAA,uBAAsB,IAAI,AAAD,IAAzBA,KAAAA,IAAAA,4BAA4B,GAAG,EAC/B1B;YAEFlB,aAAa,WAAWqC;YAExB,MAAMS,aAAa;gBACjB,GAAGT,QAAQ,IAAI,GAAGA,QAAQ,KAAK,GAAG;gBAClC,GAAGA,QAAQ,GAAG,GAAGA,QAAQ,MAAM,GAAG;YACpC;YACA,IAAIU,UAAUC,iCAAiClC,QAAQ,IAAI,EAAEgC;YAE7D,MAAMG,mBAAmBF,UACrBG,SAAS;gBAAE,GAAGH,QAAQ,MAAM,CAAC,EAAE;gBAAE,GAAGA,QAAQ,MAAM,CAAC,EAAE;YAAC,GAAGD,cACzD;YAEJ,IAAI,CAACC,WAAWE,mBAAmBE,mBACjCJ,UAAUzB,wBAAwBwB;YAGpC,IAAIC,SAAS;gBACXT,kBAAkB;oBAACS;iBAAQ;gBAC3BR,SAAS,EAAE;YACb;QACF;IACF,EAAE,OAAOa,GAAG;QACV,MAAMC,MACJD,aAAaE,QACT,CAAC,sBAAsB,EAAEF,EAAE,OAAO,EAAE,GACpC;QACN,IAAI,AAACb,UAAUA,AAAAA,CAAAA,QAAAA,SAAAA,KAAAA,IAAAA,OAAQ,MAAM,AAAD,MAAM,GAGhCA,OAAO,IAAI,CAAC,CAAC,CAAC,EAAEc,IAAI,CAAC,CAAC;aAFtBd,SAAS;YAACc;SAAI;IAIlB;IAEA,OAAO;QACL,MAAMhB;QACN,aAAa;YACX,UAAUC;YACVC;QACF;QACAJ;QACAd;QACA,OAAOY,IAAI,KAAK;QAChB,kBACE,AAAuB,YAAvB,OAAOA,IAAI,OAAO,IAClBA,AAAgB,SAAhBA,IAAI,OAAO,IACX,sBAAsBA,IAAI,OAAO,GAC5BA,IAAI,OAAO,CAAS,gBAAgB,GACrCsB;IACR;AACF;AAEO,eAAeC,gBAAgB3C,OAIrC;IAOC,MAAM,EAAEC,OAAO,EAAE2C,kBAAkB,EAAExC,WAAW,EAAE,GAAGJ;IACrD,MAAM,EAAEK,MAAM,EAAE,GAAGD;IACnB,MAAM,EAAEE,gBAAgB,EAAE,GAAGL;IAE7B,MAAMa,eAAe+B,4BAA4BxC;IACjD,MAAMyC,gCAAgC,MAAMC,0BAA0B,MAAM,CAAC;QAC3E,oBAAoBzD,wBAAwBsD;IAC9C;IACA,MAAMjD,OAAe;QACnB;YAAE,MAAM;YAAU,SAASmB;QAAa;QACxC;YACE,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,WAAW;wBACT,KAAKR;wBACL,QAAQ;oBACV;gBACF;gBACA;oBACE,MAAM;oBACN,MAAMwC;gBACR;aACD;QACH;KACD;IAED,IAAI,AAA8B,YAA9B,OAAOF,oBAAiC;QAC1C,MAAMzB,SAAS,MAAM3B,mBAAmB;YACtC,QAAQoD,mBAAmB,MAAM;YACjC,yBAAyBA,mBAAmB,uBAAuB;QACrE;QACAjD,KAAK,IAAI,IAAIwB;IACf;IAEA,MAAM6B,SAAS,MAAMC,yBACnBtD,MACA0B,aAAa,YAAY,EACzBjB;IAGF,IAAI8C;IACJ,MAAMC,cAAcH,OAAO,OAAO,CAAC,IAAI;IACvC,IAAIG,aAAa;QACf,MAAMC,aAAapB,gBACjBmB,aACAlD,QAAQ,IAAI,CAAC,KAAK,EAClBA,QAAQ,IAAI,CAAC,MAAM,EACnB,GACA,GACAI;QAEFhB,aAAa,0BAA0B+D;QAEvC,MAAMC,oBAAoBL,OAAO,OAAO,CAAC,eAAe,IAAI,EAAE;QAC9D3D,aAAa,wBAAwBgE;QAErC,MAAMC,iBAAiBD,kBACpB,MAAM,CAAC,CAACE,OAAS5B,MAAM,OAAO,CAAC4B,OAC/B,GAAG,CAAC,CAACA,OACGvB,gBACLuB,MACAtD,QAAQ,IAAI,CAAC,KAAK,EAClBA,QAAQ,IAAI,CAAC,MAAM,EACnB,GACA,GACAI;QAGNhB,aAAa,qBAAqBiE;QAGlC,MAAME,aAAaC,WAAW;YAACL;eAAeE;SAAe;QAC7DjE,aAAa,iBAAiBmE;QAG9BN,cAAcQ,iBAAiBF,YAAYvD,QAAQ,IAAI,EAAEI;QACzDhB,aAAa,2BAA2B6D;IAC1C;IAEA,IAAIS,cAAcrD;IAClB,IAAI4C,aACFS,cAAc,MAAMC,WAClBtD,kBACA4C,aACA7C,AAAW,cAAXA;IAIJ,OAAO;QACL,MAAM6C;QACNS;QACA,OAAOX,OAAO,OAAO,CAAC,KAAK;QAC3B,aAAazB,KAAK,SAAS,CAACyB,OAAO,OAAO;QAC1C,OAAOA,OAAO,KAAK;IACrB;AACF;AAEO,eAAea,qBAGpB7D,OAMD;QA8CK8D;IA7CJ,MAAM,EAAEC,SAAS,EAAE9D,OAAO,EAAE+D,aAAa,EAAEvE,gBAAgB,EAAEW,WAAW,EAAE,GACxEJ;IACF,MAAM,EAAEK,MAAM,EAAE,GAAGD;IACnB,MAAMU,eAAemD;IAErB,MAAM,EAAE3D,gBAAgB,EAAE,GAAGL;IAE7B,MAAM,EAAEM,WAAW,EAAEC,WAAW,EAAE,GAAG,MAAME,iBAAiBT,SAAS;QACnE,oBAAoB;QACpB,sBAAsB;QACtB,aAAa;QACb,aAAa+D,QAAAA,gBAAAA,KAAAA,IAAAA,cAAe,WAAW;QACvC3D;IACF;IAEA,MAAM6D,wBAAwB,MAAMC,uBAClC5D,aACAwD;IAGF,MAAMK,cAAyD,EAAE