@midscene/core 0.26.7-beta-20250818034910.0 → 0.26.7-beta-20250818035341.0
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/ai-model/common.mjs +37 -16
- package/dist/es/ai-model/common.mjs.map +1 -1
- package/dist/es/ai-model/index.mjs +3 -3
- package/dist/es/ai-model/inspect.mjs +51 -2
- package/dist/es/ai-model/inspect.mjs.map +1 -1
- package/dist/es/ai-model/llm-planning.mjs +1 -1
- package/dist/es/ai-model/llm-planning.mjs.map +1 -1
- package/dist/es/ai-model/prompt/assertion.mjs +25 -1
- package/dist/es/ai-model/prompt/assertion.mjs.map +1 -1
- package/dist/es/index.mjs +2 -2
- package/dist/es/index.mjs.map +1 -1
- package/dist/es/insight/index.mjs +36 -1
- package/dist/es/insight/index.mjs.map +1 -1
- package/dist/es/types.mjs.map +1 -1
- package/dist/es/utils.mjs +2 -2
- package/dist/lib/ai-model/common.js +36 -18
- package/dist/lib/ai-model/common.js.map +1 -1
- package/dist/lib/ai-model/index.js +3 -3
- package/dist/lib/ai-model/inspect.js +54 -2
- package/dist/lib/ai-model/inspect.js.map +1 -1
- package/dist/lib/ai-model/llm-planning.js +1 -1
- package/dist/lib/ai-model/llm-planning.js.map +1 -1
- package/dist/lib/ai-model/prompt/assertion.js +29 -2
- package/dist/lib/ai-model/prompt/assertion.js.map +1 -1
- package/dist/lib/index.js +3 -0
- package/dist/lib/index.js.map +1 -1
- package/dist/lib/insight/index.js +35 -0
- package/dist/lib/insight/index.js.map +1 -1
- package/dist/lib/types.js.map +1 -1
- package/dist/lib/utils.js +2 -2
- package/dist/types/ai-model/common.d.ts +2 -3
- package/dist/types/ai-model/index.d.ts +1 -2
- package/dist/types/ai-model/inspect.d.ts +8 -1
- package/dist/types/ai-model/prompt/assertion.d.ts +3 -0
- package/dist/types/index.d.ts +1 -1
- package/dist/types/insight/index.d.ts +2 -1
- package/dist/types/types.d.ts +0 -1
- package/dist/types/yaml.d.ts +6 -6
- package/package.json +3 -3
|
@@ -24,7 +24,6 @@ var __webpack_require__ = {};
|
|
|
24
24
|
var __webpack_exports__ = {};
|
|
25
25
|
__webpack_require__.r(__webpack_exports__);
|
|
26
26
|
__webpack_require__.d(__webpack_exports__, {
|
|
27
|
-
actionSpaceTypePrefix: ()=>actionSpaceTypePrefix,
|
|
28
27
|
markupImageForLLM: ()=>markupImageForLLM,
|
|
29
28
|
adaptBboxToRect: ()=>adaptBboxToRect,
|
|
30
29
|
expandSearchArea: ()=>expandSearchArea,
|
|
@@ -54,7 +53,6 @@ var common_AIActionType = /*#__PURE__*/ function(AIActionType) {
|
|
|
54
53
|
AIActionType[AIActionType["DESCRIBE_ELEMENT"] = 4] = "DESCRIBE_ELEMENT";
|
|
55
54
|
return AIActionType;
|
|
56
55
|
}({});
|
|
57
|
-
const actionSpaceTypePrefix = 'action_space_';
|
|
58
56
|
async function callAiFn(msgs, AIActionTypeValue) {
|
|
59
57
|
const jsonObject = await (0, index_js_namespaceObject.callToGetJSONObject)(msgs, AIActionTypeValue);
|
|
60
58
|
return {
|
|
@@ -211,31 +209,52 @@ async function markupImageForLLM(screenshotBase64, tree, size) {
|
|
|
211
209
|
});
|
|
212
210
|
return imagePayload;
|
|
213
211
|
}
|
|
214
|
-
function buildYamlFlowFromPlans(plans,
|
|
212
|
+
function buildYamlFlowFromPlans(plans, sleep) {
|
|
215
213
|
const flow = [];
|
|
216
214
|
for (const plan of plans){
|
|
217
215
|
var _plan_locate;
|
|
218
|
-
const
|
|
219
|
-
const action = actionSpace.find((action)=>action.name === verb);
|
|
220
|
-
if (!action) {
|
|
221
|
-
console.warn(`Cannot convert action ${verb} to yaml flow. Will ignore it.`);
|
|
222
|
-
continue;
|
|
223
|
-
}
|
|
216
|
+
const type = plan.type;
|
|
224
217
|
const locate = null == (_plan_locate = plan.locate) ? void 0 : _plan_locate.prompt;
|
|
225
|
-
|
|
226
|
-
|
|
227
|
-
|
|
228
|
-
|
|
229
|
-
|
|
230
|
-
|
|
218
|
+
if ('Tap' === type) flow.push({
|
|
219
|
+
aiTap: locate
|
|
220
|
+
});
|
|
221
|
+
else if ('Hover' === type) flow.push({
|
|
222
|
+
aiHover: locate
|
|
223
|
+
});
|
|
224
|
+
else if ('Input' === type) {
|
|
225
|
+
const param = plan.param;
|
|
226
|
+
flow.push({
|
|
227
|
+
aiInput: param.value,
|
|
228
|
+
locate
|
|
229
|
+
});
|
|
230
|
+
} else if ('KeyboardPress' === type) {
|
|
231
|
+
const param = plan.param;
|
|
232
|
+
flow.push({
|
|
233
|
+
aiKeyboardPress: param.value,
|
|
234
|
+
locate
|
|
235
|
+
});
|
|
236
|
+
} else if ('Scroll' === type) {
|
|
237
|
+
const param = plan.param;
|
|
238
|
+
flow.push({
|
|
239
|
+
aiScroll: null,
|
|
240
|
+
locate,
|
|
241
|
+
direction: param.direction,
|
|
242
|
+
scrollType: param.scrollType,
|
|
243
|
+
distance: param.distance
|
|
244
|
+
});
|
|
245
|
+
} else if ('Sleep' === type) {
|
|
246
|
+
const param = plan.param;
|
|
247
|
+
flow.push({
|
|
248
|
+
sleep: param.timeMs
|
|
249
|
+
});
|
|
250
|
+
} else 'AndroidBackButton' === type || 'AndroidHomeButton' === type || 'AndroidRecentAppsButton' === type || 'AndroidLongPress' === type || 'AndroidPull' === type || 'Error' === type || 'Assert' === type || 'AssertWithoutThrow' === type || 'Finished' === type || console.warn(`Cannot convert action ${type} to yaml flow. This should be a bug of Midscene.`);
|
|
231
251
|
}
|
|
232
252
|
if (sleep) flow.push({
|
|
233
|
-
sleep
|
|
253
|
+
sleep: sleep
|
|
234
254
|
});
|
|
235
255
|
return flow;
|
|
236
256
|
}
|
|
237
257
|
exports.AIActionType = __webpack_exports__.AIActionType;
|
|
238
|
-
exports.actionSpaceTypePrefix = __webpack_exports__.actionSpaceTypePrefix;
|
|
239
258
|
exports.adaptBbox = __webpack_exports__.adaptBbox;
|
|
240
259
|
exports.adaptBboxToRect = __webpack_exports__.adaptBboxToRect;
|
|
241
260
|
exports.adaptDoubaoBbox = __webpack_exports__.adaptDoubaoBbox;
|
|
@@ -250,7 +269,6 @@ exports.mergeRects = __webpack_exports__.mergeRects;
|
|
|
250
269
|
exports.warnGPT4oSizeLimit = __webpack_exports__.warnGPT4oSizeLimit;
|
|
251
270
|
for(var __webpack_i__ in __webpack_exports__)if (-1 === [
|
|
252
271
|
"AIActionType",
|
|
253
|
-
"actionSpaceTypePrefix",
|
|
254
272
|
"adaptBbox",
|
|
255
273
|
"adaptBboxToRect",
|
|
256
274
|
"adaptDoubaoBbox",
|
|
@@ -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 {\n ChatCompletionSystemMessageParam,\n ChatCompletionUserMessageParam,\n} from 'openai/resources/index';\nimport {\n call,\n callToGetJSONObject,\n getModelName,\n} from './service-caller/index';\n\nimport type { PlanningLocateParam } from '@/types';\nimport { NodeType } from '@midscene/shared/constants';\nimport { vlLocateMode } from '@midscene/shared/env';\nimport { treeToList } from '@midscene/shared/extractor';\nimport { compositeElementInfoImg } from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\n\nexport type AIArgs = [\n ChatCompletionSystemMessageParam,\n ...ChatCompletionUserMessageParam[],\n];\n\nexport enum AIActionType {\n ASSERT = 0,\n INSPECT_ELEMENT = 1,\n EXTRACT_DATA = 2,\n PLAN = 3,\n DESCRIBE_ELEMENT = 4,\n}\n\nexport const actionSpaceTypePrefix = 'action_space_';\n\nexport async function callAiFn<T>(\n msgs: AIArgs,\n AIActionTypeValue: AIActionType,\n): Promise<{ content: T; usage?: AIUsageInfo }> {\n const jsonObject = await callToGetJSONObject<T>(msgs, AIActionTypeValue);\n\n return {\n content: jsonObject.content,\n usage: jsonObject.usage,\n };\n}\n\nconst defaultBboxSize = 20; // must be even number\nconst debugInspectUtils = getDebug('ai:common');\n\n// transform the param of locate from qwen mode\nexport function fillBboxParam(\n locate: PlanningLocateParam,\n width: number,\n height: number,\n) {\n // The Qwen model might have hallucinations of naming bbox as bbox_2d.\n if ((locate as any).bbox_2d && !locate?.bbox) {\n locate.bbox = (locate as any).bbox_2d;\n // biome-ignore lint/performance/noDelete: <explanation>\n delete (locate as any).bbox_2d;\n }\n\n if (locate?.bbox) {\n locate.bbox = adaptBbox(locate.bbox, width, height);\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): [number, number, number, number] {\n if (vlLocateMode() === 'doubao-vision' || vlLocateMode() === 'vlm-ui-tars') {\n return adaptDoubaoBbox(bbox, width, height);\n }\n\n if (vlLocateMode() === 'gemini') {\n return adaptGeminiBbox(bbox, width, height);\n }\n\n return adaptQwenBbox(bbox);\n}\n\nexport function adaptGeminiBbox(\n bbox: number[],\n width: number,\n height: number,\n): [number, number, number, number] {\n const left = Math.round((bbox[1] * width) / 1000);\n const top = Math.round((bbox[0] * height) / 1000);\n const right = Math.round((bbox[3] * width) / 1000);\n const bottom = Math.round((bbox[2] * height) / 1000);\n return [left, top, right, bottom];\n}\n\nexport function adaptBboxToRect(\n bbox: number[],\n width: number,\n height: number,\n offsetX = 0,\n offsetY = 0,\n): Rect {\n debugInspectUtils('adaptBboxToRect', bbox, width, height, offsetX, offsetY);\n const [left, top, right, bottom] = adaptBbox(bbox, width, height);\n const rect = {\n left: left + offsetX,\n top: top + offsetY,\n width: right - left,\n height: bottom - top,\n };\n debugInspectUtils('adaptBboxToRect, result=', rect);\n return rect;\n}\n\nlet warned = false;\nexport function warnGPT4oSizeLimit(size: Size) {\n if (warned) return;\n if (getModelName()?.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 page 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(rect: Rect, screenSize: Size) {\n const minEdgeSize = vlLocateMode() === 'doubao-vision' ? 500 : 300;\n const defaultPadding = 160;\n\n const paddingSizeHorizontal =\n rect.width < minEdgeSize\n ? Math.ceil((minEdgeSize - rect.width) / 2)\n : defaultPadding;\n const paddingSizeVertical =\n rect.height < minEdgeSize\n ? Math.ceil((minEdgeSize - rect.height) / 2)\n : defaultPadding;\n rect.left = Math.max(0, rect.left - paddingSizeHorizontal);\n rect.width = Math.min(\n rect.width + paddingSizeHorizontal * 2,\n screenSize.width - rect.left,\n );\n rect.top = Math.max(0, rect.top - paddingSizeVertical);\n rect.height = Math.min(\n rect.height + paddingSizeVertical * 2,\n screenSize.height - rect.top,\n );\n return rect;\n}\n\nexport async function markupImageForLLM(\n screenshotBase64: string,\n tree: ElementTreeNode<BaseElement>,\n size: Size,\n) {\n const elementsInfo = treeToList(tree);\n const elementsPositionInfoWithoutText = elementsInfo!.filter(\n (elementInfo) => {\n if (elementInfo.attributes.nodeType === NodeType.TEXT) {\n return false;\n }\n return true;\n },\n );\n\n const imagePayload = await compositeElementInfoImg({\n inputImgBase64: screenshotBase64,\n elementsPositionInfo: elementsPositionInfoWithoutText,\n size,\n });\n return imagePayload;\n}\n\nexport function buildYamlFlowFromPlans(\n plans: PlanningAction[],\n actionSpace: DeviceAction[],\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 ${verb} to yaml flow. Will ignore it.`,\n );\n continue;\n }\n\n const locate = plan.locate?.prompt;\n const flowKey = action.interfaceAlias || `${actionSpaceTypePrefix}${verb}`;\n\n const flowItem: MidsceneYamlFlowItem = {\n [flowKey]: locate || '',\n ...(plan.param || {}),\n };\n\n flow.push(flowItem);\n }\n\n if (sleep) {\n flow.push({\n sleep,\n });\n }\n\n return