@midscene/core 0.30.5 → 1.0.1-beta-20251021060907.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/es/agent/agent.mjs +41 -33
- package/dist/es/agent/agent.mjs.map +1 -1
- package/dist/es/agent/execution-session.mjs +41 -0
- package/dist/es/agent/execution-session.mjs.map +1 -0
- package/dist/es/agent/task-builder.mjs +303 -0
- package/dist/es/agent/task-builder.mjs.map +1 -0
- package/dist/es/agent/tasks.mjs +68 -391
- package/dist/es/agent/tasks.mjs.map +1 -1
- package/dist/es/agent/ui-utils.mjs.map +1 -1
- package/dist/es/agent/utils.mjs +6 -6
- package/dist/es/agent/utils.mjs.map +1 -1
- package/dist/es/ai-model/common.mjs +1 -15
- package/dist/es/ai-model/common.mjs.map +1 -1
- package/dist/es/ai-model/inspect.mjs +2 -3
- package/dist/es/ai-model/inspect.mjs.map +1 -1
- package/dist/es/ai-model/llm-planning.mjs +6 -24
- package/dist/es/ai-model/llm-planning.mjs.map +1 -1
- package/dist/es/ai-model/prompt/llm-locator.mjs +3 -204
- package/dist/es/ai-model/prompt/llm-locator.mjs.map +1 -1
- package/dist/es/ai-model/service-caller/index.mjs +101 -231
- package/dist/es/ai-model/service-caller/index.mjs.map +1 -1
- package/dist/es/index.mjs +3 -2
- package/dist/es/index.mjs.map +1 -1
- package/dist/es/insight/index.mjs +18 -19
- package/dist/es/insight/index.mjs.map +1 -1
- package/dist/es/insight/utils.mjs +3 -3
- package/dist/es/insight/utils.mjs.map +1 -1
- package/dist/es/report.mjs.map +1 -1
- package/dist/es/{ai-model/action-executor.mjs → task-runner.mjs} +69 -10
- package/dist/es/task-runner.mjs.map +1 -0
- package/dist/es/types.mjs +18 -1
- package/dist/es/types.mjs.map +1 -1
- package/dist/es/utils.mjs +2 -2
- package/dist/es/yaml/player.mjs +18 -14
- package/dist/es/yaml/player.mjs.map +1 -1
- package/dist/lib/agent/agent.js +41 -33
- package/dist/lib/agent/agent.js.map +1 -1
- package/dist/lib/agent/execution-session.js +75 -0
- package/dist/lib/agent/execution-session.js.map +1 -0
- package/dist/lib/agent/task-builder.js +340 -0
- package/dist/lib/agent/task-builder.js.map +1 -0
- package/dist/lib/agent/tasks.js +68 -391
- package/dist/lib/agent/tasks.js.map +1 -1
- package/dist/lib/agent/ui-utils.js.map +1 -1
- package/dist/lib/agent/utils.js +6 -6
- package/dist/lib/agent/utils.js.map +1 -1
- package/dist/lib/ai-model/common.js +2 -19
- package/dist/lib/ai-model/common.js.map +1 -1
- package/dist/lib/ai-model/inspect.js +1 -2
- package/dist/lib/ai-model/inspect.js.map +1 -1
- package/dist/lib/ai-model/llm-planning.js +5 -23
- package/dist/lib/ai-model/llm-planning.js.map +1 -1
- package/dist/lib/ai-model/prompt/llm-locator.js +2 -206
- package/dist/lib/ai-model/prompt/llm-locator.js.map +1 -1
- package/dist/lib/ai-model/service-caller/index.js +236 -384
- package/dist/lib/ai-model/service-caller/index.js.map +1 -1
- package/dist/lib/index.js +9 -5
- package/dist/lib/index.js.map +1 -1
- package/dist/lib/insight/index.js +17 -18
- package/dist/lib/insight/index.js.map +1 -1
- package/dist/lib/insight/utils.js +5 -5
- package/dist/lib/insight/utils.js.map +1 -1
- package/dist/lib/report.js.map +1 -1
- package/dist/lib/{ai-model/action-executor.js → task-runner.js} +71 -12
- package/dist/lib/task-runner.js.map +1 -0
- package/dist/lib/types.js +22 -1
- package/dist/lib/types.js.map +1 -1
- package/dist/lib/utils.js +2 -2
- package/dist/lib/yaml/player.js +18 -14
- package/dist/lib/yaml/player.js.map +1 -1
- package/dist/types/agent/agent.d.ts +16 -0
- package/dist/types/agent/execution-session.d.ts +27 -0
- package/dist/types/agent/task-builder.d.ts +24 -0
- package/dist/types/agent/tasks.d.ts +8 -11
- package/dist/types/agent/ui-utils.d.ts +2 -2
- package/dist/types/agent/utils.d.ts +5 -2
- package/dist/types/ai-model/common.d.ts +0 -1
- package/dist/types/ai-model/prompt/llm-locator.d.ts +0 -2
- package/dist/types/index.d.ts +4 -3
- package/dist/types/insight/index.d.ts +5 -10
- package/dist/types/insight/utils.d.ts +2 -2
- package/dist/types/{ai-model/action-executor.d.ts → task-runner.d.ts} +14 -3
- package/dist/types/types.d.ts +47 -4
- package/dist/types/yaml.d.ts +3 -1
- package/package.json +4 -7
- package/dist/es/ai-model/action-executor.mjs.map +0 -1
- package/dist/lib/ai-model/action-executor.js.map +0 -1
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"ai-model/common.js","sources":["webpack://@midscene/core/webpack/runtime/define_property_getters","webpack://@midscene/core/webpack/runtime/has_own_property","webpack://@midscene/core/webpack/runtime/make_namespace_object","webpack://@midscene/core/./src/ai-model/common.ts"],"sourcesContent":["__webpack_require__.d = (exports, definition) => {\n\tfor(var key in definition) {\n if(__webpack_require__.o(definition, key) && !__webpack_require__.o(exports, key)) {\n Object.defineProperty(exports, key, { enumerable: true, get: definition[key] });\n }\n }\n};","__webpack_require__.o = (obj, prop) => (Object.prototype.hasOwnProperty.call(obj, prop))","// define __esModule on exports\n__webpack_require__.r = (exports) => {\n\tif(typeof Symbol !== 'undefined' && Symbol.toStringTag) {\n\t\tObject.defineProperty(exports, Symbol.toStringTag, { value: 'Module' });\n\t}\n\tObject.defineProperty(exports, '__esModule', { value: true });\n};","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 { ChatCompletionMessageParam } 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 = ChatCompletionMessageParam[];\n\nexport enum AIActionType {\n ASSERT = 0,\n INSPECT_ELEMENT = 1,\n EXTRACT_DATA = 2,\n PLAN = 3,\n DESCRIBE_ELEMENT = 4,\n TEXT = 5,\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 rightLimit: number,\n bottomLimit: 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(\n locate.bbox,\n width,\n height,\n rightLimit,\n bottomLimit,\n vlMode,\n );\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 rightLimit: number,\n bottomLimit: number,\n vlMode: TVlModeTypes | undefined,\n): [number, number, number, number] {\n let result: [number, number, number, number] = [0, 0, 0, 0];\n if (vlMode === 'doubao-vision' || vlMode === 'vlm-ui-tars') {\n result = adaptDoubaoBbox(bbox, width, height);\n } else if (vlMode === 'gemini') {\n result = adaptGeminiBbox(bbox, width, height);\n } else if (vlMode === 'qwen3-vl') {\n result = normalized01000(bbox, width, height);\n } else {\n result = adaptQwenBbox(bbox);\n }\n\n result[2] = Math.min(result[2], rightLimit);\n result[3] = Math.min(result[3], bottomLimit);\n\n return result;\n}\n\n// x1, y1, x2, y2 -> 0-1000\nexport function normalized01000(\n bbox: number[],\n width: number,\n height: number,\n): [number, number, number, number] {\n return [\n Math.round((bbox[0] * width) / 1000),\n Math.round((bbox[1] * height) / 1000),\n Math.round((bbox[2] * width) / 1000),\n Math.round((bbox[3] * height) / 1000),\n ];\n}\n\n// y1, x1, y2, x2 -> 0-1000\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 rightLimit = width,\n bottomLimit = height,\n vlMode?: TVlModeTypes | undefined,\n): Rect {\n debugInspectUtils(\n 'adaptBboxToRect',\n bbox,\n width,\n height,\n 'offset',\n offsetX,\n offsetY,\n 'limit',\n rightLimit,\n bottomLimit,\n 'vlMode',\n vlMode,\n );\n const [left, top, right, bottom] = adaptBbox(\n bbox,\n width,\n height,\n rightLimit,\n bottomLimit,\n vlMode,\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\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 // Calculate padding needed to reach minimum edge size\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\n // Calculate new dimensions (ensure minimum edge size)\n let newWidth = Math.max(minEdgeSize, rect.width + paddingSizeHorizontal * 2);\n let newHeight = Math.max(minEdgeSize, rect.height + paddingSizeVertical * 2);\n\n // Calculate initial position with padding\n let newLeft = rect.left - paddingSizeHorizontal;\n let newTop = rect.top - paddingSizeVertical;\n\n // Ensure the rect doesn't exceed screen boundaries by adjusting position\n // If the rect goes beyond the right edge, shift it left\n if (newLeft + newWidth > screenSize.width) {\n newLeft = screenSize.width - newWidth;\n }\n\n // If the rect goes beyond the bottom edge, shift it up\n if (newTop + newHeight > screenSize.height) {\n newTop = screenSize.height - newHeight;\n }\n\n // Ensure the rect doesn't go beyond the left/top edges\n newLeft = Math.max(0, newLeft);\n newTop = Math.max(0, newTop);\n\n // If after position adjustment, the rect still exceeds screen boundaries,\n // clamp the dimensions to fit within screen\n if (newLeft + newWidth > screenSize.width) {\n newWidth = screenSize.width - newLeft;\n }\n if (newTop + newHeight > screenSize.height) {\n newHeight = screenSize.height - newTop;\n }\n\n rect.left = newLeft;\n rect.top = newTop;\n rect.width = newWidth;\n rect.height = newHeight;\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\n// Schema for locator field input (when users provide locate parameters)\nconst MidsceneLocationInput = z\n .object({\n prompt: TUserPromptSchema,\n deepThink: z.boolean().optional(),\n cacheable: z.boolean().optional(),\n xpath: z.union([z.string(), z.boolean()]).optional(),\n })\n .passthrough();\n\n// Schema for locator field result (when AI returns locate results)\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\n// Export the result type - this is used for runtime results that include center and rect\nexport type MidsceneLocationResultType = z.infer<typeof MidsceneLocationResult>;\n\n// Export the input type - this is the inferred type from getMidsceneLocationSchema()\nexport type MidsceneLocationInputType = z.infer<typeof MidsceneLocationInput>;\n\n/**\n * Returns the schema for locator fields.\n * This now returns the input schema which is more permissive and suitable for validation.\n */\nexport const getMidsceneLocationSchema = () => {\n return MidsceneLocationInput;\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 ZodObject\n if (actualField._def?.typeName === 'ZodObject') {\n const shape = actualField._def.shape();\n\n // Method 1: Check for the location field flag (for result schema)\n if (locateFieldFlagName in shape) {\n return true;\n }\n\n // Method 2: Check if it's the input schema by checking for 'prompt' field\n // Input schema has 'prompt' as a required field\n if ('prompt' in shape && shape.prompt) {\n return true;\n }\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\n/**\n * Parse and validate action parameters using Zod schema.\n * All fields are validated through Zod, EXCEPT locator fields which are skipped.\n * Default values defined in the schema are automatically applied.\n *\n * Locator fields are special business logic fields with complex validation requirements,\n * so they are intentionally excluded from Zod parsing and use existing validation logic.