@midscene/core 1.12.5-beta-20260908034641.0 → 1.12.5
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/es/agent/agent.mjs +6 -1
- package/dist/es/agent/agent.mjs.map +1 -1
- package/dist/es/agent/progress/ai-act-progress.mjs +2 -23
- package/dist/es/agent/progress/ai-act-progress.mjs.map +1 -1
- package/dist/es/agent/task-builder.mjs +21 -24
- package/dist/es/agent/task-builder.mjs.map +1 -1
- package/dist/es/agent/utils.mjs +24 -23
- package/dist/es/agent/utils.mjs.map +1 -1
- package/dist/es/ai-model/connectivity/run-connectivity-test.mjs +1 -8
- package/dist/es/ai-model/connectivity/run-connectivity-test.mjs.map +1 -1
- package/dist/es/ai-model/model-adapter/codex-app-server.mjs +23 -0
- package/dist/es/ai-model/model-adapter/codex-app-server.mjs.map +1 -0
- package/dist/es/ai-model/model-adapter/custom-planning-action.mjs +3 -3
- package/dist/es/ai-model/model-adapter/custom-planning-action.mjs.map +1 -1
- package/dist/es/ai-model/model-adapter/resolve.mjs +3 -0
- package/dist/es/ai-model/model-adapter/resolve.mjs.map +1 -1
- package/dist/es/ai-model/models/auto-glm/locate.mjs +2 -2
- package/dist/es/ai-model/models/auto-glm/locate.mjs.map +1 -1
- package/dist/es/ai-model/models/deepseek/adapter.mjs +4 -2
- package/dist/es/ai-model/models/deepseek/adapter.mjs.map +1 -1
- package/dist/es/ai-model/models/doubao.mjs +4 -2
- package/dist/es/ai-model/models/doubao.mjs.map +1 -1
- package/dist/es/ai-model/models/gpt.mjs +60 -2
- package/dist/es/ai-model/models/gpt.mjs.map +1 -1
- package/dist/es/ai-model/models/kimi.mjs +4 -2
- package/dist/es/ai-model/models/kimi.mjs.map +1 -1
- package/dist/es/ai-model/models/qwen.mjs +8 -28
- package/dist/es/ai-model/models/qwen.mjs.map +1 -1
- package/dist/es/ai-model/models/ui-tars/adapter.mjs +4 -2
- package/dist/es/ai-model/models/ui-tars/adapter.mjs.map +1 -1
- package/dist/es/ai-model/service-caller/codex-app-server.mjs +7 -17
- package/dist/es/ai-model/service-caller/codex-app-server.mjs.map +1 -1
- package/dist/es/ai-model/service-caller/index.mjs +5 -4
- package/dist/es/ai-model/service-caller/index.mjs.map +1 -1
- package/dist/es/ai-model/shared/model-locate-result/coordinate-distance.mjs +2 -1
- package/dist/es/ai-model/shared/model-locate-result/coordinate-distance.mjs.map +1 -1
- package/dist/es/ai-model/shared/model-locate-result/factory.mjs +21 -21
- package/dist/es/ai-model/shared/model-locate-result/factory.mjs.map +1 -1
- package/dist/es/ai-model/shared/model-locate-result/pixel-mapper.mjs +87 -0
- package/dist/es/ai-model/shared/model-locate-result/pixel-mapper.mjs.map +1 -0
- package/dist/es/ai-model/shared/model-locate-result/types.mjs.map +1 -1
- package/dist/es/ai-model/shared/model-locate-result/validation.mjs +55 -0
- package/dist/es/ai-model/shared/model-locate-result/validation.mjs.map +1 -0
- package/dist/es/ai-model/workflows/grounding/locate.mjs +11 -13
- package/dist/es/ai-model/workflows/grounding/locate.mjs.map +1 -1
- package/dist/es/ai-model/workflows/grounding/planning-action-locate.mjs +3 -3
- package/dist/es/ai-model/workflows/grounding/planning-action-locate.mjs.map +1 -1
- package/dist/es/ai-model/workflows/grounding/search-area-mapping.mjs +27 -11
- package/dist/es/ai-model/workflows/grounding/search-area-mapping.mjs.map +1 -1
- package/dist/es/ai-model/workflows/grounding/search-area.mjs +26 -1
- package/dist/es/ai-model/workflows/grounding/search-area.mjs.map +1 -1
- package/dist/es/ai-model/workflows/grounding/section-locate.mjs +4 -8
- package/dist/es/ai-model/workflows/grounding/section-locate.mjs.map +1 -1
- package/dist/es/ai-model/workflows/planning/locate-normalization.mjs +2 -1
- package/dist/es/ai-model/workflows/planning/locate-normalization.mjs.map +1 -1
- package/dist/es/ai-model/workflows/planning/standard-planning.mjs.map +1 -1
- package/dist/es/common.mjs +2 -9
- package/dist/es/common.mjs.map +1 -1
- package/dist/es/element-describer.mjs +2 -2
- package/dist/es/element-describer.mjs.map +1 -1
- package/dist/es/report-html-template.mjs +1 -1
- package/dist/es/service/index.mjs +13 -11
- package/dist/es/service/index.mjs.map +1 -1
- package/dist/es/types.mjs.map +1 -1
- package/dist/es/utils.mjs +1 -1
- package/dist/lib/agent/agent.js +6 -1
- package/dist/lib/agent/agent.js.map +1 -1
- package/dist/lib/agent/progress/ai-act-progress.js +2 -23
- package/dist/lib/agent/progress/ai-act-progress.js.map +1 -1
- package/dist/lib/agent/task-builder.js +20 -23
- package/dist/lib/agent/task-builder.js.map +1 -1
- package/dist/lib/agent/utils.js +26 -28
- package/dist/lib/agent/utils.js.map +1 -1
- package/dist/lib/ai-model/connectivity/run-connectivity-test.js +1 -8
- package/dist/lib/ai-model/connectivity/run-connectivity-test.js.map +1 -1
- package/dist/lib/ai-model/{workflows/grounding/locate-result-rect.js → model-adapter/codex-app-server.js} +22 -25
- package/dist/lib/ai-model/model-adapter/codex-app-server.js.map +1 -0
- package/dist/lib/ai-model/model-adapter/custom-planning-action.js +5 -5
- package/dist/lib/ai-model/model-adapter/custom-planning-action.js.map +1 -1
- package/dist/lib/ai-model/model-adapter/resolve.js +3 -0
- package/dist/lib/ai-model/model-adapter/resolve.js.map +1 -1
- package/dist/lib/ai-model/models/auto-glm/locate.js +1 -1
- package/dist/lib/ai-model/models/auto-glm/locate.js.map +1 -1
- package/dist/lib/ai-model/models/deepseek/adapter.js +4 -2
- package/dist/lib/ai-model/models/deepseek/adapter.js.map +1 -1
- package/dist/lib/ai-model/models/doubao.js +4 -2
- package/dist/lib/ai-model/models/doubao.js.map +1 -1
- package/dist/lib/ai-model/models/gpt.js +60 -2
- package/dist/lib/ai-model/models/gpt.js.map +1 -1
- package/dist/lib/ai-model/models/kimi.js +4 -2
- package/dist/lib/ai-model/models/kimi.js.map +1 -1
- package/dist/lib/ai-model/models/qwen.js +7 -27
- package/dist/lib/ai-model/models/qwen.js.map +1 -1
- package/dist/lib/ai-model/models/ui-tars/adapter.js +4 -2
- package/dist/lib/ai-model/models/ui-tars/adapter.js.map +1 -1
- package/dist/lib/ai-model/service-caller/codex-app-server.js +7 -20
- package/dist/lib/ai-model/service-caller/codex-app-server.js.map +1 -1
- package/dist/lib/ai-model/service-caller/index.js +5 -4
- package/dist/lib/ai-model/service-caller/index.js.map +1 -1
- package/dist/lib/ai-model/shared/model-locate-result/coordinate-distance.js +2 -1
- package/dist/lib/ai-model/shared/model-locate-result/coordinate-distance.js.map +1 -1
- package/dist/lib/ai-model/shared/model-locate-result/factory.js +21 -21
- package/dist/lib/ai-model/shared/model-locate-result/factory.js.map +1 -1
- package/dist/lib/ai-model/shared/model-locate-result/pixel-mapper.js +139 -0
- package/dist/lib/ai-model/shared/model-locate-result/pixel-mapper.js.map +1 -0
- package/dist/lib/ai-model/shared/model-locate-result/types.js.map +1 -1
- package/dist/lib/ai-model/shared/model-locate-result/validation.js +92 -0
- package/dist/lib/ai-model/shared/model-locate-result/validation.js.map +1 -0
- package/dist/lib/ai-model/workflows/grounding/locate.js +10 -12
- package/dist/lib/ai-model/workflows/grounding/locate.js.map +1 -1
- package/dist/lib/ai-model/workflows/grounding/planning-action-locate.js +3 -3
- package/dist/lib/ai-model/workflows/grounding/planning-action-locate.js.map +1 -1
- package/dist/lib/ai-model/workflows/grounding/search-area-mapping.js +35 -13
- package/dist/lib/ai-model/workflows/grounding/search-area-mapping.js.map +1 -1
- package/dist/lib/ai-model/workflows/grounding/search-area.js +30 -2
- package/dist/lib/ai-model/workflows/grounding/search-area.js.map +1 -1
- package/dist/lib/ai-model/workflows/grounding/section-locate.js +3 -7
- package/dist/lib/ai-model/workflows/grounding/section-locate.js.map +1 -1
- package/dist/lib/ai-model/workflows/planning/locate-normalization.js +2 -1
- package/dist/lib/ai-model/workflows/planning/locate-normalization.js.map +1 -1
- package/dist/lib/ai-model/workflows/planning/standard-planning.js.map +1 -1
- package/dist/lib/common.js +2 -9
- package/dist/lib/common.js.map +1 -1
- package/dist/lib/element-describer.js +2 -2
- package/dist/lib/element-describer.js.map +1 -1
- package/dist/lib/report-html-template.js +1 -1
- package/dist/lib/service/index.js +13 -11
- package/dist/lib/service/index.js.map +1 -1
- package/dist/lib/types.js.map +1 -1
- package/dist/lib/utils.js +1 -1
- package/dist/types/agent/agent.d.ts +7 -12
- package/dist/types/agent/utils.d.ts +8 -9
- package/dist/types/ai-model/model-adapter/codex-app-server.d.ts +2 -0
- package/dist/types/ai-model/model-adapter/custom-planning-action.d.ts +3 -2
- package/dist/types/ai-model/model-adapter/resolve.d.ts +2 -1
- package/dist/types/ai-model/model-adapter/types.d.ts +18 -3
- package/dist/types/ai-model/models/deepseek/adapter.d.ts +2 -0
- package/dist/types/ai-model/models/gpt.d.ts +21 -2
- package/dist/types/ai-model/models/kimi.d.ts +2 -0
- package/dist/types/ai-model/models/qwen.d.ts +2 -3
- package/dist/types/ai-model/models/registry.d.ts +25 -2
- package/dist/types/ai-model/service-caller/codex-app-server.d.ts +3 -7
- package/dist/types/ai-model/shared/model-locate-result/index.d.ts +1 -1
- package/dist/types/ai-model/shared/model-locate-result/pixel-mapper.d.ts +24 -0
- package/dist/types/ai-model/shared/model-locate-result/types.d.ts +33 -71
- package/dist/types/ai-model/shared/model-locate-result/validation.d.ts +3 -0
- package/dist/types/ai-model/workflows/grounding/search-area-mapping.d.ts +8 -3
- package/dist/types/ai-model/workflows/grounding/search-area.d.ts +6 -3
- package/dist/types/ai-model/workflows/grounding/types.d.ts +4 -3
- package/dist/types/types.d.ts +2 -3
- package/package.json +2 -2
- package/dist/es/ai-model/shared/model-locate-result/bbox.mjs +0 -60
- package/dist/es/ai-model/shared/model-locate-result/bbox.mjs.map +0 -1
- package/dist/es/ai-model/shared/model-locate-result/pixel-bbox-mapper.mjs +0 -67
- package/dist/es/ai-model/shared/model-locate-result/pixel-bbox-mapper.mjs.map +0 -1
- package/dist/es/ai-model/workflows/grounding/locate-result-rect.mjs +0 -23
- package/dist/es/ai-model/workflows/grounding/locate-result-rect.mjs.map +0 -1
- package/dist/lib/ai-model/shared/model-locate-result/bbox.js +0 -106
- package/dist/lib/ai-model/shared/model-locate-result/bbox.js.map +0 -1
- package/dist/lib/ai-model/shared/model-locate-result/pixel-bbox-mapper.js +0 -101
- package/dist/lib/ai-model/shared/model-locate-result/pixel-bbox-mapper.js.map +0 -1
- package/dist/lib/ai-model/workflows/grounding/locate-result-rect.js.map +0 -1
- package/dist/types/ai-model/shared/model-locate-result/bbox.d.ts +0 -6
- package/dist/types/ai-model/shared/model-locate-result/pixel-bbox-mapper.d.ts +0 -7
- package/dist/types/ai-model/workflows/grounding/locate-result-rect.d.ts +0 -5
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"ai-model/workflows/grounding/search-area.mjs","sources":["../../../../../src/ai-model/workflows/grounding/search-area.ts"],"sourcesContent":["import type { Rect, Size, UIContext } from '@/types';\nimport { cropByRect, scaleImage } from '@midscene/shared/img';\nimport type { SearchAreaConfig } from './types';\n\n/**\n * Expand the search
|
|
1
|
+