;IAEjE,IAAIJ,AAAAA,CAAAA,QAAAA,gBAAAA,KAAAA,IAAAA,cAAe,kBAAkB,AAAD,MAAM,OACxCI,YAAY,IAAI,CAAC;QACf,MAAM;QACN,WAAW;YACT,KAAK9D;YACL,QAAQ;QACV;IACF;IAGF8D,YAAY,IAAI,CAAC;QACf,MAAM;QACN,MAAMF;IACR;IAEA,MAAMvE,OAAe;QACnB;YAAE,MAAM;YAAU,SAASmB;QAAa;QACxC;YACE,MAAM;YACN,SAASsD;QACX;KACD;IAED,IAAI,QAAAN,CAAAA,yBAAAA,QAAQ,aAAa,AAAD,IAApBA,KAAAA,IAAAA,uBAAuB,aAAa,EACtCnE,KAAK,IAAI,CAAC;QACR,MAAM;QACN,SAAS;IACX;IAGF,IAAIF,kBAAkB;QACpB,MAAM0B,SAAS,MAAM3B,mBAAmB;YACtC,QAAQC,iBAAiB,MAAM;YAC/B,yBAAyBA,iBAAiB,uBAAuB;QACnE;QACAE,KAAK,IAAI,IAAIwB;IACf;IAEA,MAAM6B,SAAS,MAAMC,yBACnBtD,MACA0B,aAAa,YAAY,EACzBjB;IAEF,OAAO;QACL,aAAa4C,OAAO,OAAO;QAC3BxC;QACA,OAAOwC,OAAO,KAAK;IACrB;AACF"}
|
|
1
|
+
{"version":3,"file":"ai-model/inspect.mjs","sources":["webpack://@midscene/core/./src/ai-model/inspect.ts"],"sourcesContent":["import type {\n AIDataExtractionResponse,\n AIElementLocatorResponse,\n AIElementResponse,\n AISectionLocatorResponse,\n AIUsageInfo,\n BaseElement,\n ElementById,\n InsightExtractOption,\n Rect,\n ReferenceImage,\n UIContext,\n} from '@/types';\nimport {\n type IModelPreferences,\n getIsUseQwenVl,\n vlLocateMode,\n} from '@midscene/shared/env';\nimport {\n cropByRect,\n paddingToMatchBlockByBase64,\n preProcessImageUrl,\n} from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type {\n ChatCompletionSystemMessageParam,\n ChatCompletionUserMessageParam,\n} from 'openai/resources/index';\nimport type { TMultimodalPrompt, TUserPrompt } from './common';\nimport {\n AIActionType,\n adaptBboxToRect,\n callAiFn,\n expandSearchArea,\n markupImageForLLM,\n mergeRects,\n} from './common';\nimport {\n extractDataQueryPrompt,\n systemPromptToExtract,\n} from './prompt/extraction';\nimport {\n findElementPrompt,\n systemPromptToLocateElement,\n} from './prompt/llm-locator';\nimport {\n sectionLocatorInstruction,\n systemPromptToLocateSection,\n} from './prompt/llm-section-locator';\nimport {\n describeUserPage,\n distance,\n distanceThreshold,\n elementByPositionWithElementInfo,\n} from './prompt/util';\nimport { callToGetJSONObject } from './service-caller/index';\n\nexport type AIArgs = [\n ChatCompletionSystemMessageParam,\n ...ChatCompletionUserMessageParam[],\n];\n\nconst debugInspect = getDebug('ai:inspect');\nconst debugSection = getDebug('ai:section');\n\nconst extraTextFromUserPrompt = (prompt: TUserPrompt): string => {\n if (typeof prompt === 'string') {\n return prompt;\n } else {\n return prompt.prompt;\n }\n};\n\nconst promptsToChatParam = async (\n multimodalPrompt: TMultimodalPrompt,\n): Promise<ChatCompletionUserMessageParam[]> => {\n const msgs: ChatCompletionUserMessageParam[] = [];\n if (multimodalPrompt?.images?.length) {\n msgs.push({\n role: 'user',\n content: [\n {\n type: 'text',\n text: 'Next, I will provide all the reference images.',\n },\n ],\n });\n\n for (const item of