flow;\n}\n"],"names":["__webpack_require__","definition","key","Object","obj","prop","Symbol","AIActionType","actionSpaceTypePrefix","callAiFn","msgs","AIActionTypeValue","jsonObject","callToGetJSONObject","defaultBboxSize","debugInspectUtils","getDebug","fillBboxParam","locate","width","height","adaptBbox","adaptQwenBbox","bbox","msg","JSON","Error","result","Math","adaptDoubaoBbox","assert","splitted","Number","Array","bboxList","item","x","y","vlLocateMode","adaptGeminiBbox","left","top","right","bottom","adaptBboxToRect","offsetX","offsetY","rect","warned","warnGPT4oSizeLimit","size","_getModelName","warningMsg","console","mergeRects","rects","minLeft","r","minTop","maxRight","maxBottom","expandSearchArea","screenSize","minEdgeSize","defaultPadding","paddingSizeHorizontal","paddingSizeVertical","markupImageForLLM","screenshotBase64","tree","elementsInfo","treeToList","elementsPositionInfoWithoutText","elementInfo","NodeType","imagePayload","compositeElementInfoImg","buildYamlFlowFromPlans","plans","actionSpace","sleep","flow","plan","_plan_locate","verb","action","flowKey","flowItem"],"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;;;;;;;;;;;;;;;;;;;;;;;;;;;AC4BO,IAAKI,sBAAYA,WAAAA,GAAAA,SAAZA,YAAY;;;;;;WAAZA;;AAQL,MAAMC,wBAAwB;AAE9B,eAAeC,SACpBC,IAAY,EACZC,iBAA+B;IAE/B,MAAMC,aAAa,MAAMC,AAAAA,IAAAA,yBAAAA,mBAAAA,AAAAA,EAAuBH,MAAMC;IAEtD,OAAO;QACL,SAASC,WAAW,OAAO;QAC3B,OAAOA,WAAW,KAAK;IACzB;AACF;AAEA,MAAME,kBAAkB;AACxB,MAAMC,oBAAoBC,AAAAA,IAAAA,uBAAAA,QAAAA,AAAAA,EAAS;AAG5B,SAASC,cACdC,MAA2B,EAC3BC,KAAa,EACbC,MAAc;IAGd,IAAKF,OAAe,OAAO,IAAI,CAACA,CAAAA,QAAAA,SAAAA,KAAAA,IAAAA,OAAQ,IAAI,AAAD,GAAG;QAC5CA,OAAO,IAAI,GAAIA,OAAe,OAAO;QAErC,OAAQA,OAAe,OAAO;IAChC;IAEA,IAAIA,QAAAA,SAAAA,KAAAA,IAAAA,OAAQ,IAAI,EACdA,OAAO,IAAI,GAAGG,UAAUH,OAAO,IAAI,EAAEC,OAAOC;IAG9C,OAAOF;AACT;AAEO,SAASI,cACdC,IAAc;IAEd,IAAIA,KAAK,MAAM,GAAG,GAAG;QACnB,MAAMC,MAAM,CAAC,oCAAoC,EAAEC,KAAK,SAAS,CAACF,MAAM,CAAC,CAAC;QAC1E,MAAM,IAAIG,MAAMF;IAClB;IAEA,MAAMG,SAA2C;QAC/CC,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE;QAClBK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE;QACC,YAAnB,OAAOA,IAAI,CAAC,EAAE,GACVK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,IAClBK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,GAAGT;QACN,YAAnB,OAAOS,IAAI,CAAC,EAAE,GACVK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,IAClBK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,GAAGT;KAC1B;IACD,OAAOa;AACT;AAEO,SAASE,gBACdN,IAAkC,EAClCJ,KAAa,EACbC,MAAc;IAEdU,IAAAA,sBAAAA,MAAAA,AAAAA,EACEX,QAAQ,KAAKC,SAAS,GACtB;IAGF,IAAI,AAAgB,YAAhB,OAAOG,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,IAAIZ,QAAS;YAC3CS,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAIX,SAAU;YAC5CQ,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAIZ,QAAS;YAC3CS,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAIX,SAAU;SAC7C;QAEH,MAAM,IAAIM,MAAM,CAAC,iDAAiD,EAAEH,MAAM;IAC5E;IAEA,IAAIU,MAAM,OAAO,CAACV,SAASU,MAAM,OAAO,CAACV,IAAI,CAAC,EAAE,GAC9CA,OAAOA,IAAI,CAAC,EAAE;IAGhB,IAAIW,WAAqB,EAAE;IAC3B,IAAID,MAAM,OAAO,CAACV,SAAS,AAAmB,YAAnB,OAAOA,IAAI,CAAC,EAAE,EACvCA,KAAK,OAAO,CAAC,CAACY;QACZ,IAAI,AAAgB,YAAhB,OAAOA,QAAqBA,KAAK,QAAQ,CAAC,MAAM;YAClD,MAAM,CAACC,GAAGC,EAAE,GAAGF,KAAK,KAAK,CAAC;YAC1BD,SAAS,IAAI,CAACF,OAAOI,EAAE,IAAI,KAAKJ,OAAOK,EAAE,IAAI;QAC/C,OAAO,IAAI,AAAgB,YAAhB,OAAOF,QAAqBA,KAAK,QAAQ,CAAC,MAAM;YACzD,MAAM,CAACC,GAAGC,EAAE,GAAGF,KAAK,KAAK,CAAC;YAC1BD,SAAS,IAAI,CAACF,OAAOI,EAAE,IAAI,KAAKJ,OAAOK,EAAE,IAAI;QAC/C,OACEH,SAAS,IAAI,CAACF,OAAOG;IAEzB;SAEAD,WAAWX;IAGb,IAAIW,AAAoB,MAApBA,SAAS,MAAM,IAAUA,AAAoB,MAApBA,SAAS,MAAM,EAC1C,OAAO;QACLN,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,QAAS;QACnCS,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGd,SAAU;QACpCQ,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,QAAS;QACnCS,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGd,SAAU;KACrC;IAIH,IACEc,AAAoB,MAApBA,SAAS,MAAM,IACfA,AAAoB,MAApBA,SAAS,MAAM,IACfA,AAAoB,MAApBA,SAAS,MAAM,IACfA,AAAoB,MAApBA,SAAS,MAAM,EAEf,OAAO;QACLN,KAAK,GAAG,CACN,GACAA,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,QAAS,QAAQL,kBAAkB;QAE/Dc,KAAK,GAAG,CACN,GACAA,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGd,SAAU,QAAQN,kBAAkB;QAEhEc,KAAK,GAAG,CACNT,OACAS,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,QAAS,QAAQL,kBAAkB;QAE/Dc,KAAK,GAAG,CACNR,QACAQ,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGd,SAAU,QAAQN,kBAAkB;KAEjE;IAGH,IAAIS,AAAgB,MAAhBA,KAAK,MAAM,EACb,OAAO;QACLK,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,QAAS;QACnCS,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGd,SAAU;QACpCQ,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,QAAS;QACnCS,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGd,SAAU;KACrC;IAGH,MAAMI,MAAM,CAAC,0CAA0C,EAAEC,KAAK,SAAS,CAACF,MAAM,CAAC,CAAC;IAChF,MAAM,IAAIG,MAAMF;AAClB;AAEO,SAASH,UACdE,IAAc,EACdJ,KAAa,EACbC,MAAc;IAEd,IAAIkB,AAAmB,oBAAnBA,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,OAAsCA,AAAmB,kBAAnBA,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,KACxC,OAAOT,gBAAgBN,MAAMJ,OAAOC;IAGtC,IAAIkB,AAAmB,aAAnBA,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,KACF,OAAOC,gBAAgBhB,MAAMJ,OAAOC;IAGtC,OAAOE,cAAcC;AACvB;AAEO,SAASgB,gBACdhB,IAAc,EACdJ,KAAa,EACbC,MAAc;IAEd,MAAMoB,OAAOZ,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGJ,QAAS;IAC5C,MAAMsB,MAAMb,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGH,SAAU;IAC5C,MAAMsB,QAAQd,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGJ,QAAS;IAC7C,MAAMwB,SAASf,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGH,SAAU;IAC/C,OAAO;QAACoB;QAAMC;QAAKC;QAAOC;KAAO;AACnC;AAEO,SAASC,gBACdrB,IAAc,EACdJ,KAAa,EACbC,MAAc,EACdyB,UAAU,CAAC,EACXC,UAAU,CAAC;IAEX/B,kBAAkB,mBAAmBQ,MAAMJ,OAAOC,QAAQyB,SAASC;IACnE,MAAM,CAACN,MAAMC,KAAKC,OAAOC,OAAO,GAAGtB,UAAUE,MAAMJ,OAAOC;IAC1D,MAAM2B,OAAO;QACX,MAAMP,OAAOK;QACb,KAAKJ,MAAMK;QACX,OAAOJ,QAAQF;QACf,QAAQG,SAASF;IACnB;IACA1B,kBAAkB,4BAA4BgC;IAC9C,OAAOA;AACT;AAEA,IAAIC,SAAS;AACN,SAASC,mBAAmBC,IAAU;QAEvCC;IADJ,IAAIH,QAAQ;IACZ,IAAI,QAAAG,CAAAA,gBAAAA,AAAAA,IAAAA,yBAAAA,YAAAA,AAAAA,GAAa,IAAbA,KAAAA,IAAAA,cAAgB,WAAW,GAAG,QAAQ,CAAC,WAAW;QACpD,MAAMC,aAAa,CAAC,uEAAuE,EAAEF,KAAK,KAAK,CAAC,CAAC,EAAEA,KAAK,MAAM,CAAC,wFAAwF,CAAC;QAEhN,IACEtB,KAAK,GAAG,CAACsB,KAAK,KAAK,EAAEA,KAAK,MAAM,IAAI,QACpCtB,KAAK,GAAG,CAACsB,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,UAAU5B,KAAK,GAAG,IAAI2B,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,IAAI;IACnD,MAAMC,SAAS9B,KAAK,GAAG,IAAI2B,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,GAAG;IACjD,MAAME,WAAW/B,KAAK,GAAG,IAAI2B,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,IAAI,GAAGA,EAAE,KAAK;IAC9D,MAAMG,YAAYhC,KAAK,GAAG,IAAI2B,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,iBAAiBd,IAAU,EAAEe,UAAgB;IAC3D,MAAMC,cAAczB,AAAmB,oBAAnBA,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,MAAqC,MAAM;IAC/D,MAAM0B,iBAAiB;IAEvB,MAAMC,wBACJlB,KAAK,KAAK,GAAGgB,cACTnC,KAAK,IAAI,CAAEmC,AAAAA,CAAAA,cAAchB,KAAK,KAAI,IAAK,KACvCiB;IACN,MAAME,sBACJnB,KAAK,MAAM,GAAGgB,cACVnC,KAAK,IAAI,CAAEmC,AAAAA,CAAAA,cAAchB,KAAK,MAAK,IAAK,KACxCiB;IACNjB,KAAK,IAAI,GAAGnB,KAAK,GAAG,CAAC,GAAGmB,KAAK,IAAI,GAAGkB;IACpClB,KAAK,KAAK,GAAGnB,KAAK,GAAG,CACnBmB,KAAK,KAAK,GAAGkB,AAAwB,IAAxBA,uBACbH,WAAW,KAAK,GAAGf,KAAK,IAAI;IAE9BA,KAAK,GAAG,GAAGnB,KAAK,GAAG,CAAC,GAAGmB,KAAK,GAAG,GAAGmB;IAClCnB,KAAK,MAAM,GAAGnB,KAAK,GAAG,CACpBmB,KAAK,MAAM,GAAGmB,AAAsB,IAAtBA,qBACdJ,WAAW,MAAM,GAAGf,KAAK,GAAG;IAE9B,OAAOA;AACT;AAEO,eAAeoB,kBACpBC,gBAAwB,EACxBC,IAAkC,EAClCnB,IAAU;IAEV,MAAMoB,eAAeC,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;QACtBtB;IACF;IACA,OAAOyB;AACT;AAEO,SAASE,uBACdC,KAAuB,EACvBC,WAA2B,EAC3BC,KAAc;IAEd,MAAMC,OAA+B,EAAE;IAEvC,KAAK,MAAMC,QAAQJ,MAAO;YAWTK;QAVf,MAAMC,OAAOF,KAAK,IAAI;QAEtB,MAAMG,SAASN,YAAY,IAAI,CAAC,CAACM,SAAWA,OAAO,IAAI,KAAKD;QAC5D,IAAI,CAACC,QAAQ;YACXhC,QAAQ,IAAI,CACV,CAAC,sBAAsB,EAAE+B,KAAK,8BAA8B,CAAC;YAE/D;QACF;QAEA,MAAMlE,SAAS,QAAAiE,CAAAA,eAAAA,KAAK,MAAM,AAAD,IAAVA,KAAAA,IAAAA,aAAa,MAAM;QAClC,MAAMG,UAAUD,OAAO,cAAc,IAAI,GAAG7E,wBAAwB4E,MAAM;QAE1E,MAAMG,WAAiC;YACrC,CAACD,QAAQ,EAAEpE,UAAU;YACrB,GAAIgE,KAAK,KAAK,IAAI,CAAC,CAAC;QACtB;QAEAD,KAAK,IAAI,CAACM;IACZ;IAEA,IAAIP,OACFC,KAAK,IAAI,CAAC;QACRD;IACF;IAGF,OAAOC;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 ElementTreeNode,\n MidsceneYamlFlowItem,\n PlanningAction,\n PlanningActionParamInputOrKeyPress,\n PlanningActionParamSleep,\n Rect,\n ScrollParam,\n Size,\n} from '@/types';\nimport { assert } from '@midscene/shared/utils';\n\nimport type {\n ChatCompletionSystemMessageParam,\n ChatCompletionUserMessageParam,\n} from 'openai/resources/index';\nimport {\n call,\n callToGetJSONObject,\n getModelName,\n} from './service-caller/index';\n\nimport type { PlanningLocateParam } from '@/types';\nimport { NodeType } from '@midscene/shared/constants';\nimport { vlLocateMode } from '@midscene/shared/env';\nimport { treeToList } from '@midscene/shared/extractor';\nimport { compositeElementInfoImg } from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\n\nexport type AIArgs = [\n ChatCompletionSystemMessageParam,\n ...ChatCompletionUserMessageParam[],\n];\n\nexport enum AIActionType {\n ASSERT = 0,\n INSPECT_ELEMENT = 1,\n EXTRACT_DATA = 2,\n PLAN = 3,\n DESCRIBE_ELEMENT = 4,\n}\n\nexport async function callAiFn<T>(\n msgs: AIArgs,\n AIActionTypeValue: AIActionType,\n): Promise<{ content: T; usage?: AIUsageInfo }> {\n const jsonObject = await callToGetJSONObject<T>(msgs, AIActionTypeValue);\n\n return {\n content: jsonObject.content,\n usage: jsonObject.usage,\n };\n}\n\nconst defaultBboxSize = 20; // must be even number\nconst debugInspectUtils = getDebug('ai:common');\n\n// transform the param of locate from qwen mode\nexport function fillBboxParam(\n locate: PlanningLocateParam,\n width: number,\n height: number,\n) {\n // The Qwen model might have hallucinations of naming bbox as bbox_2d.\n if ((locate as any).bbox_2d && !locate?.bbox) {\n locate.bbox = (locate as any).bbox_2d;\n // biome-ignore lint/performance/noDelete: <explanation>\n delete (locate as any).bbox_2d;\n }\n\n if (locate?.bbox) {\n locate.bbox = adaptBbox(locate.bbox, width, height);\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): [number, number, number, number] {\n if (vlLocateMode() === 'doubao-vision' || vlLocateMode() === 'vlm-ui-tars') {\n return adaptDoubaoBbox(bbox, width, height);\n }\n\n if (vlLocateMode() === 'gemini') {\n return adaptGeminiBbox(bbox, width, height);\n }\n\n return adaptQwenBbox(bbox);\n}\n\nexport function adaptGeminiBbox(\n bbox: number[],\n width: number,\n height: number,\n): [number, number, number, number] {\n const left = Math.round((bbox[1] * width) / 1000);\n const top = Math.round((bbox[0] * height) / 1000);\n const right = Math.round((bbox[3] * width) / 1000);\n const bottom = Math.round((bbox[2] * height) / 1000);\n return [left, top, right, bottom];\n}\n\nexport function adaptBboxToRect(\n bbox: number[],\n width: number,\n height: number,\n offsetX = 0,\n offsetY = 0,\n): Rect {\n debugInspectUtils('adaptBboxToRect', bbox, width, height, offsetX, offsetY);\n const [left, top, right, bottom] = adaptBbox(bbox, width, height);\n const rect = {\n left: left + offsetX,\n top: top + offsetY,\n width: right - left,\n height: bottom - top,\n };\n debugInspectUtils('adaptBboxToRect, result=', rect);\n return rect;\n}\n\nlet warned = false;\nexport function warnGPT4oSizeLimit(size: Size) {\n if (warned) return;\n if (getModelName()?.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 page 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(rect: Rect, screenSize: Size) {\n const minEdgeSize = vlLocateMode() === 'doubao-vision' ? 500 : 300;\n const defaultPadding = 160;\n\n const paddingSizeHorizontal =\n rect.width < minEdgeSize\n ? Math.ceil((minEdgeSize - rect.width) / 2)\n : defaultPadding;\n const paddingSizeVertical =\n rect.height < minEdgeSize\n ? Math.ceil((minEdgeSize - rect.height) / 2)\n : defaultPadding;\n rect.left = Math.max(0, rect.left - paddingSizeHorizontal);\n rect.width = Math.min(\n rect.width + paddingSizeHorizontal * 2,\n screenSize.width - rect.left,\n );\n rect.top = Math.max(0, rect.top - paddingSizeVertical);\n rect.height = Math.min(\n rect.height + paddingSizeVertical * 2,\n screenSize.height - rect.top,\n );\n return rect;\n}\n\nexport async function markupImageForLLM(\n screenshotBase64: string,\n tree: ElementTreeNode<BaseElement>,\n size: Size,\n) {\n const elementsInfo = treeToList(tree);\n const elementsPositionInfoWithoutText = elementsInfo!.filter(\n (elementInfo) => {\n if (elementInfo.attributes.nodeType === NodeType.TEXT) {\n return false;\n }\n return true;\n },\n );\n\n const imagePayload = await compositeElementInfoImg({\n inputImgBase64: screenshotBase64,\n elementsPositionInfo: elementsPositionInfoWithoutText,\n size,\n });\n return imagePayload;\n}\n\nexport function buildYamlFlowFromPlans(\n plans: PlanningAction[],\n sleep?: number,\n): MidsceneYamlFlowItem[] {\n const flow: MidsceneYamlFlowItem[] = [];\n\n for (const plan of plans) {\n const type = plan.type;\n const locate = plan.locate?.prompt!; // TODO: check if locate is null\n\n if (type === 'Tap') {\n flow.push({\n aiTap: locate!,\n });\n } else if (type === 'Hover') {\n flow.push({\n aiHover: locate!,\n });\n } else if (type === 'Input') {\n const param = plan.param as PlanningActionParamInputOrKeyPress;\n flow.push({\n aiInput: param.value,\n locate,\n });\n } else if (type === 'KeyboardPress') {\n const param = plan.param as PlanningActionParamInputOrKeyPress;\n flow.push({\n aiKeyboardPress: param.value,\n locate,\n });\n } else if (type === 'Scroll') {\n const param = plan.param as ScrollParam;\n flow.push({\n aiScroll: null,\n locate,\n direction: param.direction,\n scrollType: param.scrollType,\n distance: param.distance,\n });\n } else if (type === 'Sleep') {\n const param = plan.param as PlanningActionParamSleep;\n flow.push({\n sleep: param.timeMs,\n });\n } else if (\n type === 'AndroidBackButton' ||\n type === 'AndroidHomeButton' ||\n type === 'AndroidRecentAppsButton' ||\n type === 'AndroidLongPress' ||\n type === 'AndroidPull'\n ) {\n // not implemented in yaml yet\n } else if (\n type === 'Error' ||\n type === 'Assert' ||\n type === 'AssertWithoutThrow' ||\n type === 'Finished'\n ) {\n // do nothing\n } else {\n console.warn(\n `Cannot convert action ${type} to yaml flow. This should be a bug of Midscene.