\n */\nexport const parseActionParam = (\n rawParam: Record<string, any>,\n zodSchema: z.ZodType<any>,\n): Record<string, any> => {\n // Handle undefined or null rawParam by providing an empty object\n const param = rawParam ?? {};\n\n // Find all locate fields in the schema\n const locateFields = findAllMidsceneLocatorField(zodSchema);\n\n // If there are no locate fields, just do normal validation\n if (locateFields.length === 0) {\n return zodSchema.parse(param);\n }\n\n // Extract locate field values to restore later\n const locateFieldValues: Record<string, any> = {};\n for (const fieldName of locateFields) {\n if (fieldName in param) {\n locateFieldValues[fieldName] = param[fieldName];\n }\n }\n\n // Build params for validation - skip locate fields and use dummy values\n const paramsForValidation: Record<string, any> = {};\n for (const key in param) {\n if (locateFields.includes(key)) {\n // Use dummy value to satisfy schema validation\n paramsForValidation[key] = { prompt: '_dummy_' };\n } else {\n paramsForValidation[key] = param[key];\n }\n }\n\n // Validate with dummy locate values\n const validated = zodSchema.parse(paramsForValidation);\n\n // Restore the actual locate field values (unvalidated, as per business requirement)\n for (const fieldName in locateFieldValues) {\n validated[fieldName] = locateFieldValues[fieldName];\n }\n\n return validated;\n};\n"],"names":["__webpack_require__","definition","key","Object","obj","prop","Symbol","AIActionType","defaultBboxSize","debugInspectUtils","getDebug","fillBboxParam","locate","width","height","rightLimit","bottomLimit","vlMode","adaptBbox","adaptQwenBbox","bbox","msg","JSON","Error","result","Math","adaptDoubaoBbox","assert","splitted","Number","Array","bboxList","item","x","y","adaptGeminiBbox","normalized01000","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","newWidth","newHeight","newLeft","newTop","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","MidsceneLocationInput","getMidsceneLocationSchema","ifMidsceneLocatorField","field","_actualField__def","_actualField__def1","actualField","shape","dumpMidsceneLocatorField","String","findAllMidsceneLocatorField","zodType","requiredOnly","_zodObject__def","zodObject","keys","_field__def","jsonObject","zodSchema","locatorFields","fieldName","fieldValue","loadActionParam","parseActionParam","rawParam","param","locateFields","locateFieldValues","paramsForValidation","validated"],"mappings":";;;IAAAA,oBAAoB,CAAC,GAAG,CAAC,UAASC;QACjC,IAAI,IAAIC,OAAOD,WACR,IAAGD,oBAAoB,CAAC,CAACC,YAAYC,QAAQ,CAACF,oBAAoB,CAAC,CAAC,UAASE,MACzEC,OAAO,cAAc,CAAC,UAASD,KAAK;YAAE,YAAY;YAAM,KAAKD,UAAU,CAACC,IAAI;QAAC;IAGzF;;;ICNAF,oBAAoB,CAAC,GAAG,CAACI,KAAKC,OAAUF,OAAO,SAAS,CAAC,cAAc,CAAC,IAAI,CAACC,KAAKC;;;ICClFL,oBAAoB,CAAC,GAAG,CAAC;QACxB,IAAG,AAAkB,eAAlB,OAAOM,UAA0BA,OAAO,WAAW,EACrDH,OAAO,cAAc,CAAC,UAASG,OAAO,WAAW,EAAE;YAAE,OAAO;QAAS;QAEtEH,OAAO,cAAc,CAAC,UAAS,cAAc;YAAE,OAAO;QAAK;IAC5D;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;ACkBO,IAAKI,sBAAYA,WAAAA,GAAAA,SAAZA,YAAY;;;;;;;WAAZA;;AASZ,MAAMC,kBAAkB;AACxB,MAAMC,oBAAoBC,AAAAA,IAAAA,uBAAAA,QAAAA,AAAAA,EAAS;AAG5B,SAASC,cACdC,MAA2B,EAC3BC,KAAa,EACbC,MAAc,EACdC,UAAkB,EAClBC,WAAmB,EACnBC,MAAgC;IAGhC,IAAKL,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,GAAGM,UACZN,OAAO,IAAI,EACXC,OACAC,QACAC,YACAC,aACAC;IAIJ,OAAOL;AACT;AAEO,SAASO,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,GAAGZ;QACN,YAAnB,OAAOY,IAAI,CAAC,EAAE,GACVK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,IAClBK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,GAAGZ;KAC1B;IACD,OAAOgB;AACT;AAEO,SAASE,gBACdN,IAAkC,EAClCP,KAAa,EACbC,MAAc;IAEda,IAAAA,sBAAAA,MAAAA,AAAAA,EACEd,QAAQ,KAAKC,SAAS,GACtB;IAGF,IAAI,AAAgB,YAAhB,OAAOM,MAAmB;QAC5BO,IAAAA,sBAAAA,MAAAA,AAAAA,EACE,+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,IAAIf,QAAS;YAC3CY,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAId,SAAU;YAC5CW,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAIf,QAAS;YAC3CY,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAId,SAAU;SAC7C;QAEH,MAAM,IAAIS,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,GAAGlB,QAAS;QACnCY,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGjB,SAAU;QACpCW,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGlB,QAAS;QACnCY,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGjB,SAAU;KACrC;IAIH,IACEiB,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,GAAGlB,QAAS,QAAQL,kBAAkB;QAE/DiB,KAAK,GAAG,CACN,GACAA,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGjB,SAAU,QAAQN,kBAAkB;QAEhEiB,KAAK,GAAG,CACNZ,OACAY,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGlB,QAAS,QAAQL,kBAAkB;QAE/DiB,KAAK,GAAG,CACNX,QACAW,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGjB,SAAU,QAAQN,kBAAkB;KAEjE;IAGH,IAAIY,AAAgB,MAAhBA,KAAK,MAAM,EACb,OAAO;QACLK,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGlB,QAAS;QACnCY,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGjB,SAAU;QACpCW,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGlB,QAAS;QACnCY,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGjB,SAAU;KACrC;IAGH,MAAMO,MAAM,CAAC,0CAA0C,EAAEC,KAAK,SAAS,CAACF,MAAM,CAAC,CAAC;IAChF,MAAM,IAAIG,MAAMF;AAClB;AAEO,SAASH,UACdE,IAAc,EACdP,KAAa,EACbC,MAAc,EACdC,UAAkB,EAClBC,WAAmB,EACnBC,MAAgC;IAEhC,IAAIO,SAA2C;QAAC;QAAG;QAAG;QAAG;KAAE;IAEzDA,SADEP,AAAW,oBAAXA,UAA8BA,AAAW,kBAAXA,SACvBS,gBAAgBN,MAAMP,OAAOC,UAC7BG,AAAW,aAAXA,SACAkB,gBAAgBf,MAAMP,OAAOC,UAC7BG,AAAW,eAAXA,SACAmB,gBAAgBhB,MAAMP,OAAOC,UAE7BK,cAAcC;IAGzBI,MAAM,CAAC,EAAE,GAAGC,KAAK,GAAG,CAACD,MAAM,CAAC,EAAE,EAAET;IAChCS,MAAM,CAAC,EAAE,GAAGC,KAAK,GAAG,CAACD,MAAM,CAAC,EAAE,EAAER;IAEhC,OAAOQ;AACT;AAGO,SAASY,gBACdhB,IAAc,EACdP,KAAa,EACbC,MAAc;IAEd,OAAO;QACLW,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGP,QAAS;QAC/BY,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGN,SAAU;QAChCW,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGP,QAAS;QAC/BY,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGN,SAAU;KACjC;AACH;AAGO,SAASqB,gBACdf,IAAc,EACdP,KAAa,EACbC,MAAc;IAEd,MAAMuB,OAAOZ,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGP,QAAS;IAC5C,MAAMyB,MAAMb,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGN,SAAU;IAC5C,MAAMyB,QAAQd,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGP,QAAS;IAC7C,MAAM2B,SAASf,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGN,SAAU;IAC/C,OAAO;QAACuB;QAAMC;QAAKC;QAAOC;KAAO;AACnC;AAEO,SAASC,gBACdrB,IAAc,EACdP,KAAa,EACbC,MAAc,EACd4B,UAAU,CAAC,EACXC,UAAU,CAAC,EACX5B,aAAaF,KAAK,EAClBG,cAAcF,MAAM,EACpBG,MAAiC;IAEjCR,kBACE,mBACAW,MACAP,OACAC,QACA,UACA4B,SACAC,SACA,SACA5B,YACAC,aACA,UACAC;IAEF,MAAM,CAACoB,MAAMC,KAAKC,OAAOC,OAAO,GAAGtB,UACjCE,MACAP,OACAC,QACAC,YACAC,aACAC;IAIF,MAAM2B,WAAWP;IACjB,MAAMQ,UAAUP;IAChB,IAAIQ,YAAYP,QAAQF;IACxB,IAAIU,aAAaP,SAASF;IAI1B,IAAIM,WAAWE,YAAYjC,OACzBiC,YAAYjC,QAAQ+B;IAItB,IAAIC,UAAUE,aAAajC,QACzBiC,aAAajC,SAAS+B;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;IACAtC,kBAAkB,4BAA4BuC;IAE9C,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,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,EAChB9C,MAAgC;IAEhC,MAAM+C,cAAc/C,AAAW,oBAAXA,SAA6B,MAAM;IACvD,MAAMgD,iBAAiB;IAGvB,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;IAGN,IAAIG,WAAW3C,KAAK,GAAG,CAACuC,aAAahB,KAAK,KAAK,GAAGkB,AAAwB,IAAxBA;IAClD,IAAIG,YAAY5C,KAAK,GAAG,CAACuC,aAAahB,KAAK,MAAM,GAAGmB,AAAsB,IAAtBA;IAGpD,IAAIG,UAAUtB,KAAK,IAAI,GAAGkB;IAC1B,IAAIK,SAASvB,KAAK,GAAG,GAAGmB;IAIxB,IAAIG,UAAUF,WAAWL,WAAW,KAAK,EACvCO,UAAUP,WAAW,KAAK,GAAGK;IAI/B,IAAIG,SAASF,YAAYN,WAAW,MAAM,EACxCQ,SAASR,WAAW,MAAM,GAAGM;IAI/BC,UAAU7C,KAAK,GAAG,CAAC,GAAG6C;IACtBC,SAAS9C,KAAK,GAAG,CAAC,GAAG8C;IAIrB,IAAID,UAAUF,WAAWL,WAAW,KAAK,EACvCK,WAAWL,WAAW,KAAK,GAAGO;IAEhC,IAAIC,SAASF,YAAYN,WAAW,MAAM,EACxCM,YAAYN,WAAW,MAAM,GAAGQ;IAGlCvB,KAAK,IAAI,GAAGsB;IACZtB,KAAK,GAAG,GAAGuB;IACXvB,KAAK,KAAK,GAAGoB;IACbpB,KAAK,MAAM,GAAGqB;IAEd,OAAOrB;AACT;AAEO,eAAewB,kBACpBC,gBAAwB,EACxBC,IAAkC,EAClCvB,IAAU;IAEV,MAAMwB,eAAeC,AAAAA,IAAAA,0BAAAA,UAAAA,AAAAA,EAAWF;IAChC,MAAMG,kCAAkCF,aAAc,MAAM,CAC1D,CAACG;QACC,IAAIA,YAAY,UAAU,CAAC,QAAQ,KAAKC,0BAAAA,QAAAA,CAAAA,IAAa,EACnD,OAAO;QAET,OAAO;IACT;IAGF,MAAMC,eAAe,MAAMC,AAAAA,IAAAA,oBAAAA,uBAAAA,AAAAA,EAAwB;QACjD,gBAAgBR;QAChB,sBAAsBI;QACtB1B;IACF;IACA,OAAO6B;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;YACXnC,QAAQ,IAAI,CACV,CAAC,sBAAsB,EAAEkC,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,6BAAAA,CAAAA,CAAAA,MAAQ,CAAC;IAClC,MAAMA,6BAAAA,CAAAA,CAAAA,MAAQ;IACd,KAAKA,6BAAAA,CAAAA,CAAAA,MAAQ;AACf;AAEO,MAAMC,aAAaD,6BAAAA,CAAAA,CAAAA,MAAQ,CAAC;IACjC,OAAOA,6BAAAA,CAAAA,CAAAA,MAAQ;IACf,QAAQA,6BAAAA,CAAAA,CAAAA,MAAQ;IAChB,KAAKA,6BAAAA,CAAAA,CAAAA,MAAQ,GAAG,QAAQ;AAC1B;AAEO,MAAME,aAAaH,YAAY,GAAG,CAACE,YAAY,GAAG,CACvDD,6BAAAA,CAAAA,CAAAA,MAAQ,CAAC;IACP,MAAMA,6BAAAA,CAAAA,CAAAA,MAAQ,GAAG,QAAQ;AAC3B;AAIK,MAAMG,0BAA0BH,6BAAAA,CAAAA,CAAAA,MAAQ,CAAC;IAC9C,QAAQA,6BAAAA,CAAAA,CAAAA,KACA,CACJA,6BAAAA,CAAAA,CAAAA,MAAQ,CAAC;QACP,MAAMA,6BAAAA,CAAAA,CAAAA,MAAQ;QACd,KAAKA,6BAAAA,CAAAA,CAAAA,MAAQ;IACf,IAED,QAAQ;IACX,yBAAyBA,6BAAAA,CAAAA,CAAAA,OAAS,GAAG,QAAQ;AAC/C;AAGO,MAAMI,oBAAoBJ,6BAAAA,CAAAA,CAAAA,KAAO,CAAC;IACvCA,6BAAAA,CAAAA,CAAAA,MAAQ;IACRA,6BAAAA,CAAAA,CAAAA,MACS,CAAC;QACN,QAAQA,6BAAAA,CAAAA,CAAAA,MAAQ;IAClB,GACC,GAAG,CAACG,wBAAwB,OAAO;CACvC;AAMD,MAAME,sBAAsB;AAG5B,MAAMC,wBAAwBN,6BAAAA,CAAAA,CAAAA,MACrB,CAAC;IACN,QAAQI;IACR,WAAWJ,6BAAAA,CAAAA,CAAAA,OAAS,GAAG,QAAQ;IAC/B,WAAWA,6BAAAA,CAAAA,CAAAA,OAAS,GAAG,QAAQ;IAC/B,OAAOA,6BAAAA,CAAAA,CAAAA,KAAO,CAAC;QAACA,6BAAAA,CAAAA,CAAAA,MAAQ;QAAIA,6BAAAA,CAAAA,CAAAA,OAAS;KAAG,EAAE,QAAQ;AACpD,GACC,WAAW;AAGiBA,6BAAAA,CAAAA,CAAAA,MACtB,CAAC;IACN,CAACK,oBAAoB,EAAEL,6BAAAA,CAAAA,CAAAA,OAAS,CAAC;IACjC,QAAQI;IAGR,WAAWJ,6BAAAA,CAAAA,CAAAA,OAAS,GAAG,QAAQ;IAC/B,WAAWA,6BAAAA,CAAAA,CAAAA,OAAS,GAAG,QAAQ;IAC/B,OAAOA,6BAAAA,CAAAA,CAAAA,OAAS,GAAG,QAAQ;IAG3B,QAAQA,6BAAAA,CAAAA,CAAAA,KAAO,CAAC;QAACA,6BAAAA,CAAAA,CAAAA,MAAQ;QAAIA,6BAAAA,CAAAA,CAAAA,MAAQ;KAAG;IACxC,MAAME;AACR,GACC,WAAW;AAYP,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;QAGpC,IAAIP,uBAAuBQ,OACzB,OAAO;QAKT,IAAI,YAAYA,SAASA,MAAM,MAAM,EACnC,OAAO;IAEX;IAEA,OAAO;AACT;AAEO,MAAMC,2BAA2B,CAACL;IACvC7E,IAAAA,sBAAAA,MAAAA,AAAAA,EACE4E,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,OAAOjH,OAAO,IAAI,CAACgH,UAAU,KAAK;QACxC,OAAOC,KAAK,MAAM,CAAC,CAAClH;YAClB,MAAMsG,QAAQW,UAAU,KAAK,CAACjH,IAAI;YAClC,IAAI,CAACqG,uBAAuBC,QAC1B,OAAO;YAIT,IAAIS,cAAc;oBACTI;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,MAAMzB,kBAAkB,CAC7B0B,YACAC;IAEA,MAAMC,gBAAgBT,4BAA4BQ;IAClD,MAAM/F,SAAS;QAAE,GAAG8F,UAAU;IAAC;IAE/B,KAAK,MAAMG,aAAaD,cAAe;QACrC,MAAME,aAAalG,MAAM,CAACiG,UAAU;QACpC,IAAIC,YAEF;YAAA,IAAI,AAAsB,YAAtB,OAAOA,YACTlG,MAAM,CAACiG,UAAU,GAAGC;iBACf,IAAI,AAAsB,YAAtB,OAAOA,YAEhB;gBAAA,IAAIA,WAAW,MAAM,EAEnB;oBAAA,IAAI,AAA6B,YAA7B,OAAOA,WAAW,MAAM,EAC1BlG,MAAM,CAACiG,UAAU,GAAGC,WAAW,MAAM;yBAChC,IACL,AAA6B,YAA7B,OAAOA,WAAW,MAAM,IACxBA,WAAW,MAAM,CAAC,MAAM,EAGxBlG,MAAM,CAACiG,UAAU,GAAGC,WAAW,MAAM,CAAC,MAAM;gBAC9C;YACF;QACF;IAEJ;IAEA,OAAOlG;AACT;AAEO,MAAMmG,kBAAkB,CAC7BL,YACAC;IAEA,MAAMC,gBAAgBT,4BAA4BQ;IAClD,MAAM/F,SAAS;QAAE,GAAG8F,UAAU;IAAC;IAE/B,KAAK,MAAMG,aAAaD,cAAe;QACrC,MAAME,aAAalG,MAAM,CAACiG,UAAU;QACpC,IAAIC,cAAc,AAAsB,YAAtB,OAAOA,YACvBlG,MAAM,CAACiG,UAAU,GAAG;YAClB,CAACrB,oBAAoB,EAAE;YACvB,QAAQsB;QACV;IAEJ;IAEA,OAAOlG;AACT;AAUO,MAAMoG,mBAAmB,CAC9BC,UACAN;IAGA,MAAMO,QAAQD,YAAY,CAAC;IAG3B,MAAME,eAAehB,4BAA4BQ;IAGjD,IAAIQ,AAAwB,MAAxBA,aAAa,MAAM,EACrB,OAAOR,UAAU,KAAK,CAACO;IAIzB,MAAME,oBAAyC,CAAC;IAChD,KAAK,MAAMP,aAAaM,aACtB,IAAIN,aAAaK,OACfE,iBAAiB,CAACP,UAAU,GAAGK,KAAK,CAACL,UAAU;IAKnD,MAAMQ,sBAA2C,CAAC;IAClD,IAAK,MAAM/H,OAAO4H,MAChB,IAAIC,aAAa,QAAQ,CAAC7H,MAExB+H,mBAAmB,CAAC/H,IAAI,GAAG;QAAE,QAAQ;IAAU;SAE/C+H,mBAAmB,CAAC/H,IAAI,GAAG4H,KAAK,CAAC5H,IAAI;IAKzC,MAAMgI,YAAYX,UAAU,KAAK,CAACU;IAGlC,IAAK,MAAMR,aAAaO,kBACtBE,SAAS,CAACT,UAAU,GAAGO,iBAAiB,CAACP,UAAU;IAGrD,OAAOS;AACT"}