multimodalPrompt.images) {\n const base64 = await preProcessImageUrl(\n item.url,\n !!multimodalPrompt.convertHttpImage2Base64,\n );\n\n msgs.push({\n role: 'user',\n content: [\n {\n type: 'text',\n text: `reference image ${item.name}:`,\n },\n ],\n });\n\n msgs.push({\n role: 'user',\n content: [\n {\n type: 'image_url',\n image_url: {\n url: base64,\n detail: 'high',\n },\n },\n ],\n });\n }\n }\n return msgs;\n};\n\nexport async function AiLocateElement<\n ElementType extends BaseElement = BaseElement,\n>(options: {\n context: UIContext<ElementType>;\n targetElementDescription: TUserPrompt;\n referenceImage?: ReferenceImage;\n callAI?: typeof callAiFn<AIElementResponse | [number, number]>;\n searchConfig?: Awaited<ReturnType<typeof AiLocateSection>>;\n}): Promise<{\n parseResult: AIElementLocatorResponse;\n rect?: Rect;\n rawResponse: string;\n elementById: ElementById;\n usage?: AIUsageInfo;\n isOrderSensitive?: boolean;\n}> {\n const { context, targetElementDescription, callAI } = options;\n const { screenshotBase64 } = context;\n\n const modelPreferences: IModelPreferences = {\n intent: 'grounding',\n };\n\n const { description, elementById, insertElementByPosition } =\n await describeUserPage(context, modelPreferences);\n\n assert(\n targetElementDescription,\n 'cannot find the target element description',\n );\n\n const userInstructionPrompt = await findElementPrompt.format({\n pageDescription: description,\n targetElementDescription: extraTextFromUserPrompt(targetElementDescription),\n });\n const systemPrompt = systemPromptToLocateElement(\n vlLocateMode(modelPreferences),\n );\n\n let imagePayload = screenshotBase64;\n\n if (options.searchConfig) {\n assert(\n options.searchConfig.rect,\n 'searchArea is provided but its rect cannot be found. Failed to locate element',\n );\n assert(\n options.searchConfig.imageBase64,\n 'searchArea is provided but its imageBase64 cannot be found. Failed to locate element',\n );\n\n imagePayload = options.searchConfig.imageBase64;\n } else if (vlLocateMode(modelPreferences) === 'qwen-vl') {\n imagePayload = await paddingToMatchBlockByBase64(imagePayload);\n } else if (!vlLocateMode(modelPreferences)) {\n imagePayload = await markupImageForLLM(\n screenshotBase64,\n context.tree,\n context.size,\n );\n }\n\n const msgs: AIArgs = [\n { role: 'system', content: systemPrompt },\n {\n role: 'user',\n content: [\n {\n type: 'image_url',\n image_url: {\n url: imagePayload,\n detail: 'high',\n },\n },\n {\n type: 'text',\n text: userInstructionPrompt,\n },\n ],\n },\n ];\n\n if (typeof targetElementDescription !