`,\n );\n }\n }\n\n if (sleep) {\n flow.push({\n sleep: sleep,\n });\n }\n\n return flow;\n}\n"],"names":["__webpack_require__","definition","key","Object","obj","prop","Symbol","AIActionType","callAiFn","msgs","AIActionTypeValue","jsonObject","callToGetJSONObject","defaultBboxSize","debugInspectUtils","getDebug","fillBboxParam","locate","width","height","adaptBbox","adaptQwenBbox","bbox","msg","JSON","Error","result","Math","adaptDoubaoBbox","assert","splitted","Number","Array","bboxList","item","x","y","vlLocateMode","adaptGeminiBbox","left","top","right","bottom","adaptBboxToRect","offsetX","offsetY","rect","warned","warnGPT4oSizeLimit","size","_getModelName","warningMsg","console","mergeRects","rects","minLeft","r","minTop","maxRight","maxBottom","expandSearchArea","screenSize","minEdgeSize","defaultPadding","paddingSizeHorizontal","paddingSizeVertical","markupImageForLLM","screenshotBase64","tree","elementsInfo","treeToList","elementsPositionInfoWithoutText","elementInfo","NodeType","imagePayload","compositeElementInfoImg","buildYamlFlowFromPlans","plans","sleep","flow","plan","_plan_locate","type","param"],"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;;;;;;;;;;;;;;;;;;;;;;;;;;AC8BO,IAAKI,sBAAYA,WAAAA,GAAAA,SAAZA,YAAY;;;;;;WAAZA;;AAQL,eAAeC,SACpBC,IAAY,EACZC,iBAA+B;IAE/B,MAAMC,aAAa,MAAMC,AAAAA,IAAAA,yBAAAA,mBAAAA,AAAAA,EAAuBH,MAAMC;IAEtD,OAAO;QACL,SAASC,WAAW,OAAO;QAC3B,OAAOA,WAAW,KAAK;IACzB;AACF;AAEA,MAAME,kBAAkB;AACxB,MAAMC,oBAAoBC,AAAAA,IAAAA,uBAAAA,QAAAA,AAAAA,EAAS;AAG5B,SAASC,cACdC,MAA2B,EAC3BC,KAAa,EACbC,MAAc;IAGd,IAAKF,OAAe,OAAO,IAAI,CAACA,CAAAA,QAAAA,SAAAA,KAAAA,IAAAA,OAAQ,IAAI,AAAD,GAAG;QAC5CA,OAAO,IAAI,GAAIA,OAAe,OAAO;QAErC,OAAQA,OAAe,OAAO;IAChC;IAEA,IAAIA,QAAAA,SAAAA,KAAAA,IAAAA,OAAQ,IAAI,EACdA,OAAO,IAAI,GAAGG,UAAUH,OAAO,IAAI,EAAEC,OAAOC;IAG9C,OAAOF;AACT;AAEO,SAASI,cACdC,IAAc;IAEd,IAAIA,KAAK,MAAM,GAAG,GAAG;QACnB,MAAMC,MAAM,CAAC,oCAAoC,EAAEC,KAAK,SAAS,CAACF,MAAM,CAAC,CAAC;QAC1E,MAAM,IAAIG,MAAMF;IAClB;IAEA,MAAMG,SAA2C;QAC/CC,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE;QAClBK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE;QACC,YAAnB,OAAOA,IAAI,CAAC,EAAE,GACVK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,IAClBK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,GAAGT;QACN,YAAnB,OAAOS,IAAI,CAAC,EAAE,GACVK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,IAClBK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,GAAGT;KAC1B;IACD,OAAOa;AACT;AAEO,SAASE,gBACdN,IAAkC,EAClCJ,KAAa,EACbC,MAAc;IAEdU,IAAAA,sBAAAA,MAAAA,AAAAA,EACEX,QAAQ,KAAKC,SAAS,GACtB;IAGF,IAAI,AAAgB,YAAhB,OAAOG,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,IAAIZ,QAAS;YAC3CS,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAIX,SAAU;YAC5CQ,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAIZ,QAAS;YAC3CS,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAIX,SAAU;SAC7C;QAEH,MAAM,IAAIM,MAAM,CAAC,iDAAiD,EAAEH,MAAM;IAC5E;IAEA,IAAIU,MAAM,OAAO,CAACV,SAASU,MAAM,OAAO,CAACV,IAAI,CAAC,EAAE,GAC9CA,OAAOA,IAAI,CAAC,EAAE;IAGhB,IAAIW,WAAqB,EAAE;IAC3B,IAAID,MAAM,OAAO,CAACV,SAAS,AAAmB,YAAnB,OAAOA,IAAI,CAAC,EAAE,EACvCA,KAAK,OAAO,CAAC,CAACY;QACZ,IAAI,AAAgB,YAAhB,OAAOA,QAAqBA,KAAK,QAAQ,CAAC,MAAM;YAClD,MAAM,CAACC,GAAGC,EAAE,GAAGF,KAAK,KAAK,CAAC;YAC1BD,SAAS,IAAI,CAACF,OAAOI,EAAE,IAAI,KAAKJ,OAAOK,EAAE,IAAI;QAC/C,OAAO,IAAI,AAAgB,YAAhB,OAAOF,QAAqBA,KAAK,QAAQ,CAAC,MAAM;YACzD,MAAM,CAACC,GAAGC,EAAE,GAAGF,KAAK,KAAK,CAAC;YAC1BD,SAAS,IAAI,CAACF,OAAOI,EAAE,IAAI,KAAKJ,OAAOK,EAAE,IAAI;QAC/C,OACEH,SAAS,IAAI,CAACF,OAAOG;IAEzB;SAEAD,WAAWX;IAGb,IAAIW,AAAoB,MAApBA,SAAS,MAAM,IAAUA,AAAoB,MAApBA,SAAS,MAAM,EAC1C,OAAO;QACLN,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,QAAS;QACnCS,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGd,SAAU;QACpCQ,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,QAAS;QACnCS,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGd,SAAU;KACrC;IAIH,IACEc,AAAoB,MAApBA,SAAS,MAAM,IACfA,AAAoB,MAApBA,SAAS,MAAM,IACfA,AAAoB,MAApBA,SAAS,MAAM,IACfA,AAAoB,MAApBA,SAAS,MAAM,EAEf,OAAO;QACLN,KAAK,GAAG,CACN,GACAA,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,QAAS,QAAQL,kBAAkB;QAE/Dc,KAAK,GAAG,CACN,GACAA,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGd,SAAU,QAAQN,kBAAkB;QAEhEc,KAAK,GAAG,CACNT,OACAS,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,QAAS,QAAQL,kBAAkB;QAE/Dc,KAAK,GAAG,CACNR,QACAQ,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGd,SAAU,QAAQN,kBAAkB;KAEjE;IAGH,IAAIS,AAAgB,MAAhBA,KAAK,MAAM,EACb,OAAO;QACLK,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,QAAS;QACnCS,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGd,SAAU;QACpCQ,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGf,QAAS;QACnCS,KAAK,KAAK,CAAEM,QAAQ,CAAC,EAAE,GAAGd,SAAU;KACrC;IAGH,MAAMI,MAAM,CAAC,0CAA0C,EAAEC,KAAK,SAAS,CAACF,MAAM,CAAC,CAAC;IAChF,MAAM,IAAIG,MAAMF;AAClB;AAEO,SAASH,UACdE,IAAc,EACdJ,KAAa,EACbC,MAAc;IAEd,IAAIkB,AAAmB,oBAAnBA,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,OAAsCA,AAAmB,kBAAnBA,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,KACxC,OAAOT,gBAAgBN,MAAMJ,OAAOC;IAGtC,IAAIkB,AAAmB,aAAnBA,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,KACF,OAAOC,gBAAgBhB,MAAMJ,OAAOC;IAGtC,OAAOE,cAAcC;AACvB;AAEO,SAASgB,gBACdhB,IAAc,EACdJ,KAAa,EACbC,MAAc;IAEd,MAAMoB,OAAOZ,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGJ,QAAS;IAC5C,MAAMsB,MAAMb,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGH,SAAU;IAC5C,MAAMsB,QAAQd,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGJ,QAAS;IAC7C,MAAMwB,SAASf,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGH,SAAU;IAC/C,OAAO;QAACoB;QAAMC;QAAKC;QAAOC;KAAO;AACnC;AAEO,SAASC,gBACdrB,IAAc,EACdJ,KAAa,EACbC,MAAc,EACdyB,UAAU,CAAC,EACXC,UAAU,CAAC;IAEX/B,kBAAkB,mBAAmBQ,MAAMJ,OAAOC,QAAQyB,SAASC;IACnE,MAAM,CAACN,MAAMC,KAAKC,OAAOC,OAAO,GAAGtB,UAAUE,MAAMJ,OAAOC;IAC1D,MAAM2B,OAAO;QACX,MAAMP,OAAOK;QACb,KAAKJ,MAAMK;QACX,OAAOJ,QAAQF;QACf,QAAQG,SAASF;IACnB;IACA1B,kBAAkB,4BAA4BgC;IAC9C,OAAOA;AACT;AAEA,IAAIC,SAAS;AACN,SAASC,mBAAmBC,IAAU;QAEvCC;IADJ,IAAIH,QAAQ;IACZ,IAAI,QAAAG,CAAAA,gBAAAA,AAAAA,IAAAA,yBAAAA,YAAAA,AAAAA,GAAa,IAAbA,KAAAA,IAAAA,cAAgB,WAAW,GAAG,QAAQ,CAAC,WAAW;QACpD,MAAMC,aAAa,CAAC,uEAAuE,EAAEF,KAAK,KAAK,CAAC,CAAC,EAAEA,KAAK,MAAM,CAAC,wFAAwF,CAAC;QAEhN,IACEtB,KAAK,GAAG,CAACsB,KAAK,KAAK,EAAEA,KAAK,MAAM,IAAI,QACpCtB,KAAK,GAAG,CAACsB,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,UAAU5B,KAAK,GAAG,IAAI2B,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,IAAI;IACnD,MAAMC,SAAS9B,KAAK,GAAG,IAAI2B,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,GAAG;IACjD,MAAME,WAAW/B,KAAK,GAAG,IAAI2B,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,IAAI,GAAGA,EAAE,KAAK;IAC9D,MAAMG,YAAYhC,KAAK,GAAG,IAAI2B,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,iBAAiBd,IAAU,EAAEe,UAAgB;IAC3D,MAAMC,cAAczB,AAAmB,oBAAnBA,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,MAAqC,MAAM;IAC/D,MAAM0B,iBAAiB;IAEvB,MAAMC,wBACJlB,KAAK,KAAK,GAAGgB,cACTnC,KAAK,IAAI,CAAEmC,AAAAA,CAAAA,cAAchB,KAAK,KAAI,IAAK,KACvCiB;IACN,MAAME,sBACJnB,KAAK,MAAM,GAAGgB,cACVnC,KAAK,IAAI,CAAEmC,AAAAA,CAAAA,cAAchB,KAAK,MAAK,IAAK,KACxCiB;IACNjB,KAAK,IAAI,GAAGnB,KAAK,GAAG,CAAC,GAAGmB,KAAK,IAAI,GAAGkB;IACpClB,KAAK,KAAK,GAAGnB,KAAK,GAAG,CACnBmB,KAAK,KAAK,GAAGkB,AAAwB,IAAxBA,uBACbH,WAAW,KAAK,GAAGf,KAAK,IAAI;IAE9BA,KAAK,GAAG,GAAGnB,KAAK,GAAG,CAAC,GAAGmB,KAAK,GAAG,GAAGmB;IAClCnB,KAAK,MAAM,GAAGnB,KAAK,GAAG,CACpBmB,KAAK,MAAM,GAAGmB,AAAsB,IAAtBA,qBACdJ,WAAW,MAAM,GAAGf,KAAK,GAAG;IAE9B,OAAOA;AACT;AAEO,eAAeoB,kBACpBC,gBAAwB,EACxBC,IAAkC,EAClCnB,IAAU;IAEV,MAAMoB,eAAeC,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;QACtBtB;IACF;IACA,OAAOyB;AACT;AAEO,SAASE,uBACdC,KAAuB,EACvBC,KAAc;IAEd,MAAMC,OAA+B,EAAE;IAEvC,KAAK,MAAMC,QAAQH,MAAO;YAETI;QADf,MAAMC,OAAOF,KAAK,IAAI;QACtB,MAAM/D,SAAS,QAAAgE,CAAAA,eAAAA,KAAK,MAAM,AAAD,IAAVA,KAAAA,IAAAA,aAAa,MAAM;QAElC,IAAIC,AAAS,UAATA,MACFH,KAAK,IAAI,CAAC;YACR,OAAO9D;QACT;aACK,IAAIiE,AAAS,YAATA,MACTH,KAAK,IAAI,CAAC;YACR,SAAS9D;QACX;aACK,IAAIiE,AAAS,YAATA,MAAkB;YAC3B,MAAMC,QAAQH,KAAK,KAAK;YACxBD,KAAK,IAAI,CAAC;gBACR,SAASI,MAAM,KAAK;gBACpBlE;YACF;QACF,OAAO,IAAIiE,AAAS,oBAATA,MAA0B;YACnC,MAAMC,QAAQH,KAAK,KAAK;YACxBD,KAAK,IAAI,CAAC;gBACR,iBAAiBI,MAAM,KAAK;gBAC5BlE;YACF;QACF,OAAO,IAAIiE,AAAS,aAATA,MAAmB;YAC5B,MAAMC,QAAQH,KAAK,KAAK;YACxBD,KAAK,IAAI,CAAC;gBACR,UAAU;gBACV9D;gBACA,WAAWkE,MAAM,SAAS;gBAC1B,YAAYA,MAAM,UAAU;gBAC5B,UAAUA,MAAM,QAAQ;YAC1B;QACF,OAAO,IAAID,AAAS,YAATA,MAAkB;YAC3B,MAAMC,QAAQH,KAAK,KAAK;YACxBD,KAAK,IAAI,CAAC;gBACR,OAAOI,MAAM,MAAM;YACrB;QACF,OACW,wBAATD,QACAA,AAAS,wBAATA,QACAA,AAAS,8BAATA,QACAA,AAAS,uBAATA,QACAA,AAAS,kBAATA,QAIAA,AAAS,YAATA,QACAA,AAAS,aAATA,QACAA,AAAS,yBAATA,QACAA,AAAS,eAATA,QAIA9B,QAAQ,IAAI,CACV,CAAC,sBAAsB,EAAE8B,KAAK,gDAAgD,CAAC;IAGrF;IAEA,IAAIJ,OACFC,KAAK,IAAI,CAAC;QACR,OAAOD;IACT;IAGF,OAAOC;AACT"}
|
|
@@ -37,7 +37,7 @@ __webpack_require__.d(__webpack_exports__, {
|
|
|
37
37
|
resizeImageForUiTars: ()=>external_ui_tars_planning_js_namespaceObject.resizeImageForUiTars,
|
|
38
38
|
systemPromptToLocateElement: ()=>llm_locator_js_namespaceObject.systemPromptToLocateElement,
|
|
39
39
|
vlmPlanning: ()=>external_ui_tars_planning_js_namespaceObject.vlmPlanning,
|
|
40
|
-
|
|
40
|
+
AiAssert: ()=>external_inspect_js_namespaceObject.AiAssert,
|
|
41
41
|
adaptBboxToRect: ()=>external_common_js_namespaceObject.adaptBboxToRect,
|
|
42
42
|
AiLocateElement: ()=>external_inspect_js_namespaceObject.AiLocateElement,
|
|
43
43
|
callAiFn: ()=>external_common_js_namespaceObject.callAiFn,
|
|
@@ -55,10 +55,10 @@ const external_llm_planning_js_namespaceObject = require("./llm-planning.js");
|
|
|
55
55
|
const external_common_js_namespaceObject = require("./common.js");
|
|
56
56
|
const external_ui_tars_planning_js_namespaceObject = require("./ui-tars-planning.js");
|
|
57
57
|
exports.AIActionType = __webpack_exports__.AIActionType;
|
|
58
|
+
exports.AiAssert = __webpack_exports__.AiAssert;
|
|
58
59
|
exports.AiExtractElementInfo = __webpack_exports__.AiExtractElementInfo;
|
|
59
60
|
exports.AiLocateElement = __webpack_exports__.AiLocateElement;
|
|
60
61
|
exports.AiLocateSection = __webpack_exports__.AiLocateSection;
|
|
61
|
-
exports.actionSpaceTypePrefix = __webpack_exports__.actionSpaceTypePrefix;
|
|
62
62
|
exports.adaptBboxToRect = __webpack_exports__.adaptBboxToRect;
|
|
63
63
|
exports.callAi = __webpack_exports__.callAi;
|
|
64
64
|
exports.callAiFn = __webpack_exports__.callAiFn;
|
|
@@ -76,10 +76,10 @@ exports.systemPromptToLocateElement = __webpack_exports__.systemPromptToLocateEl
|
|
|
76
76
|
exports.vlmPlanning = __webpack_exports__.vlmPlanning;
|
|
77
77
|
for(var __webpack_i__ in __webpack_exports__)if (-1 === [
|
|
78
78
|
"AIActionType",
|
|
79
|
+
"AiAssert",
|
|
79
80
|
"AiExtractElementInfo",
|
|
80
81
|
"AiLocateElement",
|
|
81
82
|
"AiLocateSection",
|
|
82
|
-
"actionSpaceTypePrefix",
|
|
83
83
|
"adaptBboxToRect",
|
|
84
84
|
"callAi",
|
|
85
85
|
"callAiFn",
|
|
@@ -24,15 +24,17 @@ var __webpack_require__ = {};
|
|
|
24
24
|
var __webpack_exports__ = {};
|
|
25
25
|
__webpack_require__.r(__webpack_exports__);
|
|
26
26
|
__webpack_require__.d(__webpack_exports__, {
|
|
27
|
-
|
|
27
|
+
AiAssert: ()=>AiAssert,
|
|
28
|
+
AiExtractElementInfo: ()=>AiExtractElementInfo,
|
|
28
29
|
AiLocateSection: ()=>AiLocateSection,
|
|
29
|
-
|
|
30
|
+
AiLocateElement: ()=>AiLocateElement
|
|
30
31
|
});
|
|
31
32
|
const env_namespaceObject = require("@midscene/shared/env");
|
|
32
33
|
const img_namespaceObject = require("@midscene/shared/img");
|
|
33
34
|
const logger_namespaceObject = require("@midscene/shared/logger");
|
|
34
35
|
const utils_namespaceObject = require("@midscene/shared/utils");
|
|
35
36
|
const external_common_js_namespaceObject = require("./common.js");
|
|
37
|
+
const assertion_js_namespaceObject = require("./prompt/assertion.js");
|
|
36
38
|
const extraction_js_namespaceObject = require("./prompt/extraction.js");
|
|
37
39
|
const llm_locator_js_namespaceObject = require("./prompt/llm-locator.js");
|
|
38
40
|
const llm_section_locator_js_namespaceObject = require("./prompt/llm-section-locator.js");
|
|
@@ -293,10 +295,60 @@ async function AiExtractElementInfo(options) {
|
|
|
293
295
|
usage: result.usage
|
|
294
296
|
};
|
|
295
297
|
}
|
|
298
|
+
async function AiAssert(options) {
|
|
299
|
+
const { assertion, context } = options;
|
|
300
|
+
(0, utils_namespaceObject.assert)(assertion, 'assertion should not be empty');
|
|
301
|
+
const { screenshotBase64 } = context;
|
|
302
|
+
const systemPrompt = (0, assertion_js_namespaceObject.systemPromptToAssert)({
|
|
303
|
+
isUITars: (0, env_namespaceObject.getAIConfigInBoolean)(env_namespaceObject.MIDSCENE_USE_VLM_UI_TARS)
|
|
304
|
+
});
|
|
305
|
+
const assertionText = extraTextFromUserPrompt(assertion);
|
|
306
|
+
const msgs = [
|
|
307
|
+
{
|
|
308
|
+
role: 'system',
|
|
309
|
+
content: systemPrompt
|
|
310
|
+
},
|
|
311
|
+
{
|
|
312
|
+
role: 'user',
|
|
313
|
+
content: [
|
|
314
|
+
{
|
|
315
|
+
type: 'image_url',
|
|
316
|
+
image_url: {
|
|
317
|
+
url: screenshotBase64,
|
|
318
|
+
detail: 'high'
|
|
319
|
+
}
|
|
320
|
+
},
|
|
321
|
+
{
|
|
322
|
+
type: 'text',
|
|
323
|
+
text: `
|
|
324
|
+
Here is the assertion. Please tell whether it is truthy according to the screenshot.