|
|
1
|
+
{"version":3,"file":"ai-model/common.js","sources":["webpack://@midscene/core/webpack/runtime/define_property_getters","webpack://@midscene/core/webpack/runtime/has_own_property","webpack://@midscene/core/webpack/runtime/make_namespace_object","webpack://@midscene/core/./src/ai-model/common.ts"],"sourcesContent":["__webpack_require__.d = (exports, definition) => {\n\tfor(var key in definition) {\n if(__webpack_require__.o(definition, key) && !__webpack_require__.o(exports, key)) {\n Object.defineProperty(exports, key, { enumerable: true, get: definition[key] });\n }\n }\n};","__webpack_require__.o = (obj, prop) => (Object.prototype.hasOwnProperty.call(obj, prop))","// define __esModule on exports\n__webpack_require__.r = (exports) => {\n\tif(typeof Symbol !== 'undefined' && Symbol.toStringTag) {\n\t\tObject.defineProperty(exports, Symbol.toStringTag, { value: 'Module' });\n\t}\n\tObject.defineProperty(exports, '__esModule', { value: true });\n};","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 { ChatCompletionMessageParam } 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 = ChatCompletionMessageParam[];\n\nexport enum AIActionType {\n ASSERT = 0,\n INSPECT_ELEMENT = 1,\n EXTRACT_DATA = 2,\n PLAN = 3,\n DESCRIBE_ELEMENT = 4,\n TEXT = 5,\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 rightLimit: number,\n bottomLimit: 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(\n locate.bbox,\n width,\n height,\n rightLimit,\n bottomLimit,\n vlMode,\n );\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 rightLimit: number,\n bottomLimit: number,\n vlMode: TVlModeTypes | undefined,\n): [number, number, number, number] {\n let result: [number, number, number, number] = [0, 0, 0, 0];\n if (vlMode === 'doubao-vision' || vlMode === 'vlm-ui-tars') {\n result = adaptDoubaoBbox(bbox, width, height);\n } else if (vlMode === 'gemini') {\n result = adaptGeminiBbox(bbox, width, height);\n } else if (vlMode === 'qwen3-vl') {\n result = normalized01000(bbox, width, height);\n } else {\n result = adaptQwenBbox(bbox);\n }\n\n result[2] = Math.min(result[2], rightLimit);\n result[3] = Math.min(result[3], bottomLimit);\n\n return result;\n}\n\n// x1, y1, x2, y2 -> 0-1000\nexport function normalized01000(\n bbox: number[],\n width: number,\n height: number,\n): [number, number, number, number] {\n return [\n Math.round((bbox[0] * width) / 1000),\n Math.round((bbox[1] * height) / 1000),\n Math.round((bbox[2] * width) / 1000),\n Math.round((bbox[3] * height) / 1000),\n ];\n}\n\n// y1, x1, y2, x2 -> 0-1000\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 rightLimit = width,\n bottomLimit = height,\n vlMode?: TVlModeTypes | undefined,\n): Rect {\n debugInspectUtils(\n 'adaptBboxToRect',\n bbox,\n width,\n height,\n 'offset',\n offsetX,\n offsetY,\n 'limit',\n rightLimit,\n bottomLimit,\n 'vlMode',\n vlMode,\n );\n const [left, top, right, bottom] = adaptBbox(\n bbox,\n width,\n height,\n rightLimit,\n bottomLimit,\n vlMode,\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\n return rect;\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 // Calculate padding needed to reach minimum edge size\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\n // Calculate new dimensions (ensure minimum edge size)\n let newWidth = Math.max(minEdgeSize, rect.width + paddingSizeHorizontal * 2);\n let newHeight = Math.max(minEdgeSize, rect.height + paddingSizeVertical * 2);\n\n // Calculate initial position with padding\n let newLeft = rect.left - paddingSizeHorizontal;\n let newTop = rect.top - paddingSizeVertical;\n\n // Ensure the rect doesn't exceed screen boundaries by adjusting position\n // If the rect goes beyond the right edge, shift it left\n if (newLeft + newWidth > screenSize.width) {\n newLeft = screenSize.width - newWidth;\n }\n\n // If the rect goes beyond the bottom edge, shift it up\n if (newTop + newHeight > screenSize.height) {\n newTop = screenSize.height - newHeight;\n }\n\n // Ensure the rect doesn't go beyond the left/top edges\n newLeft = Math.max(0, newLeft);\n newTop = Math.max(0, newTop);\n\n // If after position adjustment, the rect still exceeds screen boundaries,\n // clamp the dimensions to fit within screen\n if (newLeft + newWidth > screenSize.width) {\n newWidth = screenSize.width - newLeft;\n }\n if (newTop + newHeight > screenSize.height) {\n newHeight = screenSize.height - newTop;\n }\n\n rect.left = newLeft;\n rect.top = newTop;\n rect.width = newWidth;\n rect.height = newHeight;\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\n// Schema for locator field input (when users provide locate parameters)\nconst MidsceneLocationInput = z\n .object({\n prompt: TUserPromptSchema,\n deepThink: z.boolean().optional(),\n cacheable: z.boolean().optional(),\n xpath: z.union([z.string(), z.boolean()]).optional(),\n })\n .passthrough();\n\n// Schema for locator field result (when AI returns locate results)\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\n// Export the result type - this is used for runtime results that include center and rect\nexport type MidsceneLocationResultType = z.infer<typeof MidsceneLocationResult>;\n\n// Export the input type - this is the inferred type from getMidsceneLocationSchema()\nexport type MidsceneLocationInputType = z.infer<typeof MidsceneLocationInput>;\n\n/**\n * Returns the schema for locator fields.\n * This now returns the input schema which is more permissive and suitable for validation.\n */\nexport const getMidsceneLocationSchema = () => {\n return MidsceneLocationInput;\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 ZodObject\n if (actualField._def?.typeName === 'ZodObject') {\n const shape = actualField._def.shape();\n\n // Method 1: Check for the location field flag (for result schema)\n if (locateFieldFlagName in shape) {\n return true;\n }\n\n // Method 2: Check if it's the input schema by checking for 'prompt' field\n // Input schema has 'prompt' as a required field\n if ('prompt' in shape && shape.prompt) {\n return true;\n }\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\n/**\n * Parse and validate action parameters using Zod schema.\n * All fields are validated through Zod, EXCEPT locator fields which are skipped.\n * Default values defined in the schema are automatically applied.\n *\n * Locator fields are special business logic fields with complex validation requirements,\n * so they are intentionally excluded from Zod parsing and use existing validation logic.\n */\nexport const parseActionParam = (\n rawParam: Record<string, any>,\n zodSchema: z.ZodType<any>,\n): Record<string, any> => {\n // Handle undefined or null rawParam by providing an empty object\n const param = rawParam ?? {};\n\n // Find all locate fields in the schema\n const locateFields = findAllMidsceneLocatorField(zodSchema);\n\n // If there are no locate fields, just do normal validation\n if (locateFields.length === 0) {\n return zodSchema.parse(param);\n }\n\n // Extract locate field values to restore later\n const locateFieldValues: Record<string, any> = {};\n for (const fieldName of locateFields) {\n if (fieldName in param) {\n locateFieldValues[fieldName] = param[fieldName];\n }\n }\n\n // Build params for validation - skip locate fields and use dummy values\n const paramsForValidation: Record<string, any> = {};\n for (const key in param) {\n if (locateFields.includes(key)) {\n // Use dummy value to satisfy schema validation\n paramsForValidation[key] = { prompt: '_dummy_' };\n } else {\n paramsForValidation[key] = param[key];\n }\n }\n\n // Validate with dummy locate values\n const validated = zodSchema.parse(paramsForValidation);\n\n // Restore the actual locate field values (unvalidated, as per business requirement)\n for (const fieldName in locateFieldValues) {\n validated[fieldName] = locateFieldValues[fieldName];\n }\n\n return