{"version":3,"file":"ai-model/workflows/grounding/search-area.mjs","sources":["../../../../../src/ai-model/workflows/grounding/search-area.ts"],"sourcesContent":["import type { Rect, Size, UIContext } from '@/types';\nimport { cropByRect, scaleImage } from '@midscene/shared/img';\nimport type { PixelLocateResult } from '../../shared/model-locate-result';\nimport type { SearchAreaConfig } from './types';\n\n/** Merge target and reference regions for DeepLocate; point-only results cover one pixel. */\nexport function mergeSearchAreaResults(\n target: PixelLocateResult,\n references: PixelLocateResult[] = [],\n): Rect {\n const bounds = [target, ...references].map(({ rect, center }) => {\n const left = rect?.left ?? Math.round(center[0]);\n const top = rect?.top ?? Math.round(center[1]);\n return {\n left,\n top,\n right: left + (rect?.width ?? 1),\n bottom: top + (rect?.height ?? 1),\n };\n });\n const left = Math.min(...bounds.map((bound) => bound.left));\n const top = Math.min(...bounds.map((bound) => bound.top));\n const right = Math.max(...bounds.map((bound) => bound.right));\n const bottom = Math.max(...bounds.map((bound) => bound.bottom));\n return { left, top, width: right - left, height: bottom - top };\n}\n\n/**\n * Expand the search region toward a minimum area of 160,000 square pixels.\n *\n * Step 1: Extend 100px on each side (top, right, bottom, left)\n * - If the element is near a boundary, expansion on that side will be limited\n * - No compensation is made for boundary limitations (this is intentional)\n *\n * Step 2: Scale proportionally toward the minimum area.\n * - Scale up proportionally from the center if needed\n * - Clamping to screen boundaries may leave the final area below this target.\n */\nexport function expandSearchArea(rect: Rect, screenSize: Size): Rect {\n const minArea = 400 * 400;\n const expandSize = 100;\n\n // Step 1: Extend each side by expandSize (100px), clamped to screen boundaries\n // Note: If element is near boundary, actual expansion may be less than 100px on that side\n const expandedLeft = Math.max(rect.left - expandSize, 0);\n const expandedTop = Math.max(rect.top - expandSize, 0);\n\n const expandRect = {\n left: expandedLeft,\n top: expandedTop,\n width: Math.min(\n rect.left - expandedLeft + rect.width + expandSize,\n screenSize.width - expandedLeft,\n ),\n height: Math.min(\n rect.top - expandedTop + rect.height + expandSize,\n screenSize.height - expandedTop,\n ),\n };\n\n // Step 2: Check if area is already >= 400x400\n const currentArea = expandRect.width * expandRect.height;\n\n if (currentArea >= minArea) {\n return expandRect;\n }\n\n // Step 2: Scale up from center to reach minimum 400x400 area\n const centerX = expandRect.left + expandRect.width / 2;\n const centerY = expandRect.top + expandRect.height / 2;\n\n // Calculate scale factor needed to reach minimum area\n const scaleFactor = Math.sqrt(minArea / currentArea);\n const newWidth = Math.round(expandRect.width * scaleFactor);\n const newHeight = Math.round(expandRect.height * scaleFactor);\n\n // Calculate new position based on center point\n const newLeft = Math.round(centerX - newWidth / 2);\n const newTop = Math.round(centerY - newHeight / 2);\n\n // Clamp to screen boundaries\n const left = Math.max(newLeft, 0);\n const top = Math.max(newTop, 0);\n\n return {\n left,\n top,\n width: Math.min(newWidth, screenSize.width - left),\n height: Math.min(newHeight, screenSize.height - top),\n };\n}\n\nexport async function buildSearchAreaConfig(options: {\n context: UIContext;\n baseRect: Rect;\n}): Promise<SearchAreaConfig> {\n const { context, baseRect } = options;\n const scaleRatio = 2;\n const sectionRect = expandSearchArea(baseRect, context.shotSize);\n\n const croppedResult = await cropByRect(\n context.screenshot.base64,\n sectionRect,\n );\n\n const scaledResult = await scaleImage(croppedResult.imageBase64, scaleRatio);\n return {\n sourceRect: sectionRect,\n image: {\n imageBase64: scaledResult.imageBase64,\n width: scaledResult.width,\n height: scaledResult.height,\n },\n mapping: {\n offset: {\n x: sectionRect.left,\n y: sectionRect.top,\n },\n scale: scaleRatio,\n },\n };\n}\n"],"names":["mergeSearchAreaResults","target","references","bounds","rect","center","left","Math","top","bound","right","bottom","expandSearchArea","screenSize","minArea","expandSize","expandedLeft","expandedTop","expandRect","currentArea","centerX","centerY","scaleFactor","newWidth","newHeight","newLeft","newTop","buildSearchAreaConfig","options","context","baseRect","scaleRatio","sectionRect","croppedResult","cropByRect","scaledResult","scaleImage"],"mappings":";AAMO,SAASA,uBACdC,MAAyB,EACzBC,aAAkC,EAAE;IAEpC,MAAMC,SAAS;QAACF;WAAWC;KAAW,CAAC,GAAG,CAAC,CAAC,EAAEE,IAAI,EAAEC,MAAM,EAAE;QAC1D,MAAMC,OAAOF,MAAM,QAAQG,KAAK,KAAK,CAACF,MAAM,CAAC,EAAE;QAC/C,MAAMG,MAAMJ,MAAM,OAAOG,KAAK,KAAK,CAACF,MAAM,CAAC,EAAE;QAC7C,OAAO;YACLC;YACAE;YACA,OAAOF,OAAQF,CAAAA,MAAM,SAAS;YAC9B,QAAQI,MAAOJ,CAAAA,MAAM,UAAU;QACjC;IACF;IACA,MAAME,OAAOC,KAAK,GAAG,IAAIJ,OAAO,GAAG,CAAC,CAACM,QAAUA,MAAM,IAAI;IACzD,MAAMD,MAAMD,KAAK,GAAG,IAAIJ,OAAO,GAAG,CAAC,CAACM,QAAUA,MAAM,GAAG;IACvD,MAAMC,QAAQH,KAAK,GAAG,IAAIJ,OAAO,GAAG,CAAC,CAACM,QAAUA,MAAM,KAAK;IAC3D,MAAME,SAASJ,KAAK,GAAG,IAAIJ,OAAO,GAAG,CAAC,CAACM,QAAUA,MAAM,MAAM;IAC7D,OAAO;QAAEH;QAAME;QAAK,OAAOE,QAAQJ;QAAM,QAAQK,SAASH;IAAI;AAChE;AAaO,SAASI,iBAAiBR,IAAU,EAAES,UAAgB;IAC3D,MAAMC,UAAU;IAChB,MAAMC,aAAa;IAInB,MAAMC,eAAeT,KAAK,GAAG,CAACH,KAAK,IAAI,GAAGW,YAAY;IACtD,MAAME,cAAcV,KAAK,GAAG,CAACH,KAAK,GAAG,GAAGW,YAAY;IAEpD,MAAMG,aAAa;QACjB,MAAMF;QACN,KAAKC;QACL,OAAOV,KAAK,GAAG,CACbH,KAAK,IAAI,GAAGY,eAAeZ,KAAK,KAAK,GAAGW,YACxCF,WAAW,KAAK,GAAGG;QAErB,QAAQT,KAAK,GAAG,CACdH,KAAK,GAAG,GAAGa,cAAcb,KAAK,MAAM,GAAGW,YACvCF,WAAW,MAAM,GAAGI;IAExB;IAGA,MAAME,cAAcD,WAAW,KAAK,GAAGA,WAAW,MAAM;IAExD,IAAIC,eAAeL,SACjB,OAAOI;IAIT,MAAME,UAAUF,WAAW,IAAI,GAAGA,WAAW,KAAK,GAAG;IACrD,MAAMG,UAAUH,WAAW,GAAG,GAAGA,WAAW,MAAM,GAAG;IAGrD,MAAMI,cAAcf,KAAK,IAAI,CAACO,UAAUK;IACxC,MAAMI,WAAWhB,KAAK,KAAK,CAACW,WAAW,KAAK,GAAGI;IAC/C,MAAME,YAAYjB,KAAK,KAAK,CAACW,WAAW,MAAM,GAAGI;IAGjD,MAAMG,UAAUlB,KAAK,KAAK,CAACa,UAAUG,WAAW;IAChD,MAAMG,SAASnB,KAAK,KAAK,CAACc,UAAUG,YAAY;IAGhD,MAAMlB,OAAOC,KAAK,GAAG,CAACkB,SAAS;IAC/B,MAAMjB,MAAMD,KAAK,GAAG,CAACmB,QAAQ;IAE7B,OAAO;QACLpB;QACAE;QACA,OAAOD,KAAK,GAAG,CAACgB,UAAUV,WAAW,KAAK,GAAGP;QAC7C,QAAQC,KAAK,GAAG,CAACiB,WAAWX,WAAW,MAAM,GAAGL;IAClD;AACF;AAEO,eAAemB,sBAAsBC,OAG3C;IACC,MAAM,EAAEC,OAAO,EAAEC,QAAQ,EAAE,GAAGF;IAC9B,MAAMG,aAAa;IACnB,MAAMC,cAAcpB,iBAAiBkB,UAAUD,QAAQ,QAAQ;IAE/D,MAAMI,gBAAgB,MAAMC,WAC1BL,QAAQ,UAAU,CAAC,MAAM,EACzBG;IAGF,MAAMG,eAAe,MAAMC,WAAWH,cAAc,WAAW,EAAEF;IACjE,OAAO;QACL,YAAYC;QACZ,OAAO;YACL,aAAaG,aAAa,WAAW;YACrC,OAAOA,aAAa,KAAK;YACzB,QAAQA,aAAa,MAAM;QAC7B;QACA,SAAS;YACP,QAAQ;gBACN,GAAGH,YAAY,IAAI;gBACnB,GAAGA,YAAY,GAAG;YACpB;YACA,OAAOD;QACT;IACF;AACF"}
|
|
@@ -4,8 +4,7 @@ import { userPromptToString } from "../../../common.mjs";
|
|
|
4
4
|
import { buildSearchAreaLocateSystemPrompt } from "../../prompt/locate/index.mjs";
|
|
5
5
|
import { AIResponseParseError, callAI } from "../../service-caller/index.mjs";
|
|
6
6
|
import { callAiAndParseWithRetry, withSemanticRetryFeedback } from "../../service-caller/semantic-retry.mjs";
|
|
7
|
-
import {
|
|
8
|
-
import { buildSearchAreaConfig, expandSearchArea } from "./search-area.mjs";
|
|
7
|
+
import { buildSearchAreaConfig, expandSearchArea, mergeSearchAreaResults } from "./search-area.mjs";
|
|
9
8
|
import { formatLocateModelContext, prepareLocateModelInput } from "./utils.mjs";
|
|
10
9
|
const debugSection = getDebug('ai:grounding:section');
|
|
11
10
|
async function AiLocateSection(options) {
|
|
@@ -54,12 +53,9 @@ async function AiLocateSection(options) {
|
|
|
54
53
|
preparedSize: preparedImage.preparedSize,
|
|
55
54
|
contentSize: preparedImage.contentSize
|
|
56
55
|
};
|
|
57
|
-
const target = resultCodec.
|
|
58
|
-
const references = parsedLocateResult.references?.map((reference)=>resultCodec.
|
|
59
|
-
const mergedRect =
|
|
60
|
-
target,
|
|
61
|
-
...references ?? []
|
|
62
|
-
]);
|
|
56
|
+
const target = resultCodec.toPixelResult(parsedLocateResult.target, locateResultContext);
|
|
57
|
+
const references = parsedLocateResult.references?.map((reference)=>resultCodec.toPixelResult(reference, locateResultContext));
|
|
58
|
+
const mergedRect = mergeSearchAreaResults(target, references);
|
|
63
59
|
debugSection('mergedRect %j', mergedRect);
|
|
64
60
|
return {
|
|
65
61
|
result: response,
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"ai-model/workflows/grounding/section-locate.mjs","sources":["../../../../../src/ai-model/workflows/grounding/section-locate.ts"],"sourcesContent":["import type { AIUsageInfo, Rect, UIContext } from '@/types';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type { TUserPrompt } from '../../../common';\nimport { userPromptToString } from '../../../common';\nimport type { ModelRuntime } from '../../models';\nimport { buildSearchAreaLocateSystemPrompt } from '../../prompt/locate';\nimport { AIResponseParseError, callAI } from '../../service-caller/index';\nimport {\n callAiAndParseWithRetry,\n withSemanticRetryFeedback,\n} from '../../service-caller/semantic-retry';\nimport { mergePixelBboxesToRect } from './locate-result-rect';\nimport { buildSearchAreaConfig, expandSearchArea } from './search-area';\nimport type { SearchAreaConfig } from './types';\nimport { formatLocateModelContext, prepareLocateModelInput } from './utils';\n\nconst debugSection = getDebug('ai:grounding:section');\n\ntype SectionLocateRawResponse = Awaited<ReturnType<typeof callAI>>;\n\nexport async function AiLocateSection(options: {\n context: UIContext;\n sectionDescription: TUserPrompt;\n modelRuntime: ModelRuntime;\n abortSignal?: AbortSignal;\n}): Promise<{\n searchAreaConfig?: SearchAreaConfig;\n error?: string;\n rawResponse: string;\n rawChoiceMessage?: unknown;\n usage?: AIUsageInfo;\n}> {\n const { context, sectionDescription } = options;\n const modelRuntime = options.modelRuntime;\n const { adapter } = modelRuntime;\n assert(\n adapter.locate.kind === 'standard',\n 'section locate requires a standard locate adapter',\n );\n const searchArea = adapter.locate.searchArea;\n assert(searchArea, 'section locate requires a search area operation');\n const { protocol: searchAreaProtocol, resultCodec } = searchArea;\n const screenshotBase64 = context.screenshot.base64;\n\n const systemPrompt = buildSearchAreaLocateSystemPrompt({\n systemPromptIntroduction: searchAreaProtocol.systemPromptIntroduction,\n responseInstructions: searchAreaProtocol.buildResponseInstructions(\n resultCodec.promptSpec,\n ),\n });\n\n const userInstructionPrompt = searchAreaProtocol.buildUserPrompt(\n userPromptToString(sectionDescription),\n );\n\n const { messages, preparedImage } = await prepareLocateModelInput({\n systemPrompt,\n userPrompt: userInstructionPrompt,\n locateImage: {\n imageBase64: screenshotBase64,\n width: context.shotSize.width,\n height: context.shotSize.height,\n },\n imagePreprocess: adapter.imagePreprocess,\n targetDescription: sectionDescription,\n userMessageContentOrder: adapter.locate.userMessageContentOrder,\n });\n\n let parsedResult:\n | {\n result: SectionLocateRawResponse;\n sectionError?: string;\n mergedRect?: undefined;\n }\n | {\n result: SectionLocateRawResponse;\n sectionError?: string;\n mergedRect: Rect;\n };\n\n try {\n parsedResult = await callAiAndParseWithRetry({\n callAi: (retryAttempt, previousParseError) =>\n callAI(\n withSemanticRetryFeedback(messages, previousParseError),\n modelRuntime,\n {\n abortSignal: options.abortSignal,\n expectedJsonObjectResponse:\n searchAreaProtocol.expectedJsonObjectResponse,\n semanticRetryAttempt: retryAttempt,\n },\n ),\n parseResponse: (response) => {\n const parsedLocateResult = searchAreaProtocol.parseRawResponse(\n response.content,\n resultCodec.promptSpec,\n );\n const sectionError = parsedLocateResult.error;\n if (parsedLocateResult.kind === 'not-found') {\n return { result: response, sectionError };\n }\n\n try {\n const locateResultContext = {\n preparedSize: preparedImage.preparedSize,\n contentSize: preparedImage.contentSize,\n };\n const target = resultCodec.toPixelBbox(\n parsedLocateResult.target,\n locateResultContext,\n );\n const references = parsedLocateResult.references?.map((reference) =>\n resultCodec.toPixelBbox(reference, locateResultContext),\n );\n const mergedRect = mergePixelBboxesToRect([\n target,\n ...