|
|
325
|
+
=====================================
|
|
326
|
+
${assertionText}
|
|
327
|
+
=====================================
|
|
328
|
+
`
|
|
329
|
+
}
|
|
330
|
+
]
|
|
331
|
+
}
|
|
332
|
+
];
|
|
333
|
+
if ('string' != typeof assertion) {
|
|
334
|
+
const addOns = await promptsToChatParam({
|
|
335
|
+
images: assertion.images,
|
|
336
|
+
convertHttpImage2Base64: assertion.convertHttpImage2Base64
|
|
337
|
+
});
|
|
338
|
+
msgs.push(...addOns);
|
|
339
|
+
}
|
|
340
|
+
const { content: assertResult, usage } = await (0, external_common_js_namespaceObject.callAiFn)(msgs, external_common_js_namespaceObject.AIActionType.ASSERT);
|
|
341
|
+
return {
|
|
342
|
+
content: assertResult,
|
|
343
|
+
usage
|
|
344
|
+
};
|
|
345
|
+
}
|
|
346
|
+
exports.AiAssert = __webpack_exports__.AiAssert;
|
|
296
347
|
exports.AiExtractElementInfo = __webpack_exports__.AiExtractElementInfo;
|
|
297
348
|
exports.AiLocateElement = __webpack_exports__.AiLocateElement;
|
|
298
349
|
exports.AiLocateSection = __webpack_exports__.AiLocateSection;
|
|
299
350
|
for(var __webpack_i__ in __webpack_exports__)if (-1 === [
|
|
351
|
+
"AiAssert",
|
|
300
352
|
"AiExtractElementInfo",
|
|
301
353
|
"AiLocateElement",
|
|
302
354
|
"AiLocateSection"
|
|
@@ -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 TMultimodalPrompt,\n TUserPrompt,\n UIContext,\n} from '@/types';\nimport {\n MIDSCENE_USE_QWEN_VL,\n getAIConfigInBoolean,\n vlLocateMode,\n} from '@midscene/shared/env';\nimport {\n cropByRect,\n paddingToMatchBlockByBase64,\n preProcessImageUrl,\n} from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type {\n ChatCompletionSystemMessageParam,\n ChatCompletionUserMessageParam,\n} from 'openai/resources/index';\nimport {\n AIActionType,\n adaptBboxToRect,\n callAiFn,\n expandSearchArea,\n markupImageForLLM,\n mergeRects,\n} from './common';\nimport {\n extractDataQueryPrompt,\n systemPromptToExtract,\n} from './prompt/extraction';\nimport {\n findElementPrompt,\n systemPromptToLocateElement,\n} from './prompt/llm-locator';\nimport {\n sectionLocatorInstruction,\n systemPromptToLocateSection,\n} from './prompt/llm-section-locator';\nimport {\n describeUserPage,\n distance,\n distanceThreshold,\n elementByPositionWithElementInfo,\n} from './prompt/util';\nimport { callToGetJSONObject } from './service-caller/index';\n\nexport type AIArgs = [\n ChatCompletionSystemMessageParam,\n ...ChatCompletionUserMessageParam[],\n];\n\nconst debugInspect = getDebug('ai:inspect');\nconst debugSection = getDebug('ai:section');\n\nconst extraTextFromUserPrompt = (prompt: TUserPrompt): string => {\n if (typeof prompt === 'string') {\n return prompt;\n } else {\n return prompt.prompt;\n }\n};\n\nconst promptsToChatParam = async (\n multimodalPrompt: TMultimodalPrompt,\n): Promise<ChatCompletionUserMessageParam[]> => {\n const msgs: ChatCompletionUserMessageParam[] = [];\n if (multimodalPrompt?.images?.length) {\n msgs.push({\n role: 'user',\n content: [\n {\n type: 'text',\n text: 'Next, I will provide all the reference images.',\n },\n ],\n });\n\n for (const item of multimodalPrompt.images) {\n const base64 = await preProcessImageUrl(\n item.url,\n !!multimodalPrompt.convertHttpImage2Base64,\n );\n\n msgs.push({\n role: 'user',\n content: [\n {\n type: 'text',\n text: `reference image ${item.name}:`,\n },\n ],\n });\n\n msgs.push({\n role: 'user',\n content: [\n {\n type: 'image_url',\n image_url: {\n url: base64,\n detail: 'high',\n },\n },\n ],\n });\n }\n }\n return msgs;\n};\n\nexport async function AiLocateElement<\n ElementType extends BaseElement = BaseElement,\n>(options: {\n context: UIContext<ElementType>;\n targetElementDescription: TUserPrompt;\n referenceImage?: ReferenceImage;\n callAI?: typeof callAiFn<AIElementResponse | [number, number]>;\n searchConfig?: Awaited<ReturnType<typeof AiLocateSection>>;\n}): Promise<{\n parseResult: AIElementLocatorResponse;\n rect?: Rect;\n rawResponse: string;\n elementById: ElementById;\n usage?: AIUsageInfo;\n isOrderSensitive?: boolean;\n}> {\n const { context, targetElementDescription, callAI } = options;\n const { screenshotBase64 } = context;\n const { description, elementById, insertElementByPosition } =\n await describeUserPage(context);\n\n assert(\n targetElementDescription,\n 'cannot find the target element description',\n );\n\n const userInstructionPrompt = await findElementPrompt.format({\n pageDescription: description,\n targetElementDescription: extraTextFromUserPrompt(targetElementDescription),\n });\n const systemPrompt = systemPromptToLocateElement(vlLocateMode());\n\n let imagePayload = screenshotBase64;\n\n if (options.searchConfig) {\n assert(\n options.searchConfig.rect,\n 'searchArea is provided but its rect cannot be found. Failed to locate element',\n );\n assert(\n options.searchConfig.imageBase64,\n 'searchArea is provided but its imageBase64 cannot be found. Failed to locate element',\n );\n\n imagePayload = options.searchConfig.imageBase64;\n } else if (vlLocateMode() === 'qwen-vl') {\n imagePayload = await paddingToMatchBlockByBase64(imagePayload);\n } else if (!vlLocateMode()) {\n imagePayload = await markupImageForLLM(\n screenshotBase64,\n context.tree,\n context.size,\n );\n }\n\n const msgs: AIArgs = [\n { role: 'system', content: systemPrompt },\n {\n role: 'user',\n content: [\n {\n type: 'image_url',\n image_url: {\n url: imagePayload,\n detail: 'high',\n },\n },\n {\n type: 'text',\n text: userInstructionPrompt,\n },\n ],\n },\n ];\n\n if (typeof targetElementDescription !== 'string') {\n const addOns = await promptsToChatParam({\n images: targetElementDescription.images,\n convertHttpImage2Base64: targetElementDescription.convertHttpImage2Base64,\n });\n msgs.push(...addOns);\n }\n\n const callAIFn =\n callAI || callToGetJSONObject<AIElementResponse | [number, number]>;\n\n const res = await callAIFn(msgs, AIActionType.INSPECT_ELEMENT);\n\n const rawResponse = JSON.stringify(res.content);\n\n let resRect: Rect | undefined;\n let matchedElements: AIElementLocatorResponse['elements'] =\n 'elements' in res.content ? res.content.elements : [];\n let errors: AIElementLocatorResponse['errors'] | undefined =\n 'errors' in res.content ? res.content.errors : [];\n try {\n if ('bbox' in res.content && Array.isArray(res.content.bbox)) {\n resRect = adaptBboxToRect(\n res.content.bbox,\n options.searchConfig?.rect?.width || context.size.width,\n options.searchConfig?.rect?.height || context.size.height,\n options.searchConfig?.rect?.left,\n options.searchConfig?.rect?.top,\n );\n debugInspect('resRect', resRect);\n\n const rectCenter = {\n x: resRect.left + resRect.width / 2,\n y: resRect.top + resRect.height / 2,\n };\n let element = elementByPositionWithElementInfo(context.tree, rectCenter);\n\n const distanceToCenter = element\n ? distance({ x: element.center[0], y: element.center[1] }, rectCenter)\n : 0;\n\n if (!element || distanceToCenter > distanceThreshold) {\n element = insertElementByPosition(rectCenter);\n }\n\n if (element) {\n matchedElements = [element];\n errors = [];\n }\n }\n } catch (e) {\n const msg =\n e instanceof Error\n ? `Failed to parse bbox: ${e.message}`\n : 'unknown error in locate';\n if (!errors || errors?.length === 0) {\n errors = [msg];\n } else {\n errors.push(`(${msg})`);\n }\n }\n\n return {\n rect: resRect,\n parseResult: {\n elements: matchedElements,\n errors,\n },\n rawResponse,\n elementById,\n usage: res.usage,\n isOrderSensitive:\n typeof res.content === 'object' &&\n res.content !== null &&\n 'isOrderSensitive' in res.content\n ? (res.content as any).isOrderSensitive\n : undefined,\n };\n}\n\nexport async function AiLocateSection(options: {\n context: UIContext<BaseElement>;\n sectionDescription: TUserPrompt;\n callAI?: typeof callAiFn<AISectionLocatorResponse>;\n}): Promise<{\n rect?: Rect;\n imageBase64?: string;\n error?: string;\n rawResponse: string;\n usage?: AIUsageInfo;\n}> {\n const { context, sectionDescription } = options;\n const { screenshotBase64 } = context;\n\n const systemPrompt = systemPromptToLocateSection(vlLocateMode());\n const sectionLocatorInstructionText = await sectionLocatorInstruction.format({\n sectionDescription: extraTextFromUserPrompt(sectionDescription),\n });\n const msgs: AIArgs = [\n { role: 'system', content: systemPrompt },\n {\n role: 'user',\n content: [\n {\n type: 'image_url',\n image_url: {\n url: screenshotBase64,\n detail: 'high',\n },\n },\n {\n type: 'text',\n text: sectionLocatorInstructionText,\n },\n ],\n },\n ];\n\n if (typeof sectionDescription !== 'string') {\n const addOns = await promptsToChatParam({\n images: sectionDescription.images,\n convertHttpImage2Base64: sectionDescription.convertHttpImage2Base64,\n });\n msgs.push(...addOns);\n }\n\n const result = await callAiFn<AISectionLocatorResponse>(\n msgs,\n AIActionType.EXTRACT_DATA,\n );\n\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 );\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(bbox, context.size.width, context.size.height);\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);\n debugSection('expanded sectionRect %j', sectionRect);\n }\n\n let imageBase64 = screenshotBase64;\n if (sectionRect) {\n imageBase64 = await cropByRect(\n screenshotBase64,\n sectionRect,\n getAIConfigInBoolean(MIDSCENE_USE_QWEN_VL),\n );\n }\n\n return {\n rect: sectionRect,\n imageBase64,\n error: result.content.error,\n rawResponse: JSON.stringify(result.content),\n usage: result.usage,\n };\n}\n\nexport async function AiExtractElementInfo<\n T,\n ElementType extends BaseElement = BaseElement,\n>(options: {\n dataQuery: string | Record<string, string>;\n multimodalPrompt?: TMultimodalPrompt;\n context: UIContext<ElementType>;\n extractOption?: InsightExtractOption;\n}) {\n const { dataQuery, context, extractOption, multimodalPrompt } = options;\n const systemPrompt = systemPromptToExtract();\n\n const { screenshotBase64 } = context;\n const { description, elementById } = await describeUserPage(context, {\n truncateTextLength: 200,\n filterNonTextContent: false,\n visibleOnly: false,\n domIncluded: extractOption?.domIncluded,\n });\n\n const extractDataPromptText = await extractDataQueryPrompt(\n description,\n dataQuery,\n );\n\n const userContent: ChatCompletionUserMessageParam['content'] = [];\n\n if (extractOption?.screenshotIncluded !