== 'string') {\n const addOns = await promptsToChatParam({\n images: targetElementDescription.images,\n convertHttpImage2Base64: targetElementDescription.convertHttpImage2Base64,\n });\n msgs.push(...addOns);\n }\n\n const callAIFn =\n callAI || callToGetJSONObject<AIElementResponse | [number, number]>;\n\n const res = await callAIFn(msgs, AIActionType.INSPECT_ELEMENT, {\n intent: 'grounding',\n });\n\n const rawResponse = JSON.stringify(res.content);\n\n let resRect: Rect | undefined;\n let matchedElements: AIElementLocatorResponse['elements'] =\n 'elements' in res.content ? res.content.elements : [];\n let errors: AIElementLocatorResponse['errors'] | undefined =\n 'errors' in res.content ? res.content.errors : [];\n try {\n if ('bbox' in res.content && Array.isArray(res.content.bbox)) {\n resRect = adaptBboxToRect(\n res.content.bbox,\n options.searchConfig?.rect?.width || context.size.width,\n options.searchConfig?.rect?.height || context.size.height,\n modelPreferences,\n options.searchConfig?.rect?.left,\n options.searchConfig?.rect?.top,\n );\n debugInspect('resRect', resRect);\n\n const rectCenter = {\n x: resRect.left + resRect.width / 2,\n y: resRect.top + resRect.height / 2,\n };\n let element = elementByPositionWithElementInfo(context.tree, rectCenter);\n\n const distanceToCenter = element\n ? distance({ x: element.center[0], y: element.center[1] }, rectCenter)\n : 0;\n\n if (!element || distanceToCenter > distanceThreshold) {\n element = insertElementByPosition(rectCenter);\n }\n\n if (element) {\n matchedElements = [element];\n errors = [];\n }\n }\n } catch (e) {\n const msg =\n e instanceof Error\n ? `Failed to parse bbox: ${e.message}`\n : 'unknown error in locate';\n if (!errors || errors?.length === 0) {\n errors = [msg];\n } else {\n errors.push(`(${msg})`);\n }\n }\n\n return {\n rect: resRect,\n parseResult: {\n elements: matchedElements,\n errors,\n },\n rawResponse,\n elementById,\n usage: res.usage,\n isOrderSensitive:\n typeof res.content === 'object' &&\n res.content !== null &&\n 'isOrderSensitive' in res.content\n ? (res.content as any).isOrderSensitive\n : undefined,\n };\n}\n\nexport async function AiLocateSection(options: {\n context: UIContext<BaseElement>;\n sectionDescription: TUserPrompt;\n callAI?: typeof callAiFn<AISectionLocatorResponse>;\n}): Promise<{\n rect?: Rect;\n imageBase64?: string;\n error?: string;\n rawResponse: string;\n usage?: AIUsageInfo;\n}> {\n const { context, sectionDescription } = options;\n const { screenshotBase64 } = context;\n\n const modelPreferences: IModelPreferences = {\n intent: 'grounding',\n };\n\n const systemPrompt = systemPromptToLocateSection(\n vlLocateMode(modelPreferences),\n );\n const sectionLocatorInstructionText = await sectionLocatorInstruction.format({\n sectionDescription: extraTextFromUserPrompt(sectionDescription),\n });\n const msgs: AIArgs = [\n { role: 'system', content: systemPrompt },\n {\n role: 'user',\n content: [\n {\n type: 'image_url',\n image_url: {\n url: screenshotBase64,\n detail: 'high',\n },\n },\n {\n type: 'text',\n text: sectionLocatorInstructionText,\n },\n ],\n },\n ];\n\n if (typeof sectionDescription !