(references ?? []),\n ]);\n debugSection('mergedRect %j', mergedRect);\n return {\n result: response,\n sectionError,\n mergedRect,\n };\n } catch (error) {\n const errorMessage =\n error instanceof Error ? error.message : String(error);\n const message = [\n sectionError && `error in section locate result: ${sectionError}`,\n `coordinate parsing error: ${errorMessage}`,\n formatLocateModelContext(modelRuntime),\n ]\n .filter(Boolean)\n .join('\\n');\n throw new Error(message, { cause: error });\n }\n },\n toParseError: (error, response) => {\n const parseErrorMessage =\n error instanceof Error\n ? `Failed to parse section locate result: ${error.message}`\n : 'unknown error in section locate';\n return new AIResponseParseError(\n parseErrorMessage,\n response.content,\n response.usage,\n response.rawChoiceMessage,\n response.reasoning_content,\n );\n },\n parseRetryTimes: modelRuntime.config.retryCount,\n parseRetryInterval: modelRuntime.config.retryInterval,\n abortSignal: options.abortSignal,\n onParseRetry: (error) => {\n debugSection(\n 'retrying section locate after coordinate parsing failed: %s',\n error instanceof Error ? error.message : String(error),\n );\n },\n });\n } catch (callError) {\n if (callError instanceof AIResponseParseError) {\n return {\n searchAreaConfig: undefined,\n error: callError.message,\n rawResponse: callError.rawResponse,\n rawChoiceMessage: callError.rawChoiceMessage,\n usage: callError.usage,\n };\n }\n\n const errorMessage =\n callError instanceof Error ? callError.message : String(callError);\n return {\n searchAreaConfig: undefined,\n error: `AI call error: ${errorMessage}`,\n rawResponse: errorMessage,\n };\n }\n\n const { result, sectionError, mergedRect } = parsedResult;\n if (!mergedRect) {\n return {\n searchAreaConfig: undefined,\n error: sectionError,\n rawResponse: result.content,\n rawChoiceMessage: result.rawChoiceMessage,\n usage: result.usage,\n };\n }\n\n try {\n const expandedRect = expandSearchArea(mergedRect, context.shotSize);\n const originalWidth = expandedRect.width;\n const originalHeight = expandedRect.height;\n debugSection('expanded sectionRect %j', expandedRect);\n\n const searchAreaConfig = await buildSearchAreaConfig({\n context,\n baseRect: mergedRect,\n });\n\n debugSection(\n 'scaled section image from %dx%d to %dx%d (scale=%d)',\n originalWidth,\n originalHeight,\n searchAreaConfig.image.width,\n searchAreaConfig.image.height,\n searchAreaConfig.mapping.scale,\n );\n return {\n searchAreaConfig,\n error: sectionError,\n rawResponse: result.content,\n rawChoiceMessage: result.rawChoiceMessage,\n usage: result.usage,\n };\n } catch (error) {\n const parseErrorMessage =\n error instanceof Error\n ? `Failed to parse section locate result: ${error.message}`\n : 'unknown error in section locate';\n const errorMessage = sectionError\n ? `${sectionError} (${parseErrorMessage})`\n : parseErrorMessage;\n return {\n searchAreaConfig: undefined,\n error: errorMessage,\n rawResponse: result.content,\n rawChoiceMessage: result.rawChoiceMessage,\n usage: result.usage,\n };\n }\n}\n"],"names":["debugSection","getDebug","AiLocateSection","options","context","sectionDescription","modelRuntime","adapter","assert","searchArea","searchAreaProtocol","resultCodec","screenshotBase64","systemPrompt","buildSearchAreaLocateSystemPrompt","userInstructionPrompt","userPromptToString","messages","preparedImage","prepareLocateModelInput","parsedResult","callAiAndParseWithRetry","retryAttempt","previousParseError","callAI","withSemanticRetryFeedback","response","parsedLocateResult","sectionError","locateResultContext","target","references","reference","mergedRect","mergePixelBboxesToRect","error","errorMessage","Error","String","message","formatLocateModelContext","Boolean","parseErrorMessage","AIResponseParseError","callError","undefined","result","expandedRect","expandSearchArea","originalWidth","originalHeight","searchAreaConfig","buildSearchAreaConfig"],"mappings":";;;;;;;;;AAiBA,MAAMA,eAAeC,SAAS;AAIvB,eAAeC,gBAAgBC,OAKrC;IAOC,MAAM,EAAEC,OAAO,EAAEC,kBAAkB,EAAE,GAAGF;IACxC,MAAMG,eAAeH,QAAQ,YAAY;IACzC,MAAM,EAAEI,OAAO,EAAE,GAAGD;IACpBE,OACED,AAAwB,eAAxBA,QAAQ,MAAM,CAAC,IAAI,EACnB;IAEF,MAAME,aAAaF,QAAQ,MAAM,CAAC,UAAU;IAC5CC,OAAOC,YAAY;IACnB,MAAM,EAAE,UAAUC,kBAAkB,EAAEC,WAAW,EAAE,GAAGF;IACtD,MAAMG,mBAAmBR,QAAQ,UAAU,CAAC,MAAM;IAElD,MAAMS,eAAeC,kCAAkC;QACrD,0BAA0BJ,mBAAmB,wBAAwB;QACrE,sBAAsBA,mBAAmB,yBAAyB,CAChEC,YAAY,UAAU;IAE1B;IAEA,MAAMI,wBAAwBL,mBAAmB,eAAe,CAC9DM,mBAAmBX;IAGrB,MAAM,EAAEY,QAAQ,EAAEC,aAAa,EAAE,GAAG,MAAMC,wBAAwB;QAChEN;QACA,YAAYE;QACZ,aAAa;YACX,aAAaH;YACb,OAAOR,QAAQ,QAAQ,CAAC,KAAK;YAC7B,QAAQA,QAAQ,QAAQ,CAAC,MAAM;QACjC;QACA,iBAAiBG,QAAQ,eAAe;QACxC,mBAAmBF;QACnB,yBAAyBE,QAAQ,MAAM,CAAC,uBAAuB;IACjE;IAEA,IAAIa;IAYJ,IAAI;QACFA,eAAe,MAAMC,wBAAwB;YAC3C,QAAQ,CAACC,cAAcC,qBACrBC,OACEC,0BAA0BR,UAAUM,qBACpCjB,cACA;oBACE,aAAaH,QAAQ,WAAW;oBAChC,4BACEO,mBAAmB,0BAA0B;oBAC/C,sBAAsBY;gBACxB;YAEJ,eAAe,CAACI;gBACd,MAAMC,qBAAqBjB,mBAAmB,gBAAgB,CAC5DgB,SAAS,OAAO,EAChBf,YAAY,UAAU;gBAExB,MAAMiB,eAAeD,mBAAmB,KAAK;gBAC7C,IAAIA,AAA4B,gBAA5BA,mBAAmB,IAAI,EACzB,OAAO;oBAAE,QAAQD;oBAAUE;gBAAa;gBAG1C,IAAI;oBACF,MAAMC,sBAAsB;wBAC1B,cAAcX,cAAc,YAAY;wBACxC,aAAaA,cAAc,WAAW;oBACxC;oBACA,MAAMY,SAASnB,YAAY,WAAW,CACpCgB,mBAAmB,MAAM,EACzBE;oBAEF,MAAME,aAAaJ,mBAAmB,UAAU,EAAE,IAAI,CAACK,YACrDrB,YAAY,WAAW,CAACqB,WAAWH;oBAErC,MAAMI,aAAaC,uBAAuB;wBACxCJ;2BACIC,cAAc,EAAE;qBACrB;oBACD/B,aAAa,iBAAiBiC;oBAC9B,OAAO;wBACL,QAAQP;wBACRE;wBACAK;oBACF;gBACF,EAAE,OAAOE,OAAO;oBACd,MAAMC,eACJD,iBAAiBE,QAAQF,MAAM,OAAO,GAAGG,OAAOH;oBAClD,MAAMI,UAAU;wBACdX,gBAAgB,CAAC,gCAAgC,EAAEA,cAAc;wBACjE,CAAC,0BAA0B,EAAEQ,cAAc;wBAC3CI,yBAAyBlC;qBAC1B,CACE,MAAM,CAACmC,SACP,IAAI,CAAC;oBACR,MAAM,IAAIJ,MAAME,SAAS;wBAAE,OAAOJ;oBAAM;gBAC1C;YACF;YACA,cAAc,CAACA,OAAOT;gBACpB,MAAMgB,oBACJP,iBAAiBE,QACb,CAAC,uCAAuC,EAAEF,MAAM,OAAO,EAAE,GACzD;gBACN,OAAO,IAAIQ,qBACTD,mBACAhB,SAAS,OAAO,EAChBA,SAAS,KAAK,EACdA,SAAS,gBAAgB,EACzBA,SAAS,iBAAiB;YAE9B;YACA,iBAAiBpB,aAAa,MAAM,CAAC,UAAU;YAC/C,oBAAoBA,aAAa,MAAM,CAAC,aAAa;YACrD,aAAaH,QAAQ,WAAW;YAChC,cAAc,CAACgC;gBACbnC,aACE,+DACAmC,iBAAiBE,QAAQF,MAAM,OAAO,GAAGG,OAAOH;YAEpD;QACF;IACF,EAAE,OAAOS,WAAW;QAClB,IAAIA,qBAAqBD,sBACvB,OAAO;YACL,kBAAkBE;YAClB,OAAOD,UAAU,OAAO;YACxB,aAAaA,UAAU,WAAW;YAClC,kBAAkBA,UAAU,gBAAgB;YAC5C,OAAOA,UAAU,KAAK;QACxB;QAGF,MAAMR,eACJQ,qBAAqBP,QAAQO,UAAU,OAAO,GAAGN,OAAOM;QAC1D,OAAO;YACL,kBAAkBC;YAClB,OAAO,CAAC,eAAe,EAAET,cAAc;YACvC,aAAaA;QACf;IACF;IAEA,MAAM,EAAEU,MAAM,EAAElB,YAAY,EAAEK,UAAU,EAAE,GAAGb;IAC7C,IAAI,CAACa,YACH,OAAO;QACL,kBAAkBY;QAClB,OAAOjB;QACP,aAAakB,OAAO,OAAO;QAC3B,kBAAkBA,OAAO,gBAAgB;QACzC,OAAOA,OAAO,KAAK;IACrB;IAGF,IAAI;QACF,MAAMC,eAAeC,iBAAiBf,YAAY7B,QAAQ,QAAQ;QAClE,MAAM6C,gBAAgBF,aAAa,KAAK;QACxC,MAAMG,iBAAiBH,aAAa,MAAM;QAC1C/C,aAAa,2BAA2B+C;QAExC,MAAMI,mBAAmB,MAAMC,sBAAsB;YACnDhD;YACA,UAAU6B;QACZ;QAEAjC,aACE,uDACAiD,eACAC,gBACAC,iBAAiB,KAAK,CAAC,KAAK,EAC5BA,iBAAiB,KAAK,CAAC,MAAM,EAC7BA,iBAAiB,OAAO,CAAC,KAAK;QAEhC,OAAO;YACLA;YACA,OAAOvB;YACP,aAAakB,OAAO,OAAO;YAC3B,kBAAkBA,OAAO,gBAAgB;YACzC,OAAOA,OAAO,KAAK;QACrB;IACF,EAAE,OAAOX,OAAO;QACd,MAAMO,oBACJP,iBAAiBE,QACb,CAAC,uCAAuC,EAAEF,MAAM,OAAO,EAAE,GACzD;QACN,MAAMC,eAAeR,eACjB,GAAGA,aAAa,EAAE,EAAEc,kBAAkB,CAAC,CAAC,GACxCA;QACJ,OAAO;YACL,kBAAkBG;YAClB,OAAOT;YACP,aAAaU,OAAO,OAAO;YAC3B,kBAAkBA,OAAO,gBAAgB;YACzC,OAAOA,OAAO,KAAK;QACrB;IACF;AACF"}
|
|
1
|
+
{"version":3,"file":"ai-model/workflows/grounding/section-locate.mjs","sources":["../../../../../src/ai-model/workflows/grounding/section-locate.ts"],"sourcesContent":["import type { AIUsageInfo, Rect, UIContext } from '@/types';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type { TUserPrompt } from '../../../common';\nimport { userPromptToString } from '../../../common';\nimport type { ModelRuntime } from '../../models';\nimport { buildSearchAreaLocateSystemPrompt } from '../../prompt/locate';\nimport { AIResponseParseError, callAI } from '../../service-caller/index';\nimport {\n callAiAndParseWithRetry,\n withSemanticRetryFeedback,\n} from '../../service-caller/semantic-retry';\nimport {\n buildSearchAreaConfig,\n expandSearchArea,\n mergeSearchAreaResults,\n} from './search-area';\nimport type { SearchAreaConfig } from './types';\nimport { formatLocateModelContext, prepareLocateModelInput } from './utils';\n\nconst debugSection = getDebug('ai:grounding:section');\n\ntype SectionLocateRawResponse = Awaited<ReturnType<typeof callAI>>;\n\nexport async function AiLocateSection(options: {\n context: UIContext;\n sectionDescription: TUserPrompt;\n modelRuntime: ModelRuntime;\n abortSignal?: AbortSignal;\n}): Promise<{\n searchAreaConfig?: SearchAreaConfig;\n error?: string;\n rawResponse: string;\n rawChoiceMessage?: unknown;\n usage?: AIUsageInfo;\n}> {\n const { context, sectionDescription } = options;\n const modelRuntime = options.modelRuntime;\n const { adapter } = modelRuntime;\n assert(\n adapter.locate.kind === 'standard',\n 'section locate requires a standard locate adapter',\n );\n const searchArea = adapter.locate.searchArea;\n assert(searchArea, 'section locate requires a search area operation');\n const { protocol: searchAreaProtocol, resultCodec } = searchArea;\n const screenshotBase64 = context.screenshot.base64;\n\n const systemPrompt = buildSearchAreaLocateSystemPrompt({\n systemPromptIntroduction: searchAreaProtocol.systemPromptIntroduction,\n responseInstructions: searchAreaProtocol.buildResponseInstructions(\n resultCodec.promptSpec,\n ),\n });\n\n const userInstructionPrompt = searchAreaProtocol.buildUserPrompt(\n userPromptToString(sectionDescription),\n );\n\n const { messages, preparedImage } = await prepareLocateModelInput({\n systemPrompt,\n userPrompt: userInstructionPrompt,\n locateImage: {\n imageBase64: screenshotBase64,\n width: context.shotSize.width,\n height: context.shotSize.height,\n },\n imagePreprocess: adapter.imagePreprocess,\n targetDescription: sectionDescription,\n userMessageContentOrder: adapter.locate.userMessageContentOrder,\n });\n\n let parsedResult:\n | {\n result: SectionLocateRawResponse;\n sectionError?: string;\n mergedRect?: undefined;\n }\n | {\n result: SectionLocateRawResponse;\n sectionError?: string;\n mergedRect: Rect;\n };\n\n try {\n parsedResult = await callAiAndParseWithRetry({\n callAi: (retryAttempt, previousParseError) =>\n callAI(\n withSemanticRetryFeedback(messages, previousParseError),\n modelRuntime,\n {\n abortSignal: options.abortSignal,\n expectedJsonObjectResponse:\n searchAreaProtocol.expectedJsonObjectResponse,\n semanticRetryAttempt: retryAttempt,\n },\n ),\n parseResponse: (response) => {\n const parsedLocateResult = searchAreaProtocol.parseRawResponse(\n response.content,\n resultCodec.promptSpec,\n );\n const sectionError = parsedLocateResult.error;\n if (parsedLocateResult.kind === 'not-found') {\n return { result: response, sectionError };\n }\n\n try {\n const locateResultContext = {\n preparedSize: preparedImage.preparedSize,\n contentSize: preparedImage.contentSize,\n };\n const target = resultCodec.toPixelResult(\n parsedLocateResult.target,\n locateResultContext,\n );\n const references = parsedLocateResult.references?.map((reference) =>\n resultCodec.toPixelResult(reference, locateResultContext),\n );\n const mergedRect = mergeSearchAreaResults(target, references);\n debugSection('mergedRect %j', mergedRect);\n return {\n result: response,\n sectionError,\n mergedRect,\n };\n } catch (error) {\n const errorMessage =\n error instanceof Error ? error.message : String(error);\n const message = [\n sectionError && `error in