== false) {\n userContent.push({\n type: 'image_url',\n image_url: {\n url: screenshotBase64,\n detail: 'high',\n },\n });\n }\n\n userContent.push({\n type: 'text',\n text: extractDataPromptText,\n });\n\n const msgs: AIArgs = [\n { role: 'system', content: systemPrompt },\n {\n role: 'user',\n content: userContent,\n },\n ];\n\n if (options.extractOption?.returnThought) {\n msgs.push({\n role: 'user',\n content: 'Please provide reasons.',\n });\n }\n\n if (multimodalPrompt) {\n const addOns = await promptsToChatParam({\n images: multimodalPrompt.images,\n convertHttpImage2Base64: multimodalPrompt.convertHttpImage2Base64,\n });\n msgs.push(...addOns);\n }\n\n const result = await callAiFn<AIDataExtractionResponse<T>>(\n msgs,\n AIActionType.EXTRACT_DATA,\n );\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","callAI","screenshotBase64","description","elementById","insertElementByPosition","describeUserPage","assert","userInstructionPrompt","findElementPrompt","systemPrompt","systemPromptToLocateElement","vlLocateMode","imagePayload","paddingToMatchBlockByBase64","markupImageForLLM","addOns","callAIFn","callToGetJSONObject","res","AIActionType","rawResponse","JSON","resRect","matchedElements","errors","Array","_options_searchConfig_rect","_options_searchConfig_rect1","_options_searchConfig_rect2","_options_searchConfig_rect3","adaptBboxToRect","rectCenter","element","elementByPositionWithElementInfo","distanceToCenter","distance","distanceThreshold","e","msg","Error","undefined","AiLocateSection","sectionDescription","systemPromptToLocateSection","sectionLocatorInstructionText","sectionLocatorInstruction","result","callAiFn","sectionRect","sectionBbox","targetRect","referenceBboxList","referenceRects","bbox","mergedRect","mergeRects","expandSearchArea","imageBase64","cropByRect","getAIConfigInBoolean","MIDSCENE_USE_QWEN_VL","AiExtractElementInfo","_options_extractOption","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;;;;;;;;;;;;;;;;;;;AC0DA,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,OAMD;IAQC,MAAM,EAAEC,OAAO,EAAEC,wBAAwB,EAAEC,MAAM,EAAE,GAAGH;IACtD,MAAM,EAAEI,gBAAgB,EAAE,GAAGH;IAC7B,MAAM,EAAEI,WAAW,EAAEC,WAAW,EAAEC,uBAAuB,EAAE,GACzD,MAAMC,AAAAA,IAAAA,wBAAAA,gBAAAA,AAAAA,EAAiBP;IAEzBQ,IAAAA,sBAAAA,MAAAA,AAAAA,EACEP,0BACA;IAGF,MAAMQ,wBAAwB,MAAMC,+BAAAA,iBAAAA,CAAAA,MAAwB,CAAC;QAC3D,iBAAiBN;QACjB,0BAA0Bf,wBAAwBY;IACpD;IACA,MAAMU,eAAeC,AAAAA,IAAAA,+BAAAA,2BAAAA,AAAAA,EAA4BC,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA;IAEjD,IAAIC,eAAeX;IAEnB,IAAIJ,QAAQ,YAAY,EAAE;QACxBS,IAAAA,sBAAAA,MAAAA,AAAAA,EACET,QAAQ,YAAY,CAAC,IAAI,EACzB;QAEFS,IAAAA,sBAAAA,MAAAA,AAAAA,EACET,QAAQ,YAAY,CAAC,WAAW,EAChC;QAGFe,eAAef,QAAQ,YAAY,CAAC,WAAW;IACjD,OAAO,IAAIc,AAAmB,cAAnBA,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,KACTC,eAAe,MAAMC,AAAAA,IAAAA,oBAAAA,2BAAAA,AAAAA,EAA4BD;SAC5C,IAAI,CAACD,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,KACVC,eAAe,MAAME,AAAAA,IAAAA,mCAAAA,iBAAAA,AAAAA,EACnBb,kBACAH,QAAQ,IAAI,EACZA,QAAQ,IAAI;IAIhB,MAAMN,OAAe;QACnB;YAAE,MAAM;YAAU,SAASiB;QAAa;QACxC;YACE,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,WAAW;wBACT,KAAKG;wBACL,QAAQ;oBACV;gBACF;gBACA;oBACE,MAAM;oBACN,MAAML;gBACR;aACD;QACH;KACD;IAED,IAAI,AAAoC,YAApC,OAAOR,0BAAuC;QAChD,MAAMgB,SAAS,MAAM1B,mBAAmB;YACtC,QAAQU,yBAAyB,MAAM;YACvC,yBAAyBA,yBAAyB,uBAAuB;QAC3E;QACAP,KAAK,IAAI,IAAIuB;IACf;IAEA,MAAMC,WACJhB,UAAUiB,yBAAAA,mBAAmBA;IAE/B,MAAMC,MAAM,MAAMF,SAASxB,MAAM2B,mCAAAA,YAAAA,CAAAA,eAA4B;IAE7D,MAAMC,cAAcC,KAAK,SAAS,CAACH,IAAI,OAAO;IAE9C,IAAII;IACJ,IAAIC,kBACF,cAAcL,IAAI,OAAO,GAAGA,IAAI,OAAO,CAAC,QAAQ,GAAG,EAAE;IACvD,IAAIM,SACF,YAAYN,IAAI,OAAO,GAAGA,IAAI,OAAO,CAAC,MAAM,GAAG,EAAE;IACnD,IAAI;QACF,IAAI,UAAUA,IAAI,OAAO,IAAIO,MAAM,OAAO,CAACP,IAAI,OAAO,CAAC,IAAI,GAAG;gBAG1DQ,4BAAAA,uBACAC,6BAAAA,wBACAC,6BAAAA,wBACAC,6BAAAA;YALFP,UAAUQ,AAAAA,IAAAA,mCAAAA,eAAAA,AAAAA,EACRZ,IAAI,OAAO,CAAC,IAAI,EAChBQ,AAAAA,SAAAA,CAAAA,wBAAAA,QAAQ,YAAY,AAAD,IAAnBA,KAAAA,IAAAA,QAAAA,CAAAA,6BAAAA,sBAAsB,IAAI,AAAD,IAAzBA,KAAAA,IAAAA,2BAA4B,KAAK,AAAD,KAAK5B,QAAQ,IAAI,CAAC,KAAK,EACvD6B,AAAAA,SAAAA,CAAAA,yBAAAA,QAAQ,YAAY,AAAD,IAAnBA,KAAAA,IAAAA,QAAAA,CAAAA,8BAAAA,uBAAsB,IAAI,AAAD,IAAzBA,KAAAA,IAAAA,4BAA4B,MAAM,AAAD,KAAK7B,QAAQ,IAAI,CAAC,MAAM,UACzD8B,CAAAA,yBAAAA,QAAQ,YAAY,AAAD,IAAnBA,KAAAA,IAAAA,QAAAA,CAAAA,8BAAAA,uBAAsB,IAAI,AAAD,IAAzBA,KAAAA,IAAAA,4BAA4B,IAAI,UAChCC,CAAAA,yBAAAA,QAAQ,YAAY,AAAD,IAAnBA,KAAAA,IAAAA,QAAAA,CAAAA,8BAAAA,uBAAsB,IAAI,AAAD,IAAzBA,KAAAA,IAAAA,4BAA4B,GAAG;YAEjC7C,aAAa,WAAWsC;YAExB,MAAMS,aAAa;gBACjB,GAAGT,QAAQ,IAAI,GAAGA,QAAQ,KAAK,GAAG;gBAClC,GAAGA,QAAQ,GAAG,GAAGA,QAAQ,MAAM,GAAG;YACpC;YACA,IAAIU,UAAUC,AAAAA,IAAAA,wBAAAA,gCAAAA,AAAAA,EAAiCnC,QAAQ,IAAI,EAAEiC;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,UAAU5B,wBAAwB2B;YAGpC,IAAIC,SAAS;gBACXT,kBAAkB;oBAACS;iBAAQ;gBAC3BR,SAAS,EAAE;YACb;QACF;IACF,EAAE,OAAOa,GAAG;QACV,MAAMC,MACJD,aAAaE,QACT,CAAC,sBAAsB,EAAEF,EAAE,OAAO,EAAE,GACpC;QACN,IAAI,AAACb,UAAUA,AAAAA,CAAAA,QAAAA,SAAAA,KAAAA,IAAAA,OAAQ,MAAM,AAAD,MAAM,GAGhCA,OAAO,IAAI,CAAC,CAAC,CAAC,EAAEc,IAAI,CAAC,CAAC;aAFtBd,SAAS;YAACc;SAAI;IAIlB;IAEA,OAAO;QACL,MAAMhB;QACN,aAAa;YACX,UAAUC;YACVC;QACF;QACAJ;QACAjB;QACA,OAAOe,IAAI,KAAK;QAChB,kBACE,AAAuB,YAAvB,OAAOA,IAAI,OAAO,IAClBA,AAAgB,SAAhBA,IAAI,OAAO,IACX,sBAAsBA,IAAI,OAAO,GAC5BA,IAAI,OAAO,CAAS,gBAAgB,GACrCsB;IACR;AACF;AAEO,eAAeC,gBAAgB5C,OAIrC;IAOC,MAAM,EAAEC,OAAO,EAAE4C,kBAAkB,EAAE,GAAG7C;IACxC,MAAM,EAAEI,gBAAgB,EAAE,GAAGH;IAE7B,MAAMW,eAAekC,AAAAA,IAAAA,uCAAAA,2BAAAA,AAAAA,EAA4BhC,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA;IACjD,MAAMiC,gCAAgC,MAAMC,uCAAAA,yBAAAA,CAAAA,MAAgC,CAAC;QAC3E,oBAAoB1D,wBAAwBuD;IAC9C;IACA,MAAMlD,OAAe;QACnB;YAAE,MAAM;YAAU,SAASiB;QAAa;QACxC;YACE,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,WAAW;wBACT,KAAKR;wBACL,QAAQ;oBACV;gBACF;gBACA;oBACE,MAAM;oBACN,MAAM2C;gBACR;aACD;QACH;KACD;IAED,IAAI,AAA8B,YAA9B,OAAOF,oBAAiC;QAC1C,MAAM3B,SAAS,MAAM1B,mBAAmB;YACtC,QAAQqD,mBAAmB,MAAM;YACjC,yBAAyBA,mBAAmB,uBAAuB;QACrE;QACAlD,KAAK,IAAI,IAAIuB;IACf;IAEA,MAAM+B,SAAS,MAAMC,AAAAA,IAAAA,mCAAAA,QAAAA,AAAAA,EACnBvD,MACA2B,mCAAAA,YAAAA,CAAAA,YAAyB;IAG3B,IAAI6B;IACJ,MAAMC,cAAcH,OAAO,OAAO,CAAC,IAAI;IACvC,IAAIG,aAAa;QACf,MAAMC,aAAapB,AAAAA,IAAAA,mCAAAA,eAAAA,AAAAA,EACjBmB,aACAnD,QAAQ,IAAI,CAAC,KAAK,EAClBA,QAAQ,IAAI,CAAC,MAAM;QAErBZ,aAAa,0BAA0BgE;QAEvC,MAAMC,oBAAoBL,OAAO,OAAO,CAAC,eAAe,IAAI,EAAE;QAC9D5D,aAAa,wBAAwBiE;QAErC,MAAMC,iBAAiBD,kBACpB,MAAM,CAAC,CAACE,OAAS5B,MAAM,OAAO,CAAC4B,OAC/B,GAAG,CAAC,CAACA,OACGvB,AAAAA,IAAAA,mCAAAA,eAAAA,AAAAA,EAAgBuB,MAAMvD,QAAQ,IAAI,CAAC,KAAK,EAAEA,QAAQ,IAAI,CAAC,MAAM;QAExEZ,aAAa,qBAAqBkE;QAGlC,MAAME,aAAaC,AAAAA,IAAAA,mCAAAA,UAAAA,AAAAA,EAAW;YAACL;eAAeE;SAAe;QAC7DlE,aAAa,iBAAiBoE;QAG9BN,cAAcQ,AAAAA,IAAAA,mCAAAA,gBAAAA,AAAAA,EAAiBF,YAAYxD,QAAQ,IAAI;QACvDZ,aAAa,2BAA2B8D;IAC1C;IAEA,IAAIS,cAAcxD;IAClB,IAAI+C,aACFS,cAAc,MAAMC,AAAAA,IAAAA,oBAAAA,UAAAA,AAAAA,EAClBzD,kBACA+C,aACAW,AAAAA,IAAAA,oBAAAA,oBAAAA,AAAAA,EAAqBC,oBAAAA,oBAAoBA;IAI7C,OAAO;QACL,MAAMZ;QACNS;QACA,OAAOX,OAAO,OAAO,CAAC,KAAK;QAC3B,aAAazB,KAAK,SAAS,CAACyB,OAAO,OAAO;QAC1C,OAAOA,OAAO,KAAK;IACrB;AACF;AAEO,eAAee,qBAGpBhE,OAKD;QA0CKiE;IAzCJ,MAAM,EAAEC,SAAS,EAAEjE,OAAO,EAAEkE,aAAa,EAAE1E,gBAAgB,EAAE,GAAGO;IAChE,MAAMY,eAAewD,AAAAA,IAAAA,8BAAAA,qBAAAA,AAAAA;IAErB,MAAM,EAAEhE,gBAAgB,EAAE,GAAGH;IAC7B,MAAM,EAAEI,WAAW,EAAEC,WAAW,EAAE,GAAG,MAAME,AAAAA,IAAAA,wBAAAA,gBAAAA,AAAAA,EAAiBP,SAAS;QACnE,oBAAoB;QACpB,sBAAsB;QACtB,aAAa;QACb,aAAakE,QAAAA,gBAAAA,KAAAA,IAAAA,cAAe,WAAW;IACzC;IAEA,MAAME,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,MAAM1E,OAAe;QACnB;YAAE,MAAM;YAAU,SAASiB;QAAa;QACxC;YACE,MAAM;YACN,SAAS2D;QACX;KACD;IAED,IAAI,QAAAN,CAAAA,yBAAAA,QAAQ,aAAa,AAAD,IAApBA,KAAAA,IAAAA,uBAAuB,aAAa,EACtCtE,KAAK,IAAI,CAAC;QACR,MAAM;QACN,SAAS;IACX;IAGF,IAAIF,kBAAkB;QACpB,MAAMyB,SAAS,MAAM1B,mBAAmB;YACtC,QAAQC,iBAAiB,MAAM;YAC/B,yBAAyBA,iBAAiB,uBAAuB;QACnE;QACAE,KAAK,IAAI,IAAIuB;IACf;IAEA,MAAM+B,SAAS,MAAMC,AAAAA,IAAAA,mCAAAA,QAAAA,AAAAA,EACnBvD,MACA2B,mCAAAA,YAAAA,CAAAA,YAAyB;IAE3B,OAAO;QACL,aAAa2B,OAAO,OAAO;QAC3B3C;QACA,OAAO2C,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 AIAssertionResponse,\n AIDataExtractionResponse,\n AIElementLocatorResponse,\n AIElementResponse,\n AISectionLocatorResponse,\n AIUsageInfo,\n BaseElement,\n ElementById,\n InsightExtractOption,\n Rect,\n ReferenceImage,\n TMultimodalPrompt,\n TUserPrompt,\n UIContext,\n} from '@/types';\nimport {\n MIDSCENE_USE_QWEN_VL,\n MIDSCENE_USE_VLM_UI_TARS,\n getAIConfigInBoolean,\n vlLocateMode,\n} from '@midscene/shared/env';\nimport {\n cropByRect,\n paddingToMatchBlockByBase64,\n preProcessImageUrl,\n} from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type {\n ChatCompletionSystemMessageParam,\n ChatCompletionUserMessageParam,\n} from 'openai/resources/index';\nimport {\n AIActionType,\n adaptBboxToRect,\n callAiFn,\n expandSearchArea,\n markupImageForLLM,\n mergeRects,\n} from './common';\nimport { systemPromptToAssert } from './prompt/assertion';\nimport {\n extractDataQueryPrompt,\n systemPromptToExtract,\n} from './prompt/extraction';\nimport {\n findElementPrompt,\n systemPromptToLocateElement,\n} from './prompt/llm-locator';\nimport {\n sectionLocatorInstruction,\n systemPromptToLocateSection,\n} from './prompt/llm-section-locator';\nimport {\n describeUserPage,\n distance,\n distanceThreshold,\n elementByPositionWithElementInfo,\n} from './prompt/util';\nimport { callToGetJSONObject } from './service-caller/index';\n\nexport type AIArgs = [\n ChatCompletionSystemMessageParam,\n ...ChatCompletionUserMessageParam[],\n];\n\nconst debugInspect = getDebug('ai:inspect');\nconst debugSection = getDebug('ai:section');\n\nconst extraTextFromUserPrompt = (prompt: TUserPrompt): string => {\n if (typeof prompt === 'string') {\n return prompt;\n } else {\n return prompt.prompt;\n }\n};\n\nconst promptsToChatParam = async (\n multimodalPrompt: TMultimodalPrompt,\n): Promise<ChatCompletionUserMessageParam[]> => {\n const msgs: ChatCompletionUserMessageParam[] = [];\n if (multimodalPrompt?.images?.length) {\n msgs.push({\n role: 'user',\n content: [\n {\n type: 'text',\n text: 'Next, I will provide all the reference images.',\n },\n ],\n });\n\n for (const item of multimodalPrompt.images) {\n const base64 = await preProcessImageUrl(\n item.url,\n !!multimodalPrompt.convertHttpImage2Base64,\n );\n\n msgs.push({\n role: 'user',\n content: [\n {\n type: 'text',\n text: `reference image ${item.name}:`,\n },\n ],\n });\n\n msgs.push({\n role: 'user',\n content: [\n {\n type: 'image_url',\n image_url: {\n url: base64,\n detail: 'high',\n },\n },\n ],\n });\n }\n }\n return msgs;\n};\n\nexport async function AiLocateElement<\n ElementType extends BaseElement = BaseElement,\n>(options: {\n context: UIContext<ElementType>;\n targetElementDescription: TUserPrompt;\n referenceImage?: ReferenceImage;\n callAI?: typeof callAiFn<AIElementResponse | [number, number]>;\n searchConfig?: Awaited<ReturnType<typeof AiLocateSection>>;\n}): Promise<{\n parseResult: AIElementLocatorResponse;\n rect?: Rect;\n rawResponse: string;\n elementById: ElementById;\n usage?: AIUsageInfo;\n isOrderSensitive?: boolean;\n}> {\n const { context, targetElementDescription, callAI } = options;\n const { screenshotBase64 } = context;\n const { description, elementById, insertElementByPosition } =\n await describeUserPage(context);\n\n assert(\n targetElementDescription,\n 'cannot find the target element description',\n );\n\n const userInstructionPrompt = await findElementPrompt.format({\n pageDescription: description,\n targetElementDescription: extraTextFromUserPrompt(targetElementDescription),\n });\n const systemPrompt = systemPromptToLocateElement(vlLocateMode());\n\n let imagePayload = screenshotBase64;\n\n if (options.searchConfig) {\n assert(\n options.searchConfig.rect,\n 'searchArea is provided but its rect cannot be found. Failed to locate element',\n );\n assert(\n options.searchConfig.imageBase64,\n 'searchArea is provided but its imageBase64 cannot be found. Failed to locate element',\n );\n\n imagePayload = options.searchConfig.imageBase64;\n } else if (vlLocateMode() === 'qwen-vl') {\n imagePayload = await paddingToMatchBlockByBase64(imagePayload);\n } else if (!vlLocateMode()) {\n imagePayload = await markupImageForLLM(\n screenshotBase64,\n context.tree,\n context.size,\n );\n }\n\n const msgs: AIArgs = [\n { role: 'system', content: systemPrompt },\n {\n role: 'user',\n content: [\n {\n type: 'image_url',\n image_url: {\n url: imagePayload,\n detail: 'high',\n },\n },\n {\n type: 'text',\n text: userInstructionPrompt,\n },\n ],\n },\n ];\n\n if (typeof targetElementDescription !