== 'string') {\n const addOns = await promptsToChatParam({\n images: sectionDescription.images,\n convertHttpImage2Base64: sectionDescription.convertHttpImage2Base64,\n });\n msgs.push(...addOns);\n }\n\n const result = await callAiFn<AISectionLocatorResponse>(\n msgs,\n AIActionType.EXTRACT_DATA,\n {\n intent: 'grounding',\n },\n );\n\n let sectionRect: Rect | undefined;\n const sectionBbox = result.content.bbox;\n if (sectionBbox) {\n const targetRect = adaptBboxToRect(\n sectionBbox,\n context.size.width,\n context.size.height,\n modelPreferences,\n );\n debugSection('original targetRect %j', targetRect);\n\n const referenceBboxList = result.content.references_bbox || [];\n debugSection('referenceBboxList %j', referenceBboxList);\n\n const referenceRects = referenceBboxList\n .filter((bbox) => Array.isArray(bbox))\n .map((bbox) => {\n return adaptBboxToRect(\n bbox,\n context.size.width,\n context.size.height,\n modelPreferences,\n );\n });\n debugSection('referenceRects %j', referenceRects);\n\n // merge the sectionRect and referenceRects\n const mergedRect = mergeRects([targetRect, ...referenceRects]);\n debugSection('mergedRect %j', mergedRect);\n\n // expand search area to at least 200 x 200\n sectionRect = expandSearchArea(mergedRect, context.size, modelPreferences);\n debugSection('expanded sectionRect %j', sectionRect);\n }\n\n let imageBase64 = screenshotBase64;\n if (sectionRect) {\n imageBase64 = await cropByRect(\n screenshotBase64,\n sectionRect,\n getIsUseQwenVl({\n intent: 'grounding',\n }),\n );\n }\n\n return {\n rect: sectionRect,\n imageBase64,\n error: result.content.error,\n rawResponse: JSON.stringify(result.content),\n usage: result.usage,\n };\n}\n\nexport async function AiExtractElementInfo<\n T,\n ElementType extends BaseElement = BaseElement,\n>(options: {\n dataQuery: string | Record<string, string>;\n multimodalPrompt?: TMultimodalPrompt;\n context: UIContext<ElementType>;\n extractOption?: InsightExtractOption;\n modelPreferences: IModelPreferences;\n}) {\n const {\n dataQuery,\n context,\n extractOption,\n multimodalPrompt,\n modelPreferences,\n } = options;\n const systemPrompt = systemPromptToExtract();\n\n const { screenshotBase64 } = context;\n\n const { description, elementById } = await describeUserPage(\n context,\n modelPreferences,\n {\n truncateTextLength: 200,\n filterNonTextContent: false,\n visibleOnly: false,\n domIncluded: extractOption?.domIncluded,\n },\n );\n\n const extractDataPromptText = await extractDataQueryPrompt(\n description,\n dataQuery,\n );\n\n const userContent: ChatCompletionUserMessageParam['content'] = [];\n\n if (extractOption?.screenshotIncluded !