section locate result: ${sectionError}`,\n `coordinate parsing error: ${errorMessage}`,\n formatLocateModelContext(modelRuntime),\n ]\n .filter(Boolean)\n .join('\\n');\n throw new Error(message, { cause: error });\n }\n },\n toParseError: (error, response) => {\n const parseErrorMessage =\n error instanceof Error\n ? `Failed to parse section locate result: ${error.message}`\n : 'unknown error in section locate';\n return new AIResponseParseError(\n parseErrorMessage,\n response.content,\n response.usage,\n response.rawChoiceMessage,\n response.reasoning_content,\n );\n },\n parseRetryTimes: modelRuntime.config.retryCount,\n parseRetryInterval: modelRuntime.config.retryInterval,\n abortSignal: options.abortSignal,\n onParseRetry: (error) => {\n debugSection(\n 'retrying section locate after coordinate parsing failed: %s',\n error instanceof Error ? error.message : String(error),\n );\n },\n });\n } catch (callError) {\n if (callError instanceof AIResponseParseError) {\n return {\n searchAreaConfig: undefined,\n error: callError.message,\n rawResponse: callError.rawResponse,\n rawChoiceMessage: callError.rawChoiceMessage,\n usage: callError.usage,\n };\n }\n\n const errorMessage =\n callError instanceof Error ? callError.message : String(callError);\n return {\n searchAreaConfig: undefined,\n error: `AI call error: ${errorMessage}`,\n rawResponse: errorMessage,\n };\n }\n\n const { result, sectionError, mergedRect } = parsedResult;\n if (!mergedRect) {\n return {\n searchAreaConfig: undefined,\n error: sectionError,\n rawResponse: result.content,\n rawChoiceMessage: result.rawChoiceMessage,\n usage: result.usage,\n };\n }\n\n try {\n const expandedRect = expandSearchArea(mergedRect, context.shotSize);\n const originalWidth = expandedRect.width;\n const originalHeight = expandedRect.height;\n debugSection('expanded sectionRect %j', expandedRect);\n\n const searchAreaConfig = await buildSearchAreaConfig({\n context,\n baseRect: mergedRect,\n });\n\n debugSection(\n 'scaled section image from %dx%d to %dx%d (scale=%d)',\n originalWidth,\n originalHeight,\n searchAreaConfig.image.width,\n searchAreaConfig.image.height,\n searchAreaConfig.mapping.scale,\n );\n return {\n searchAreaConfig,\n error: sectionError,\n rawResponse: result.content,\n rawChoiceMessage: result.rawChoiceMessage,\n usage: result.usage,\n };\n } catch (error) {\n const parseErrorMessage =\n error instanceof Error\n ? `Failed to parse section locate result: ${error.message}`\n : 'unknown error in section locate';\n const errorMessage = sectionError\n ? `${sectionError} (${parseErrorMessage})`\n : parseErrorMessage;\n return {\n searchAreaConfig: undefined,\n error: errorMessage,\n rawResponse: result.content,\n rawChoiceMessage: result.rawChoiceMessage,\n usage: result.usage,\n };\n }\n}\n"],"names":["debugSection","getDebug","AiLocateSection","options","context","sectionDescription","modelRuntime","adapter","assert","searchArea","searchAreaProtocol","resultCodec","screenshotBase64","systemPrompt","buildSearchAreaLocateSystemPrompt","userInstructionPrompt","userPromptToString","messages","preparedImage","prepareLocateModelInput","parsedResult","callAiAndParseWithRetry","retryAttempt","previousParseError","callAI","withSemanticRetryFeedback","response","parsedLocateResult","sectionError","locateResultContext","target","references","reference","mergedRect","mergeSearchAreaResults","error","errorMessage","Error","String","message","formatLocateModelContext","Boolean","parseErrorMessage","AIResponseParseError","callError","undefined","result","expandedRect","expandSearchArea","originalWidth","originalHeight","searchAreaConfig","buildSearchAreaConfig"],"mappings":";;;;;;;;AAoBA,MAAMA,eAAeC,SAAS;AAIvB,eAAeC,gBAAgBC,OAKrC;IAOC,MAAM,EAAEC,OAAO,EAAEC,kBAAkB,EAAE,GAAGF;IACxC,MAAMG,eAAeH,QAAQ,YAAY;IACzC,MAAM,EAAEI,OAAO,EAAE,GAAGD;IACpBE,OACED,AAAwB,eAAxBA,QAAQ,MAAM,CAAC,IAAI,EACnB;IAEF,MAAME,aAAaF,QAAQ,MAAM,CAAC,UAAU;IAC5CC,OAAOC,YAAY;IACnB,MAAM,EAAE,UAAUC,kBAAkB,EAAEC,WAAW,EAAE,GAAGF;IACtD,MAAMG,mBAAmBR,QAAQ,UAAU,CAAC,MAAM;IAElD,MAAMS,eAAeC,kCAAkC;QACrD,0BAA0BJ,mBAAmB,wBAAwB;QACrE,sBAAsBA,mBAAmB,yBAAyB,CAChEC,YAAY,UAAU;IAE1B;IAEA,MAAMI,wBAAwBL,mBAAmB,eAAe,CAC9DM,mBAAmBX;IAGrB,MAAM,EAAEY,QAAQ,EAAEC,aAAa,EAAE,GAAG,MAAMC,wBAAwB;QAChEN;QACA,YAAYE;QACZ,aAAa;YACX,aAAaH;YACb,OAAOR,QAAQ,QAAQ,CAAC,KAAK;YAC7B,QAAQA,QAAQ,QAAQ,CAAC,MAAM;QACjC;QACA,iBAAiBG,QAAQ,eAAe;QACxC,mBAAmBF;QACnB,yBAAyBE,QAAQ,MAAM,CAAC,uBAAuB;IACjE;IAEA,IAAIa;IAYJ,IAAI;QACFA,eAAe,MAAMC,wBAAwB;YAC3C,QAAQ,CAACC,cAAcC,qBACrBC,OACEC,0BAA0BR,UAAUM,qBACpCjB,cACA;oBACE,aAAaH,QAAQ,WAAW;oBAChC,4BACEO,mBAAmB,0BAA0B;oBAC/C,sBAAsBY;gBACxB;YAEJ,eAAe,CAACI;gBACd,MAAMC,qBAAqBjB,mBAAmB,gBAAgB,CAC5DgB,SAAS,OAAO,EAChBf,YAAY,UAAU;gBAExB,MAAMiB,eAAeD,mBAAmB,KAAK;gBAC7C,IAAIA,AAA4B,gBAA5BA,mBAAmB,IAAI,EACzB,OAAO;oBAAE,QAAQD;oBAAUE;gBAAa;gBAG1C,IAAI;oBACF,MAAMC,sBAAsB;wBAC1B,cAAcX,cAAc,YAAY;wBACxC,aAAaA,cAAc,WAAW;oBACxC;oBACA,MAAMY,SAASnB,YAAY,aAAa,CACtCgB,mBAAmB,MAAM,EACzBE;oBAEF,MAAME,aAAaJ,mBAAmB,UAAU,EAAE,IAAI,CAACK,YACrDrB,YAAY,aAAa,CAACqB,WAAWH;oBAEvC,MAAMI,aAAaC,uBAAuBJ,QAAQC;oBAClD/B,aAAa,iBAAiBiC;oBAC9B,OAAO;wBACL,QAAQP;wBACRE;wBACAK;oBACF;gBACF,EAAE,OAAOE,OAAO;oBACd,MAAMC,eACJD,iBAAiBE,QAAQF,MAAM,OAAO,GAAGG,OAAOH;oBAClD,MAAMI,UAAU;wBACdX,gBAAgB,CAAC,gCAAgC,EAAEA,cAAc;wBACjE,CAAC,0BAA0B,EAAEQ,cAAc;wBAC3CI,yBAAyBlC;qBAC1B,CACE,MAAM,CAACmC,SACP,IAAI,CAAC;oBACR,MAAM,IAAIJ,MAAME,SAAS;wBAAE,OAAOJ;oBAAM;gBAC1C;YACF;YACA,cAAc,CAACA,OAAOT;gBACpB,MAAMgB,oBACJP,iBAAiBE,QACb,CAAC,uCAAuC,EAAEF,MAAM,OAAO,EAAE,GACzD;gBACN,OAAO,IAAIQ,qBACTD,mBACAhB,SAAS,OAAO,EAChBA,SAAS,KAAK,EACdA,SAAS,gBAAgB,EACzBA,SAAS,iBAAiB;YAE9B;YACA,iBAAiBpB,aAAa,MAAM,CAAC,UAAU;YAC/C,oBAAoBA,aAAa,MAAM,CAAC,aAAa;YACrD,aAAaH,QAAQ,WAAW;YAChC,cAAc,CAACgC;gBACbnC,aACE,+DACAmC,iBAAiBE,QAAQF,MAAM,OAAO,GAAGG,OAAOH;YAEpD;QACF;IACF,EAAE,OAAOS,WAAW;QAClB,IAAIA,qBAAqBD,sBACvB,OAAO;YACL,kBAAkBE;YAClB,OAAOD,UAAU,OAAO;YACxB,aAAaA,UAAU,WAAW;YAClC,kBAAkBA,UAAU,gBAAgB;YAC5C,OAAOA,UAAU,KAAK;QACxB;QAGF,MAAMR,eACJQ,qBAAqBP,QAAQO,UAAU,OAAO,GAAGN,OAAOM;QAC1D,OAAO;YACL,kBAAkBC;YAClB,OAAO,CAAC,eAAe,EAAET,cAAc;YACvC,aAAaA;QACf;IACF;IAEA,MAAM,EAAEU,MAAM,EAAElB,YAAY,EAAEK,UAAU,EAAE,GAAGb;IAC7C,IAAI,CAACa,YACH,OAAO;QACL,kBAAkBY;QAClB,OAAOjB;QACP,aAAakB,OAAO,OAAO;QAC3B,kBAAkBA,OAAO,gBAAgB;QACzC,OAAOA,OAAO,KAAK;IACrB;IAGF,IAAI;QACF,MAAMC,eAAeC,iBAAiBf,YAAY7B,QAAQ,QAAQ;QAClE,MAAM6C,gBAAgBF,aAAa,KAAK;QACxC,MAAMG,iBAAiBH,aAAa,MAAM;QAC1C/C,aAAa,2BAA2B+C;QAExC,MAAMI,mBAAmB,MAAMC,sBAAsB;YACnDhD;YACA,UAAU6B;QACZ;QAEAjC,aACE,uDACAiD,eACAC,gBACAC,iBAAiB,KAAK,CAAC,KAAK,EAC5BA,iBAAiB,KAAK,CAAC,MAAM,EAC7BA,iBAAiB,OAAO,CAAC,KAAK;QAEhC,OAAO;YACLA;YACA,OAAOvB;YACP,aAAakB,OAAO,OAAO;YAC3B,kBAAkBA,OAAO,gBAAgB;YACzC,OAAOA,OAAO,KAAK;QACrB;IACF,EAAE,OAAOX,OAAO;QACd,MAAMO,oBACJP,iBAAiBE,QACb,CAAC,uCAAuC,EAAEF,MAAM,OAAO,EAAE,GACzD;QACN,MAAMC,eAAeR,eACjB,GAAGA,aAAa,EAAE,EAAEc,kBAAkB,CAAC,CAAC,GACxCA;QACJ,OAAO;YACL,kBAAkBG;YAClB,OAAOT;YACP,aAAaU,OAAO,OAAO;YAC3B,kBAAkBA,OAAO,gBAAgB;YACzC,OAAOA,OAAO,KAAK;QACrB;IACF;AACF"}
|
|
@@ -29,9 +29,10 @@ function normalizePlanningActionLocateFields(actions, { actionSpace, includeLoca
|
|
|
29
29
|
] : []
|
|
30
30
|
]);
|
|
31
31
|
const locateParamWithoutRawCoordinates = Object.fromEntries(Object.entries(locateParameter).filter(([key])=>!rawCoordinateKeys.has(key)));
|
|
32
|
+
const result = locateResultCodec.toPixelResult(rawLocateValue, locateResultContext);
|
|
32
33
|
action.param[field] = {
|
|
33
34
|
...locateParamWithoutRawCoordinates,
|
|
34
|
-
|
|
35
|
+
locatedPixelResult: result
|
|
35
36
|
};
|
|
36
37
|
});
|
|
37
38
|
});
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"ai-model/workflows/planning/locate-normalization.mjs","sources":["../../../../../src/ai-model/workflows/planning/locate-normalization.ts"],"sourcesContent":["import { findAllMidsceneLocatorField } from '@/common';\nimport type { DeviceAction } from '@/device';\nimport type { PlanningAction } from '@/types';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type { ParsedPlanningLocateParameter } from '../../model-adapter/planning-protocol';\nimport type {\n LocateResultCodec,\n LocateResultContext,\n} from '../../shared/model-locate-result/types';\n\nconst debug = getDebug('planning');\n\nexport function normalizePlanningActionLocateFields(\n actions: PlanningAction[],\n {\n actionSpace,\n includeLocateInPlanning,\n locateResultCodec,\n locateResultContext,\n acceptBbox2dAlias = false,\n parseRawLocateParameter,\n }: {\n actionSpace: DeviceAction[];\n includeLocateInPlanning: boolean;\n locateResultCodec?: LocateResultCodec;\n locateResultContext: LocateResultContext;\n acceptBbox2dAlias?: boolean;\n parseRawLocateParameter: (value: unknown) => ParsedPlanningLocateParameter;\n },\n): void {\n actions.forEach((action) => {\n const actionInActionSpace = actionSpace.find(\n (actionInSpace) => actionInSpace.name === action.type,\n );\n if (!actionInActionSpace) {\n debug('skip locate normalization for action outside actionSpace', action);\n return;\n }\n\n debug('actionInActionSpace matched', actionInActionSpace);\n const locateFields = findAllMidsceneLocatorField(\n actionInActionSpace.paramSchema,\n );\n\n debug('locateFields', locateFields);\n\n locateFields.forEach((field) => {\n const rawLocateParameter = action.param?.[field];\n if (!rawLocateParameter) {\n return;\n }\n\n const locateParameter = parseRawLocateParameter(rawLocateParameter);\n\n if (!includeLocateInPlanning) {\n // In prompt-only planning mode, ignore any accidental coordinates from the model.\n action.param[field] = { prompt: locateParameter.prompt };\n return;\n }\n\n assert(\n locateResultCodec,\n 'planning locate normalization requires a locate result codec',\n );\n\n const resultKey = locateResultCodec.promptSpec.resultKey;\n const rawLocateValue =\n locateParameter[resultKey] !== undefined\n ? locateParameter[resultKey]\n : acceptBbox2dAlias && resultKey === 'bbox'\n ? locateParameter.bbox_2d\n : undefined;\n\n // The raw result field is replaced by
|
|
1
|
+