== 'string') {\n const addOns = await promptsToChatParam({\n images: targetElementDescription.images,\n convertHttpImage2Base64: targetElementDescription.convertHttpImage2Base64,\n });\n msgs.push(...addOns);\n }\n\n const callAIFn =\n callAI || callToGetJSONObject<AIElementResponse | [number, number]>;\n\n const res = await callAIFn(msgs, AIActionType.INSPECT_ELEMENT);\n\n const rawResponse = JSON.stringify(res.content);\n\n let resRect: Rect | undefined;\n let matchedElements: AIElementLocatorResponse['elements'] =\n 'elements' in res.content ? res.content.elements : [];\n let errors: AIElementLocatorResponse['errors'] | undefined =\n 'errors' in res.content ? res.content.errors : [];\n try {\n if ('bbox' in res.content && Array.isArray(res.content.bbox)) {\n resRect = adaptBboxToRect(\n res.content.bbox,\n options.searchConfig?.rect?.width || context.size.width,\n options.searchConfig?.rect?.height || context.size.height,\n options.searchConfig?.rect?.left,\n options.searchConfig?.rect?.top,\n );\n debugInspect('resRect', resRect);\n\n const rectCenter = {\n x: resRect.left + resRect.width / 2,\n y: resRect.top + resRect.height / 2,\n };\n let element = elementByPositionWithElementInfo(context.tree, rectCenter);\n\n const distanceToCenter = element\n ? distance({ x: element.center[0], y: element.center[1] }, rectCenter)\n : 0;\n\n if (!element || distanceToCenter > distanceThreshold) {\n element = insertElementByPosition(rectCenter);\n }\n\n if (element) {\n matchedElements = [element];\n errors = [];\n }\n }\n } catch (e) {\n const msg =\n e instanceof Error\n ? `Failed to parse bbox: ${e.message}`\n : 'unknown error in locate';\n if (!errors || errors?.length === 0) {\n errors = [msg];\n } else {\n errors.push(`(${msg})`);\n }\n }\n\n return {\n rect: resRect,\n parseResult: {\n elements: matchedElements,\n errors,\n },\n rawResponse,\n elementById,\n usage: res.usage,\n isOrderSensitive:\n typeof res.content === 'object' &&\n res.content !== null &&\n 'isOrderSensitive' in res.content\n ? (res.content as any).isOrderSensitive\n : undefined,\n };\n}\n\nexport async function AiLocateSection(options: {\n context: UIContext<BaseElement>;\n sectionDescription: TUserPrompt;\n callAI?: typeof callAiFn<AISectionLocatorResponse>;\n}): Promise<{\n rect?: Rect;\n imageBase64?: string;\n error?: string;\n rawResponse: string;\n usage?: AIUsageInfo;\n}> {\n const { context, sectionDescription } = options;\n const { screenshotBase64 } = context;\n\n const systemPrompt = systemPromptToLocateSection(vlLocateMode());\n const sectionLocatorInstructionText = await sectionLocatorInstruction.format({\n sectionDescription: extraTextFromUserPrompt(sectionDescription),\n });\n const msgs: AIArgs = [\n { role: 'system', content: systemPrompt },\n {\n role: 'user',\n content: [\n {\n type: 'image_url',\n image_url: {\n url: screenshotBase64,\n detail: 'high',\n },\n },\n {\n type: 'text',\n text: sectionLocatorInstructionText,\n },\n ],\n },\n ];\n\n if (typeof sectionDescription !== 'string') {\n const addOns = await promptsToChatParam({\n images: sectionDescription.images,\n convertHttpImage2Base64: sectionDescription.convertHttpImage2Base64,\n });\n msgs.push(...addOns);\n }\n\n const result = await callAiFn<AISectionLocatorResponse>(\n msgs,\n AIActionType.EXTRACT_DATA,\n );\n\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 );\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(bbox, context.size.width, context.size.height);\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);\n debugSection('expanded sectionRect %j', sectionRect);\n }\n\n let imageBase64 = screenshotBase64;\n if (sectionRect) {\n imageBase64 = await cropByRect(\n screenshotBase64,\n sectionRect,\n getAIConfigInBoolean(MIDSCENE_USE_QWEN_VL),\n );\n }\n\n return {\n rect: sectionRect,\n imageBase64,\n error: result.content.error,\n rawResponse: JSON.stringify(result.content),\n usage: result.usage,\n };\n}\n\nexport async function AiExtractElementInfo<\n T,\n ElementType extends BaseElement = BaseElement,\n>(options: {\n dataQuery: string | Record<string, string>;\n multimodalPrompt?: TMultimodalPrompt;\n context: UIContext<ElementType>;\n extractOption?: InsightExtractOption;\n}) {\n const { dataQuery, context, extractOption, multimodalPrompt } = options;\n const systemPrompt = systemPromptToExtract();\n\n const { screenshotBase64 } = context;\n const { description, elementById } = await describeUserPage(context, {\n truncateTextLength: 200,\n filterNonTextContent: false,\n visibleOnly: false,\n domIncluded: extractOption?.domIncluded,\n });\n\n const extractDataPromptText = await extractDataQueryPrompt(\n description,\n dataQuery,\n );\n\n const userContent: ChatCompletionUserMessageParam['content'] = [];\n\n if (extractOption?.screenshotIncluded !== false) {\n userContent.push({\n type: 'image_url',\n image_url: {\n url: screenshotBase64,\n detail: 'high',\n },\n });\n }\n\n userContent.push({\n type: 'text',\n text: extractDataPromptText,\n });\n\n const msgs: AIArgs = [\n { role: 'system', content: systemPrompt },\n {\n role: 'user',\n content: userContent,\n },\n ];\n\n if (options.extractOption?.returnThought) {\n msgs.push({\n role: 'user',\n content: 'Please provide reasons.',\n });\n }\n\n if (multimodalPrompt) {\n const addOns = await promptsToChatParam({\n images: multimodalPrompt.images,\n convertHttpImage2Base64: multimodalPrompt.convertHttpImage2Base64,\n });\n msgs.push(...addOns);\n }\n\n const result = await callAiFn<AIDataExtractionResponse<T>>(\n msgs,\n AIActionType.EXTRACT_DATA,\n );\n return {\n parseResult: result.content,\n elementById,\n usage: result.usage,\n };\n}\n\nexport async function AiAssert<\n ElementType extends BaseElement = BaseElement,\n>(options: { assertion: TUserPrompt; context: UIContext<ElementType> }) {\n const { assertion, context } = options;\n\n assert(assertion, 'assertion should not be empty');\n\n const { screenshotBase64 } = context;\n\n const systemPrompt = systemPromptToAssert({\n isUITars: getAIConfigInBoolean(MIDSCENE_USE_VLM_UI_TARS),\n });\n\n const assertionText = extraTextFromUserPrompt(assertion);\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: `\nHere is the assertion. Please tell whether it is truthy according to the screenshot.\n=====================================\n${assertionText}\n=====================================\n `,\n },\n ],\n },\n ];\n\n if (typeof assertion !== 'string') {\n const addOns = await promptsToChatParam({\n images: assertion.images,\n convertHttpImage2Base64: assertion.convertHttpImage2Base64,\n });\n msgs.push(...addOns);\n }\n\n const { content: assertResult, usage } = await callAiFn<AIAssertionResponse>(\n msgs,\n AIActionType.ASSERT,\n );\n return {\n content: assertResult,\n 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","callAI","screenshotBase64","description","elementById","insertElementByPosition","describeUserPage","assert","userInstructionPrompt","findElementPrompt","systemPrompt","systemPromptToLocateElement","vlLocateMode","imagePayload","paddingToMatchBlockByBase64","markupImageForLLM","addOns","callAIFn","callToGetJSONObject","res","AIActionType","rawResponse","JSON","resRect","matchedElements","errors","Array","_options_searchConfig_rect","_options_searchConfig_rect1","_options_searchConfig_rect2","_options_searchConfig_rect3","adaptBboxToRect","rectCenter","element","elementByPositionWithElementInfo","distanceToCenter","distance","distanceThreshold","e","msg","Error","undefined","AiLocateSection","sectionDescription","systemPromptToLocateSection","sectionLocatorInstructionText","sectionLocatorInstruction","result","callAiFn","sectionRect","sectionBbox","targetRect","referenceBboxList","referenceRects","bbox","mergedRect","mergeRects","expandSearchArea","imageBase64","cropByRect","getAIConfigInBoolean","MIDSCENE_USE_QWEN_VL","AiExtractElementInfo","_options_extractOption","dataQuery","extractOption","systemPromptToExtract","extractDataPromptText","extractDataQueryPrompt","userContent","AiAssert","assertion","systemPromptToAssert","MIDSCENE_USE_VLM_UI_TARS","assertionText","assertResult","usage"],"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;;;;;;;;;;;;;;;;;;;;;AC6DA,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,OAMD;IAQC,MAAM,EAAEC,OAAO,EAAEC,wBAAwB,EAAEC,MAAM,EAAE,GAAGH;IACtD,MAAM,EAAEI,gBAAgB,EAAE,GAAGH;IAC7B,MAAM,EAAEI,WAAW,EAAEC,WAAW,EAAEC,uBAAuB,EAAE,GACzD,MAAMC,AAAAA,IAAAA,wBAAAA,gBAAAA,AAAAA,EAAiBP;IAEzBQ,IAAAA,sBAAAA,MAAAA,AAAAA,EACEP,0BACA;IAGF,MAAMQ,wBAAwB,MAAMC,+BAAAA,iBAAAA,CAAAA,MAAwB,CAAC;QAC3D,iBAAiBN;QACjB,0BAA0Bf,wBAAwBY;IACpD;IACA,MAAMU,eAAeC,AAAAA,IAAAA,+BAAAA,2BAAAA,AAAAA,EAA4BC,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA;IAEjD,IAAIC,eAAeX;IAEnB,IAAIJ,QAAQ,YAAY,EAAE;QACxBS,IAAAA,sBAAAA,MAAAA,AAAAA,EACET,QAAQ,YAAY,CAAC,IAAI,EACzB;QAEFS,IAAAA,sBAAAA,MAAAA,AAAAA,EACET,QAAQ,YAAY,CAAC,WAAW,EAChC;QAGFe,eAAef,QAAQ,YAAY,CAAC,WAAW;IACjD,OAAO,IAAIc,AAAmB,cAAnBA,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,KACTC,eAAe,MAAMC,AAAAA,IAAAA,oBAAAA,2BAAAA,AAAAA,EAA4BD;SAC5C,IAAI,CAACD,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,KACVC,eAAe,MAAME,AAAAA,IAAAA,mCAAAA,iBAAAA,AAAAA,EACnBb,kBACAH,QAAQ,IAAI,EACZA,QAAQ,IAAI;IAIhB,MAAMN,OAAe;QACnB;YAAE,MAAM;YAAU,SAASiB;QAAa;QACxC;YACE,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,WAAW;wBACT,KAAKG;wBACL,QAAQ;oBACV;gBACF;gBACA;oBACE,MAAM;oBACN,MAAML;gBACR;aACD;QACH;KACD;IAED,IAAI,AAAoC,YAApC,OAAOR,0BAAuC;QAChD,MAAMgB,SAAS,MAAM1B,mBAAmB;YACtC,QAAQU,yBAAyB,MAAM;YACvC,yBAAyBA,yBAAyB,uBAAuB;QAC3E;QACAP,KAAK,IAAI,IAAIuB;IACf;IAEA,MAAMC,WACJhB,UAAUiB,yBAAAA,mBAAmBA;IAE/B,MAAMC,MAAM,MAAMF,SAASxB,MAAM2B,mCAAAA,YAAAA,CAAAA,eAA4B;IAE7D,MAAMC,cAAcC,KAAK,SAAS,CAACH,IAAI,OAAO;IAE9C,IAAII;IACJ,IAAIC,kBACF,cAAcL,IAAI,OAAO,GAAGA,IAAI,OAAO,CAAC,QAAQ,GAAG,EAAE;IACvD,IAAIM,SACF,YAAYN,IAAI,OAAO,GAAGA,IAAI,OAAO,CAAC,MAAM,GAAG,EAAE;IACnD,IAAI;QACF,IAAI,UAAUA,IAAI,OAAO,IAAIO,MAAM,OAAO,CAACP,IAAI,OAAO,CAAC,IAAI,GAAG;gBAG1DQ,4BAAAA,uBACAC,6BAAAA,wBACAC,6BAAAA,wBACAC,6BAAAA;YALFP,UAAUQ,AAAAA,IAAAA,mCAAAA,eAAAA,AAAAA,EACRZ,IAAI,OAAO,CAAC,IAAI,EAChBQ,AAAAA,SAAAA,CAAAA,wBAAAA,QAAQ,YAAY,AAAD,IAAnBA,KAAAA,IAAAA,QAAAA,CAAAA,6BAAAA,sBAAsB,IAAI,AAAD,IAAzBA,KAAAA,IAAAA,2BAA4B,KAAK,AAAD,KAAK5B,QAAQ,IAAI,CAAC,KAAK,EACvD6B,AAAAA,SAAAA,CAAAA,yBAAAA,QAAQ,YAAY,AAAD,IAAnBA,KAAAA,IAAAA,QAAAA,CAAAA,8BAAAA,uBAAsB,IAAI,AAAD,IAAzBA,KAAAA,IAAAA,4BAA4B,MAAM,AAAD,KAAK7B,QAAQ,IAAI,CAAC,MAAM,UACzD8B,CAAAA,yBAAAA,QAAQ,YAAY,AAAD,IAAnBA,KAAAA,IAAAA,QAAAA,CAAAA,8BAAAA,uBAAsB,IAAI,AAAD,IAAzBA,KAAAA,IAAAA,4BAA4B,IAAI,UAChCC,CAAAA,yBAAAA,QAAQ,YAAY,AAAD,IAAnBA,KAAAA,IAAAA,QAAAA,CAAAA,8BAAAA,uBAAsB,IAAI,AAAD,IAAzBA,KAAAA,IAAAA,4BAA4B,GAAG;YAEjC7C,aAAa,WAAWsC;YAExB,MAAMS,aAAa;gBACjB,GAAGT,QAAQ,IAAI,GAAGA,QAAQ,KAAK,GAAG;gBAClC,GAAGA,QAAQ,GAAG,GAAGA,QAAQ,MAAM,GAAG;YACpC;YACA,IAAIU,UAAUC,AAAAA,IAAAA,wBAAAA,gCAAAA,AAAAA,EAAiCnC,QAAQ,IAAI,EAAEiC;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,UAAU5B,wBAAwB2B;YAGpC,IAAIC,SAAS;gBACXT,kBAAkB;oBAACS;iBAAQ;gBAC3BR,SAAS,EAAE;YACb;QACF;IACF,EAAE,OAAOa,GAAG;QACV,MAAMC,MACJD,aAAaE,QACT,CAAC,sBAAsB,EAAEF,EAAE,OAAO,EAAE,GACpC;QACN,IAAI,AAACb,UAAUA,AAAAA,CAAAA,QAAAA,SAAAA,KAAAA,IAAAA,OAAQ,MAAM,AAAD,MAAM,GAGhCA,OAAO,IAAI,CAAC,CAAC,CAAC,EAAEc,IAAI,CAAC,CAAC;aAFtBd,SAAS;YAACc;SAAI;IAIlB;IAEA,OAAO;QACL,MAAMhB;QACN,aAAa;YACX,UAAUC;YACVC;QACF;QACAJ;QACAjB;QACA,OAAOe,IAAI,KAAK;QAChB,kBACE,AAAuB,YAAvB,OAAOA,IAAI,OAAO,IAClBA,AAAgB,SAAhBA,IAAI,OAAO,IACX,sBAAsBA,IAAI,OAAO,GAC5BA,IAAI,OAAO,CAAS,gBAAgB,GACrCsB;IACR;AACF;AAEO,eAAeC,gBAAgB5C,OAIrC;IAOC,MAAM,EAAEC,OAAO,EAAE4C,kBAAkB,EAAE,GAAG7C;IACxC,MAAM,EAAEI,gBAAgB,EAAE,GAAGH;IAE7B,MAAMW,eAAekC,AAAAA,IAAAA,uCAAAA,2BAAAA,AAAAA,EAA4BhC,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA;IACjD,MAAMiC,gCAAgC,MAAMC,uCAAAA,yBAAAA,CAAAA,MAAgC,CAAC;QAC3E,oBAAoB1D,wBAAwBuD;IAC9C;IACA,MAAMlD,OAAe;QACnB;YAAE,MAAM;YAAU,SAASiB;QAAa;QACxC;YACE,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,WAAW;wBACT,KAAKR;wBACL,QAAQ;oBACV;gBACF;gBACA;oBACE,MAAM;oBACN,MAAM2C;gBACR;aACD;QACH;KACD;IAED,IAAI,AAA8B,YAA9B,OAAOF,oBAAiC;QAC1C,MAAM3B,SAAS,MAAM1B,mBAAmB;YACtC,QAAQqD,mBAAmB,MAAM;YACjC,yBAAyBA,mBAAmB,uBAAuB;QACrE;QACAlD,KAAK,IAAI,IAAIuB;IACf;IAEA,MAAM+B,SAAS,MAAMC,AAAAA,IAAAA,mCAAAA,QAAAA,AAAAA,EACnBvD,MACA2B,mCAAAA,YAAAA,CAAAA,YAAyB;IAG3B,IAAI6B;IACJ,MAAMC,cAAcH,OAAO,OAAO,CAAC,IAAI;IACvC,IAAIG,aAAa;QACf,MAAMC,aAAapB,AAAAA,IAAAA,mCAAAA,eAAAA,AAAAA,EACjBmB,aACAnD,QAAQ,IAAI,CAAC,KAAK,EAClBA,QAAQ,IAAI,CAAC,MAAM;QAErBZ,aAAa,0BAA0BgE;QAEvC,MAAMC,oBAAoBL,OAAO,OAAO,CAAC,eAAe,IAAI,EAAE;QAC9D5D,aAAa,wBAAwBiE;QAErC,MAAMC,iBAAiBD,kBACpB,MAAM,CAAC,CAACE,OAAS5B,MAAM,OAAO,CAAC4B,OAC/B,GAAG,CAAC,CAACA,OACGvB,AAAAA,IAAAA,mCAAAA,eAAAA,AAAAA,EAAgBuB,MAAMvD,QAAQ,IAAI,CAAC,KAAK,EAAEA,QAAQ,IAAI,CAAC,MAAM;QAExEZ,aAAa,qBAAqBkE;QAGlC,MAAME,aAAaC,AAAAA,IAAAA,mCAAAA,UAAAA,AAAAA,EAAW;YAACL;eAAeE;SAAe;QAC7DlE,aAAa,iBAAiBoE;QAG9BN,cAAcQ,AAAAA,IAAAA,mCAAAA,gBAAAA,AAAAA,EAAiBF,YAAYxD,QAAQ,IAAI;QACvDZ,aAAa,2BAA2B8D;IAC1C;IAEA,IAAIS,cAAcxD;IAClB,IAAI+C,aACFS,cAAc,MAAMC,AAAAA,IAAAA,oBAAAA,UAAAA,AAAAA,EAClBzD,kBACA+C,aACAW,AAAAA,IAAAA,oBAAAA,oBAAAA,AAAAA,EAAqBC,oBAAAA,oBAAoBA;IAI7C,OAAO;QACL,MAAMZ;QACNS;QACA,OAAOX,OAAO,OAAO,CAAC,KAAK;QAC3B,aAAazB,KAAK,SAAS,CAACyB,OAAO,OAAO;QAC1C,OAAOA,OAAO,KAAK;IACrB;AACF;AAEO,eAAee,qBAGpBhE,OAKD;QA0CKiE;IAzCJ,MAAM,EAAEC,SAAS,EAAEjE,OAAO,EAAEkE,aAAa,EAAE1E,gBAAgB,EAAE,GAAGO;IAChE,MAAMY,eAAewD,AAAAA,IAAAA,8BAAAA,qBAAAA,AAAAA;IAErB,MAAM,EAAEhE,gBAAgB,EAAE,GAAGH;IAC7B,MAAM,EAAEI,WAAW,EAAEC,WAAW,EAAE,GAAG,MAAME,AAAAA,IAAAA,wBAAAA,gBAAAA,AAAAA,EAAiBP,SAAS;QACnE,oBAAoB;QACpB,sBAAsB;QACtB,aAAa;QACb,aAAakE,QAAAA,gBAAAA,KAAAA,IAAAA,cAAe,WAAW;IACzC;IAEA,MAAME,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,MAAM1E,OAAe;QACnB;YAAE,MAAM;YAAU,SAASiB;QAAa;QACxC;YACE,MAAM;YACN,SAAS2D;QACX;KACD;IAED,IAAI,QAAAN,CAAAA,yBAAAA,QAAQ,aAAa,AAAD,IAApBA,KAAAA,IAAAA,uBAAuB,aAAa,EACtCtE,KAAK,IAAI,CAAC;QACR,MAAM;QACN,SAAS;IACX;IAGF,IAAIF,kBAAkB;QACpB,MAAMyB,SAAS,MAAM1B,mBAAmB;YACtC,QAAQC,iBAAiB,MAAM;YAC/B,yBAAyBA,iBAAiB,uBAAuB;QACnE;QACAE,KAAK,IAAI,IAAIuB;IACf;IAEA,MAAM+B,SAAS,MAAMC,AAAAA,IAAAA,mCAAAA,QAAAA,AAAAA,EACnBvD,MACA2B,mCAAAA,YAAAA,CAAAA,YAAyB;IAE3B,OAAO;QACL,aAAa2B,OAAO,OAAO;QAC3B3C;QACA,OAAO2C,OAAO,KAAK;IACrB;AACF;AAEO,eAAeuB,SAEpBxE,OAAoE;IACpE,MAAM,EAAEyE,SAAS,EAAExE,OAAO,EAAE,GAAGD;IAE/BS,IAAAA,sBAAAA,MAAAA,AAAAA,EAAOgE,WAAW;IAElB,MAAM,EAAErE,gBAAgB,EAAE,GAAGH;IAE7B,MAAMW,eAAe8D,AAAAA,IAAAA,6BAAAA,oBAAAA,AAAAA,EAAqB;QACxC,UAAUZ,AAAAA,IAAAA,oBAAAA,oBAAAA,AAAAA,EAAqBa,oBAAAA,wBAAwBA;IACzD;IAEA,MAAMC,gBAAgBtF,wBAAwBmF;IAE9C,MAAM9E,OAAe;QACnB;YAAE,MAAM;YAAU,SAASiB;QAAa;QACxC;YACE,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,WAAW;wBACT,KAAKR;wBACL,QAAQ;oBACV;gBACF;gBACA;oBACE,MAAM;oBACN,MAAM,CAAC;;;AAGjB,EAAEwE,cAAc;;EAEd,CAAC;gBACK;aACD;QACH;KACD;IAED,IAAI,AAAqB,YAArB,OAAOH,WAAwB;QACjC,MAAMvD,SAAS,MAAM1B,mBAAmB;YACtC,QAAQiF,UAAU,MAAM;YACxB,yBAAyBA,UAAU,uBAAuB;QAC5D;QACA9E,KAAK,IAAI,IAAIuB;IACf;IAEA,MAAM,EAAE,SAAS2D,YAAY,EAAEC,KAAK,EAAE,GAAG,MAAM5B,AAAAA,IAAAA,mCAAAA,QAAAA,AAAAA,EAC7CvD,MACA2B,mCAAAA,YAAAA,CAAAA,MAAmB;IAErB,OAAO;QACL,SAASuD;QACTC;IACF;AACF"}
|
|
@@ -84,7 +84,7 @@ async function plan(userInstruction, opts) {
|
|
|
84
84
|
actions,
|
|
85
85
|
rawResponse,
|
|
86
86
|
usage,
|
|
87
|
-
yamlFlow: (0, external_common_js_namespaceObject.buildYamlFlowFromPlans)(actions,
|
|
87
|
+
yamlFlow: (0, external_common_js_namespaceObject.buildYamlFlowFromPlans)(actions, planFromAI.sleep)
|
|
88
88
|
};
|
|
89
89
|
(0, utils_namespaceObject.assert)(planFromAI, "can't get plans from AI");
|
|
90
90
|
if ((0, env_namespaceObject.vlLocateMode)()) {
|
|
@@ -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 PageType,\n PlanningAIResponse,\n UIContext,\n} from '@/types';\nimport { vlLocateMode } from '@midscene/shared/env';\nimport { paddingToMatchBlockByBase64 } from '@midscene/shared/img';\nimport { assert } from '@midscene/shared/utils';\nimport {\n AIActionType,\n type AIArgs,\n buildYamlFlowFromPlans,\n callAiFn,\n fillBboxParam,\n markupImageForLLM,\n warnGPT4oSizeLimit,\n} from './common';\nimport {\n automationUserPrompt,\n generateTaskBackgroundContext,\n systemPromptToTaskPlanning,\n} from './prompt/llm-planning';\nimport { describeUserPage } from './prompt/util';\n\nexport async function plan(\n userInstruction: string,\n opts: {\n context: UIContext;\n pageType: PageType;\n actionSpace: DeviceAction[];\n callAI?: typeof callAiFn<PlanningAIResponse>;\n log?: string;\n actionContext?: string;\n },\n): Promise<PlanningAIResponse> {\n const { callAI, context } = opts || {};\n const { screenshotBase64, size } = context;\n const { description: pageDescription, elementById } =\n await describeUserPage(context);\n\n const systemPrompt = await systemPromptToTaskPlanning({\n actionSpace: opts.actionSpace,\n vlMode: vlLocateMode(),\n });\n const taskBackgroundContextText = generateTaskBackgroundContext(\n userInstruction,\n opts.log,\n opts.actionContext,\n );\n const userInstructionPrompt = await automationUserPrompt(\n vlLocateMode(),\n ).format({\n pageDescription,\n taskBackgroundContext: taskBackgroundContextText,\n });\n\n let imagePayload = screenshotBase64;\n if (vlLocateMode() === 'qwen-vl') {\n imagePayload = await paddingToMatchBlockByBase64(imagePayload);\n } else if (!vlLocateMode()) {\n imagePayload = await markupImageForLLM(\n screenshotBase64,\n context.tree,\n context.size,\n );\n }\n\n warnGPT4oSizeLimit(size);\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 const call = callAI || callAiFn;\n const { content, usage } = await call(msgs, AIActionType.PLAN);\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(
|
|
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 PageType,\n PlanningAIResponse,\n UIContext,\n} from '@/types';\nimport { vlLocateMode } from '@midscene/shared/env';\nimport { paddingToMatchBlockByBase64 } from '@midscene/shared/img';\nimport { assert } from '@midscene/shared/utils';\nimport {\n AIActionType,\n type AIArgs,\n buildYamlFlowFromPlans,\n callAiFn,\n fillBboxParam,\n markupImageForLLM,\n warnGPT4oSizeLimit,\n} from './common';\nimport {\n automationUserPrompt,\n generateTaskBackgroundContext,\n systemPromptToTaskPlanning,\n} from './prompt/llm-planning';\nimport { describeUserPage } from './prompt/util';\n\nexport async function plan(\n userInstruction: string,\n opts: {\n context: UIContext;\n pageType: PageType;\n actionSpace: DeviceAction[];\n callAI?: typeof callAiFn<PlanningAIResponse>;\n log?: string;\n actionContext?: string;\n },\n): Promise<PlanningAIResponse> {\n const { callAI, context } = opts || {};\n const { screenshotBase64, size } = context;\n const { description: pageDescription, elementById } =\n await describeUserPage(context);\n\n const systemPrompt = await systemPromptToTaskPlanning({\n actionSpace: opts.actionSpace,\n vlMode: vlLocateMode(),\n });\n const taskBackgroundContextText = generateTaskBackgroundContext(\n userInstruction,\n opts.log,\n opts.actionContext,\n );\n const userInstructionPrompt = await automationUserPrompt(\n vlLocateMode(),\n ).format({\n pageDescription,\n taskBackgroundContext: taskBackgroundContextText,\n });\n\n let imagePayload = screenshotBase64;\n if (vlLocateMode() === 'qwen-vl') {\n imagePayload = await paddingToMatchBlockByBase64(imagePayload);\n } else if (!vlLocateMode()) {\n imagePayload = await markupImageForLLM(\n screenshotBase64,\n context.tree,\n context.size,\n );\n }\n\n warnGPT4oSizeLimit(size);\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 const call = callAI || callAiFn;\n const { content, usage } = await call(msgs, AIActionType.PLAN);\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(actions, planFromAI.sleep),\n };\n\n assert(planFromAI, \"can't get plans from AI\");\n\n if (vlLocateMode()) {\n actions.forEach((action) => {\n if (action.locate) {\n try {\n action.locate = fillBboxParam(action.locate, size.width, size.height);\n } catch (e) {\n throw new Error(\n `Failed to fill locate param: ${planFromAI.error} (${\n e instanceof Error ? e.message : 'unknown error'\n })`,\n {\n cause: e,\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 } else {\n actions.forEach((action) => {\n if (action.locate?.id) {\n // The model may return indexId, need to perform a query correction to avoid exceptions\n const element = elementById(action.locate.id);\n if (element) {\n action.locate.id = element.id;\n }\n }\n });\n }\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 return returnValue;\n}\n"],"names":["__webpack_require__","definition","key","Object","obj","prop","Symbol","plan","userInstruction","opts","_planFromAI_action","callAI","context","screenshotBase64","size","pageDescription","elementById","describeUserPage","systemPrompt","systemPromptToTaskPlanning","vlLocateMode","taskBackgroundContextText","generateTaskBackgroundContext","userInstructionPrompt","automationUserPrompt","imagePayload","paddingToMatchBlockByBase64","markupImageForLLM","warnGPT4oSizeLimit","msgs","call","callAiFn","content","usage","AIActionType","rawResponse","JSON","undefined","planFromAI","actions","returnValue","buildYamlFlowFromPlans","assert","action","fillBboxParam","e","Error","_action_locate","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;;;;;;;;;;;;;ACmBO,eAAeI,KACpBC,eAAuB,EACvBC,IAOC;QA8DEC;IA5DH,MAAM,EAAEC,MAAM,EAAEC,OAAO,EAAE,GAAGH,QAAQ,CAAC;IACrC,MAAM,EAAEI,gBAAgB,EAAEC,IAAI,EAAE,GAAGF;IACnC,MAAM,EAAE,aAAaG,eAAe,EAAEC,WAAW,EAAE,GACjD,MAAMC,AAAAA,IAAAA,wBAAAA,gBAAAA,AAAAA,EAAiBL;IAEzB,MAAMM,eAAe,MAAMC,AAAAA,IAAAA,gCAAAA,0BAAAA,AAAAA,EAA2B;QACpD,aAAaV,KAAK,WAAW;QAC7B,QAAQW,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA;IACV;IACA,MAAMC,4BAA4BC,AAAAA,IAAAA,gCAAAA,6BAAAA,AAAAA,EAChCd,iBACAC,KAAK,GAAG,EACRA,KAAK,aAAa;IAEpB,MAAMc,wBAAwB,MAAMC,AAAAA,IAAAA,gCAAAA,oBAAAA,AAAAA,EAClCJ,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,KACA,MAAM,CAAC;QACPL;QACA,uBAAuBM;IACzB;IAEA,IAAII,eAAeZ;IACnB,IAAIO,AAAmB,cAAnBA,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,KACFK,eAAe,MAAMC,AAAAA,IAAAA,oBAAAA,2BAAAA,AAAAA,EAA4BD;SAC5C,IAAI,CAACL,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,KACVK,eAAe,MAAME,AAAAA,IAAAA,mCAAAA,iBAAAA,AAAAA,EACnBd,kBACAD,QAAQ,IAAI,EACZA,QAAQ,IAAI;IAIhBgB,IAAAA,mCAAAA,kBAAAA,AAAAA,EAAmBd;IAEnB,MAAMe,OAAe;QACnB;YAAE,MAAM;YAAU,SAASX;QAAa;QACxC;YACE,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,WAAW;wBACT,KAAKO;wBACL,QAAQ;oBACV;gBACF;gBACA;oBACE,MAAM;oBACN,MAAMF;gBACR;aACD;QACH;KACD;IAED,MAAMO,OAAOnB,UAAUoB,mCAAAA,QAAQA;IAC/B,MAAM,EAAEC,OAAO,EAAEC,KAAK,EAAE,GAAG,MAAMH,KAAKD,MAAMK,mCAAAA,YAAAA,CAAAA,IAAiB;IAC7D,MAAMC,cAAcC,KAAK,SAAS,CAACJ,SAASK,QAAW;IACvD,MAAMC,aAAaN;IAEnB,MAAMO,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;QACAF;QACA,UAAUQ,AAAAA,IAAAA,mCAAAA,sBAAAA,AAAAA,EAAuBF,SAASD,WAAW,KAAK;IAC5D;IAEAI,IAAAA,sBAAAA,MAAAA,AAAAA,EAAOJ,YAAY;IAEnB,IAAIlB,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,KAAgB;QAClBmB,QAAQ,OAAO,CAAC,CAACI;YACf,IAAIA,OAAO,MAAM,EACf,IAAI;gBACFA,OAAO,MAAM,GAAGC,AAAAA,IAAAA,mCAAAA,aAAAA,AAAAA,EAAcD,OAAO,MAAM,EAAE7B,KAAK,KAAK,EAAEA,KAAK,MAAM;YACtE,EAAE,OAAO+B,GAAG;gBACV,MAAM,IAAIC,MACR,CAAC,6BAA6B,EAAER,WAAW,KAAK,CAAC,EAAE,EACjDO,aAAaC,QAAQD,EAAE,OAAO,GAAG,gBAClC,CAAC,CAAC,EACH;oBACE,OAAOA;gBACT;YAEJ;QAEJ;QAEAH,IAAAA,sBAAAA,MAAAA,AAAAA,EAAO,CAACJ,WAAW,KAAK,EAAE,CAAC,wBAAwB,EAAEA,WAAW,KAAK,EAAE;IACzE,OACEC,QAAQ,OAAO,CAAC,CAACI;YACXI;QAAJ,IAAI,QAAAA,CAAAA,iBAAAA,OAAO,MAAM,AAAD,IAAZA,KAAAA,IAAAA,eAAe,EAAE,EAAE;YAErB,MAAMC,UAAUhC,YAAY2B,OAAO,MAAM,CAAC,EAAE;YAC5C,IAAIK,SACFL,OAAO,MAAM,CAAC,EAAE,GAAGK,QAAQ,EAAE;QAEjC;IACF;IAGF,IACET,AAAmB,MAAnBA,QAAQ,MAAM,IACdC,YAAY,kCAAkC,IAC9C,CAACA,YAAY,KAAK,EAElBS,QAAQ,IAAI,CACV,8EACAzC;IAIJ,OAAOgC;AACT"}
|
|
@@ -24,8 +24,33 @@ var __webpack_require__ = {};
|
|
|
24
24
|
var __webpack_exports__ = {};
|
|
25
25
|
__webpack_require__.r(__webpack_exports__);
|
|
26
26
|
__webpack_require__.d(__webpack_exports__, {
|
|
27
|
-
assertSchema: ()=>assertSchema
|
|
27
|
+
assertSchema: ()=>assertSchema,
|
|
28
|
+
systemPromptToAssert: ()=>systemPromptToAssert
|
|
28
29
|
});
|
|
30
|
+
const env_namespaceObject = require("@midscene/shared/env");
|
|
31
|
+
const defaultAssertionPrompt = 'You are a senior testing engineer. User will give an assertion and a screenshot of a page. By carefully viewing the screenshot, please tell whether the assertion is truthy.';
|
|
32
|
+
const defaultAssertionResponseJsonFormat = `Return in the following JSON format:
|
|
33
|
+
{
|
|
34
|
+
pass: boolean, // whether the assertion is truthy
|
|
35
|
+
thought: string | null, // string, if the result is falsy, give the reason why it is falsy. Otherwise, put null.