== false) {\n userContent.push({\n type: 'image_url',\n image_url: {\n url: screenshotBase64,\n detail: 'high',\n },\n });\n }\n\n userContent.push({\n type: 'text',\n text: extractDataPromptText,\n });\n\n const msgs: AIArgs = [\n { role: 'system', content: systemPrompt },\n {\n role: 'user',\n content: userContent,\n },\n ];\n\n if (options.extractOption?.returnThought) {\n msgs.push({\n role: 'user',\n content: 'Please provide reasons.',\n });\n }\n\n if (multimodalPrompt) {\n const addOns = await promptsToChatParam({\n images: multimodalPrompt.images,\n convertHttpImage2Base64: multimodalPrompt.convertHttpImage2Base64,\n });\n msgs.push(...addOns);\n }\n\n const result = await callAiFn<AIDataExtractionResponse<T>>(\n msgs,\n AIActionType.EXTRACT_DATA,\n modelPreferences,\n );\n return {\n parseResult: result.content,\n elementById,\n usage: result.usage,\n };\n}\n"],"names":["debugInspect","getDebug","debugSection","extraTextFromUserPrompt","prompt","promptsToChatParam","multimodalPrompt","_multimodalPrompt_images","msgs","item","base64","preProcessImageUrl","AiLocateElement","options","context","targetElementDescription","callAI","screenshotBase64","modelPreferences","description","elementById","insertElementByPosition","describeUserPage","assert","userInstructionPrompt","findElementPrompt","systemPrompt","systemPromptToLocateElement","vlLocateMode","imagePayload","paddingToMatchBlockByBase64","markupImageForLLM","addOns","callAIFn","callToGetJSONObject","res","AIActionType","rawResponse","JSON","resRect","matchedElements","errors","Array","_options_searchConfig_rect","_options_searchConfig_rect1","_options_searchConfig_rect2","_options_searchConfig_rect3","adaptBboxToRect","rectCenter","element","elementByPositionWithElementInfo","distanceToCenter","distance","distanceThreshold","e","msg","Error","undefined","AiLocateSection","sectionDescription","systemPromptToLocateSection","sectionLocatorInstructionText","sectionLocatorInstruction","result","callAiFn","sectionRect","sectionBbox","targetRect","referenceBboxList","referenceRects","bbox","mergedRect","mergeRects","expandSearchArea","imageBase64","cropByRect","getIsUseQwenVl","AiExtractElementInfo","_options_extractOption","dataQuery","extractOption","systemPromptToExtract","extractDataPromptText","extractDataQueryPrompt","userContent"],"mappings":";;;;;;;;;;AA+DA,MAAMA,eAAeC,SAAS;AAC9B,MAAMC,eAAeD,SAAS;AAE9B,MAAME,0BAA0B,CAACC;IAC/B,IAAI,AAAkB,YAAlB,OAAOA,QACT,OAAOA;IAEP,OAAOA,OAAO,MAAM;AAExB;AAEA,MAAMC,qBAAqB,OACzBC;QAGIC;IADJ,MAAMC,OAAyC,EAAE;IACjD,IAAID,QAAAA,mBAAAA,KAAAA,IAAAA,QAAAA,CAAAA,2BAAAA,iBAAkB,MAAM,AAAD,IAAvBA,KAAAA,IAAAA,yBAA0B,MAAM,EAAE;QACpCC,KAAK,IAAI,CAAC;YACR,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,MAAM;gBACR;aACD;QACH;QAEA,KAAK,MAAMC,QAAQH,iBAAiB,MAAM,CAAE;YAC1C,MAAMI,SAAS,MAAMC,mBACnBF,KAAK,GAAG,EACR,CAAC,CAACH,iBAAiB,uBAAuB;YAG5CE,KAAK,IAAI,CAAC;gBACR,MAAM;gBACN,SAAS;oBACP;wBACE,MAAM;wBACN,MAAM,CAAC,gBAAgB,