{"version":3,"file":"ai-model/workflows/planning/locate-normalization.mjs","sources":["../../../../../src/ai-model/workflows/planning/locate-normalization.ts"],"sourcesContent":["import { findAllMidsceneLocatorField } from '@/common';\nimport type { DeviceAction } from '@/device';\nimport type { PlanningAction } from '@/types';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type { ParsedPlanningLocateParameter } from '../../model-adapter/planning-protocol';\nimport type {\n LocateResultCodec,\n LocateResultContext,\n} from '../../shared/model-locate-result/types';\n\nconst debug = getDebug('planning');\n\nexport function normalizePlanningActionLocateFields(\n actions: PlanningAction[],\n {\n actionSpace,\n includeLocateInPlanning,\n locateResultCodec,\n locateResultContext,\n acceptBbox2dAlias = false,\n parseRawLocateParameter,\n }: {\n actionSpace: DeviceAction[];\n includeLocateInPlanning: boolean;\n locateResultCodec?: LocateResultCodec;\n locateResultContext: LocateResultContext;\n acceptBbox2dAlias?: boolean;\n parseRawLocateParameter: (value: unknown) => ParsedPlanningLocateParameter;\n },\n): void {\n actions.forEach((action) => {\n const actionInActionSpace = actionSpace.find(\n (actionInSpace) => actionInSpace.name === action.type,\n );\n if (!actionInActionSpace) {\n debug('skip locate normalization for action outside actionSpace', action);\n return;\n }\n\n debug('actionInActionSpace matched', actionInActionSpace);\n const locateFields = findAllMidsceneLocatorField(\n actionInActionSpace.paramSchema,\n );\n\n debug('locateFields', locateFields);\n\n locateFields.forEach((field) => {\n const rawLocateParameter = action.param?.[field];\n if (!rawLocateParameter) {\n return;\n }\n\n const locateParameter = parseRawLocateParameter(rawLocateParameter);\n\n if (!includeLocateInPlanning) {\n // In prompt-only planning mode, ignore any accidental coordinates from the model.\n action.param[field] = { prompt: locateParameter.prompt };\n return;\n }\n\n assert(\n locateResultCodec,\n 'planning locate normalization requires a locate result codec',\n );\n\n const resultKey = locateResultCodec.promptSpec.resultKey;\n const rawLocateValue =\n locateParameter[resultKey] !== undefined\n ? locateParameter[resultKey]\n : acceptBbox2dAlias && resultKey === 'bbox'\n ? locateParameter.bbox_2d\n : undefined;\n\n // The raw result field is replaced by locatedPixelResult, so it should not\n // remain in the normalized locate parameter.\n const rawCoordinateKeys = new Set([\n resultKey,\n ...(acceptBbox2dAlias && resultKey === 'bbox' ? ['bbox_2d'] : []),\n ]);\n\n const locateParamWithoutRawCoordinates = Object.fromEntries(\n Object.entries(locateParameter).filter(\n ([key]) => !rawCoordinateKeys.has(key),\n ),\n );\n\n const result = locateResultCodec.toPixelResult(\n rawLocateValue,\n locateResultContext,\n );\n action.param[field] = {\n ...locateParamWithoutRawCoordinates,\n locatedPixelResult: result,\n };\n });\n });\n}\n"],"names":["debug","getDebug","normalizePlanningActionLocateFields","actions","actionSpace","includeLocateInPlanning","locateResultCodec","locateResultContext","acceptBbox2dAlias","parseRawLocateParameter","action","actionInActionSpace","actionInSpace","locateFields","findAllMidsceneLocatorField","field","rawLocateParameter","locateParameter","assert","resultKey","rawLocateValue","undefined","rawCoordinateKeys","Set","locateParamWithoutRawCoordinates","Object","key","result"],"mappings":";;;AAWA,MAAMA,QAAQC,SAAS;AAEhB,SAASC,oCACdC,OAAyB,EACzB,EACEC,WAAW,EACXC,uBAAuB,EACvBC,iBAAiB,EACjBC,mBAAmB,EACnBC,oBAAoB,KAAK,EACzBC,uBAAuB,EAQxB;IAEDN,QAAQ,OAAO,CAAC,CAACO;QACf,MAAMC,sBAAsBP,YAAY,IAAI,CAC1C,CAACQ,gBAAkBA,cAAc,IAAI,KAAKF,OAAO,IAAI;QAEvD,IAAI,CAACC,qBAAqB,YACxBX,MAAM,4DAA4DU;QAIpEV,MAAM,+BAA+BW;QACrC,MAAME,eAAeC,4BACnBH,oBAAoB,WAAW;QAGjCX,MAAM,gBAAgBa;QAEtBA,aAAa,OAAO,CAAC,CAACE;YACpB,MAAMC,qBAAqBN,OAAO,KAAK,EAAE,CAACK,MAAM;YAChD,IAAI,CAACC,oBACH;YAGF,MAAMC,kBAAkBR,wBAAwBO;YAEhD,IAAI,CAACX,yBAAyB;gBAE5BK,OAAO,KAAK,CAACK,MAAM,GAAG;oBAAE,QAAQE,gBAAgB,MAAM;gBAAC;gBACvD;YACF;YAEAC,OACEZ,mBACA;YAGF,MAAMa,YAAYb,kBAAkB,UAAU,CAAC,SAAS;YACxD,MAAMc,iBACJH,AAA+BI,WAA/BJ,eAAe,CAACE,UAAU,GACtBF,eAAe,CAACE,UAAU,GAC1BX,qBAAqBW,AAAc,WAAdA,YACnBF,gBAAgB,OAAO,GACvBI;YAIR,MAAMC,oBAAoB,IAAIC,IAAI;gBAChCJ;mBACIX,qBAAqBW,AAAc,WAAdA,YAAuB;oBAAC;iBAAU,GAAG,EAAE;aACjE;YAED,MAAMK,mCAAmCC,OAAO,WAAW,CACzDA,OAAO,OAAO,CAACR,iBAAiB,MAAM,CACpC,CAAC,CAACS,IAAI,GAAK,CAACJ,kBAAkB,GAAG,CAACI;YAItC,MAAMC,SAASrB,kBAAkB,aAAa,CAC5Cc,gBACAb;YAEFG,OAAO,KAAK,CAACK,MAAM,GAAG;gBACpB,GAAGS,gCAAgC;gBACnC,oBAAoBG;YACtB;QACF;IACF;AACF"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"ai-model/workflows/planning/standard-planning.mjs","sources":["../../../../../src/ai-model/workflows/planning/standard-planning.ts"],"sourcesContent":["import type {\n PlanningAIResponse,\n PlanningAction,\n RawResponsePlanningAIResponse,\n} from '@/types';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type { ChatCompletionMessageParam } from 'openai/resources/index';\nimport { buildYamlFlowFromPlans } from '../../../common';\nimport { prepareModelImage } from '../../model-adapter/image-preprocess';\nimport { buildStandardPlanningSystemPrompt } from '../../prompt/planning';\nimport { AIResponseParseError, callAI } from '../../service-caller/index';\nimport {\n callAiAndParseWithRetry,\n withSemanticRetryFeedback,\n} from '../../service-caller/semantic-retry';\nimport type {\n LocateResultCodec,\n LocateResultContext,\n} from '../../shared/model-locate-result';\nimport { planningModelFamilyRequiredForLocateMessage } from '../../shared/model-locate-result/errors';\nimport {\n type PreparedUserPrompt,\n preparedReferenceImagesToChatMessages,\n} from '../../shared/multimodal-prompt';\nimport { normalizePlanningActionLocateFields } from './locate-normalization';\nimport { parseStandardPlanningResponse } from './standard-planning-parser';\nimport type { PlanOptions } from './types';\n\nconst debug = getDebug('planning');\nconst warnLog = getDebug('planning', { console: true });\n\nconst noPreviousActionsText =\n 'No previous actions have been executed in this aiAct execution yet. If the instruction asks for actions, choose the first action to execute.';\ntype PlanningCallResponse = Awaited<ReturnType<typeof callAI>>;\n\ntype CallAndParsePlanningResponseOptions = {\n messages: ChatCompletionMessageParam[];\n modelRuntime: PlanOptions['modelRuntime'];\n abortSignal?: AbortSignal;\n includeLocateInPlanning: boolean;\n actionSpace: PlanOptions['actionSpace'];\n locateResultCodec?: LocateResultCodec;\n locateResultContext: LocateResultContext;\n includeThought: boolean;\n includeLog: boolean;\n};\n\nasync function callAndParsePlanningResponse(\n options: CallAndParsePlanningResponseOptions,\n): Promise<{\n response: PlanningCallResponse;\n planFromAI: RawResponsePlanningAIResponse;\n actions: PlanningAction[];\n yamlFlow: ReturnType<typeof buildYamlFlowFromPlans>;\n}> {\n const {\n messages,\n modelRuntime,\n abortSignal,\n includeLocateInPlanning,\n actionSpace,\n locateResultCodec,\n locateResultContext,\n includeThought,\n includeLog,\n } = options;\n assert(\n modelRuntime.adapter.planning.kind === 'standard',\n 'callAndParsePlanningResponse requires a standard planning adapter',\n );\n const actionOutputProtocol =\n modelRuntime.adapter.planning.protocol.actionOutputProtocol;\n\n return callAiAndParseWithRetry({\n callAi: (retryAttempt, previousParseError) =>\n callAI(\n withSemanticRetryFeedback(messages, previousParseError),\n modelRuntime,\n {\n abortSignal,\n requiresOriginalImageDetail: includeLocateInPlanning,\n semanticRetryAttempt: retryAttempt,\n },\n ),\n parseResponse: (response) => {\n const planFromAI = parseStandardPlanningResponse(response.content, {\n includeThought,\n actionOutputProtocol,\n actionSpace,\n logSource: includeLog ? 'model' : 'action',\n });\n if (planFromAI.action && planFromAI.finalizeSuccess !== undefined) {\n warnLog(\n 'Planning response included both an action and <complete>; ignoring <complete> output.',\n );\n planFromAI.finalizeMessage = undefined;\n planFromAI.finalizeSuccess = undefined;\n }\n\n const actions = planFromAI.action ? [planFromAI.action] : [];\n normalizePlanningActionLocateFields(actions, {\n actionSpace,\n includeLocateInPlanning,\n locateResultCodec,\n locateResultContext,\n acceptBbox2dAlias: modelRuntime.adapter.acceptBbox2dAlias,\n parseRawLocateParameter: actionOutputProtocol.parseRawLocateParameter,\n });\n // dumpActionParam keeps only the locator prompt, so runtime-only\n // locatedPixelBbox fields added during normalization are not serialized.\n const yamlFlow = buildYamlFlowFromPlans(actions, actionSpace);\n return { response, planFromAI, actions, yamlFlow };\n },\n toParseError: (parseError, response) => {\n const errorMessage =\n parseError instanceof Error ? parseError.message : String(parseError);\n return new AIResponseParseError(\n `XML parse error: ${errorMessage}`,\n response.content,\n response.usage,\n response.rawChoiceMessage,\n response.reasoning_content,\n );\n },\n parseRetryTimes: modelRuntime.config.retryCount,\n parseRetryInterval: modelRuntime.config.retryInterval,\n abortSignal,\n onParseRetry: (parseError) => {\n debug(\n 'retrying plan after response parsing failed: %s',\n parseError instanceof Error ? parseError.message : String(parseError),\n );\n },\n });\n}\n\nexport async function standardPlan(\n userInstruction: PreparedUserPrompt,\n opts: PlanOptions,\n): Promise<PlanningAIResponse> {\n const { context, conversationHistory } = opts;\n const modelRuntime = opts.modelRuntime;\n const { adapter } = modelRuntime;\n const { shotSize } = context;\n const screenshotBase64 = context.screenshot.base64;\n assert(\n adapter.planning.kind === 'standard',\n 'standardPlan requires a standard planning adapter',\n );\n const planningProtocol = adapter.planning.protocol;\n\n if (opts.includeLocateInPlanning && !modelRuntime.config.modelFamily) {\n throw new Error(\n planningModelFamilyRequiredForLocateMessage(modelRuntime.config.slot),\n );\n }\n\n const locateResultCodec = modelRuntime.config.modelFamily\n ? adapter.planning.locateResultCodec\n : undefined;\n\n // Only enable sub-goals when aiAct is in deep-thinking planning mode.\n const includeSubGoals = opts.effort === 'deepThink';\n const includeThought = opts.effort !== 'fast';\n const includeLog = opts.effort !== 'fast';\n\n if (opts.includeLocateInPlanning && !locateResultCodec) {\n throw new Error(\n planningModelFamilyRequiredForLocateMessage(modelRuntime.config.slot),\n );\n }\n\n const systemPrompt = await buildStandardPlanningSystemPrompt({\n actionSpace: opts.actionSpace,\n includeThought,\n includeLog,\n includeSubGoals,\n planningProtocol,\n ...(opts.includeLocateInPlanning && locateResultCodec\n ? {\n includeLocateInPlanning: true,\n locatePromptSpec: locateResultCodec.promptSpec,\n }\n : { includeLocateInPlanning: false }),\n });\n\n const preparedImage = await prepareModelImage({\n imageBase64: screenshotBase64,\n width: shotSize.width,\n height: shotSize.height,\n policy: adapter.imagePreprocess,\n });\n const imagePayload = preparedImage.imageBase64;\n\n const actionContext = opts.actionContext ? `${opts.actionContext}\\n` : '';\n\n const referenceImageMessages = preparedReferenceImagesToChatMessages(\n userInstruction.referenceImages,\n );\n const instruction: ChatCompletionMessageParam[] = [\n {\n role: 'user',\n content: [\n {\n type: 'text',\n text: `${actionContext}<user_instruction>${userInstruction.text}</user_instruction>`,\n },\n ],\n },\n ...referenceImageMessages,\n ];\n\n let latestFeedbackMessage: ChatCompletionMessageParam;\n\n // Build sub-goal status text to include in the message\n // In planning deep-think mode: show full sub-goals with logs\n // Otherwise: show historical execution logs\n const executionProgressText = includeSubGoals\n ? conversationHistory.subGoalsToText()\n : conversationHistory.historicalLogsToText();\n const executionProgressSection = executionProgressText\n ? `\\n\\n${executionProgressText}`\n : conversationHistory.pendingFeedbackMessage\n ? ''\n : `\\n\\n${noPreviousActionsText}`;\n\n // Build memories text to include in the message\n const memoriesText = conversationHistory.memoriesToText();\n const memoriesSection = memoriesText ? `\\n\\n${memoriesText}` : '';\n\n if (conversationHistory.pendingFeedbackMessage) {\n latestFeedbackMessage = {\n role: 'user',\n content: [\n {\n type: 'text',\n text: `${conversationHistory.pendingFeedbackMessage}. The previous action has been executed, here is the latest screenshot. Please continue according to the instruction.${memoriesSection}${executionProgressSection}`,\n },\n {\n type: 'image_url',\n image_url: {\n url: imagePayload,\n detail: 'high',\n },\n },\n ],\n };\n\n conversationHistory.resetPendingFeedbackMessageIfExists();\n } else {\n latestFeedbackMessage = {\n role: 'user',\n content: [\n {\n type: 'text',\n text: `This is the current screenshot.