|
|
36
|
+
}`;
|
|
37
|
+
const getUiTarsAssertionResponseJsonFormat = ()=>`## Output Json String Format
|
|
38
|
+
\`\`\`
|
|
39
|
+
"{
|
|
40
|
+
"pass": <<is a boolean value from the enum [true, false], true means the assertion is truthy>>,
|
|
41
|
+
"thought": "<<is a string, give the reason why the assertion is falsy or truthy. Otherwise.>>"
|
|
42
|
+
}"
|
|
43
|
+
\`\`\`
|
|
44
|
+
|
|
45
|
+
## Rules **MUST** follow
|
|
46
|
+
- Make sure to return **only** the JSON, with **no additional** text or explanations.
|
|
47
|
+
- Use ${(0, env_namespaceObject.getPreferredLanguage)()} in \`thought\` part.
|
|
48
|
+
- You **MUST** strictly follow up the **Output Json String Format**.`;
|
|
49
|
+
function systemPromptToAssert(model) {
|
|
50
|
+
return `${defaultAssertionPrompt}
|
|
51
|
+
|
|
52
|
+
${model.isUITars ? getUiTarsAssertionResponseJsonFormat() : defaultAssertionResponseJsonFormat}`;
|
|
53
|
+
}
|
|
29
54
|
const assertSchema = {
|
|
30
55
|
type: 'json_schema',
|
|
31
56
|
json_schema: {
|
|
@@ -55,8 +80,10 @@ const assertSchema = {
|
|
|
55
80
|
}
|
|
56
81
|
};
|
|
57
82
|
exports.assertSchema = __webpack_exports__.assertSchema;
|
|
83
|
+
exports.systemPromptToAssert = __webpack_exports__.systemPromptToAssert;
|
|
58
84
|
for(var __webpack_i__ in __webpack_exports__)if (-1 === [
|
|
59
|
-
"assertSchema"
|
|
85
|
+
"assertSchema",
|
|
86
|
+
"systemPromptToAssert"
|
|
60
87
|
].indexOf(__webpack_i__)) exports[__webpack_i__] = __webpack_exports__[__webpack_i__];
|
|
61
88
|
Object.defineProperty(exports, '__esModule', {
|
|
62
89
|
value: true
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"ai-model/prompt/assertion.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/prompt/assertion.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 { ResponseFormatJSONSchema } from 'openai/resources/index';\n\nexport const assertSchema: ResponseFormatJSONSchema = {\n type: 'json_schema',\n json_schema: {\n name: 'assert',\n strict: true,\n schema: {\n type: 'object',\n properties: {\n pass: {\n type: 'boolean',\n description: 'Whether the assertion passed or failed',\n },\n thought: {\n type: ['string', 'null'],\n description: 'The thought process behind the assertion',\n },\n },\n required: ['pass', 'thought'],\n additionalProperties: false,\n },\n },\n};\n"],"names":["__webpack_require__","definition","key","Object","obj","prop","Symbol","assertSchema"],"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
|
|
1
|
+
{"version":3,"file":"ai-model/prompt/assertion.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/prompt/assertion.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 { getPreferredLanguage } from '@midscene/shared/env';\nimport type { ResponseFormatJSONSchema } from 'openai/resources/index';\n\nconst defaultAssertionPrompt =\n 'You are a senior testing engineer. User will give an assertion and a screenshot of a page. By carefully viewing the screenshot, please tell whether the assertion is truthy.';\n\nconst defaultAssertionResponseJsonFormat = `Return in the following JSON format:\n{\n pass: boolean, // whether the assertion is truthy\n thought: string | null, // string, if the result is falsy, give the reason why it is falsy. Otherwise, put null.\n}`;\n\nconst getUiTarsAssertionResponseJsonFormat = () => `## Output Json String Format\n\\`\\`\\`\n\"{\n \"pass\": <<is a boolean value from the enum [true, false], true means the assertion is truthy>>, \n \"thought\": \"<<is a string, give the reason why the assertion is falsy or truthy. Otherwise.>>\"\n}\"\n\\`\\`\\`\n\n## Rules **MUST** follow\n- Make sure to return **only** the JSON, with **no additional** text or explanations.\n- Use ${getPreferredLanguage()} in \\`thought\\` part.\n- You **MUST** strictly follow up the **Output Json String Format**.`;\n\nexport function systemPromptToAssert(model: { isUITars: boolean }) {\n return `${defaultAssertionPrompt}\n\n${model.isUITars ? getUiTarsAssertionResponseJsonFormat() : defaultAssertionResponseJsonFormat}`;\n}\n\nexport const assertSchema: ResponseFormatJSONSchema = {\n type: 'json_schema',\n json_schema: {\n name: 'assert',\n strict: true,\n schema: {\n type: 'object',\n properties: {\n pass: {\n type: 'boolean',\n description: 'Whether the assertion passed or failed',\n },\n thought: {\n type: ['string', 'null'],\n description: 'The thought process behind the assertion',\n },\n },\n required: ['pass', 'thought'],\n additionalProperties: false,\n },\n },\n};\n"],"names":["__webpack_require__","definition","key","Object","obj","prop","Symbol","defaultAssertionPrompt","defaultAssertionResponseJsonFormat","getUiTarsAssertionResponseJsonFormat","getPreferredLanguage","systemPromptToAssert","model","assertSchema"],"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;;;;;;;;;ACHA,MAAMI,yBACJ;AAEF,MAAMC,qCAAqC,CAAC;;;;CAI3C,CAAC;AAEF,MAAMC,uCAAuC,IAAM,CAAC;;;;;;;;;;MAU9C,EAAEC,AAAAA,IAAAA,oBAAAA,oBAAAA,AAAAA,IAAuB;oEACqC,CAAC;AAE9D,SAASC,qBAAqBC,KAA4B;IAC/D,OAAO,GAAGL,uBAAuB;;AAEnC,EAAEK,MAAM,QAAQ,GAAGH,yCAAyCD,oCAAoC;AAChG;AAEO,MAAMK,eAAyC;IACpD,MAAM;IACN,aAAa;QACX,MAAM;QACN,QAAQ;QACR,QAAQ;YACN,MAAM;YACN,YAAY;gBACV,MAAM;oBACJ,MAAM;oBACN,aAAa;gBACf;gBACA,SAAS;oBACP,MAAM;wBAAC;wBAAU;qBAAO;oBACxB,aAAa;gBACf;YACF;YACA,UAAU;gBAAC;gBAAQ;aAAU;YAC7B,sBAAsB;QACxB;IACF;AACF"}
|
package/dist/lib/index.js
CHANGED
|
@@ -33,6 +33,7 @@ var __webpack_require__ = {};
|
|
|
33
33
|
var __webpack_exports__ = {};
|
|
34
34
|
__webpack_require__.r(__webpack_exports__);
|
|
35
35
|
__webpack_require__.d(__webpack_exports__, {
|
|
36
|
+
AiAssert: ()=>external_ai_model_index_js_namespaceObject.AiAssert,
|
|
36
37
|
MIDSCENE_MODEL_NAME: ()=>env_namespaceObject.MIDSCENE_MODEL_NAME,
|
|
37
38
|
default: ()=>src,
|
|
38
39
|
getVersion: ()=>external_utils_js_namespaceObject.getVersion,
|
|
@@ -50,6 +51,7 @@ const external_utils_js_namespaceObject = require("./utils.js");
|
|
|
50
51
|
const external_ai_model_index_js_namespaceObject = require("./ai-model/index.js");
|
|
51
52
|
const env_namespaceObject = require("@midscene/shared/env");
|
|
52
53
|
const src = index_js_default();
|
|
54
|
+
exports.AiAssert = __webpack_exports__.AiAssert;
|
|
53
55
|
exports.AiLocateElement = __webpack_exports__.AiLocateElement;
|
|
54
56
|
exports.Executor = __webpack_exports__.Executor;
|
|
55
57
|
exports.Insight = __webpack_exports__.Insight;
|
|
@@ -60,6 +62,7 @@ exports.getAIConfig = __webpack_exports__.getAIConfig;
|
|
|
60
62
|
exports.getVersion = __webpack_exports__.getVersion;
|
|
61
63
|
exports.plan = __webpack_exports__.plan;
|
|
62
64
|
for(var __webpack_i__ in __webpack_exports__)if (-1 === [
|
|
65
|
+
"AiAssert",
|
|
63
66
|
"AiLocateElement",
|
|
64
67
|
"Executor",
|
|
65
68
|
"Insight",
|
package/dist/lib/index.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.js","sources":["webpack://@midscene/core/webpack/runtime/compat_get_default_export","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/index.ts"],"sourcesContent":["// getDefaultExport function for compatibility with non-ESM modules\n__webpack_require__.n = (module) => {\n\tvar getter = module && module.__esModule ?\n\t\t() => (module['default']) :\n\t\t() => (module);\n\t__webpack_require__.d(getter, { a: getter });\n\treturn getter;\n};\n","__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 { Executor } from './ai-model/action-executor';\nimport Insight from './insight/index';\nimport { getVersion } from './utils';\n\nexport {\n plan,\n describeUserPage,\n AiLocateElement,\n} from './ai-model/index';\n\nexport { getAIConfig, MIDSCENE_MODEL_NAME } from '@midscene/shared/env';\n\nexport type * from './types';\nexport default Insight;\nexport { Executor, Insight, getVersion };\n\nexport type {\n MidsceneYamlScript,\n MidsceneYamlTask,\n MidsceneYamlFlowItem,\n MidsceneYamlFlowItemAIRightClick,\n MidsceneYamlConfigResult,\n LocateOption,\n DetailedLocateParam,\n} from './yaml';\n"],"names":["__webpack_require__","module","getter","definition","key","Object","obj","prop","Symbol","Insight"],"mappings":";;;IACAA,oBAAoB,CAAC,GAAG,CAACC;QACxB,IAAIC,SAASD,UAAUA,OAAO,UAAU,GACvC,IAAOA,MAAM,CAAC,UAAU,GACxB,IAAOA;QACRD,oBAAoB,CAAC,CAACE,QAAQ;YAAE,GAAGA;QAAO;QAC1C,OAAOA;IACR;;;ICPAF,oBAAoB,CAAC,GAAG,CAAC,UAASG;QACjC,IAAI,IAAIC,OAAOD,WACR,IAAGH,oBAAoB,CAAC,CAACG,YAAYC,QAAQ,CAACJ,oBAAoB,CAAC,CAAC,UAASI,MACzEC,OAAO,cAAc,CAAC,UAASD,KAAK;YAAE,YAAY;YAAM,KAAKD,UAAU,CAACC,IAAI;QAAC;IAGzF;;;ICNAJ,oBAAoB,CAAC,GAAG,CAACM,KAAKC,OAAUF,OAAO,SAAS,CAAC,cAAc,CAAC,IAAI,CAACC,KAAKC;;;ICClFP,oBAAoB,CAAC,GAAG,CAAC;QACxB,IAAG,AAAkB,eAAlB,OAAOQ,UAA0BA,OAAO,WAAW,EACrDH,OAAO,cAAc,CAAC,UAASG,OAAO,WAAW,EAAE;YAAE,OAAO;QAAS;QAEtEH,OAAO,cAAc,CAAC,UAAS,cAAc;YAAE,OAAO;QAAK;IAC5D
|
|
1
|
+
{"version":3,"file":"index.js","sources":["webpack://@midscene/core/webpack/runtime/compat_get_default_export","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/index.ts"],"sourcesContent":["// getDefaultExport function for compatibility with non-ESM modules\n__webpack_require__.n = (module) => {\n\tvar getter = module && module.__esModule ?\n\t\t() => (module['default']) :\n\t\t() => (module);\n\t__webpack_require__.d(getter, { a: getter });\n\treturn getter;\n};\n","__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 { Executor } from './ai-model/action-executor';\nimport Insight from './insight/index';\nimport { getVersion } from './utils';\n\nexport {\n plan,\n describeUserPage,\n AiLocateElement,\n AiAssert,\n} from './ai-model/index';\n\nexport { getAIConfig, MIDSCENE_MODEL_NAME } from '@midscene/shared/env';\n\nexport type * from './types';\nexport default Insight;\nexport { Executor, Insight, getVersion };\n\nexport type {\n MidsceneYamlScript,\n MidsceneYamlTask,\n MidsceneYamlFlowItem,\n MidsceneYamlFlowItemAIRightClick,\n MidsceneYamlConfigResult,\n LocateOption,\n DetailedLocateParam,\n} from './yaml';\n"],"names":["__webpack_require__","module","getter","definition","key","Object","obj","prop","Symbol","Insight"],"mappings":";;;IACAA,oBAAoB,CAAC,GAAG,CAACC;QACxB,IAAIC,SAASD,UAAUA,OAAO,UAAU,GACvC,IAAOA,MAAM,CAAC,UAAU,GACxB,IAAOA;QACRD,oBAAoB,CAAC,CAACE,QAAQ;YAAE,GAAGA;QAAO;QAC1C,OAAOA;IACR;;;ICPAF,oBAAoB,CAAC,GAAG,CAAC,UAASG;QACjC,IAAI,IAAIC,OAAOD,WACR,IAAGH,oBAAoB,CAAC,CAACG,YAAYC,QAAQ,CAACJ,oBAAoB,CAAC,CAAC,UAASI,MACzEC,OAAO,cAAc,CAAC,UAASD,KAAK;YAAE,YAAY;YAAM,KAAKD,UAAU,CAACC,IAAI;QAAC;IAGzF;;;ICNAJ,oBAAoB,CAAC,GAAG,CAACM,KAAKC,OAAUF,OAAO,SAAS,CAAC,cAAc,CAAC,IAAI,CAACC,KAAKC;;;ICClFP,oBAAoB,CAAC,GAAG,CAAC;QACxB,IAAG,AAAkB,eAAlB,OAAOQ,UAA0BA,OAAO,WAAW,EACrDH,OAAO,cAAc,CAAC,UAASG,OAAO,WAAW,EAAE;YAAE,OAAO;QAAS;QAEtEH,OAAO,cAAc,CAAC,UAAS,cAAc;YAAE,OAAO;QAAK;IAC5D;;;;;;;;;;;;;;;;;;;;;;ACQA,YAAeI"}
|