validated;\n};\n"],"names":["__webpack_require__","definition","key","Object","obj","prop","Symbol","AIActionType","defaultBboxSize","debugInspectUtils","getDebug","fillBboxParam","locate","width","height","rightLimit","bottomLimit","vlMode","adaptBbox","adaptQwenBbox","bbox","msg","JSON","Error","result","Math","adaptDoubaoBbox","assert","splitted","Number","Array","bboxList","item","x","y","adaptGeminiBbox","normalized01000","left","top","right","bottom","adaptBboxToRect","offsetX","offsetY","rectLeft","rectTop","rectWidth","rectHeight","rect","mergeRects","rects","minLeft","r","minTop","maxRight","maxBottom","expandSearchArea","screenSize","minEdgeSize","defaultPadding","paddingSizeHorizontal","paddingSizeVertical","newWidth","newHeight","newLeft","newTop","markupImageForLLM","screenshotBase64","tree","size","elementsInfo","treeToList","elementsPositionInfoWithoutText","elementInfo","NodeType","imagePayload","compositeElementInfoImg","buildYamlFlowFromPlans","plans","actionSpace","sleep","flow","plan","verb","action","console","flowKey","flowParam","dumpActionParam","flowItem","PointSchema","z","SizeSchema","RectSchema","TMultimodalPromptSchema","TUserPromptSchema","locateFieldFlagName","MidsceneLocationInput","getMidsceneLocationSchema","ifMidsceneLocatorField","field","_actualField__def","_actualField__def1","actualField","shape","dumpMidsceneLocatorField","String","findAllMidsceneLocatorField","zodType","requiredOnly","_zodObject__def","zodObject","keys","_field__def","jsonObject","zodSchema","locatorFields","fieldName","fieldValue","loadActionParam","parseActionParam","rawParam","param","locateFields","locateFieldValues","paramsForValidation","validated"],"mappings":";;;IAAAA,oBAAoB,CAAC,GAAG,CAAC,UAASC;QACjC,IAAI,IAAIC,OAAOD,WACR,IAAGD,oBAAoB,CAAC,CAACC,YAAYC,QAAQ,CAACF,oBAAoB,CAAC,CAAC,UAASE,MACzEC,OAAO,cAAc,CAAC,UAASD,KAAK;YAAE,YAAY;YAAM,KAAKD,UAAU,CAACC,IAAI;QAAC;IAGzF;;;ICNAF,oBAAoB,CAAC,GAAG,CAACI,KAAKC,OAAUF,OAAO,SAAS,CAAC,cAAc,CAAC,IAAI,CAACC,KAAKC;;;ICClFL,oBAAoB,CAAC,GAAG,CAAC;QACxB,IAAG,AAAkB,eAAlB,OAAOM,UAA0BA,OAAO,WAAW,EACrDH,OAAO,cAAc,CAAC,UAASG,OAAO,WAAW,EAAE;YAAE,OAAO;QAAS;QAEtEH,OAAO,cAAc,CAAC,UAAS,cAAc;YAAE,OAAO;QAAK;IAC5D;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;ACkBO,IAAKI,sBAAYA,WAAAA,GAAAA,SAAZA,YAAY;;;;;;;WAAZA;;AASZ,MAAMC,kBAAkB;AACxB,MAAMC,oBAAoBC,AAAAA,IAAAA,uBAAAA,QAAAA,AAAAA,EAAS;AAG5B,SAASC,cACdC,MAA2B,EAC3BC,KAAa,EACbC,MAAc,EACdC,UAAkB,EAClBC,WAAmB,EACnBC,MAAgC;IAGhC,IAAKL,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,GAAGM,UACZN,OAAO,IAAI,EACXC,OACAC,QACAC,YACAC,aACAC;IAIJ,OAAOL;AACT;AAEO,SAASO,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,GAAGZ;QACN,YAAnB,OAAOY,IAAI,CAAC,EAAE,GACVK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,IAClBK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,GAAGZ;KAC1B;IACD,OAAOgB;AACT;AAEO,SAASE,gBACdN,IAAkC,EAClCP,KAAa,EACbC,MAAc;IAEda,IAAAA,sBAAAA,MAAAA,AAAAA,EACEd,QAAQ,KAAKC,SAAS,GACtB;IAGF,IAAI,AAAgB,YAAhB,OAAOM,MAAmB;QAC5BO,IAAAA,sBAAAA,MAAAA,AAAAA,EACE,+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,IAAIf,QAAS;YAC3CY,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAId,SAAU;YAC5CW,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAIf,QAAS;YAC3CY,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAId,SAAU;SAC7C;QAEH,MAAM,IAAIS,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,GAAGlB,QAAS;QACnCY,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGjB,SAAU;QACpCW,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGlB,QAAS;QACnCY,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGjB,SAAU;KACrC;IAIH,IACEiB,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,GAAGlB,QAAS,QAAQL,kBAAkB;QAE/DiB,KAAK,GAAG,CACN,GACAA,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGjB,SAAU,QAAQN,kBAAkB;QAEhEiB,KAAK,GAAG,CACNZ,OACAY,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGlB,QAAS,QAAQL,kBAAkB;QAE/DiB,KAAK,GAAG,CACNX,QACAW,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGjB,SAAU,QAAQN,kBAAkB;KAEjE;IAGH,IAAIY,AAAgB,MAAhBA,KAAK,MAAM,EACb,OAAO;QACLK,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGlB,QAAS;QACnCY,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGjB,SAAU;QACpCW,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGlB,QAAS;QACnCY,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGjB,SAAU;KACrC;IAGH,MAAMO,MAAM,CAAC,0CAA0C,EAAEC,KAAK,SAAS,CAACF,MAAM,CAAC,CAAC;IAChF,MAAM,IAAIG,MAAMF;AAClB;AAEO,SAASH,UACdE,IAAc,EACdP,KAAa,EACbC,MAAc,EACdC,UAAkB,EAClBC,WAAmB,EACnBC,MAAgC;IAEhC,IAAIO,SAA2C;QAAC;QAAG;QAAG;QAAG;KAAE;IAEzDA,SADEP,AAAW,oBAAXA,UAA8BA,AAAW,kBAAXA,SACvBS,gBAAgBN,MAAMP,OAAOC,UAC7BG,AAAW,aAAXA,SACAkB,gBAAgBf,MAAMP,OAAOC,UAC7BG,AAAW,eAAXA,SACAmB,gBAAgBhB,MAAMP,OAAOC,UAE7BK,cAAcC;IAGzBI,MAAM,CAAC,EAAE,GAAGC,KAAK,GAAG,CAACD,MAAM,CAAC,EAAE,EAAET;IAChCS,MAAM,CAAC,EAAE,GAAGC,KAAK,GAAG,CAACD,MAAM,CAAC,EAAE,EAAER;IAEhC,OAAOQ;AACT;AAGO,SAASY,gBACdhB,IAAc,EACdP,KAAa,EACbC,MAAc;IAEd,OAAO;QACLW,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGP,QAAS;QAC/BY,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGN,SAAU;QAChCW,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGP,QAAS;QAC/BY,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGN,SAAU;KACjC;AACH;AAGO,SAASqB,gBACdf,IAAc,EACdP,KAAa,EACbC,MAAc;IAEd,MAAMuB,OAAOZ,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGP,QAAS;IAC5C,MAAMyB,MAAMb,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGN,SAAU;IAC5C,MAAMyB,QAAQd,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGP,QAAS;IAC7C,MAAM2B,SAASf,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGN,SAAU;IAC/C,OAAO;QAACuB;QAAMC;QAAKC;QAAOC;KAAO;AACnC;AAEO,SAASC,gBACdrB,IAAc,EACdP,KAAa,EACbC,MAAc,EACd4B,UAAU,CAAC,EACXC,UAAU,CAAC,EACX5B,aAAaF,KAAK,EAClBG,cAAcF,MAAM,EACpBG,MAAiC;IAEjCR,kBACE,mBACAW,MACAP,OACAC,QACA,UACA4B,SACAC,SACA,SACA5B,YACAC,aACA,UACAC;IAEF,MAAM,CAACoB,MAAMC,KAAKC,OAAOC,OAAO,GAAGtB,UACjCE,MACAP,OACAC,QACAC,YACAC,aACAC;IAIF,MAAM2B,WAAWP;IACjB,MAAMQ,UAAUP;IAChB,IAAIQ,YAAYP,QAAQF;IACxB,IAAIU,aAAaP,SAASF;IAI1B,IAAIM,WAAWE,YAAYjC,OACzBiC,YAAYjC,QAAQ+B;IAItB,IAAIC,UAAUE,aAAajC,QACzBiC,aAAajC,SAAS+B;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;IACAtC,kBAAkB,4BAA4BuC;IAE9C,OAAOA;AACT;AAEO,SAASC,WAAWC,KAAa;IACtC,MAAMC,UAAU1B,KAAK,GAAG,IAAIyB,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,IAAI;IACnD,MAAMC,SAAS5B,KAAK,GAAG,IAAIyB,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,GAAG;IACjD,MAAME,WAAW7B,KAAK,GAAG,IAAIyB,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,IAAI,GAAGA,EAAE,KAAK;IAC9D,MAAMG,YAAY9B,KAAK,GAAG,IAAIyB,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,iBACdR,IAAU,EACVS,UAAgB,EAChBxC,MAAgC;IAEhC,MAAMyC,cAAczC,AAAW,oBAAXA,SAA6B,MAAM;IACvD,MAAM0C,iBAAiB;IAGvB,MAAMC,wBACJZ,KAAK,KAAK,GAAGU,cACTjC,KAAK,IAAI,CAAEiC,AAAAA,CAAAA,cAAcV,KAAK,KAAI,IAAK,KACvCW;IACN,MAAME,sBACJb,KAAK,MAAM,GAAGU,cACVjC,KAAK,IAAI,CAAEiC,AAAAA,CAAAA,cAAcV,KAAK,MAAK,IAAK,KACxCW;IAGN,IAAIG,WAAWrC,KAAK,GAAG,CAACiC,aAAaV,KAAK,KAAK,GAAGY,AAAwB,IAAxBA;IAClD,IAAIG,YAAYtC,KAAK,GAAG,CAACiC,aAAaV,KAAK,MAAM,GAAGa,AAAsB,IAAtBA;IAGpD,IAAIG,UAAUhB,KAAK,IAAI,GAAGY;IAC1B,IAAIK,SAASjB,KAAK,GAAG,GAAGa;IAIxB,IAAIG,UAAUF,WAAWL,WAAW,KAAK,EACvCO,UAAUP,WAAW,KAAK,GAAGK;IAI/B,IAAIG,SAASF,YAAYN,WAAW,MAAM,EACxCQ,SAASR,WAAW,MAAM,GAAGM;IAI/BC,UAAUvC,KAAK,GAAG,CAAC,GAAGuC;IACtBC,SAASxC,KAAK,GAAG,CAAC,GAAGwC;IAIrB,IAAID,UAAUF,WAAWL,WAAW,KAAK,EACvCK,WAAWL,WAAW,KAAK,GAAGO;IAEhC,IAAIC,SAASF,YAAYN,WAAW,MAAM,EACxCM,YAAYN,WAAW,MAAM,GAAGQ;IAGlCjB,KAAK,IAAI,GAAGgB;IACZhB,KAAK,GAAG,GAAGiB;IACXjB,KAAK,KAAK,GAAGc;IACbd,KAAK,MAAM,GAAGe;IAEd,OAAOf;AACT;AAEO,eAAekB,kBACpBC,gBAAwB,EACxBC,IAAkC,EAClCC,IAAU;IAEV,MAAMC,eAAeC,AAAAA,IAAAA,0BAAAA,UAAAA,AAAAA,EAAWH;IAChC,MAAMI,kCAAkCF,aAAc,MAAM,CAC1D,CAACG;QACC,IAAIA,YAAY,UAAU,CAAC,QAAQ,KAAKC,0BAAAA,QAAAA,CAAAA,IAAa,EACnD,OAAO;QAET,OAAO;IACT;IAGF,MAAMC,eAAe,MAAMC,AAAAA,IAAAA,oBAAAA,uBAAAA,AAAAA,EAAwB;QACjD,gBAAgBT;QAChB,sBAAsBK;QACtBH;IACF;IACA,OAAOM;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;YACXC,QAAQ,IAAI,CACV,CAAC,sBAAsB,EAAEF,KAAK,8BAA8B,CAAC;YAE/D;QACF;QAEA,MAAMG,UAAUF,OAAO,cAAc,IAAID;QACzC,MAAMI,YAAYH,OAAO,WAAW,GAChCI,gBAAgBN,KAAK,KAAK,IAAI,CAAC,GAAGE,OAAO,WAAW,IACpD,CAAC;QAEL,MAAMK,WAAiC;YACrC,CAACH,QAAQ,EAAE;YACX,GAAGC,SAAS;QACd;QAEAN,KAAK,IAAI,CAACQ;IACZ;IAEA,IAAIT,OACFC,KAAK,IAAI,CAAC;QACRD;IACF;IAGF,OAAOC;AACT;AAGO,MAAMS,cAAcC,6BAAAA,CAAAA,CAAAA,MAAQ,CAAC;IAClC,MAAMA,6BAAAA,CAAAA,CAAAA,MAAQ;IACd,KAAKA,6BAAAA,CAAAA,CAAAA,MAAQ;AACf;AAEO,MAAMC,aAAaD,6BAAAA,CAAAA,CAAAA,MAAQ,CAAC;IACjC,OAAOA,6BAAAA,CAAAA,CAAAA,MAAQ;IACf,QAAQA,6BAAAA,CAAAA,CAAAA,MAAQ;IAChB,KAAKA,6BAAAA,CAAAA,CAAAA,MAAQ,GAAG,QAAQ;AAC1B;AAEO,MAAME,aAAaH,YAAY,GAAG,CAACE,YAAY,GAAG,CACvDD,6BAAAA,CAAAA,CAAAA,MAAQ,CAAC;IACP,MAAMA,6BAAAA,CAAAA,CAAAA,MAAQ,GAAG,QAAQ;AAC3B;AAIK,MAAMG,0BAA0BH,6BAAAA,CAAAA,CAAAA,MAAQ,CAAC;IAC9C,QAAQA,6BAAAA,CAAAA,CAAAA,KACA,CACJA,6BAAAA,CAAAA,CAAAA,MAAQ,CAAC;QACP,MAAMA,6BAAAA,CAAAA,CAAAA,MAAQ;QACd,KAAKA,6BAAAA,CAAAA,CAAAA,MAAQ;IACf,IAED,QAAQ;IACX,yBAAyBA,6BAAAA,CAAAA,CAAAA,OAAS,GAAG,QAAQ;AAC/C;AAGO,MAAMI,oBAAoBJ,6BAAAA,CAAAA,CAAAA,KAAO,CAAC;IACvCA,6BAAAA,CAAAA,CAAAA,MAAQ;IACRA,6BAAAA,CAAAA,CAAAA,MACS,CAAC;QACN,QAAQA,6BAAAA,CAAAA,CAAAA,MAAQ;IAClB,GACC,GAAG,CAACG,wBAAwB,OAAO;CACvC;AAMD,MAAME,sBAAsB;AAG5B,MAAMC,wBAAwBN,6BAAAA,CAAAA,CAAAA,MACrB,CAAC;IACN,QAAQI;IACR,WAAWJ,6BAAAA,CAAAA,CAAAA,OAAS,GAAG,QAAQ;IAC/B,WAAWA,6BAAAA,CAAAA,CAAAA,OAAS,GAAG,QAAQ;IAC/B,OAAOA,6BAAAA,CAAAA,CAAAA,KAAO,CAAC;QAACA,6BAAAA,CAAAA,CAAAA,MAAQ;QAAIA,6BAAAA,CAAAA,CAAAA,OAAS;KAAG,EAAE,QAAQ;AACpD,GACC,WAAW;AAGiBA,6BAAAA,CAAAA,CAAAA,MACtB,CAAC;IACN,CAACK,oBAAoB,EAAEL,6BAAAA,CAAAA,CAAAA,OAAS,CAAC;IACjC,QAAQI;IAGR,WAAWJ,6BAAAA,CAAAA,CAAAA,OAAS,GAAG,QAAQ;IAC/B,WAAWA,6BAAAA,CAAAA,CAAAA,OAAS,GAAG,QAAQ;IAC/B,OAAOA,6BAAAA,CAAAA,CAAAA,OAAS,GAAG,QAAQ;IAG3B,QAAQA,6BAAAA,CAAAA,CAAAA,KAAO,CAAC;QAACA,6BAAAA,CAAAA,CAAAA,MAAQ;QAAIA,6BAAAA,CAAAA,CAAAA,MAAQ;KAAG;IACxC,MAAME;AACR,GACC,WAAW;AAYP,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;QAGpC,IAAIP,uBAAuBQ,OACzB,OAAO;QAKT,IAAI,YAAYA,SAASA,MAAM,MAAM,EACnC,OAAO;IAEX;IAEA,OAAO;AACT;AAEO,MAAMC,2BAA2B,CAACL;IACvCzE,IAAAA,sBAAAA,MAAAA,AAAAA,EACEwE,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,OAAO7G,OAAO,IAAI,CAAC4G,UAAU,KAAK;QACxC,OAAOC,KAAK,MAAM,CAAC,CAAC9G;YAClB,MAAMkG,QAAQW,UAAU,KAAK,CAAC7G,IAAI;YAClC,IAAI,CAACiG,uBAAuBC,QAC1B,OAAO;YAIT,IAAIS,cAAc;oBACTI;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,MAAMzB,kBAAkB,CAC7B0B,YACAC;IAEA,MAAMC,gBAAgBT,4BAA4BQ;IAClD,MAAM3F,SAAS;QAAE,GAAG0F,UAAU;IAAC;IAE/B,KAAK,MAAMG,aAAaD,cAAe;QACrC,MAAME,aAAa9F,MAAM,CAAC6F,UAAU;QACpC,IAAIC,YAEF;YAAA,IAAI,AAAsB,YAAtB,OAAOA,YACT9F,MAAM,CAAC6F,UAAU,GAAGC;iBACf,IAAI,AAAsB,YAAtB,OAAOA,YAEhB;gBAAA,IAAIA,WAAW,MAAM,EAEnB;oBAAA,IAAI,AAA6B,YAA7B,OAAOA,WAAW,MAAM,EAC1B9F,MAAM,CAAC6F,UAAU,GAAGC,WAAW,MAAM;yBAChC,IACL,AAA6B,YAA7B,OAAOA,WAAW,MAAM,IACxBA,WAAW,MAAM,CAAC,MAAM,EAGxB9F,MAAM,CAAC6F,UAAU,GAAGC,WAAW,MAAM,CAAC,MAAM;gBAC9C;YACF;QACF;IAEJ;IAEA,OAAO9F;AACT;AAEO,MAAM+F,kBAAkB,CAC7BL,YACAC;IAEA,MAAMC,gBAAgBT,4BAA4BQ;IAClD,MAAM3F,SAAS;QAAE,GAAG0F,UAAU;IAAC;IAE/B,KAAK,MAAMG,aAAaD,cAAe;QACrC,MAAME,aAAa9F,MAAM,CAAC6F,UAAU;QACpC,IAAIC,cAAc,AAAsB,YAAtB,OAAOA,YACvB9F,MAAM,CAAC6F,UAAU,GAAG;YAClB,CAACrB,oBAAoB,EAAE;YACvB,QAAQsB;QACV;IAEJ;IAEA,OAAO9F;AACT;AAUO,MAAMgG,mBAAmB,CAC9BC,UACAN;IAGA,MAAMO,QAAQD,YAAY,CAAC;IAG3B,MAAME,eAAehB,4BAA4BQ;IAGjD,IAAIQ,AAAwB,MAAxBA,aAAa,MAAM,EACrB,OAAOR,UAAU,KAAK,CAACO;IAIzB,MAAME,oBAAyC,CAAC;IAChD,KAAK,MAAMP,aAAaM,aACtB,IAAIN,aAAaK,OACfE,iBAAiB,CAACP,UAAU,GAAGK,KAAK,CAACL,UAAU;IAKnD,MAAMQ,sBAA2C,CAAC;IAClD,IAAK,MAAM3H,OAAOwH,MAChB,IAAIC,aAAa,QAAQ,CAACzH,MAExB2H,mBAAmB,CAAC3H,IAAI,GAAG;QAAE,QAAQ;IAAU;SAE/C2H,mBAAmB,CAAC3H,IAAI,GAAGwH,KAAK,CAACxH,IAAI;IAKzC,MAAM4H,YAAYX,UAAU,KAAK,CAACU;IAGlC,IAAK,MAAMR,aAAaO,kBACtBE,SAAS,CAACT,UAAU,GAAGO,iBAAiB,CAACP,UAAU;IAGrD,OAAOS;AACT"}
|
|
@@ -115,8 +115,7 @@ async function AiLocateElement(options) {
|
|
|
115
115
|
imageWidth = paddedResult.width;
|
|
116
116
|
imageHeight = paddedResult.height;
|
|
117
117
|
imagePayload = paddedResult.imageBase64;
|
|
118
|
-
}
|
|
119
|
-
else if (!vlMode) imagePayload = await (0, external_common_js_namespaceObject.markupImageForLLM)(screenshotBase64, context.tree, context.size);
|
|
118
|
+
}
|
|
120
119
|
const msgs = [
|
|
121
120
|
{
|
|
122
121
|
role: 'system',