${memoriesSection}${executionProgressSection}`,\n },\n {\n type: 'image_url',\n image_url: {\n url: imagePayload,\n detail: 'high',\n },\n },\n ],\n };\n }\n conversationHistory.append(latestFeedbackMessage);\n\n // Compress history if it exceeds the threshold to avoid context overflow\n conversationHistory.compressHistory(50, 20);\n\n const historyLog = conversationHistory.snapshot(opts.imagesIncludeCount);\n\n const msgs: ChatCompletionMessageParam[] = [\n { role: 'system', content: systemPrompt },\n ...instruction,\n ...historyLog,\n ];\n\n const {\n response: {\n content: rawResponse,\n usage,\n reasoning_content,\n rawChoiceMessage,\n },\n planFromAI,\n actions,\n yamlFlow,\n } = await callAndParsePlanningResponse({\n messages: msgs,\n modelRuntime,\n abortSignal: opts.abortSignal,\n includeLocateInPlanning: opts.includeLocateInPlanning,\n actionSpace: opts.actionSpace,\n locateResultCodec,\n locateResultContext: {\n preparedSize: preparedImage.preparedSize,\n contentSize: preparedImage.contentSize,\n },\n includeThought,\n includeLog,\n });\n\n let shouldContinuePlanning = true;\n\n // Check if task is completed via <complete> tag\n if (planFromAI.finalizeSuccess !== undefined) {\n debug('task completed via <complete> tag, stop planning');\n shouldContinuePlanning = false;\n // Mark all sub-goals as finished when goal is completed in planning deep-think mode.\n if (includeSubGoals) {\n conversationHistory.markAllSubGoalsFinished();\n }\n }\n\n const returnValue: PlanningAIResponse = {\n ...planFromAI,\n actions,\n rawResponse,\n rawChoiceMessage,\n usage,\n reasoning_content,\n yamlFlow,\n shouldContinuePlanning,\n };\n\n assert(planFromAI, \"can't get plans from AI\");\n\n // TODO: The plan log is recorded before its action has executed, so a failed\n // action may still appear in the next round as an action already performed.\n // Move this write to the successful action execution path in TaskExecutor.action.\n // Update sub-goals in conversation history only in planning deep-think mode.\n if (includeSubGoals) {\n if (planFromAI.updateSubGoals?.length) {\n conversationHistory.mergeSubGoals(planFromAI.updateSubGoals);\n }\n if (planFromAI.markFinishedIndexes?.length) {\n for (const index of planFromAI.markFinishedIndexes) {\n conversationHistory.markSubGoalFinished(index);\n }\n }\n // Append the planning log to the currently running sub-goal\n if (planFromAI.log) {\n conversationHistory.appendSubGoalLog(planFromAI.log);\n }\n } else {\n // Without planning deep-think mode, accumulate logs as historical execution steps.\n if (planFromAI.log) {\n conversationHistory.appendHistoricalLog(planFromAI.log);\n }\n }\n\n // Append memory to conversation history if present\n if (planFromAI.memory) {\n conversationHistory.appendMemory(planFromAI.memory);\n }\n\n // Some model providers require opaque assistant fields to be replayed\n // verbatim in later turns. Keep this opt-in per model adapter so that an\n // unverified provider does not receive non-standard response fields.\n if (\n modelRuntime.adapter.chatCompletion.replayRawAssistantMessage &&\n rawChoiceMessage\n ) {\n conversationHistory.append(rawChoiceMessage as ChatCompletionMessageParam);\n } else {\n conversationHistory.append({\n role: 'assistant',\n content: [\n {\n type: 'text',\n text: rawResponse,\n },\n ],\n });\n }\n\n return returnValue;\n}\n"],"names":["debug","getDebug","warnLog","noPreviousActionsText","callAndParsePlanningResponse","options","messages","modelRuntime","abortSignal","includeLocateInPlanning","actionSpace","locateResultCodec","locateResultContext","includeThought","includeLog","assert","actionOutputProtocol","callAiAndParseWithRetry","retryAttempt","previousParseError","callAI","withSemanticRetryFeedback","response","planFromAI","parseStandardPlanningResponse","undefined","actions","normalizePlanningActionLocateFields","yamlFlow","buildYamlFlowFromPlans","parseError","errorMessage","Error","String","AIResponseParseError","standardPlan","userInstruction","opts","context","conversationHistory","adapter","shotSize","screenshotBase64","planningProtocol","planningModelFamilyRequiredForLocateMessage","includeSubGoals","systemPrompt","buildStandardPlanningSystemPrompt","preparedImage","prepareModelImage","imagePayload","actionContext","referenceImageMessages","preparedReferenceImagesToChatMessages","instruction","latestFeedbackMessage","executionProgressText","executionProgressSection","memoriesText","memoriesSection","historyLog","msgs","rawResponse","usage","reasoning_content","rawChoiceMessage","shouldContinuePlanning","returnValue","index"],"mappings":";;;;;;;;;;;AA6BA,MAAMA,QAAQC,SAAS;AACvB,MAAMC,UAAUD,SAAS,YAAY;IAAE,SAAS;AAAK;AAErD,MAAME,wBACJ;AAeF,eAAeC,6BACbC,OAA4C;IAO5C,MAAM,EACJC,QAAQ,EACRC,YAAY,EACZC,WAAW,EACXC,uBAAuB,EACvBC,WAAW,EACXC,iBAAiB,EACjBC,mBAAmB,EACnBC,cAAc,EACdC,UAAU,EACX,GAAGT;IACJU,OACER,AAAuC,eAAvCA,aAAa,OAAO,CAAC,QAAQ,CAAC,IAAI,EAClC;IAEF,MAAMS,uBACJT,aAAa,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAC,oBAAoB;IAE7D,OAAOU,wBAAwB;QAC7B,QAAQ,CAACC,cAAcC,qBACrBC,OACEC,0BAA0Bf,UAAUa,qBACpCZ,cACA;gBACEC;gBACA,6BAA6BC;gBAC7B,sBAAsBS;YACxB;QAEJ,eAAe,CAACI;YACd,MAAMC,aAAaC,8BAA8BF,SAAS,OAAO,EAAE;gBACjET;gBACAG;gBACAN;gBACA,WAAWI,aAAa,UAAU;YACpC;YACA,IAAIS,WAAW,MAAM,IAAIA,AAA+BE,WAA/BF,WAAW,eAAe,EAAgB;gBACjErB,QACE;gBAEFqB,WAAW,eAAe,GAAGE;gBAC7BF,WAAW,eAAe,GAAGE;YAC/B;YAEA,MAAMC,UAAUH,WAAW,MAAM,GAAG;gBAACA,WAAW,MAAM;aAAC,GAAG,EAAE;YAC5DI,oCAAoCD,SAAS;gBAC3ChB;gBACAD;gBACAE;gBACAC;gBACA,mBAAmBL,aAAa,OAAO,CAAC,iBAAiB;gBACzD,yBAAyBS,qBAAqB,uBAAuB;YACvE;YAGA,MAAMY,WAAWC,uBAAuBH,SAAShB;YACjD,OAAO;gBAAEY;gBAAUC;gBAAYG;gBAASE;YAAS;QACnD;QACA,cAAc,CAACE,YAAYR;YACzB,MAAMS,eACJD,sBAAsBE,QAAQF,WAAW,OAAO,GAAGG,OAAOH;YAC5D,OAAO,IAAII,qBACT,CAAC,iBAAiB,EAAEH,cAAc,EAClCT,SAAS,OAAO,EAChBA,SAAS,KAAK,EACdA,SAAS,gBAAgB,EACzBA,SAAS,iBAAiB;QAE9B;QACA,iBAAiBf,aAAa,MAAM,CAAC,UAAU;QAC/C,oBAAoBA,aAAa,MAAM,CAAC,aAAa;QACrDC;QACA,cAAc,CAACsB;YACb9B,MACE,mDACA8B,sBAAsBE,QAAQF,WAAW,OAAO,GAAGG,OAAOH;QAE9D;IACF;AACF;AAEO,eAAeK,aACpBC,eAAmC,EACnCC,IAAiB;IAEjB,MAAM,EAAEC,OAAO,EAAEC,mBAAmB,EAAE,GAAGF;IACzC,MAAM9B,eAAe8B,KAAK,YAAY;IACtC,MAAM,EAAEG,OAAO,EAAE,GAAGjC;IACpB,MAAM,EAAEkC,QAAQ,EAAE,GAAGH;IACrB,MAAMI,mBAAmBJ,QAAQ,UAAU,CAAC,MAAM;IAClDvB,OACEyB,AAA0B,eAA1BA,QAAQ,QAAQ,CAAC,IAAI,EACrB;IAEF,MAAMG,mBAAmBH,QAAQ,QAAQ,CAAC,QAAQ;IAElD,IAAIH,KAAK,uBAAuB,IAAI,CAAC9B,aAAa,MAAM,CAAC,WAAW,EAClE,MAAM,IAAIyB,MACRY,4CAA4CrC,aAAa,MAAM,CAAC,IAAI;IAIxE,MAAMI,oBAAoBJ,aAAa,MAAM,CAAC,WAAW,GACrDiC,QAAQ,QAAQ,CAAC,iBAAiB,GAClCf;IAGJ,MAAMoB,kBAAkBR,AAAgB,gBAAhBA,KAAK,MAAM;IACnC,MAAMxB,iBAAiBwB,AAAgB,WAAhBA,KAAK,MAAM;IAClC,MAAMvB,aAAauB,AAAgB,WAAhBA,KAAK,MAAM;IAE9B,IAAIA,KAAK,uBAAuB,IAAI,CAAC1B,mBACnC,MAAM,IAAIqB,MACRY,4CAA4CrC,aAAa,MAAM,CAAC,IAAI;IAIxE,MAAMuC,eAAe,MAAMC,kCAAkC;QAC3D,aAAaV,KAAK,WAAW;QAC7BxB;QACAC;QACA+B;QACAF;QACA,GAAIN,KAAK,uBAAuB,IAAI1B,oBAChC;YACE,yBAAyB;YACzB,kBAAkBA,kBAAkB,UAAU;QAChD,IACA;YAAE,yBAAyB;QAAM,CAAC;IACxC;IAEA,MAAMqC,gBAAgB,MAAMC,kBAAkB;QAC5C,aAAaP;QACb,OAAOD,SAAS,KAAK;QACrB,QAAQA,SAAS,MAAM;QACvB,QAAQD,QAAQ,eAAe;IACjC;IACA,MAAMU,eAAeF,cAAc,WAAW;IAE9C,MAAMG,gBAAgBd,KAAK,aAAa,GAAG,GAAGA,KAAK,aAAa,CAAC,EAAE,CAAC,GAAG;IAEvE,MAAMe,yBAAyBC,sCAC7BjB,gBAAgB,eAAe;IAEjC,MAAMkB,cAA4C;QAChD;YACE,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,MAAM,GAAGH,cAAc,kBAAkB,EAAEf,gBAAgB,IAAI,CAAC,mBAAmB,CAAC;gBACtF;aACD;QACH;WACGgB;KACJ;IAED,IAAIG;IAKJ,MAAMC,wBAAwBX,kBAC1BN,oBAAoB,cAAc,KAClCA,oBAAoB,oBAAoB;IAC5C,MAAMkB,2BAA2BD,wBAC7B,CAAC,IAAI,EAAEA,uBAAuB,GAC9BjB,oBAAoB,sBAAsB,GACxC,KACA,CAAC,IAAI,EAAEpC,uBAAuB;IAGpC,MAAMuD,eAAenB,oBAAoB,cAAc;IACvD,MAAMoB,kBAAkBD,eAAe,CAAC,IAAI,EAAEA,cAAc,GAAG;IAE/D,IAAInB,oBAAoB,sBAAsB,EAAE;QAC9CgB,wBAAwB;YACtB,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,MAAM,GAAGhB,oBAAoB,sBAAsB,CAAC,qHAAqH,EAAEoB,kBAAkBF,0BAA0B;gBACzN;gBACA;oBACE,MAAM;oBACN,WAAW;wBACT,KAAKP;wBACL,QAAQ;oBACV;gBACF;aACD;QACH;QAEAX,oBAAoB,mCAAmC;IACzD,OACEgB,wBAAwB;QACtB,MAAM;QACN,SAAS;YACP;gBACE,MAAM;gBACN,MAAM,CAAC,+BAA+B,EAAEI,kBAAkBF,0BAA0B;YACtF;YACA;gBACE,MAAM;gBACN,WAAW;oBACT,KAAKP;oBACL,QAAQ;gBACV;YACF;SACD;IACH;IAEFX,oBAAoB,MAAM,CAACgB;IAG3BhB,oBAAoB,eAAe,CAAC,IAAI;IAExC,MAAMqB,aAAarB,oBAAoB,QAAQ,CAACF,KAAK,kBAAkB;IAEvE,MAAMwB,OAAqC;QACzC;YAAE,MAAM;YAAU,SAASf;QAAa;WACrCQ;WACAM;KACJ;IAED,MAAM,EACJ,UAAU,EACR,SAASE,WAAW,EACpBC,KAAK,EACLC,iBAAiB,EACjBC,gBAAgB,EACjB,EACD1C,UAAU,EACVG,OAAO,EACPE,QAAQ,EACT,GAAG,MAAMxB,6BAA6B;QACrC,UAAUyD;QACVtD;QACA,aAAa8B,KAAK,WAAW;QAC7B,yBAAyBA,KAAK,uBAAuB;QACrD,aAAaA,KAAK,WAAW;QAC7B1B;QACA,qBAAqB;YACnB,cAAcqC,cAAc,YAAY;YACxC,aAAaA,cAAc,WAAW;QACxC;QACAnC;QACAC;IACF;IAEA,IAAIoD,yBAAyB;IAG7B,IAAI3C,AAA+BE,WAA/BF,WAAW,eAAe,EAAgB;QAC5CvB,MAAM;QACNkE,yBAAyB;QAEzB,IAAIrB,iBACFN,oBAAoB,uBAAuB;IAE/C;IAEA,MAAM4B,cAAkC;QACtC,GAAG5C,UAAU;QACbG;QACAoC;QACAG;QACAF;QACAC;QACApC;QACAsC;IACF;IAEAnD,OAAOQ,YAAY;IAMnB,IAAIsB,iBAAiB;QACnB,IAAItB,WAAW,cAAc,EAAE,QAC7BgB,oBAAoB,aAAa,CAAChB,WAAW,cAAc;QAE7D,IAAIA,WAAW,mBAAmB,EAAE,QAClC,KAAK,MAAM6C,SAAS7C,WAAW,mBAAmB,CAChDgB,oBAAoB,mBAAmB,CAAC6B;QAI5C,IAAI7C,WAAW,GAAG,EAChBgB,oBAAoB,gBAAgB,CAAChB,WAAW,GAAG;IAEvD,OAEE,IAAIA,WAAW,GAAG,EAChBgB,oBAAoB,mBAAmB,CAAChB,WAAW,GAAG;IAK1D,IAAIA,WAAW,MAAM,EACnBgB,oBAAoB,YAAY,CAAChB,WAAW,MAAM;IAMpD,IACEhB,aAAa,OAAO,CAAC,cAAc,CAAC,yBAAyB,IAC7D0D,kBAEA1B,oBAAoB,MAAM,CAAC0B;SAE3B1B,oBAAoB,MAAM,CAAC;QACzB,MAAM;QACN,SAAS;YACP;gBACE,MAAM;gBACN,MAAMuB;YACR;SACD;IACH;IAGF,OAAOK;AACT"}
|
|
1
|
+
{"version":3,"file":"ai-model/workflows/planning/standard-planning.mjs","sources":["../../../../../src/ai-model/workflows/planning/standard-planning.ts"],"sourcesContent":["import type {\n PlanningAIResponse,\n PlanningAction,\n RawResponsePlanningAIResponse,\n} from '@/types';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type { ChatCompletionMessageParam } from 'openai/resources/index';\nimport { buildYamlFlowFromPlans } from '../../../common';\nimport { prepareModelImage } from '../../model-adapter/image-preprocess';\nimport { buildStandardPlanningSystemPrompt } from '../../prompt/planning';\nimport { AIResponseParseError, callAI } from '../../service-caller/index';\nimport {\n callAiAndParseWithRetry,\n withSemanticRetryFeedback,\n} from '../../service-caller/semantic-retry';\nimport type {\n LocateResultCodec,\n LocateResultContext,\n} from '../../shared/model-locate-result';\nimport { planningModelFamilyRequiredForLocateMessage } from '../../shared/model-locate-result/errors';\nimport {\n type PreparedUserPrompt,\n preparedReferenceImagesToChatMessages,\n} from '../../shared/multimodal-prompt';\nimport { normalizePlanningActionLocateFields } from './locate-normalization';\nimport { parseStandardPlanningResponse } from './standard-planning-parser';\nimport type { PlanOptions } from './types';\n\nconst debug = getDebug('planning');\nconst warnLog = getDebug('planning', { console: true });\n\nconst noPreviousActionsText =\n 'No previous actions have been executed in this aiAct execution yet. If the instruction asks for actions, choose the first action to execute.';\ntype PlanningCallResponse = Awaited<ReturnType<typeof callAI>>;\n\ntype CallAndParsePlanningResponseOptions = {\n messages: ChatCompletionMessageParam[];\n modelRuntime: PlanOptions['modelRuntime'];\n abortSignal?: AbortSignal;\n includeLocateInPlanning: boolean;\n actionSpace: PlanOptions['actionSpace'];\n locateResultCodec?: LocateResultCodec;\n locateResultContext: LocateResultContext;\n includeThought: boolean;\n includeLog: boolean;\n};\n\nasync function callAndParsePlanningResponse(\n options: CallAndParsePlanningResponseOptions,\n): Promise<{\n response: PlanningCallResponse;\n planFromAI: RawResponsePlanningAIResponse;\n actions: PlanningAction[];\n yamlFlow: ReturnType<typeof buildYamlFlowFromPlans>;\n}> {\n const {\n messages,\n modelRuntime,\n abortSignal,\n includeLocateInPlanning,\n actionSpace,\n locateResultCodec,\n locateResultContext,\n includeThought,\n includeLog,\n } = options;\n assert(\n modelRuntime.adapter.planning.kind === 'standard',\n 'callAndParsePlanningResponse requires a standard planning adapter',\n );\n const actionOutputProtocol =\n modelRuntime.adapter.planning.protocol.actionOutputProtocol;\n\n return callAiAndParseWithRetry({\n callAi: (retryAttempt, previousParseError) =>\n callAI(\n withSemanticRetryFeedback(messages, previousParseError),\n modelRuntime,\n {\n abortSignal,\n requiresOriginalImageDetail: includeLocateInPlanning,\n semanticRetryAttempt: retryAttempt,\n },\n ),\n parseResponse: (response) => {\n const planFromAI = parseStandardPlanningResponse(response.content, {\n includeThought,\n actionOutputProtocol,\n actionSpace,\n logSource: includeLog ? 