EAAEC,KAAK,IAAI,CAAC,CAAC,CAAC;oBACvC;iBACD;YACH;YAEAD,KAAK,IAAI,CAAC;gBACR,MAAM;gBACN,SAAS;oBACP;wBACE,MAAM;wBACN,WAAW;4BACT,KAAKE;4BACL,QAAQ;wBACV;oBACF;iBACD;YACH;QACF;IACF;IACA,OAAOF;AACT;AAEO,eAAeI,gBAEpBC,OAMD;IAQC,MAAM,EAAEC,OAAO,EAAEC,wBAAwB,EAAEC,MAAM,EAAE,GAAGH;IACtD,MAAM,EAAEI,gBAAgB,EAAE,GAAGH;IAE7B,MAAMI,mBAAsC;QAC1C,QAAQ;IACV;IAEA,MAAM,EAAEC,WAAW,EAAEC,WAAW,EAAEC,uBAAuB,EAAE,GACzD,MAAMC,iBAAiBR,SAASI;IAElCK,OACER,0BACA;IAGF,MAAMS,wBAAwB,MAAMC,kBAAkB,MAAM,CAAC;QAC3D,iBAAiBN;QACjB,0BAA0BhB,wBAAwBY;IACpD;IACA,MAAMW,eAAeC,4BACnBC,aAAaV;IAGf,IAAIW,eAAeZ;IAEnB,IAAIJ,QAAQ,YAAY,EAAE;QACxBU,OACEV,QAAQ,YAAY,CAAC,IAAI,EACzB;QAEFU,OACEV,QAAQ,YAAY,CAAC,WAAW,EAChC;QAGFgB,eAAehB,QAAQ,YAAY,CAAC,WAAW;IACjD,OAAO,IAAIe,AAAmC,cAAnCA,aAAaV,mBACtBW,eAAe,MAAMC,4BAA4BD;SAC5C,IAAI,CAACD,aAAaV,mBACvBW,eAAe,MAAME,kBACnBd,kBACAH,QAAQ,IAAI,EACZA,QAAQ,IAAI;IAIhB,MAAMN,OAAe;QACnB;YAAE,MAAM;YAAU,SAASkB;QAAa;QACxC;YACE,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,WAAW;wBACT,KAAKG;wBACL,QAAQ;oBACV;gBACF;gBACA;oBACE,MAAM;oBACN,MAAML;gBACR;aACD;QACH;KACD;IAED,IAAI,AAAoC,YAApC,OAAOT,0BAAuC;QAChD,MAAMiB,SAAS,MAAM3B,mBAAmB;YACtC,QAAQU,yBAAyB,MAAM;YACvC,yBAAyBA,yBAAyB,uBAAuB;QAC3E;QACAP,KAAK,IAAI,IAAIwB;IACf;IAEA,MAAMC,WACJjB,UAAUkB;IAEZ,MAAMC,MAAM,MAAMF,SAASzB,MAAM4B,aAAa,eAAe,EAAE;QAC7D,QAAQ;IACV;IAEA,MAAMC,cAAcC,KAAK,SAAS,CAACH,IAAI,OAAO;IAE9C,IAAII;IACJ,IAAIC,kBACF,cAAcL,IAAI,OAAO,GAAGA,IAAI,OAAO,CAAC,QAAQ,GAAG,EAAE;IACvD,IAAIM,SACF,YAAYN,IAAI,OAAO,GAAGA,IAAI,OAAO,CAAC,MAAM,GAAG,EAAE;IACnD,IAAI;QACF,IAAI,UAAUA,IAAI,OAAO,IAAIO,MAAM,OAAO,CAACP,IAAI,OAAO,CAAC,IAAI,GAAG;gBAG1DQ,4BAAAA,uBACAC,6BAAAA,wBAEAC,6BAAAA,wBACAC,6BAAAA;YANFP,UAAUQ,gBACRZ,IAAI,OAAO,CAAC,IAAI,EAChBQ,AAAAA,SAAAA,CAAAA,wBAAAA,QAAQ,YAAY,AAAD,IAAnBA,KAAAA,IAAAA,QAAAA,CAAAA,6BAAAA,sBAAsB,IAAI,AAAD,IAAzBA,KAAAA,IAAAA,2BAA4B,KAAK,AAAD,KAAK7B,QAAQ,IAAI,CAAC,KAAK,EACvD8B,AAAAA,SAAAA,CAAAA,yBAAAA,QAAQ,YAAY,AAAD,IAAnBA,KAAAA,IAAAA,QAAAA,CAAAA,8BAAAA,uBAAsB,IAAI,AAAD,IAAzBA,KAAAA,IAAAA,4BAA4B,MAAM,AAAD,KAAK9B,QAAQ,IAAI,CAAC,MAAM,EACzDI,kBAAAA,QACA2B,CAAAA,yBAAAA,QAAQ,YAAY,AAAD,IAAnBA,KAAAA,IAAAA,QAAAA,CAAAA,8BAAAA,uBAAsB,IAAI,AAAD,IAAzBA,KAAAA,IAAAA,4BAA4B,IAAI,UAChCC,CAAAA,yBAAAA,QAAQ,YAAY,AAAD,IAAnBA,KAAAA,IAAAA,QAAAA,CAAAA,8BAAAA,uBAAsB,IAAI,AAAD,IAAzBA,KAAAA,IAAAA,4BAA4B,GAAG;YAEjC9C,aAAa,WAAWuC;YAExB,MAAMS,aAAa;gBACjB,GAAGT,QAAQ,IAAI,GAAGA,QAAQ,KAAK,GAAG;gBAClC,GAAGA,QAAQ,GAAG,GAAGA,QAAQ,MAAM,GAAG;YACpC;YACA,IAAIU,UAAUC,iCAAiCpC,QAAQ,IAAI,EAAEkC;YAE7D,MAAMG,mBAAmBF,UACrBG,SAAS;gBAAE,GAAGH,QAAQ,MAAM,CAAC,EAAE;gBAAE,GAAGA,QAAQ,MAAM,CAAC,EAAE;YAAC,GAAGD,cACzD;YAEJ,IAAI,CAACC,WAAWE,mBAAmBE,mBACjCJ,UAAU5B,wBAAwB2B;YAGpC,IAAIC,SAAS;gBACXT,kBAAkB;oBAACS;iBAAQ;gBAC3BR,SAAS,EAAE;YACb;QACF;IACF,EAAE,OAAOa,GAAG;QACV,MAAMC,MACJD,aAAaE,QACT,CAAC,sBAAsB,EAAEF,EAAE,OAAO,EAAE,GACpC;QACN,IAAI,AAACb,UAAUA,AAAAA,CAAAA,QAAAA,SAAAA,KAAAA,IAAAA,OAAQ,MAAM,AAAD,MAAM