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"ai-model/inspect.js","sources":["webpack://@midscene/core/webpack/runtime/define_property_getters","webpack://@midscene/core/webpack/runtime/has_own_property","webpack://@midscene/core/webpack/runtime/make_namespace_object","webpack://@midscene/core/./src/ai-model/inspect.ts"],"sourcesContent":["__webpack_require__.d = (exports, definition) => {\n\tfor(var key in definition) {\n if(__webpack_require__.o(definition, key) && !__webpack_require__.o(exports, key)) {\n Object.defineProperty(exports, key, { enumerable: true, get: definition[key] });\n }\n }\n};","__webpack_require__.o = (obj, prop) => (Object.prototype.hasOwnProperty.call(obj, prop))","// define __esModule on exports\n__webpack_require__.r = (exports) => {\n\tif(typeof Symbol !== 'undefined' && Symbol.toStringTag) {\n\t\tObject.defineProperty(exports, Symbol.toStringTag, { value: 'Module' });\n\t}\n\tObject.defineProperty(exports, '__esModule', { value: true });\n};","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 let imageWidth = context.size.width;\n let imageHeight = context.size.height;\n let originalImageWidth = imageWidth;\n let originalImageHeight = imageHeight;\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 imageWidth = options.searchConfig.rect?.width;\n imageHeight = options.searchConfig.rect?.height;\n originalImageWidth = imageWidth;\n originalImageHeight = imageHeight;\n } else if (vlMode === 'qwen-vl') {\n const paddedResult = await paddingToMatchBlockByBase64(imagePayload);\n imageWidth = paddedResult.width;\n imageHeight = paddedResult.height;\n imagePayload = paddedResult.imageBase64;\n } else if (vlMode === 'qwen3-vl') {\n // const paddedResult = await paddingToMatchBlockByBase64(imagePayload, 32);\n // imageWidth = paddedResult.width;\n // imageHeight = paddedResult.height;\n // imagePayload = paddedResult.imageBase64;\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 imageWidth,\n imageHeight,\n options.searchConfig?.rect?.left,\n options.searchConfig?.rect?.top,\n originalImageWidth,\n originalImageHeight,\n vlMode,\n );\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 context.size.width,\n context.size.height,\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 context.size.width,\n context.size.height,\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 const croppedResult = await cropByRect(\n screenshotBase64,\n sectionRect,\n vlMode === 'qwen-vl',\n );\n imageBase64 = croppedResult.imageBase64;\n sectionRect.width = croppedResult.width;\n sectionRect.height = croppedResult.height;\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 (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":["__webpack_require__","definition","key","Object","obj","prop","Symbol","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","imageWidth","imageHeight","originalImageWidth","originalImageHeight","_options_searchConfig_rect","_options_searchConfig_rect1","paddedResult","paddingToMatchBlockByBase64","markupImageForLLM","addOns","res","AIActionType","rawResponse","JSON","resRect","matchedElements","errors","Array","_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","croppedResult","cropByRect","AiExtractElementInfo","dataQuery","extractOption","systemPromptToExtract","extractDataPromptText","extractDataQueryPrompt","userContent"],"mappings":";;;IAAAA,oBAAoB,CAAC,GAAG,CAAC,UAASC;QACjC,IAAI,IAAIC,OAAOD,WACR,IAAGD,oBAAoB,CAAC,CAACC,YAAYC,QAAQ,CAACF,oBAAoB,CAAC,CAAC,UAASE,MACzEC,OAAO,cAAc,CAAC,UAASD,KAAK;YAAE,YAAY;YAAM,KAAKD,UAAU,CAACC,IAAI;QAAC;IAGzF;;;ICNAF,oBAAoB,CAAC,GAAG,CAACI,KAAKC,OAAUF,OAAO,SAAS,CAAC,cAAc,CAAC,IAAI,CAACC,KAAKC;;;ICClFL,oBAAoB,CAAC,GAAG,CAAC;QACxB,IAAG,AAAkB,eAAlB,OAAOM,UAA0BA,OAAO,WAAW,EACrDH,OAAO,cAAc,CAAC,UAASG,OAAO,WAAW,EAAE;YAAE,OAAO;QAAS;QAEtEH,OAAO,cAAc,CAAC,UAAS,cAAc;YAAE,OAAO;QAAK;IAC5D;;;;;;;;;;;;;;;;;;ACoDA,MAAMI,eAAeC,AAAAA,IAAAA,uBAAAA,QAAAA,AAAAA,EAAS;AAC9B,MAAMC,eAAeD,AAAAA,IAAAA,uBAAAA,QAAAA,AAAAA,EAAS;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,AAAAA,IAAAA,oBAAAA,kBAAAA,AAAAA,EACnBF,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,AAAAA,IAAAA,wBAAAA,gBAAAA,AAAAA,EAAiBT,SAAS;QAAEI;IAAO;IAE3CM,IAAAA,sBAAAA,MAAAA,AAAAA,EACET,0BACA;IAEF,MAAMU,wBAAwB,MAAMC,+BAAAA,iBAAAA,CAAAA,MAAwB,CAAC;QAC3D,iBAAiBN;QACjB,0BAA0BjB,wBAAwBY;IACpD;IACA,MAAMY,eAAeC,AAAAA,IAAAA,+BAAAA,2BAAAA,AAAAA,EAA4BV;IAEjD,IAAIW,eAAeV;IACnB,IAAIW,aAAahB,QAAQ,IAAI,CAAC,KAAK;IACnC,IAAIiB,cAAcjB,QAAQ,IAAI,CAAC,MAAM;IACrC,IAAIkB,qBAAqBF;IACzB,IAAIG,sBAAsBF;IAE1B,IAAIlB,QAAQ,YAAY,EAAE;YAWXqB,4BACCC;QAXdX,IAAAA,sBAAAA,MAAAA,AAAAA,EACEX,QAAQ,YAAY,CAAC,IAAI,EACzB;QAEFW,IAAAA,sBAAAA,MAAAA,AAAAA,EACEX,QAAQ,YAAY,CAAC,WAAW,EAChC;QAGFgB,eAAehB,QAAQ,YAAY,CAAC,WAAW;QAC/CiB,aAAa,QAAAI,CAAAA,6BAAAA,QAAQ,YAAY,CAAC,IAAI,AAAD,IAAxBA,KAAAA,IAAAA,2BAA2B,KAAK;QAC7CH,cAAc,QAAAI,CAAAA,8BAAAA,QAAQ,YAAY,CAAC,IAAI,AAAD,IAAxBA,KAAAA,IAAAA,4BAA2B,MAAM;QAC/CH,qBAAqBF;QACrBG,sBAAsBF;IACxB,OAAO,IAAIb,AAAW,cAAXA,QAAsB;QAC/B,MAAMkB,eAAe,MAAMC,AAAAA,IAAAA,oBAAAA,2BAAAA,AAAAA,EAA4BR;QACvDC,aAAaM,aAAa,KAAK;QAC/BL,cAAcK,aAAa,MAAM;QACjCP,eAAeO,aAAa,WAAW;IACzC,OAAO,IAAIlB,AAAW,eAAXA;SAKJ,IAAI,CAACA,QACVW,eAAe,MAAMS,AAAAA,IAAAA,mCAAAA,iBAAAA,AAAAA,EACnBnB,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,MAAMwB,SAAS,MAAMlC,mBAAmB;YACtC,QAAQU,yBAAyB,MAAM;YACvC,yBAAyBA,yBAAyB,uBAAuB;QAC3E;QACAP,KAAK,IAAI,IAAI+B;IACf;IAEA,MAAMC,MAAM,MAAMxB,SAASR,MAAMiC,mCAAAA,YAAAA,CAAAA,eAA4B,EAAExB;IAE/D,MAAMyB,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;gBAK1DQ,6BAAAA,uBACAC,6BAAAA;YALFL,UAAUM,AAAAA,IAAAA,mCAAAA,eAAAA,AAAAA,EACRV,IAAI,OAAO,CAAC,IAAI,EAChBV,YACAC,aAAAA,QACAiB,CAAAA,wBAAAA,QAAQ,YAAY,AAAD,IAAnBA,KAAAA,IAAAA,QAAAA,CAAAA,8BAAAA,sBAAsB,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/BjB,oBACAC,qBACAf;YAGFlB,aAAa,WAAW4C;YAExB,MAAMO,aAAa;gBACjB,GAAGP,QAAQ,IAAI,GAAGA,QAAQ,KAAK,GAAG;gBAClC,GAAGA,QAAQ,GAAG,GAAGA,QAAQ,MAAM,GAAG;YACpC;YACA,IAAIQ,UAAUC,AAAAA,IAAAA,wBAAAA,gCAAAA,AAAAA,EAAiCvC,QAAQ,IAAI,EAAEqC;YAE7D,MAAMG,mBAAmBF,UACrBG,AAAAA,IAAAA,wBAAAA,QAAAA,AAAAA,EAAS;gBAAE,GAAGH,QAAQ,MAAM,CAAC,EAAE;gBAAE,GAAGA,QAAQ,MAAM,CAAC,EAAE;YAAC,GAAGD,cACzD;YAEJ,IAAI,CAACC,WAAWE,mBAAmBE,wBAAAA,iBAAiBA,EAClDJ,UAAU9B,wBAAwB6B;YAGpC,IAAIC,SAAS;gBACXP,kBAAkB;oBAACO;iBAAQ;gBAC3BN,SAAS,EAAE;YACb;QACF;IACF,EAAE,OAAOW,GAAG;QACV,MAAMC,MACJD,aAAaE,QACT,CAAC,sBAAsB,EAAEF,EAAE,OAAO,EAAE,GACpC;QACN,IAAI,AAACX,UAAUA,AAAAA,CAAAA,QAAAA,SAAAA,KAAAA,IAAAA,OAAQ,MAAM,AAAD,MAAM,GAGhCA,OAAO,IAAI,CAAC,CAAC,CAAC,EAAEY,IAAI,CAAC,CAAC;aAFtBZ,SAAS;YAACY;SAAI;IAIlB;IAEA,OAAO;QACL,MAAMd;QACN,aAAa;YACX,UAAUC;YACVC;QACF;QACAJ;QACArB;QACA,OAAOmB,IAAI,KAAK;QAChB,kBACE,AAAuB,YAAvB,OAAOA,IAAI,OAAO,IAClBA,AAAgB,SAAhBA,IAAI,OAAO,IACX,sBAAsBA,IAAI,OAAO,GAC5BA,IAAI,OAAO,CAAS,gBAAgB,GACrCoB;IACR;AACF;AAEO,eAAeC,gBAAgBhD,OAIrC;IAOC,MAAM,EAAEC,OAAO,EAAEgD,kBAAkB,EAAE7C,WAAW,EAAE,GAAGJ;IACrD,MAAM,EAAEK,MAAM,EAAE,GAAGD;IACnB,MAAM,EAAEE,gBAAgB,EAAE,GAAGL;IAE7B,MAAMa,eAAeoC,AAAAA,IAAAA,uCAAAA,2BAAAA,AAAAA,EAA4B7C;IACjD,MAAM8C,gCAAgC,MAAMC,uCAAAA,yBAAAA,CAAAA,MAAgC,CAAC;QAC3E,oBAAoB9D,wBAAwB2D;IAC9C;IACA,MAAMtD,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,MAAM6C;gBACR;aACD;QACH;KACD;IAED,IAAI,AAA8B,YAA9B,OAAOF,oBAAiC;QAC1C,MAAMvB,SAAS,MAAMlC,mBAAmB;YACtC,QAAQyD,mBAAmB,MAAM;YACjC,yBAAyBA,mBAAmB,uBAAuB;QACrE;QACAtD,KAAK,IAAI,IAAI+B;IACf;IAEA,MAAM2B,SAAS,MAAMC,AAAAA,IAAAA,yBAAAA,wBAAAA,AAAAA,EACnB3D,MACAiC,mCAAAA,YAAAA,CAAAA,YAAyB,EACzBxB;IAGF,IAAImD;IACJ,MAAMC,cAAcH,OAAO,OAAO,CAAC,IAAI;IACvC,IAAIG,aAAa;QACf,MAAMC,aAAapB,AAAAA,IAAAA,mCAAAA,eAAAA,AAAAA,EACjBmB,aACAvD,QAAQ,IAAI,CAAC,KAAK,EAClBA,QAAQ,IAAI,CAAC,MAAM,EACnB,GACA,GACAA,QAAQ,IAAI,CAAC,KAAK,EAClBA,QAAQ,IAAI,CAAC,MAAM,EACnBI;QAEFhB,aAAa,0BAA0BoE;QAEvC,MAAMC,oBAAoBL,OAAO,OAAO,CAAC,eAAe,IAAI,EAAE;QAC9DhE,aAAa,wBAAwBqE;QAErC,MAAMC,iBAAiBD,kBACpB,MAAM,CAAC,CAACE,OAAS1B,MAAM,OAAO,CAAC0B,OAC/B,GAAG,CAAC,CAACA,OACGvB,AAAAA,IAAAA,mCAAAA,eAAAA,AAAAA,EACLuB,MACA3D,QAAQ,IAAI,CAAC,KAAK,EAClBA,QAAQ,IAAI,CAAC,MAAM,EACnB,GACA,GACAA,QAAQ,IAAI,CAAC,KAAK,EAClBA,QAAQ,IAAI,CAAC,MAAM,EACnBI;QAGNhB,aAAa,qBAAqBsE;QAGlC,MAAME,aAAaC,AAAAA,IAAAA,mCAAAA,UAAAA,AAAAA,EAAW;YAACL;eAAeE;SAAe;QAC7DtE,aAAa,iBAAiBwE;QAG9BN,cAAcQ,AAAAA,IAAAA,mCAAAA,gBAAAA,AAAAA,EAAiBF,YAAY5D,QAAQ,IAAI,EAAEI;QACzDhB,aAAa,2BAA2BkE;IAC1C;IAEA,IAAIS,cAAc1D;IAClB,IAAIiD,aAAa;QACf,MAAMU,gBAAgB,MAAMC,AAAAA,IAAAA,oBAAAA,UAAAA,AAAAA,EAC1B5D,kBACAiD,aACAlD,AAAW,cAAXA;QAEF2D,cAAcC,cAAc,WAAW;QACvCV,YAAY,KAAK,GAAGU,cAAc,KAAK;QACvCV,YAAY,MAAM,GAAGU,cAAc,MAAM;IAC3C;IAEA,OAAO;QACL,MAAMV;QACNS;QACA,OAAOX,OAAO,OAAO,CAAC,KAAK;QAC3B,aAAavB,KAAK,SAAS,CAACuB,OAAO,OAAO;QAC1C,OAAOA,OAAO,KAAK;IACrB;AACF;AAEO,eAAec,qBAGpBnE,OAMD;IACC,MAAM,EAAEoE,SAAS,EAAEnE,OAAO,EAAEoE,aAAa,EAAE5E,gBAAgB,EAAEW,WAAW,EAAE,GACxEJ;IACF,MAAM,EAAEK,MAAM,EAAE,GAAGD;IACnB,MAAMU,eAAewD,AAAAA,IAAAA,8BAAAA,qBAAAA,AAAAA;IAErB,MAAM,EAAEhE,gBAAgB,EAAE,GAAGL;IAE7B,MAAM,EAAEM,WAAW,EAAEC,WAAW,EAAE,GAAG,MAAME,AAAAA,IAAAA,wBAAAA,gBAAAA,AAAAA,EAAiBT,SAAS;QACnE,oBAAoB;QACpB,sBAAsB;QACtB,aAAa;QACb,aAAaoE,QAAAA,gBAAAA,KAAAA,IAAAA,cAAe,WAAW;QACvChE;IACF;IAEA,MAAMkE,wBAAwB,MAAMC,AAAAA,IAAAA,8BAAAA,sBAAAA,AAAAA,EAClCjE,aACA6D;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,MAAM5E,OAAe;QACnB;YAAE,MAAM;YAAU,SAASmB;QAAa;QACxC;YACE,MAAM;YACN,SAAS2D;QACX;KACD;IAED,IAAIhF,kBAAkB;QACpB,MAAMiC,SAAS,MAAMlC,mBAAmB;YACtC,QAAQC,iBAAiB,MAAM;YAC/B,yBAAyBA,iBAAiB,uBAAuB;QACnE;QACAE,KAAK,IAAI,IAAI+B;IACf;IAEA,MAAM2B,SAAS,MAAMC,AAAAA,IAAAA,yBAAAA,wBAAAA,AAAAA,EACnB3D,MACAiC,mCAAAA,YAAAA,CAAAA,YAAyB,EACzBxB;IAEF,OAAO;QACL,aAAaiD,OAAO,OAAO;QAC3B7C;QACA,OAAO6C,OAAO,KAAK;IACrB;AACF"}
|
|
1
|
+
{"version":3,"file":"ai-model/inspect.js","sources":["webpack://@midscene/core/webpack/runtime/define_property_getters","webpack://@midscene/core/webpack/runtime/has_own_property","webpack://@midscene/core/webpack/runtime/make_namespace_object","webpack://@midscene/core/./src/ai-model/inspect.ts"],"sourcesContent":["__webpack_require__.d = (exports, definition) => {\n\tfor(var key in definition) {\n if(__webpack_require__.o(definition, key) && !__webpack_require__.o(exports, key)) {\n Object.defineProperty(exports, key, { enumerable: true, get: definition[key] });\n }\n }\n};","__webpack_require__.o = (obj, prop) => (Object.prototype.hasOwnProperty.call(obj, prop))","// define __esModule on exports\n__webpack_require__.r = (exports) => {\n\tif(typeof Symbol !== 'undefined' && Symbol.toStringTag) {\n\t\tObject.defineProperty(exports, Symbol.toStringTag, { value: 'Module' });\n\t}\n\tObject.defineProperty(exports, '__esModule', { value: true });\n};","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 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 let imageWidth = context.size.width;\n let imageHeight = context.size.height;\n let originalImageWidth = imageWidth;\n let originalImageHeight = imageHeight;\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 imageWidth = options.searchConfig.rect?.width;\n imageHeight = options.searchConfig.rect?.height;\n originalImageWidth = imageWidth;\n originalImageHeight = imageHeight;\n } else if (vlMode === 'qwen-vl') {\n const paddedResult = await paddingToMatchBlockByBase64(imagePayload);\n imageWidth = paddedResult.width;\n imageHeight = paddedResult.height;\n imagePayload = paddedResult.imageBase64;\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 imageWidth,\n imageHeight,\n options.searchConfig?.rect?.left,\n options.searchConfig?.rect?.top,\n originalImageWidth,\n originalImageHeight,\n vlMode,\n );\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 context.size.width,\n context.size.height,\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 context.size.width,\n context.size.height,\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 const croppedResult = await cropByRect(\n screenshotBase64,\n sectionRect,\n vlMode === 'qwen-vl',\n );\n imageBase64 = croppedResult.imageBase64;\n sectionRect.width = croppedResult.width;\n sectionRect.height = croppedResult.height;\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 (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":["__webpack_require__","definition","key","Object","obj","prop","Symbol","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","imageWidth","imageHeight","originalImageWidth","originalImageHeight","_options_searchConfig_rect","_options_searchConfig_rect1","paddedResult","paddingToMatchBlockByBase64","addOns","res","AIActionType","rawResponse","JSON","resRect","matchedElements","errors","Array","_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","croppedResult","cropByRect","AiExtractElementInfo","dataQuery","extractOption","systemPromptToExtract","extractDataPromptText","extractDataQueryPrompt","userContent"],"mappings":";;;IAAAA,oBAAoB,CAAC,GAAG,CAAC,UAASC;QACjC,IAAI,IAAIC,OAAOD,WACR,IAAGD,oBAAoB,CAAC,CAACC,YAAYC,QAAQ,CAACF,oBAAoB,CAAC,CAAC,UAASE,MACzEC,OAAO,cAAc,CAAC,UAASD,KAAK;YAAE,YAAY;YAAM,KAAKD,UAAU,CAACC,IAAI;QAAC;IAGzF;;;ICNAF,oBAAoB,CAAC,GAAG,CAACI,KAAKC,OAAUF,OAAO,SAAS,CAAC,cAAc,CAAC,IAAI,CAACC,KAAKC;;;ICClFL,oBAAoB,CAAC,GAAG,CAAC;QACxB,IAAG,AAAkB,eAAlB,OAAOM,UAA0BA,OAAO,WAAW,EACrDH,OAAO,cAAc,CAAC,UAASG,OAAO,WAAW,EAAE;YAAE,OAAO;QAAS;QAEtEH,OAAO,cAAc,CAAC,UAAS,cAAc;YAAE,OAAO;QAAK;IAC5D;;;;;;;;;;;;;;;;;;ACmDA,MAAMI,eAAeC,AAAAA,IAAAA,uBAAAA,QAAAA,AAAAA,EAAS;AAC9B,MAAMC,eAAeD,AAAAA,IAAAA,uBAAAA,QAAAA,AAAAA,EAAS;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,AAAAA,IAAAA,oBAAAA,kBAAAA,AAAAA,EACnBF,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,AAAAA,IAAAA,wBAAAA,gBAAAA,AAAAA,EAAiBT,SAAS;QAAEI;IAAO;IAE3CM,IAAAA,sBAAAA,MAAAA,AAAAA,EACET,0BACA;IAEF,MAAMU,wBAAwB,MAAMC,+BAAAA,iBAAAA,CAAAA,MAAwB,CAAC;QAC3D,iBAAiBN;QACjB,0BAA0BjB,wBAAwBY;IACpD;IACA,MAAMY,eAAeC,AAAAA,IAAAA,+BAAAA,2BAAAA,AAAAA,EAA4BV;IAEjD,IAAIW,eAAeV;IACnB,IAAIW,aAAahB,QAAQ,IAAI,CAAC,KAAK;IACnC,IAAIiB,cAAcjB,QAAQ,IAAI,CAAC,MAAM;IACrC,IAAIkB,qBAAqBF;IACzB,IAAIG,sBAAsBF;IAE1B,IAAIlB,QAAQ,YAAY,EAAE;YAWXqB,4BACCC;QAXdX,IAAAA,sBAAAA,MAAAA,AAAAA,EACEX,QAAQ,YAAY,CAAC,IAAI,EACzB;QAEFW,IAAAA,sBAAAA,MAAAA,AAAAA,EACEX,QAAQ,YAAY,CAAC,WAAW,EAChC;QAGFgB,eAAehB,QAAQ,YAAY,CAAC,WAAW;QAC/CiB,aAAa,QAAAI,CAAAA,6BAAAA,QAAQ,YAAY,CAAC,IAAI,AAAD,IAAxBA,KAAAA,IAAAA,2BAA2B,KAAK;QAC7CH,cAAc,QAAAI,CAAAA,8BAAAA,QAAQ,YAAY,CAAC,IAAI,AAAD,IAAxBA,KAAAA,IAAAA,4BAA2B,MAAM;QAC/CH,qBAAqBF;QACrBG,sBAAsBF;IACxB,OAAO,IAAIb,AAAW,cAAXA,QAAsB;QAC/B,MAAMkB,eAAe,MAAMC,AAAAA,IAAAA,oBAAAA,2BAAAA,AAAAA,EAA4BR;QACvDC,aAAaM,aAAa,KAAK;QAC/BL,cAAcK,aAAa,MAAM;QACjCP,eAAeO,aAAa,WAAW;IACzC;IAEA,MAAM5B,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,MAAMuB,SAAS,MAAMjC,mBAAmB;YACtC,QAAQU,yBAAyB,MAAM;YACvC,yBAAyBA,yBAAyB,uBAAuB;QAC3E;QACAP,KAAK,IAAI,IAAI8B;IACf;IAEA,MAAMC,MAAM,MAAMvB,SAASR,MAAMgC,mCAAAA,YAAAA,CAAAA,eAA4B,EAAEvB;IAE/D,MAAMwB,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;gBAK1DQ,6BAAAA,uBACAC,6BAAAA;YALFL,UAAUM,AAAAA,IAAAA,mCAAAA,eAAAA,AAAAA,EACRV,IAAI,OAAO,CAAC,IAAI,EAChBT,YACAC,aAAAA,QACAgB,CAAAA,wBAAAA,QAAQ,YAAY,AAAD,IAAnBA,KAAAA,IAAAA,QAAAA,CAAAA,8BAAAA,sBAAsB,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/BhB,oBACAC,qBACAf;YAGFlB,aAAa,WAAW2C;YAExB,MAAMO,aAAa;gBACjB,GAAGP,QAAQ,IAAI,GAAGA,QAAQ,KAAK,GAAG;gBAClC,GAAGA,QAAQ,GAAG,GAAGA,QAAQ,MAAM,GAAG;YACpC;YACA,IAAIQ,UAAUC,AAAAA,IAAAA,wBAAAA,gCAAAA,AAAAA,EAAiCtC,QAAQ,IAAI,EAAEoC;YAE7D,MAAMG,mBAAmBF,UACrBG,AAAAA,IAAAA,wBAAAA,QAAAA,AAAAA,EAAS;gBAAE,GAAGH,QAAQ,MAAM,CAAC,EAAE;gBAAE,GAAGA,QAAQ,MAAM,CAAC,EAAE;YAAC,GAAGD,cACzD;YAEJ,IAAI,CAACC,WAAWE,mBAAmBE,wBAAAA,iBAAiBA,EAClDJ,UAAU7B,wBAAwB4B;YAGpC,IAAIC,SAAS;gBACXP,kBAAkB;oBAACO;iBAAQ;gBAC3BN,SAAS,EAAE;YACb;QACF;IACF,EAAE,OAAOW,GAAG;QACV,MAAMC,MACJD,aAAaE,QACT,CAAC,sBAAsB,EAAEF,EAAE,OAAO,EAAE,GACpC;QACN,IAAI,AAACX,UAAUA,AAAAA,CAAAA,QAAAA,SAAAA,KAAAA,IAAAA,OAAQ,MAAM,AAAD,MAAM,GAGhCA,OAAO,IAAI,CAAC,CAAC,CAAC,EAAEY,IAAI,CAAC,CAAC;aAFtBZ,SAAS;YAACY;SAAI;IAIlB;IAEA,OAAO;QACL,MAAMd;QACN,aAAa;YACX,UAAUC;YACVC;QACF;QACAJ;QACApB;QACA,OAAOkB,IAAI,KAAK;QAChB,kBACE,AAAuB,YAAvB,OAAOA,IAAI,OAAO,IAClBA,AAAgB,SAAhBA,IAAI,OAAO,IACX,sBAAsBA,IAAI,OAAO,GAC5BA,IAAI,OAAO,CAAS,gBAAgB,GACrCoB;IACR;AACF;AAEO,eAAeC,gBAAgB/C,OAIrC;IAOC,MAAM,EAAEC,OAAO,EAAE+C,kBAAkB,EAAE5C,WAAW,EAAE,GAAGJ;IACrD,MAAM,EAAEK,MAAM,EAAE,GAAGD;IACnB,MAAM,EAAEE,gBAAgB,EAAE,GAAGL;IAE7B,MAAMa,eAAemC,AAAAA,IAAAA,uCAAAA,2BAAAA,AAAAA,EAA4B5C;IACjD,MAAM6C,gCAAgC,MAAMC,uCAAAA,yBAAAA,CAAAA,MAAgC,CAAC;QAC3E,oBAAoB7D,wBAAwB0D;IAC9C;IACA,MAAMrD,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,MAAM4C;gBACR;aACD;QACH;KACD;IAED,IAAI,AAA8B,YAA9B,OAAOF,oBAAiC;QAC1C,MAAMvB,SAAS,MAAMjC,mBAAmB;YACtC,QAAQwD,mBAAmB,MAAM;YACjC,yBAAyBA,mBAAmB,uBAAuB;QACrE;QACArD,KAAK,IAAI,IAAI8B;IACf;IAEA,MAAM2B,SAAS,MAAMC,AAAAA,IAAAA,yBAAAA,wBAAAA,AAAAA,EACnB1D,MACAgC,mCAAAA,YAAAA,CAAAA,YAAyB,EACzBvB;IAGF,IAAIkD;IACJ,MAAMC,cAAcH,OAAO,OAAO,CAAC,IAAI;IACvC,IAAIG,aAAa;QACf,MAAMC,aAAapB,AAAAA,IAAAA,mCAAAA,eAAAA,AAAAA,EACjBmB,aACAtD,QAAQ,IAAI,CAAC,KAAK,EAClBA,QAAQ,IAAI,CAAC,MAAM,EACnB,GACA,GACAA,QAAQ,IAAI,CAAC,KAAK,EAClBA,QAAQ,IAAI,CAAC,MAAM,EACnBI;QAEFhB,aAAa,0BAA0BmE;QAEvC,MAAMC,oBAAoBL,OAAO,OAAO,CAAC,eAAe,IAAI,EAAE;QAC9D/D,aAAa,wBAAwBoE;QAErC,MAAMC,iBAAiBD,kBACpB,MAAM,CAAC,CAACE,OAAS1B,MAAM,OAAO,CAAC0B,OAC/B,GAAG,CAAC,CAACA,OACGvB,AAAAA,IAAAA,mCAAAA,eAAAA,AAAAA,EACLuB,MACA1D,QAAQ,IAAI,CAAC,KAAK,EAClBA,QAAQ,IAAI,CAAC,MAAM,EACnB,GACA,GACAA,QAAQ,IAAI,CAAC,KAAK,EAClBA,QAAQ,IAAI,CAAC,MAAM,EACnBI;QAGNhB,aAAa,qBAAqBqE;QAGlC,MAAME,aAAaC,AAAAA,IAAAA,mCAAAA,UAAAA,AAAAA,EAAW;YAACL;eAAeE;SAAe;QAC7DrE,aAAa,iBAAiBuE;QAG9BN,cAAcQ,AAAAA,IAAAA,mCAAAA,gBAAAA,AAAAA,EAAiBF,YAAY3D,QAAQ,IAAI,EAAEI;QACzDhB,aAAa,2BAA2BiE;IAC1C;IAEA,IAAIS,cAAczD;IAClB,IAAIgD,aAAa;QACf,MAAMU,gBAAgB,MAAMC,AAAAA,IAAAA,oBAAAA,UAAAA,AAAAA,EAC1B3D,kBACAgD,aACAjD,AAAW,cAAXA;QAEF0D,cAAcC,cAAc,WAAW;QACvCV,YAAY,KAAK,GAAGU,cAAc,KAAK;QACvCV,YAAY,MAAM,GAAGU,cAAc,MAAM;IAC3C;IAEA,OAAO;QACL,MAAMV;QACNS;QACA,OAAOX,OAAO,OAAO,CAAC,KAAK;QAC3B,aAAavB,KAAK,SAAS,CAACuB,OAAO,OAAO;QAC1C,OAAOA,OAAO,KAAK;IACrB;AACF;AAEO,eAAec,qBAGpBlE,OAMD;IACC,MAAM,EAAEmE,SAAS,EAAElE,OAAO,EAAEmE,aAAa,EAAE3E,gBAAgB,EAAEW,WAAW,EAAE,GACxEJ;IACF,MAAM,EAAEK,MAAM,EAAE,GAAGD;IACnB,MAAMU,eAAeuD,AAAAA,IAAAA,8BAAAA,qBAAAA,AAAAA;IAErB,MAAM,EAAE/D,gBAAgB,EAAE,GAAGL;IAE7B,MAAM,EAAEM,WAAW,EAAEC,WAAW,EAAE,GAAG,MAAME,AAAAA,IAAAA,wBAAAA,gBAAAA,AAAAA,EAAiBT,SAAS;QACnE,oBAAoB;QACpB,sBAAsB;QACtB,aAAa;QACb,aAAamE,QAAAA,gBAAAA,KAAAA,IAAAA,cAAe,WAAW;QACvC/D;IACF;IAEA,MAAMiE,wBAAwB,MAAMC,AAAAA,IAAAA,8BAAAA,sBAAAA,AAAAA,EAClChE,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,KAAKlE;YACL,QAAQ;QACV;IACF;IAGFkE,YAAY,IAAI,CAAC;QACf,MAAM;QACN,MAAMF;IACR;IAEA,MAAM3E,OAAe;QACnB;YAAE,MAAM;YAAU,SAASmB;QAAa;QACxC;YACE,MAAM;YACN,SAAS0D;QACX;KACD;IAED,IAAI/E,kBAAkB;QACpB,MAAMgC,SAAS,MAAMjC,mBAAmB;YACtC,QAAQC,iBAAiB,MAAM;YAC/B,yBAAyBA,iBAAiB,uBAAuB;QACnE;QACAE,KAAK,IAAI,IAAI8B;IACf;IAEA,MAAM2B,SAAS,MAAMC,AAAAA,IAAAA,yBAAAA,wBAAAA,AAAAA,EACnB1D,MACAgC,mCAAAA,YAAAA,CAAAA,YAAyB,EACzBvB;IAEF,OAAO;QACL,aAAagD,OAAO,OAAO;QAC3B5C;QACA,OAAO4C,OAAO,KAAK;IACrB;AACF"}
|
|
@@ -31,17 +31,14 @@ const logger_namespaceObject = require("@midscene/shared/logger");
|
|
|
31
31
|
const utils_namespaceObject = require("@midscene/shared/utils");
|
|
32
32
|
const external_common_js_namespaceObject = require("./common.js");
|
|
33
33
|
const llm_planning_js_namespaceObject = require("./prompt/llm-planning.js");
|
|
34
|
-
const util_js_namespaceObject = require("./prompt/util.js");
|
|
35
34
|
const index_js_namespaceObject = require("./service-caller/index.js");
|
|
36
35
|
const debug = (0, logger_namespaceObject.getDebug)('planning');
|
|
37
36
|
async function plan(userInstruction, opts) {
|
|
38
37
|
var _opts_conversationHistory, _planFromAI_action;
|
|
39
38
|
const { context, modelConfig, conversationHistory } = opts;
|
|
40
39
|
const { screenshotBase64, size } = context;
|
|
41
|
-
const {
|
|
42
|
-
|
|
43
|
-
vlMode
|
|
44
|
-
});
|
|
40
|
+
const { vlMode } = modelConfig;
|
|
41
|
+
(0, utils_namespaceObject.assert)(vlMode, 'Planning requires vlMode to be configured.');
|
|
45
42
|
const systemPrompt = await (0, llm_planning_js_namespaceObject.systemPromptToTaskPlanning)({
|
|
46
43
|
actionSpace: opts.actionSpace,
|
|
47
44
|
vlMode: vlMode
|
|
@@ -56,12 +53,7 @@ async function plan(userInstruction, opts) {
|
|
|
56
53
|
imageWidth = paddedResult.width;
|
|
57
54
|
imageHeight = paddedResult.height;
|
|
58
55
|
imagePayload = paddedResult.imageBase64;
|
|
59
|
-
}
|
|
60
|
-
else if (!vlMode) imagePayload = await (0, external_common_js_namespaceObject.markupImageForLLM)(screenshotBase64, context.tree, {
|
|
61
|
-
width: imageWidth,
|
|
62
|
-
height: imageHeight
|
|
63
|
-
});
|
|
64
|
-
(0, external_common_js_namespaceObject.warnGPT4oSizeLimit)(size, modelName);
|
|
56
|
+
}
|
|
65
57
|
const historyLog = (null == (_opts_conversationHistory = opts.conversationHistory) ? void 0 : _opts_conversationHistory.snapshot()) || [];
|
|
66
58
|
const knowledgeContext = opts.actionContext ? [
|
|
67
59
|
{
|
|
@@ -102,13 +94,7 @@ async function plan(userInstruction, opts) {
|
|
|
102
94
|
url: imagePayload,
|
|
103
95
|
detail: 'high'
|
|
104
96
|
}
|
|
105
|
-
}
|
|
106
|
-
...vlMode ? [] : [
|
|
107
|
-
{
|
|
108
|
-
type: 'text',
|
|
109
|
-
text: pageDescription
|
|
110
|
-
}
|
|
111
|
-
]
|
|
97
|
+
}
|
|
112
98
|
]
|
|
113
99
|
}
|
|
114
100
|
];
|
|
@@ -134,11 +120,7 @@ async function plan(userInstruction, opts) {
|
|
|
134
120
|
debug('locateFields', locateFields);
|
|
135
121
|
locateFields.forEach((field)=>{
|
|
136
122
|