,GAGhCA,OAAO,IAAI,CAAC,CAAC,CAAC,EAAEc,IAAI,CAAC,CAAC;aAFtBd,SAAS;YAACc;SAAI;IAIlB;IAEA,OAAO;QACL,MAAMhB;QACN,aAAa;YACX,UAAUC;YACVC;QACF;QACAJ;QACAjB;QACA,OAAOe,IAAI,KAAK;QAChB,kBACE,AAAuB,YAAvB,OAAOA,IAAI,OAAO,IAClBA,AAAgB,SAAhBA,IAAI,OAAO,IACX,sBAAsBA,IAAI,OAAO,GAC5BA,IAAI,OAAO,CAAS,gBAAgB,GACrCsB;IACR;AACF;AAEO,eAAeC,gBAAgB7C,OAIrC;IAOC,MAAM,EAAEC,OAAO,EAAE6C,kBAAkB,EAAE,GAAG9C;IACxC,MAAM,EAAEI,gBAAgB,EAAE,GAAGH;IAE7B,MAAMI,mBAAsC;QAC1C,QAAQ;IACV;IAEA,MAAMQ,eAAekC,4BACnBhC,aAAaV;IAEf,MAAM2C,gCAAgC,MAAMC,0BAA0B,MAAM,CAAC;QAC3E,oBAAoB3D,wBAAwBwD;IAC9C;IACA,MAAMnD,OAAe;QACnB;YAAE,MAAM;YAAU,SAASkB;QAAa;QACxC;YACE,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,WAAW;wBACT,KAAKT;wBACL,QAAQ;oBACV;gBACF;gBACA;oBACE,MAAM;oBACN,MAAM4C;gBACR;aACD;QACH;KACD;IAED,IAAI,AAA8B,YAA9B,OAAOF,oBAAiC;QAC1C,MAAM3B,SAAS,MAAM3B,mBAAmB;YACtC,QAAQsD,mBAAmB,MAAM;YACjC,yBAAyBA,mBAAmB,uBAAuB;QACrE;QACAnD,KAAK,IAAI,IAAIwB;IACf;IAEA,MAAM+B,SAAS,MAAMC,SACnBxD,MACA4B,aAAa,YAAY,EACzB;QACE,QAAQ;IACV;IAGF,IAAI6B;IACJ,MAAMC,cAAcH,OAAO,OAAO,CAAC,IAAI;IACvC,IAAIG,aAAa;QACf,MAAMC,aAAapB,gBACjBmB,aACApD,QAAQ,IAAI,CAAC,KAAK,EAClBA,QAAQ,IAAI,CAAC,MAAM,EACnBI;QAEFhB,aAAa,0BAA0BiE;QAEvC,MAAMC,oBAAoBL,OAAO,OAAO,CAAC,eAAe,IAAI,EAAE;QAC9D7D,aAAa,wBAAwBkE;QAErC,MAAMC,iBAAiBD,kBACpB,MAAM,CAAC,CAACE,OAAS5B,MAAM,OAAO,CAAC4B,OAC/B,GAAG,CAAC,CAACA,OACGvB,gBACLuB,MACAxD,QAAQ,IAAI,CAAC,KAAK,EAClBA,QAAQ,IAAI,CAAC,MAAM,EACnBI;QAGNhB,aAAa,qBAAqBmE;QAGlC,MAAME,aAAaC,WAAW;YAACL;eAAeE;SAAe;QAC7DnE,aAAa,iBAAiBqE;QAG9BN,cAAcQ,iBAAiBF,YAAYzD,QAAQ,IAAI,EAAEI;QACzDhB,aAAa,2BAA2B+D;IAC1C;IAEA,IAAIS,cAAczD;IAClB,IAAIgD,aACFS,cAAc,MAAMC,WAClB1D,kBACAgD,aACAW,eAAe;QACb,QAAQ;IACV;IAIJ,OAAO;QACL,MAAMX;QACNS;QACA,OAAOX,OAAO,OAAO,CAAC,KAAK;QAC3B,aAAazB,KAAK,SAAS,CAACyB,OAAO,OAAO;QAC1C,OAAOA,OAAO,KAAK;IACrB;AACF;AAEO,eAAec,qBAGpBhE,OAMD;QAqDKiE;IApDJ,MAAM,EACJC,SAAS,EACTjE,OAAO,EACPkE,aAAa,EACb1E,gBAAgB,EAChBY,gBAAgB,EACjB,GAAGL;IACJ,MAAMa,eAAeuD;IAErB,MAAM,EAAEhE,gBAAgB,EAAE,GAAGH;IAE7B,MAAM,EAAEK,WAAW,EAAEC,WAAW,EAAE,GAAG,MAAME,iBACzCR,SACAI,kBACA;QACE,oBAAoB;QACpB,sBAAsB;QACtB,aAAa;QACb,aAAa8D,QAAAA,gBAAAA,KAAAA,IAAAA,cAAe,WAAW;IACzC;IAGF,MAAME,wBAAwB,MAAMC,uBAClChE,aACA4D;IAGF,MAAMK,cAAyD,EAAE;IAEjE,IAAIJ,AAAAA,CAAAA,QAAAA,gBAAAA,KAAAA,IAAAA,cAAe,kBAAkB,AAAD,MAAM,OACxCI,YAAY,IAAI,CAAC;QACf,MAAM;QACN,WAAW;YACT,KAAKnE;YACL,QAAQ;QACV;IACF;IAGFmE,YAAY,IAAI,CAAC;QACf,MAAM;QACN,MAAMF;IACR;IAEA,MAAM1E,OAAe;QACnB;YAAE,MAAM;YAAU,SAASkB;QAAa;QACxC;YACE,MAAM;YACN,SAAS0D;QACX;KACD;IAED,IAAI,QAAAN,CAAAA,yBAAAA,QAAQ,aAAa,AAAD,IAApBA,KAAAA,IAAAA,uBAAuB,aAAa,EACtCtE,KAAK,IAAI,CAAC;QACR,MAAM;QACN,SAAS;IACX;IAGF,IAAIF,kBAAkB;QACpB,MAAM0B,SAAS,MAAM3B,mBAAmB;YACtC,QAAQC,iBAAiB,MAAM;YAC/B,yBAAyBA,iBAAiB,uBAAuB;QACnE;QACAE,KAAK,IAAI,IAAIwB;IACf;IAEA,MAAM+B,SAAS,MAAMC,SACnBxD,MACA4B,aAAa,YAAY,EACzBlB;IAEF,OAAO;QACL,aAAa6C,OAAO,OAAO;QAC3B3C;QACA,OAAO2C,OAAO,KAAK;IACrB;AACF"}
|