const locateResult = action.param[field];
|
|
137
|
-
if (locateResult)
|
|
138
|
-
else {
|
|
139
|
-
const element = elementById(locateResult);
|
|
140
|
-
if (element) action.param[field].id = element.id;
|
|
141
|
-
}
|
|
123
|
+
if (locateResult) action.param[field] = (0, external_common_js_namespaceObject.fillBboxParam)(locateResult, imageWidth, imageHeight, rightLimit, bottomLimit, vlMode);
|
|
142
124
|
});
|
|
143
125
|
});
|
|
144
126
|
(0, utils_namespaceObject.assert)(!planFromAI.error, `Failed to plan actions: ${planFromAI.error}`);
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"ai-model/llm-planning.js","sources":["webpack://@midscene/core/webpack/runtime/define_property_getters","webpack://@midscene/core/webpack/runtime/has_own_property","webpack://@midscene/core/webpack/runtime/make_namespace_object","webpack://@midscene/core/./src/ai-model/llm-planning.ts"],"sourcesContent":["__webpack_require__.d = (exports, definition) => {\n\tfor(var key in definition) {\n if(__webpack_require__.o(definition, key) && !__webpack_require__.o(exports, key)) {\n Object.defineProperty(exports, key, { enumerable: true, get: definition[key] });\n }\n }\n};","__webpack_require__.o = (obj, prop) => (Object.prototype.hasOwnProperty.call(obj, prop))","// define __esModule on exports\n__webpack_require__.r = (exports) => {\n\tif(typeof Symbol !== 'undefined' && Symbol.toStringTag) {\n\t\tObject.defineProperty(exports, Symbol.toStringTag, { value: 'Module' });\n\t}\n\tObject.defineProperty(exports, '__esModule', { value: true });\n};","import type {\n DeviceAction,\n InterfaceType,\n PlanningAIResponse,\n UIContext,\n} from '@/types';\nimport type { IModelConfig } from '@midscene/shared/env';\nimport { paddingToMatchBlockByBase64 } from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type {\n ChatCompletionContentPart,\n ChatCompletionMessageParam,\n} from 'openai/resources/index';\nimport {\n AIActionType,\n buildYamlFlowFromPlans,\n fillBboxParam,\n findAllMidsceneLocatorField,\n markupImageForLLM,\n warnGPT4oSizeLimit,\n} from './common';\nimport type { ConversationHistory } from './conversation-history';\nimport { systemPromptToTaskPlanning } from './prompt/llm-planning';\nimport { describeUserPage } from './prompt/util';\nimport { callAIWithObjectResponse } from './service-caller/index';\n\nconst debug = getDebug('planning');\n\nexport async function plan(\n userInstruction: string,\n opts: {\n context: UIContext;\n interfaceType: InterfaceType;\n actionSpace: DeviceAction<any>[];\n actionContext?: string;\n modelConfig: IModelConfig;\n conversationHistory?: ConversationHistory;\n },\n): Promise<PlanningAIResponse> {\n const { context, modelConfig, conversationHistory } = opts;\n const { screenshotBase64, size } = context;\n\n const { modelName, vlMode } = modelConfig;\n\n const { description: pageDescription, elementById } = await describeUserPage(\n context,\n { vlMode },\n );\n const systemPrompt = await systemPromptToTaskPlanning({\n actionSpace: opts.actionSpace,\n vlMode: vlMode,\n });\n\n let imagePayload = screenshotBase64;\n let imageWidth = size.width;\n let imageHeight = size.height;\n const rightLimit = imageWidth;\n const bottomLimit = imageHeight;\n if (vlMode === 'qwen-vl') {\n const paddedResult = await paddingToMatchBlockByBase64(imagePayload);\n imageWidth = paddedResult.width;\n imageHeight = paddedResult.height;\n imagePayload = paddedResult.imageBase64;\n } else if (vlMode === 'qwen3-vl') {\n // const paddedResult = await paddingToMatchBlockByBase64(imagePayload, 32);\n // imageWidth = paddedResult.width;\n // imageHeight = paddedResult.height;\n // imagePayload = paddedResult.imageBase64;\n } else if (!vlMode) {\n imagePayload = await markupImageForLLM(screenshotBase64, context.tree, {\n width: imageWidth,\n height: imageHeight,\n });\n }\n\n warnGPT4oSizeLimit(size, modelName);\n\n const historyLog = opts.conversationHistory?.snapshot() || [];\n // .filter((item) => item.role === 'assistant') || [];\n\n const knowledgeContext: ChatCompletionMessageParam[] = opts.actionContext\n ? [\n {\n role: 'user',\n content: [\n {\n type: 'text',\n text: `<high_priority_knowledge>${opts.actionContext}</high_priority_knowledge>`,\n },\n ],\n },\n ]\n : [];\n\n const instruction: ChatCompletionMessageParam[] = [\n {\n role: 'user',\n content: [\n {\n type: 'text',\n text: `<user_instruction>${userInstruction}</user_instruction>`,\n },\n ],\n },\n ];\n\n const msgs: ChatCompletionMessageParam[] = [\n { role: 'system', content: systemPrompt },\n ...knowledgeContext,\n ...instruction,\n ...historyLog,\n {\n role: 'user',\n content: [\n {\n type: 'image_url',\n image_url: {\n url: imagePayload,\n detail: 'high',\n },\n },\n ...(vlMode\n ? []\n : ([\n {\n type: 'text',\n text: pageDescription,\n },\n ] as ChatCompletionContentPart[])),\n ],\n },\n ];\n\n const { content, usage } = await callAIWithObjectResponse<PlanningAIResponse>(\n msgs,\n AIActionType.PLAN,\n modelConfig,\n );\n const rawResponse = JSON.stringify(content, undefined, 2);\n const planFromAI = content;\n\n const actions =\n (planFromAI.action?.type ? [planFromAI.action] : planFromAI.actions) || [];\n const returnValue: PlanningAIResponse = {\n ...planFromAI,\n actions,\n rawResponse,\n usage,\n yamlFlow: buildYamlFlowFromPlans(\n actions,\n opts.actionSpace,\n planFromAI.sleep,\n ),\n };\n\n assert(planFromAI, \"can't get plans from AI\");\n\n // TODO: use zod.parse to parse the action.param, and then fill the bbox param.\n actions.forEach((action) => {\n const type = action.type;\n const actionInActionSpace = opts.actionSpace.find(\n (action) => action.name === type,\n );\n\n debug('actionInActionSpace matched', actionInActionSpace);\n const locateFields = actionInActionSpace\n ? findAllMidsceneLocatorField(actionInActionSpace.paramSchema)\n : [];\n\n debug('locateFields', locateFields);\n\n locateFields.forEach((field) => {\n const locateResult = action.param[field];\n if (locateResult) {\n if (vlMode) {\n action.param[field] = fillBboxParam(\n locateResult,\n imageWidth,\n imageHeight,\n rightLimit,\n bottomLimit,\n vlMode,\n );\n } else {\n const element = elementById(locateResult);\n if (element) {\n action.param[field].id = element.id;\n }\n }\n }\n });\n });\n // in Qwen-VL, error means error. In GPT-4o, error may mean more actions are needed.\n assert(!planFromAI.error, `Failed to plan actions: ${planFromAI.error}`);\n\n if (\n actions.length === 0 &&\n returnValue.more_actions_needed_by_instruction &&\n !returnValue.sleep\n ) {\n console.warn(\n 'No actions planned for the prompt, but model said more actions are needed:',\n userInstruction,\n );\n }\n\n conversationHistory?.append({\n role: 'assistant',\n content: [\n {\n type: 'text',\n text: rawResponse,\n },\n ],\n });\n conversationHistory?.append({\n role: 'user',\n content: [\n {\n type: 'text',\n text: 'I have finished the action previously planned',\n },\n ],\n });\n\n return returnValue;\n}\n"],"names":["__webpack_require__","definition","key","Object","obj","prop","Symbol","debug","getDebug","plan","userInstruction","opts","_opts_conversationHistory","_planFromAI_action","context","modelConfig","conversationHistory","screenshotBase64","size","modelName","vlMode","pageDescription","elementById","describeUserPage","systemPrompt","systemPromptToTaskPlanning","imagePayload","imageWidth","imageHeight","rightLimit","bottomLimit","paddedResult","paddingToMatchBlockByBase64","markupImageForLLM","warnGPT4oSizeLimit","historyLog","knowledgeContext","instruction","msgs","content","usage","callAIWithObjectResponse","AIActionType","rawResponse","JSON","undefined","planFromAI","actions","returnValue","buildYamlFlowFromPlans","assert","action","type","actionInActionSpace","locateFields","findAllMidsceneLocatorField","field","locateResult","fillBboxParam","element","console"],"mappings":";;;IAAAA,oBAAoB,CAAC,GAAG,CAAC,UAASC;QACjC,IAAI,IAAIC,OAAOD,WACR,IAAGD,oBAAoB,CAAC,CAACC,YAAYC,QAAQ,CAACF,oBAAoB,CAAC,CAAC,UAASE,MACzEC,OAAO,cAAc,CAAC,UAASD,KAAK;YAAE,YAAY;YAAM,KAAKD,UAAU,CAACC,IAAI;QAAC;IAGzF;;;ICNAF,oBAAoB,CAAC,GAAG,CAACI,KAAKC,OAAUF,OAAO,SAAS,CAAC,cAAc,CAAC,IAAI,CAACC,KAAKC;;;ICClFL,oBAAoB,CAAC,GAAG,CAAC;QACxB,IAAG,AAAkB,eAAlB,OAAOM,UAA0BA,OAAO,WAAW,EACrDH,OAAO,cAAc,CAAC,UAASG,OAAO,WAAW,EAAE;YAAE,OAAO;QAAS;QAEtEH,OAAO,cAAc,CAAC,UAAS,cAAc;YAAE,OAAO;QAAK;IAC5D;;;;;;;;;;;;;;ACqBA,MAAMI,QAAQC,AAAAA,IAAAA,uBAAAA,QAAAA,AAAAA,EAAS;AAEhB,eAAeC,KACpBC,eAAuB,EACvBC,IAOC;QAwCkBC,2BAiEhBC;IAvGH,MAAM,EAAEC,OAAO,EAAEC,WAAW,EAAEC,mBAAmB,EAAE,GAAGL;IACtD,MAAM,EAAEM,gBAAgB,EAAEC,IAAI,EAAE,GAAGJ;IAEnC,MAAM,EAAEK,SAAS,EAAEC,MAAM,EAAE,GAAGL;IAE9B,MAAM,EAAE,aAAaM,eAAe,EAAEC,WAAW,EAAE,GAAG,MAAMC,AAAAA,IAAAA,wBAAAA,gBAAAA,AAAAA,EAC1DT,SACA;QAAEM;IAAO;IAEX,MAAMI,eAAe,MAAMC,AAAAA,IAAAA,gCAAAA,0BAAAA,AAAAA,EAA2B;QACpD,aAAad,KAAK,WAAW;QAC7B,QAAQS;IACV;IAEA,IAAIM,eAAeT;IACnB,IAAIU,aAAaT,KAAK,KAAK;IAC3B,IAAIU,cAAcV,KAAK,MAAM;IAC7B,MAAMW,aAAaF;IACnB,MAAMG,cAAcF;IACpB,IAAIR,AAAW,cAAXA,QAAsB;QACxB,MAAMW,eAAe,MAAMC,AAAAA,IAAAA,oBAAAA,2BAAAA,AAAAA,EAA4BN;QACvDC,aAAaI,aAAa,KAAK;QAC/BH,cAAcG,aAAa,MAAM;QACjCL,eAAeK,aAAa,WAAW;IACzC,OAAO,IAAIX,AAAW,eAAXA;SAKJ,IAAI,CAACA,QACVM,eAAe,MAAMO,AAAAA,IAAAA,mCAAAA,iBAAAA,AAAAA,EAAkBhB,kBAAkBH,QAAQ,IAAI,EAAE;QACrE,OAAOa;QACP,QAAQC;IACV;IAGFM,IAAAA,mCAAAA,kBAAAA,AAAAA,EAAmBhB,MAAMC;IAEzB,MAAMgB,aAAavB,AAAAA,SAAAA,CAAAA,4BAAAA,KAAK,mBAAmB,AAAD,IAAvBA,KAAAA,IAAAA,0BAA0B,QAAQ,EAAC,KAAK,EAAE;IAG7D,MAAMwB,mBAAiDzB,KAAK,aAAa,GACrE;QACE;YACE,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,MAAM,CAAC,yBAAyB,EAAEA,KAAK,aAAa,CAAC,0BAA0B,CAAC;gBAClF;aACD;QACH;KACD,GACD,EAAE;IAEN,MAAM0B,cAA4C;QAChD;YACE,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,MAAM,CAAC,kBAAkB,EAAE3B,gBAAgB,mBAAmB,CAAC;gBACjE;aACD;QACH;KACD;IAED,MAAM4B,OAAqC;QACzC;YAAE,MAAM;YAAU,SAASd;QAAa;WACrCY;WACAC;WACAF;QACH;YACE,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,WAAW;wBACT,KAAKT;wBACL,QAAQ;oBACV;gBACF;mBACIN,SACA,EAAE,GACD;oBACC;wBACE,MAAM;wBACN,MAAMC;oBACR;iBACD;aACN;QACH;KACD;IAED,MAAM,EAAEkB,OAAO,EAAEC,KAAK,EAAE,GAAG,MAAMC,AAAAA,IAAAA,yBAAAA,wBAAAA,AAAAA,EAC/BH,MACAI,mCAAAA,YAAAA,CAAAA,IAAiB,EACjB3B;IAEF,MAAM4B,cAAcC,KAAK,SAAS,CAACL,SAASM,QAAW;IACvD,MAAMC,aAAaP;IAEnB,MAAMQ,UACHlC,AAAAA,CAAAA,SAAAA,CAAAA,qBAAAA,WAAW,MAAM,AAAD,IAAhBA,KAAAA,IAAAA,mBAAmB,IAAI,AAAD,IAAI;QAACiC,WAAW,MAAM;KAAC,GAAGA,WAAW,OAAM,KAAM,EAAE;IAC5E,MAAME,cAAkC;QACtC,GAAGF,UAAU;QACbC;QACAJ;QACAH;QACA,UAAUS,AAAAA,IAAAA,mCAAAA,sBAAAA,AAAAA,EACRF,SACApC,KAAK,WAAW,EAChBmC,WAAW,KAAK;IAEpB;IAEAI,IAAAA,sBAAAA,MAAAA,AAAAA,EAAOJ,YAAY;IAGnBC,QAAQ,OAAO,CAAC,CAACI;QACf,MAAMC,OAAOD,OAAO,IAAI;QACxB,MAAME,sBAAsB1C,KAAK,WAAW,CAAC,IAAI,CAC/C,CAACwC,SAAWA,OAAO,IAAI,KAAKC;QAG9B7C,MAAM,+BAA+B8C;QACrC,MAAMC,eAAeD,sBACjBE,AAAAA,IAAAA,mCAAAA,2BAAAA,AAAAA,EAA4BF,oBAAoB,WAAW,IAC3D,EAAE;QAEN9C,MAAM,gBAAgB+C;QAEtBA,aAAa,OAAO,CAAC,CAACE;YACpB,MAAMC,eAAeN,OAAO,KAAK,CAACK,MAAM;YACxC,IAAIC,cACF,IAAIrC,QACF+B,OAAO,KAAK,CAACK,MAAM,GAAGE,AAAAA,IAAAA,mCAAAA,aAAAA,AAAAA,EACpBD,cACA9B,YACAC,aACAC,YACAC,aACAV;iBAEG;gBACL,MAAMuC,UAAUrC,YAAYmC;gBAC5B,IAAIE,SACFR,OAAO,KAAK,CAACK,MAAM,CAAC,EAAE,GAAGG,QAAQ,EAAE;YAEvC;QAEJ;IACF;IAEAT,IAAAA,sBAAAA,MAAAA,AAAAA,EAAO,CAACJ,WAAW,KAAK,EAAE,CAAC,wBAAwB,EAAEA,WAAW,KAAK,EAAE;IAEvE,IACEC,AAAmB,MAAnBA,QAAQ,MAAM,IACdC,YAAY,kCAAkC,IAC9C,CAACA,YAAY,KAAK,EAElBY,QAAQ,IAAI,CACV,8EACAlD;IAIJM,QAAAA,uBAAAA,oBAAqB,MAAM,CAAC;QAC1B,MAAM;QACN,SAAS;YACP;gBACE,MAAM;gBACN,MAAM2B;YACR;SACD;IACH;IACA3B,QAAAA,uBAAAA,oBAAqB,MAAM,CAAC;QAC1B,MAAM;QACN,SAAS;YACP;gBACE,MAAM;gBACN,MAAM;YACR;SACD;IACH;IAEA,OAAOgC;AACT"}
|
|
1
|
+
{"version":3,"file":"ai-model/llm-planning.js","sources":["webpack://@midscene/core/webpack/runtime/define_property_getters","webpack://@midscene/core/webpack/runtime/has_own_property","webpack://@midscene/core/webpack/runtime/make_namespace_object","webpack://@midscene/core/./src/ai-model/llm-planning.ts"],"sourcesContent":["__webpack_require__.d = (exports, definition) => {\n\tfor(var key in definition) {\n if(__webpack_require__.o(definition, key) && !__webpack_require__.o(exports, key)) {\n Object.defineProperty(exports, key, { enumerable: true, get: definition[key] });\n }\n }\n};","__webpack_require__.o = (obj, prop) => (Object.prototype.hasOwnProperty.call(obj, prop))","// define __esModule on exports\n__webpack_require__.r = (exports) => {\n\tif(typeof Symbol !== 'undefined' && Symbol.toStringTag) {\n\t\tObject.defineProperty(exports, Symbol.toStringTag, { value: 'Module' });\n\t}\n\tObject.defineProperty(exports, '__esModule', { value: true });\n};","import type {\n DeviceAction,\n InterfaceType,\n PlanningAIResponse,\n UIContext,\n} from '@/types';\nimport type { IModelConfig } from '@midscene/shared/env';\nimport { paddingToMatchBlockByBase64 } from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type { ChatCompletionMessageParam } from 'openai/resources/index';\nimport {\n AIActionType,\n buildYamlFlowFromPlans,\n fillBboxParam,\n findAllMidsceneLocatorField,\n} from './common';\nimport type { ConversationHistory } from './conversation-history';\nimport { systemPromptToTaskPlanning } from './prompt/llm-planning';\nimport { callAIWithObjectResponse } from './service-caller/index';\n\nconst debug = getDebug('planning');\n\nexport async function plan(\n userInstruction: string,\n opts: {\n context: UIContext;\n interfaceType: InterfaceType;\n actionSpace: DeviceAction<any>[];\n actionContext?: string;\n modelConfig: IModelConfig;\n conversationHistory?: ConversationHistory;\n },\n): Promise<PlanningAIResponse> {\n const { context, modelConfig, conversationHistory } = opts;\n const { screenshotBase64, size } = context;\n\n const { vlMode } = modelConfig;\n\n // Planning requires VL mode (validated by ModelConfigManager.getModelConfig)\n assert(vlMode, 'Planning requires vlMode to be configured.');\n\n const systemPrompt = await systemPromptToTaskPlanning({\n actionSpace: opts.actionSpace,\n vlMode: vlMode,\n });\n\n let imagePayload = screenshotBase64;\n let imageWidth = size.width;\n let imageHeight = size.height;\n const rightLimit = imageWidth;\n const bottomLimit = imageHeight;\n\n // Process image based on VL mode requirements\n if (vlMode === 'qwen-vl') {\n const paddedResult = await paddingToMatchBlockByBase64(imagePayload);\n imageWidth = paddedResult.width;\n imageHeight = paddedResult.height;\n imagePayload = paddedResult.imageBase64;\n }\n\n const historyLog = opts.conversationHistory?.snapshot() || [];\n // .filter((item) => item.role === 'assistant') || [];\n\n const knowledgeContext: ChatCompletionMessageParam[] = opts.actionContext\n ? [\n {\n role: 'user',\n content: [\n {\n type: 'text',\n text: `<high_priority_knowledge>${opts.actionContext}</high_priority_knowledge>`,\n },\n ],\n },\n ]\n : [];\n\n const instruction: ChatCompletionMessageParam[] = [\n {\n role: 'user',\n content: [\n {\n type: 'text',\n text: `<user_instruction>${userInstruction}</user_instruction>`,\n },\n ],\n },\n ];\n\n const msgs: ChatCompletionMessageParam[] = [\n { role: 'system', content: systemPrompt },\n ...knowledgeContext,\n ...instruction,\n ...historyLog,\n {\n role: 'user',\n content: [\n {\n type: 'image_url',\n image_url: {\n url: imagePayload,\n detail: 'high',\n },\n },\n // Planning uses pure vision mode, no DOM description needed\n ],\n },\n ];\n\n const { content, usage } = await callAIWithObjectResponse<PlanningAIResponse>(\n msgs,\n AIActionType.PLAN,\n modelConfig,\n );\n const rawResponse = JSON.stringify(content, undefined, 2);\n const planFromAI = content;\n\n const actions =\n (planFromAI.action?.type ? [planFromAI.action] : planFromAI.actions) || [];\n const returnValue: PlanningAIResponse = {\n ...planFromAI,\n actions,\n rawResponse,\n usage,\n yamlFlow: buildYamlFlowFromPlans(\n actions,\n opts.actionSpace,\n planFromAI.sleep,\n ),\n };\n\n assert(planFromAI, \"can't get plans from AI\");\n\n // TODO: use zod.parse to parse the action.param, and then fill the bbox param.\n actions.forEach((action) => {\n const type = action.type;\n const actionInActionSpace = opts.actionSpace.find(\n (action) => action.name === type,\n );\n\n debug('actionInActionSpace matched', actionInActionSpace);\n const locateFields = actionInActionSpace\n ? findAllMidsceneLocatorField(actionInActionSpace.paramSchema)\n : [];\n\n debug('locateFields', locateFields);\n\n locateFields.forEach((field) => {\n const locateResult = action.param[field];\n if (locateResult) {\n // Always use VL mode to fill bbox parameters\n action.param[field] = fillBboxParam(\n locateResult,\n imageWidth,\n imageHeight,\n rightLimit,\n bottomLimit,\n vlMode,\n );\n }\n });\n });\n // in Qwen-VL, error means error. In GPT-4o, error may mean more actions are needed.\n assert(!planFromAI.error, `Failed to plan actions: ${planFromAI.error}`);\n\n if (\n actions.length === 0 &&\n returnValue.more_actions_needed_by_instruction &&\n !returnValue.sleep\n ) {\n console.warn(\n 'No actions planned for the prompt, but model said more actions are needed:',\n userInstruction,\n );\n }\n\n conversationHistory?.append({\n role: 'assistant',\n content: [\n {\n type: 'text',\n text: rawResponse,\n },\n ],\n });\n conversationHistory?.append({\n role: 'user',\n content: [\n {\n type: 'text',\n text: 'I have finished the action previously planned',\n },\n ],\n });\n\n return returnValue;\n}\n"],"names":["__webpack_require__","definition","key","Object","obj","prop","Symbol","debug","getDebug","plan","userInstruction","opts","_opts_conversationHistory","_planFromAI_action","context","modelConfig","conversationHistory","screenshotBase64","size","vlMode","assert","systemPrompt","systemPromptToTaskPlanning","imagePayload","imageWidth","imageHeight","rightLimit","bottomLimit","paddedResult","paddingToMatchBlockByBase64","historyLog","knowledgeContext","instruction","msgs","content","usage","callAIWithObjectResponse","AIActionType","rawResponse","JSON","undefined","planFromAI","actions","returnValue","buildYamlFlowFromPlans","action","type","actionInActionSpace","locateFields","findAllMidsceneLocatorField","field","locateResult","fillBboxParam","console"],"mappings":";;;IAAAA,oBAAoB,CAAC,GAAG,CAAC,UAASC;QACjC,IAAI,IAAIC,OAAOD,WACR,IAAGD,oBAAoB,CAAC,CAACC,YAAYC,QAAQ,CAACF,oBAAoB,CAAC,CAAC,UAASE,MACzEC,OAAO,cAAc,CAAC,UAASD,KAAK;YAAE,YAAY;YAAM,KAAKD,UAAU,CAACC,IAAI;QAAC;IAGzF;;;ICNAF,oBAAoB,CAAC,GAAG,CAACI,KAAKC,OAAUF,OAAO,SAAS,CAAC,cAAc,CAAC,IAAI,CAACC,KAAKC;;;ICClFL,oBAAoB,CAAC,GAAG,CAAC;QACxB,IAAG,AAAkB,eAAlB,OAAOM,UAA0BA,OAAO,WAAW,EACrDH,OAAO,cAAc,CAAC,UAASG,OAAO,WAAW,EAAE;YAAE,OAAO;QAAS;QAEtEH,OAAO,cAAc,CAAC,UAAS,cAAc;YAAE,OAAO;QAAK;IAC5D;;;;;;;;;;;;;ACeA,MAAMI,QAAQC,AAAAA,IAAAA,uBAAAA,QAAAA,AAAAA,EAAS;AAEhB,eAAeC,KACpBC,eAAuB,EACvBC,IAOC;QA6BkBC,2BA0DhBC;IArFH,MAAM,EAAEC,OAAO,EAAEC,WAAW,EAAEC,mBAAmB,EAAE,GAAGL;IACtD,MAAM,EAAEM,gBAAgB,EAAEC,IAAI,EAAE,GAAGJ;IAEnC,MAAM,EAAEK,MAAM,EAAE,GAAGJ;IAGnBK,IAAAA,sBAAAA,MAAAA,AAAAA,EAAOD,QAAQ;IAEf,MAAME,eAAe,MAAMC,AAAAA,IAAAA,gCAAAA,0BAAAA,AAAAA,EAA2B;QACpD,aAAaX,KAAK,WAAW;QAC7B,QAAQQ;IACV;IAEA,IAAII,eAAeN;IACnB,IAAIO,aAAaN,KAAK,KAAK;IAC3B,IAAIO,cAAcP,KAAK,MAAM;IAC7B,MAAMQ,aAAaF;IACnB,MAAMG,cAAcF;IAGpB,IAAIN,AAAW,cAAXA,QAAsB;QACxB,MAAMS,eAAe,MAAMC,AAAAA,IAAAA,oBAAAA,2BAAAA,AAAAA,EAA4BN;QACvDC,aAAaI,aAAa,KAAK;QAC/BH,cAAcG,aAAa,MAAM;QACjCL,eAAeK,aAAa,WAAW;IACzC;IAEA,MAAME,aAAalB,AAAAA,SAAAA,CAAAA,4BAAAA,KAAK,mBAAmB,AAAD,IAAvBA,KAAAA,IAAAA,0BAA0B,QAAQ,EAAC,KAAK,EAAE;IAG7D,MAAMmB,mBAAiDpB,KAAK,aAAa,GACrE;QACE;YACE,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,MAAM,CAAC,yBAAyB,EAAEA,KAAK,aAAa,CAAC,0BAA0B,CAAC;gBAClF;aACD;QACH;KACD,GACD,EAAE;IAEN,MAAMqB,cAA4C;QAChD;YACE,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,MAAM,CAAC,kBAAkB,EAAEtB,gBAAgB,mBAAmB,CAAC;gBACjE;aACD;QACH;KACD;IAED,MAAMuB,OAAqC;QACzC;YAAE,MAAM;YAAU,SAASZ;QAAa;WACrCU;WACAC;WACAF;QACH;YACE,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,WAAW;wBACT,KAAKP;wBACL,QAAQ;oBACV;gBACF;aAED;QACH;KACD;IAED,MAAM,EAAEW,OAAO,EAAEC,KAAK,EAAE,GAAG,MAAMC,AAAAA,IAAAA,yBAAAA,wBAAAA,AAAAA,EAC/BH,MACAI,mCAAAA,YAAAA,CAAAA,IAAiB,EACjBtB;IAEF,MAAMuB,cAAcC,KAAK,SAAS,CAACL,SAASM,QAAW;IACvD,MAAMC,aAAaP;IAEnB,MAAMQ,UACH7B,AAAAA,CAAAA,SAAAA,CAAAA,qBAAAA,WAAW,MAAM,AAAD,IAAhBA,KAAAA,IAAAA,mBAAmB,IAAI,AAAD,IAAI;QAAC4B,WAAW,MAAM;KAAC,GAAGA,WAAW,OAAM,KAAM,EAAE;IAC5E,MAAME,cAAkC;QACtC,GAAGF,UAAU;QACbC;QACAJ;QACAH;QACA,UAAUS,AAAAA,IAAAA,mCAAAA,sBAAAA,AAAAA,EACRF,SACA/B,KAAK,WAAW,EAChB8B,WAAW,KAAK;IAEpB;IAEArB,IAAAA,sBAAAA,MAAAA,AAAAA,EAAOqB,YAAY;IAGnBC,QAAQ,OAAO,CAAC,CAACG;QACf,MAAMC,OAAOD,OAAO,IAAI;QACxB,MAAME,sBAAsBpC,KAAK,WAAW,CAAC,IAAI,CAC/C,CAACkC,SAAWA,OAAO,IAAI,KAAKC;QAG9BvC,MAAM,+BAA+BwC;QACrC,MAAMC,eAAeD,sBACjBE,AAAAA,IAAAA,mCAAAA,2BAAAA,AAAAA,EAA4BF,oBAAoB,WAAW,IAC3D,EAAE;QAENxC,MAAM,gBAAgByC;QAEtBA,aAAa,OAAO,CAAC,CAACE;YACpB,MAAMC,eAAeN,OAAO,KAAK,CAACK,MAAM;YACxC,IAAIC,cAEFN,OAAO,KAAK,CAACK,MAAM,GAAGE,AAAAA,IAAAA,mCAAAA,aAAAA,AAAAA,EACpBD,cACA3B,YACAC,aACAC,YACAC,aACAR;QAGN;IACF;IAEAC,IAAAA,sBAAAA,MAAAA,AAAAA,EAAO,CAACqB,WAAW,KAAK,EAAE,CAAC,wBAAwB,EAAEA,WAAW,KAAK,EAAE;IAEvE,IACEC,AAAmB,MAAnBA,QAAQ,MAAM,IACdC,YAAY,kCAAkC,IAC9C,CAACA,YAAY,KAAK,EAElBU,QAAQ,IAAI,CACV,8EACA3C;IAIJM,QAAAA,uBAAAA,oBAAqB,MAAM,CAAC;QAC1B,MAAM;QACN,SAAS;YACP;gBACE,MAAM;gBACN,MAAMsB;YACR;SACD;IACH;IACAtB,QAAAA,uBAAAA,oBAAqB,MAAM,CAAC;QAC1B,MAAM;QACN,SAAS;YACP;gBACE,MAAM;gBACN,MAAM;YACR;SACD;IACH;IAEA,OAAO2B;AACT"}
|