'model' : 'action',\n });\n if (planFromAI.action && planFromAI.finalizeSuccess !== undefined) {\n warnLog(\n 'Planning response included both an action and <complete>; ignoring <complete> output.',\n );\n planFromAI.finalizeMessage = undefined;\n planFromAI.finalizeSuccess = undefined;\n }\n\n const actions = planFromAI.action ? [planFromAI.action] : [];\n normalizePlanningActionLocateFields(actions, {\n actionSpace,\n includeLocateInPlanning,\n locateResultCodec,\n locateResultContext,\n acceptBbox2dAlias: modelRuntime.adapter.acceptBbox2dAlias,\n parseRawLocateParameter: actionOutputProtocol.parseRawLocateParameter,\n });\n // dumpActionParam keeps only the locator prompt, so runtime-only\n // locatedPixelResult fields added during normalization are not serialized.\n const yamlFlow = buildYamlFlowFromPlans(actions, actionSpace);\n return { response, planFromAI, actions, yamlFlow };\n },\n toParseError: (parseError, response) => {\n const errorMessage =\n parseError instanceof Error ? parseError.message : String(parseError);\n return new AIResponseParseError(\n `XML parse error: ${errorMessage}`,\n response.content,\n response.usage,\n response.rawChoiceMessage,\n response.reasoning_content,\n );\n },\n parseRetryTimes: modelRuntime.config.retryCount,\n parseRetryInterval: modelRuntime.config.retryInterval,\n abortSignal,\n onParseRetry: (parseError) => {\n debug(\n 'retrying plan after response parsing failed: %s',\n parseError instanceof Error ? parseError.message : String(parseError),\n );\n },\n });\n}\n\nexport async function standardPlan(\n userInstruction: PreparedUserPrompt,\n opts: PlanOptions,\n): Promise<PlanningAIResponse> {\n const { context, conversationHistory } = opts;\n const modelRuntime = opts.modelRuntime;\n const { adapter } = modelRuntime;\n const { shotSize } = context;\n const screenshotBase64 = context.screenshot.base64;\n assert(\n adapter.planning.kind === 'standard',\n 'standardPlan requires a standard planning adapter',\n );\n const planningProtocol = adapter.planning.protocol;\n\n if (opts.includeLocateInPlanning && !modelRuntime.config.modelFamily) {\n throw new Error(\n planningModelFamilyRequiredForLocateMessage(modelRuntime.config.slot),\n );\n }\n\n const locateResultCodec = modelRuntime.config.modelFamily\n ? adapter.planning.locateResultCodec\n : undefined;\n\n // Only enable sub-goals when aiAct is in deep-thinking planning mode.\n const includeSubGoals = opts.effort === 'deepThink';\n const includeThought = opts.effort !== 'fast';\n const includeLog = opts.effort !== 'fast';\n\n if (opts.includeLocateInPlanning && !locateResultCodec) {\n throw new Error(\n planningModelFamilyRequiredForLocateMessage(modelRuntime.config.slot),\n );\n }\n\n const systemPrompt = await buildStandardPlanningSystemPrompt({\n actionSpace: opts.actionSpace,\n includeThought,\n includeLog,\n includeSubGoals,\n planningProtocol,\n ...(opts.includeLocateInPlanning && locateResultCodec\n ? {\n includeLocateInPlanning: true,\n locatePromptSpec: locateResultCodec.promptSpec,\n }\n : { includeLocateInPlanning: false }),\n });\n\n const preparedImage = await prepareModelImage({\n imageBase64: screenshotBase64,\n width: shotSize.width,\n height: shotSize.height,\n policy: adapter.imagePreprocess,\n });\n const imagePayload = preparedImage.imageBase64;\n\n const actionContext = opts.actionContext ? `${opts.actionContext}\\n` : '';\n\n const referenceImageMessages = preparedReferenceImagesToChatMessages(\n userInstruction.referenceImages,\n );\n const instruction: ChatCompletionMessageParam[] = [\n {\n role: 'user',\n content: [\n {\n type: 'text',\n text: `${actionContext}<user_instruction>${userInstruction.text}</user_instruction>`,\n },\n ],\n },\n ...referenceImageMessages,\n ];\n\n let latestFeedbackMessage: ChatCompletionMessageParam;\n\n // Build sub-goal status text to include in the message\n // In planning deep-think mode: show full sub-goals with logs\n // Otherwise: show historical execution logs\n const executionProgressText = includeSubGoals\n ? conversationHistory.subGoalsToText()\n : conversationHistory.historicalLogsToText();\n const executionProgressSection = executionProgressText\n ? `\\n\\n${executionProgressText}`\n : conversationHistory.pendingFeedbackMessage\n ? ''\n : `\\n\\n${noPreviousActionsText}`;\n\n // Build memories text to include in the message\n const memoriesText = conversationHistory.memoriesToText();\n const memoriesSection = memoriesText ? `\\n\\n${memoriesText}` : '';\n\n if (conversationHistory.pendingFeedbackMessage) {\n latestFeedbackMessage = {\n role: 'user',\n content: [\n {\n type: 'text',\n text: `${conversationHistory.pendingFeedbackMessage}. The previous action has been executed, here is the latest screenshot. Please continue according to the instruction.${memoriesSection}${executionProgressSection}`,\n },\n {\n type: 'image_url',\n image_url: {\n url: imagePayload,\n detail: 'high',\n },\n },\n ],\n };\n\n conversationHistory.resetPendingFeedbackMessageIfExists();\n } else {\n latestFeedbackMessage = {\n role: 'user',\n content: [\n {\n type: 'text',\n text: `This is the current screenshot.${memoriesSection}${executionProgressSection}`,\n },\n {\n type: 'image_url',\n image_url: {\n url: imagePayload,\n detail: 'high',\n },\n },\n ],\n };\n }\n conversationHistory.append(latestFeedbackMessage);\n\n // Compress history if it exceeds the threshold to avoid context overflow\n conversationHistory.compressHistory(50, 20);\n\n const historyLog = conversationHistory.snapshot(opts.imagesIncludeCount);\n\n const msgs: ChatCompletionMessageParam[] = [\n { role: 'system', content: systemPrompt },\n ...instruction,\n ...historyLog,\n ];\n\n const {\n response: {\n content: rawResponse,\n usage,\n reasoning_content,\n rawChoiceMessage,\n },\n planFromAI,\n actions,\n yamlFlow,\n } = await callAndParsePlanningResponse({\n messages: msgs,\n modelRuntime,\n abortSignal: opts.abortSignal,\n includeLocateInPlanning: opts.includeLocateInPlanning,\n actionSpace: opts.actionSpace,\n locateResultCodec,\n locateResultContext: {\n preparedSize: preparedImage.preparedSize,\n contentSize: preparedImage.contentSize,\n },\n includeThought,\n includeLog,\n });\n\n let shouldContinuePlanning = true;\n\n // Check if task is completed via <complete> tag\n if (planFromAI.finalizeSuccess !== undefined) {\n debug('task completed via <complete> tag, stop planning');\n shouldContinuePlanning = false;\n // Mark all sub-goals as finished when goal is completed in planning deep-think mode.\n if (includeSubGoals) {\n conversationHistory.markAllSubGoalsFinished();\n }\n }\n\n const returnValue: PlanningAIResponse = {\n ...planFromAI,\n actions,\n rawResponse,\n rawChoiceMessage,\n usage,\n reasoning_content,\n yamlFlow,\n shouldContinuePlanning,\n };\n\n assert(planFromAI, \"can't get plans from AI\");\n\n // TODO: The plan log is recorded before its action has executed, so a failed\n // action may still appear in the next round as an action already performed.\n // Move this write to the successful action execution path in TaskExecutor.action.\n // Update sub-goals in conversation history only in planning deep-think mode.\n if (includeSubGoals) {\n if (planFromAI.updateSubGoals?.length) {\n conversationHistory.mergeSubGoals(planFromAI.updateSubGoals);\n }\n if (planFromAI.markFinishedIndexes?.length) {\n for (const index of planFromAI.markFinishedIndexes) {\n conversationHistory.markSubGoalFinished(index);\n }\n }\n // Append the planning log to the currently running sub-goal\n if (planFromAI.log) {\n conversationHistory.appendSubGoalLog(planFromAI.log);\n }\n } else {\n // Without planning deep-think mode, accumulate logs as historical execution steps.\n if (planFromAI.log) {\n conversationHistory.appendHistoricalLog(planFromAI.log);\n }\n }\n\n // Append memory to conversation history if present\n if (planFromAI.memory) {\n conversationHistory.appendMemory(planFromAI.memory);\n }\n\n // Some model providers require opaque assistant fields to be replayed\n // verbatim in later turns. Keep this opt-in per model adapter so that an\n // unverified provider does not receive non-standard response fields.\n if (\n modelRuntime.adapter.chatCompletion.replayRawAssistantMessage &&\n rawChoiceMessage\n ) {\n conversationHistory.append(rawChoiceMessage as ChatCompletionMessageParam);\n } else {\n conversationHistory.append({\n role: 'assistant',\n content: [\n {\n type: 'text',\n text: rawResponse,\n },\n ],\n });\n }\n\n return returnValue;\n}\n"],"names":["debug","getDebug","warnLog","noPreviousActionsText","callAndParsePlanningResponse","options","messages","modelRuntime","abortSignal","includeLocateInPlanning","actionSpace","locateResultCodec","locateResultContext","includeThought","includeLog","assert","actionOutputProtocol","callAiAndParseWithRetry","retryAttempt","previousParseError","callAI","withSemanticRetryFeedback","response","planFromAI","parseStandardPlanningResponse","undefined","actions","normalizePlanningActionLocateFields","yamlFlow","buildYamlFlowFromPlans","parseError","errorMessage","Error","String","AIResponseParseError","standardPlan","userInstruction","opts","context","conversationHistory","adapter","shotSize","screenshotBase64","planningProtocol","planningModelFamilyRequiredForLocateMessage","includeSubGoals","systemPrompt","buildStandardPlanningSystemPrompt","preparedImage","prepareModelImage","imagePayload","actionContext","referenceImageMessages","preparedReferenceImagesToChatMessages","instruction","latestFeedbackMessage","executionProgressText","executionProgressSection","memoriesText","memoriesSection","historyLog","msgs","rawResponse","usage","reasoning_content","rawChoiceMessage","shouldContinuePlanning","returnValue","index"],"mappings":";;;;;;;;;;;AA6BA,MAAMA,QAAQC,SAAS;AACvB,MAAMC,UAAUD,SAAS,YAAY;IAAE,SAAS;AAAK;AAErD,MAAME,wBACJ;AAeF,eAAeC,6BACbC,OAA4C;IAO5C,MAAM,EACJC,QAAQ,EACRC,YAAY,EACZC,WAAW,EACXC,uBAAuB,EACvBC,WAAW,EACXC,iBAAiB,EACjBC,mBAAmB,EACnBC,cAAc,EACdC,UAAU,EACX,GAAGT;IACJU,OACER,AAAuC,eAAvCA,aAAa,OAAO,CAAC,QAAQ,CAAC,IAAI,EAClC;IAEF,MAAMS,uBACJT,aAAa,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAC,oBAAoB;IAE7D,OAAOU,wBAAwB;QAC7B,QAAQ,CAACC,cAAcC,qBACrBC,OACEC,0BAA0Bf,UAAUa,qBACpCZ,cACA;gBACEC;gBACA,6BAA6BC;gBAC7B,sBAAsBS;YACxB;QAEJ,eAAe,CAACI;YACd,MAAMC,aAAaC,8BAA8BF,SAAS,OAAO,EAAE;gBACjET;gBACAG;gBACAN;gBACA,WAAWI,aAAa,UAAU;YACpC;YACA,IAAIS,WAAW,MAAM,IAAIA,AAA+BE,WAA/BF,WAAW,eAAe,EAAgB;gBACjErB,QACE;gBAEFqB,WAAW,eAAe,GAAGE;gBAC7BF,WAAW,eAAe,GAAGE;YAC/B;YAEA,MAAMC,UAAUH,WAAW,MAAM,GAAG;gBAACA,WAAW,MAAM;aAAC,GAAG,EAAE;YAC5DI,oCAAoCD,SAAS;gBAC3ChB;gBACAD;gBACAE;gBACAC;gBACA,mBAAmBL,aAAa,OAAO,CAAC,iBAAiB;gBACzD,yBAAyBS,qBAAqB,uBAAuB;YACvE;YAGA,MAAMY,WAAWC,uBAAuBH,SAAShB;YACjD,OAAO;gBAAEY;gBAAUC;gBAAYG;gBAASE;YAAS;QACnD;QACA,cAAc,CAACE,YAAYR;YACzB,MAAMS,eACJD,sBAAsBE,QAAQF,WAAW,OAAO,GAAGG,OAAOH;YAC5D,OAAO,IAAII,qBACT,CAAC,iBAAiB,EAAEH,cAAc,EAClCT,SAAS,OAAO,EAChBA,SAAS,KAAK,EACdA,SAAS,gBAAgB,EACzBA,SAAS,iBAAiB;QAE9B;QACA,iBAAiBf,aAAa,MAAM,CAAC,UAAU;QAC/C,oBAAoBA,aAAa,MAAM,CAAC,aAAa;QACrDC;QACA,cAAc,CAACsB;YACb9B,MACE,mDACA8B,sBAAsBE,QAAQF,WAAW,OAAO,GAAGG,OAAOH;QAE9D;IACF;AACF;AAEO,eAAeK,aACpBC,eAAmC,EACnCC,IAAiB;IAEjB,MAAM,EAAEC,OAAO,EAAEC,mBAAmB,EAAE,GAAGF;IACzC,MAAM9B,eAAe8B,KAAK,YAAY;IACtC,MAAM,EAAEG,OAAO,EAAE,GAAGjC;IACpB,MAAM,EAAEkC,QAAQ,EAAE,GAAGH;IACrB,MAAMI,mBAAmBJ,QAAQ,UAAU,CAAC,MAAM;IAClDvB,OACEyB,AAA0B,eAA1BA,QAAQ,QAAQ,CAAC,IAAI,EACrB;IAEF,MAAMG,mBAAmBH,QAAQ,QAAQ,CAAC,QAAQ;IAElD,IAAIH,KAAK,uBAAuB,IAAI,CAAC9B,aAAa,MAAM,CAAC,WAAW,EAClE,MAAM,IAAIyB,MACRY,4CAA4CrC,aAAa,MAAM,CAAC,IAAI;IAIxE,MAAMI,oBAAoBJ,aAAa,MAAM,CAAC,WAAW,GACrDiC,QAAQ,QAAQ,CAAC,iBAAiB,GAClCf;IAGJ,MAAMoB,kBAAkBR,AAAgB,gBAAhBA,KAAK,MAAM;IACnC,MAAMxB,iBAAiBwB,AAAgB,WAAhBA,KAAK,MAAM;IAClC,MAAMvB,aAAauB,AAAgB,WAAhBA,KAAK,MAAM;IAE9B,IAAIA,KAAK,uBAAuB,IAAI,CAAC1B,mBACnC,MAAM,IAAIqB,MACRY,4CAA4CrC,aAAa,MAAM,CAAC,IAAI;IAIxE,MAAMuC,eAAe,MAAMC,kCAAkC;QAC3D,aAAaV,KAAK,WAAW;QAC7BxB;QACAC;QACA+B;QACAF;QACA,GAAIN,KAAK,uBAAuB,IAAI1B,oBAChC;YACE,yBAAyB;YACzB,kBAAkBA,kBAAkB,UAAU;QAChD,IACA;YAAE,yBAAyB;QAAM,CAAC;IACxC;IAEA,MAAMqC,gBAAgB,MAAMC,kBAAkB;QAC5C,aAAaP;QACb,OAAOD,SAAS,KAAK;QACrB,QAAQA,SAAS,MAAM;QACvB,QAAQD,QAAQ,eAAe;IACjC;IACA,MAAMU,eAAeF,cAAc,WAAW;IAE9C,MAAMG,gBAAgBd,KAAK,aAAa,GAAG,GAAGA,KAAK,aAAa,CAAC,EAAE,CAAC,GAAG;IAEvE,MAAMe,yBAAyBC,sCAC7BjB,gBAAgB,eAAe;IAEjC,MAAMkB,cAA4C;QAChD;YACE,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,MAAM,GAAGH,cAAc,kBAAkB,EAAEf,gBAAgB,IAAI,CAAC,mBAAmB,CAAC;gBACtF;aACD;QACH;WACGgB;KACJ;IAED,IAAIG;IAKJ,MAAMC,wBAAwBX,kBAC1BN,oBAAoB,cAAc,KAClCA,oBAAoB,oBAAoB;IAC5C,MAAMkB,2BAA2BD,wBAC7B,CAAC,IAAI,EAAEA,uBAAuB,GAC9BjB,oBAAoB,sBAAsB,GACxC,KACA,CAAC,IAAI,EAAEpC,uBAAuB;IAGpC,MAAMuD,eAAenB,oBAAoB,cAAc;IACvD,MAAMoB,kBAAkBD,eAAe,CAAC,IAAI,EAAEA,cAAc,GAAG;IAE/D,IAAInB,oBAAoB,sBAAsB,EAAE;QAC9CgB,wBAAwB;YACtB,MAAM;YACN,SAAS;gBACP;oBACE,MAAM;oBACN,MAAM,GAAGhB,oBAAoB,sBAAsB,CAAC,qHAAqH,EAAEoB,kBAAkBF,0BAA0B;gBACzN;gBACA;oBACE,MAAM;oBACN,WAAW;wBACT,KAAKP;wBACL,QAAQ;oBACV;gBACF;aACD;QACH;QAEAX,oBAAoB,mCAAmC;IACzD,OACEgB,wBAAwB;QACtB,MAAM;QACN,SAAS;YACP;gBACE,MAAM;gBACN,MAAM,CAAC,+BAA+B,EAAEI,kBAAkBF,0BAA0B;YACtF;YACA;gBACE,MAAM;gBACN,WAAW;oBACT,KAAKP;oBACL,QAAQ;gBACV;YACF;SACD;IACH;IAEFX,oBAAoB,MAAM,CAACgB;IAG3BhB,oBAAoB,eAAe,CAAC,IAAI;IAExC,MAAMqB,aAAarB,oBAAoB,QAAQ,CAACF,KAAK,kBAAkB;IAEvE,MAAMwB,OAAqC;QACzC;YAAE,MAAM;YAAU,SAASf;QAAa;WACrCQ;WACAM;KACJ;IAED,MAAM,EACJ,UAAU,EACR,SAASE,WAAW,EACpBC,KAAK,EACLC,iBAAiB,EACjBC,gBAAgB,EACjB,EACD1C,UAAU,EACVG,OAAO,EACPE,QAAQ,EACT,GAAG,MAAMxB,6BAA6B;QACrC,UAAUyD;QACVtD;QACA,aAAa8B,KAAK,WAAW;QAC7B,yBAAyBA,KAAK,uBAAuB;QACrD,aAAaA,KAAK,WAAW;QAC7B1B;QACA,qBAAqB;YACnB,cAAcqC,cAAc,YAAY;YACxC,aAAaA,cAAc,WAAW;QACxC;QACAnC;QACAC;IACF;IAEA,IAAIoD,yBAAyB;IAG7B,IAAI3C,AAA+BE,WAA/BF,WAAW,eAAe,EAAgB;QAC5CvB,MAAM;QACNkE,yBAAyB;QAEzB,IAAIrB,iBACFN,oBAAoB,uBAAuB;IAE/C;IAEA,MAAM4B,cAAkC;QACtC,GAAG5C,UAAU;QACbG;QACAoC;QACAG;QACAF;QACAC;QACApC;QACAsC;IACF;IAEAnD,OAAOQ,YAAY;IAMnB,IAAIsB,iBAAiB;QACnB,IAAItB,WAAW,cAAc,EAAE,QAC7BgB,oBAAoB,aAAa,CAAChB,WAAW,cAAc;QAE7D,IAAIA,WAAW,mBAAmB,EAAE,QAClC,KAAK,MAAM6C,SAAS7C,WAAW,mBAAmB,CAChDgB,oBAAoB,mBAAmB,CAAC6B;QAI5C,IAAI7C,WAAW,GAAG,EAChBgB,oBAAoB,gBAAgB,CAAChB,WAAW,GAAG;IAEvD,OAEE,IAAIA,WAAW,GAAG,EAChBgB,oBAAoB,mBAAmB,CAAChB,WAAW,GAAG;IAK1D,IAAIA,WAAW,MAAM,EACnBgB,oBAAoB,YAAY,CAAChB,WAAW,MAAM;IAMpD,IACEhB,aAAa,OAAO,CAAC,cAAc,CAAC,yBAAyB,IAC7D0D,kBAEA1B,oBAAoB,MAAM,CAAC0B;SAE3B1B,oBAAoB,MAAM,CAAC;QACzB,MAAM;QACN,SAAS;YACP;gBACE,MAAM;gBACN,MAAMuB;YACR;SACD;IACH;IAGF,OAAOK;AACT"}
|
package/dist/es/common.mjs
CHANGED
|
@@ -138,19 +138,12 @@ const parseActionParam = (rawParam, zodSchema, options)=>{
|
|
|
138
138
|
const ratio = options?.shrunkShotToLogicalRatio;
|
|
139
139
|
for(const fieldName in locateFieldValues){
|
|
140
140
|
let value = locateFieldValues[fieldName];
|
|
141
|
-
if (void 0 !== ratio &&
|
|
141
|
+
if (void 0 !== ratio && value && 'object' == typeof value && value.center) value = {
|
|
142
142
|
...value,
|
|
143
143
|
center: [
|
|
144
144
|
Math.round(value.center[0] / ratio),
|
|
145
145
|
Math.round(value.center[1] / ratio)
|
|
146
|
-
]
|
|
147
|
-
rect: {
|
|
148
|
-
...value.rect,
|
|
149
|
-
left: Math.round(value.rect.left / ratio),
|
|
150
|
-
top: Math.round(value.rect.top / ratio),
|
|
151
|
-
width: Math.round(value.rect.width / ratio),
|
|
152
|
-
height: Math.round(value.rect.height / ratio)
|
|
153
|
-
}
|
|
146
|
+
]
|
|
154
147
|
};
|
|
155
148
|
validated[fieldName] = value;
|
|
156
149
|
}
|
package/dist/es/common.mjs.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"common.mjs","sources":["../../src/common.ts"],"sourcesContent":["import type {\n DeviceAction,\n MidsceneYamlFlowItem,\n PlanningAction,\n} from '@/types';\nimport { isPlainObject } from '@midscene/shared/utils';\nimport { z } from 'zod';\n\nexport function findActionInActionSpaceOrThrow(\n planType: string,\n actionSpace: DeviceAction<any>[],\n): DeviceAction<any> {\n const action = actionSpace.find((item) => item.name === planType);\n if (!action) {\n const available = actionSpace.map((item) => item.name).join(', ');\n throw new Error(\n `Action type '${planType}' is not in the current action space. Available actions: ${available || '(none)'}`,\n );\n }\n return action;\n}\n\nexport function buildYamlFlowFromPlans(\n plans: PlanningAction[],\n actionSpace: DeviceAction<any>[],\n): MidsceneYamlFlowItem[] {\n const flow: MidsceneYamlFlowItem[] = [];\n\n for (const plan of plans) {\n const verb = plan.type;\n\n const action = findActionInActionSpaceOrThrow(verb, actionSpace);\n\n const flowKey = action.interfaceAlias || verb;\n const flowParam = action.paramSchema\n ? dumpActionParam(plan.param || {}, action.paramSchema)\n : {};\n\n // For actions whose param is a single string field (e.g. Launch/Terminate's\n // `uri`, RunAdbShell's `command`), inline the value on the flowKey. Writing\n // `{ terminate: '', uri: '...' }` makes the YAML player treat the empty\n // string as the param and drop the sibling `uri`, so cache replay would\n // call the action with an empty argument.\n const shortcutField =\n action.name === 'Launch' || action.interfaceAlias === 'launch'\n ? 'uri'\n : action.name === 'Terminate' || action.interfaceAlias === 'terminate'\n ? 'uri'\n : action.name === 'RunAdbShell' ||\n action.interfaceAlias === 'runAdbShell' ||\n action.name === 'RunHdcShell' ||\n action.interfaceAlias === 'runHdcShell'\n ? 'command'\n : undefined;\n const shortcutKeys = shortcutField ? Object.keys(flowParam) : [];\n const canInlineShortcut =\n shortcutField &&\n shortcutKeys.length === 1 &&\n shortcutKeys[0] === shortcutField &&\n typeof flowParam[shortcutField] === 'string';\n\n const flowItem: MidsceneYamlFlowItem = canInlineShortcut\n ? { [flowKey]: flowParam[shortcutField as string] }\n : { [flowKey]: '', ...flowParam };\n\n flow.push(flowItem);\n }\n\n return flow;\n}\n\n// Zod schemas for shared types\nexport const PointSchema = z.object({\n left: z.number(),\n top: z.number(),\n});\n\nexport const SizeSchema = z.object({\n width: z.number(),\n height: z.number(),\n});\n\nexport const RectSchema = PointSchema.and(SizeSchema).and(\n z.object({\n zoom: z.number().optional(),\n }),\n);\n\n// Zod schema for TMultimodalPrompt\nexport const TMultimodalPromptSchema = z.object({\n images: z\n .array(\n z.object({\n name: z.string(),\n url: z.string(),\n }),\n )\n .optional(),\n convertHttpImage2Base64: z.boolean().optional(),\n});\n\n// Zod schema for TUserPrompt\nexport const TUserPromptSchema = z.union([\n z.string(),\n z\n .object({\n prompt: z.string(),\n })\n .and(TMultimodalPromptSchema.partial()),\n]);\n\n// Generate TypeScript types from Zod schemas\nexport type TMultimodalPrompt = z.infer<typeof TMultimodalPromptSchema>;\nexport type TUserPrompt = z.infer<typeof TUserPromptSchema>;\n\nexport const userPromptToString = (prompt: TUserPrompt): string => {\n return typeof prompt === 'string' ? prompt : prompt.prompt;\n};\n\nexport const userPromptToMultimodalPrompt = (\n prompt: TUserPrompt,\n): TMultimodalPrompt | undefined => {\n if (typeof prompt === 'string' || !prompt.images) {\n return undefined;\n }\n return {\n images: prompt.images,\n convertHttpImage2Base64: !!prompt.convertHttpImage2Base64,\n };\n};\n\n// Schema for locator field input (when users provide locate parameters)\nconst MidsceneLocationInput = z\n .object({\n prompt: TUserPromptSchema,\n deepLocate: z.boolean().optional(),\n deepThink: z\n .boolean()\n .optional()\n .describe('@deprecated Use `deepLocate` instead.'),\n cacheable: z.boolean().optional(),\n xpath: z.union([z.string(), z.boolean()]).optional(),\n })\n .passthrough();\n\n/**\n * Returns the schema for locator fields.\n * This now returns the input schema which is more permissive and suitable for validation.\n */\nexport const getMidsceneLocationSchema = () => {\n return MidsceneLocationInput;\n};\n\nexport const ifMidsceneLocatorField = (field: any): boolean => {\n // Handle optional fields by getting the inner type\n let actualField = field;\n if (actualField._def?.typeName === 'ZodOptional') {\n actualField = actualField._def.innerType;\n }\n\n // Check if this is a ZodObject\n if (actualField._def?.typeName === 'ZodObject') {\n const shape = actualField._def.shape();\n\n // Input schema has 'prompt' as a required field\n if ('prompt' in shape && shape.prompt) {\n return true;\n }\n }\n\n return false;\n};\n\nconst formatPromptWithImages = (\n promptObj: Exclude<TUserPrompt, string>,\n): string => {\n let promptString = promptObj.prompt;\n if (Array.isArray(promptObj.images) && promptObj.images.length > 0) {\n const imageCount = promptObj.images.length;\n promptString += ` (with ${imageCount} image${imageCount > 1 ? 's' : ''})`;\n }\n return promptString;\n};\n\nexport const findAllMidsceneLocatorField = (\n zodType?: z.ZodType<any>,\n requiredOnly?: boolean,\n): string[] => {\n if (!zodType) {\n return [];\n }\n\n // Check if this is a ZodObject by checking if it has a shape property\n const zodObject = zodType as any;\n if (zodObject._def?.typeName === 'ZodObject' && zodObject.shape) {\n const keys = Object.keys(zodObject.shape);\n return keys.filter((key) => {\n const field = zodObject.shape[key];\n // TODO: A similar locator-field check exists as isMidsceneLocatorField in\n // @midscene/shared/zod-schema-utils.\n if (!ifMidsceneLocatorField(field)) {\n return false;\n }\n\n // If requiredOnly is true, filter out optional fields\n if (requiredOnly) {\n return field._def?.typeName !== 'ZodOptional';\n }\n\n return true;\n });\n }\n\n // For other ZodType instances, we can't extract field names\n return [];\n};\n\nexport const dumpActionParam = (\n jsonObject: Record<string, any>,\n zodSchema: z.ZodType<any>,\n): Record<string, any> => {\n // Prevent spreading strings into {0: 'c', 1: 'o', ...}\n if (!isPlainObject(jsonObject)) {\n return {};\n }\n\n const locatorFields = findAllMidsceneLocatorField(zodSchema);\n const result = { ...jsonObject };\n\n for (const fieldName of locatorFields) {\n const fieldValue = result[fieldName];\n if (fieldValue) {\n // If it's already a string, keep it as is\n if (typeof fieldValue === 'string') {\n result[fieldName] = fieldValue;\n } else if (typeof fieldValue === 'object') {\n // Check if this field is actually a MidsceneLocationType object\n if (fieldValue.prompt) {\n // If prompt is a string, use it directly\n if (typeof fieldValue.prompt === 'string') {\n result[fieldName] = fieldValue.prompt;\n } else if (\n typeof fieldValue.prompt === 'object' &&\n fieldValue.prompt.prompt\n ) {\n // If prompt is a TUserPrompt object, extract the prompt string\n result[fieldName] = formatPromptWithImages(fieldValue.prompt);\n }\n }\n }\n }\n }\n\n return result;\n};\n\n/**\n * Parse and validate action parameters using Zod schema.\n * All fields are validated through Zod, EXCEPT locator fields which are skipped.\n * Default values defined in the schema are automatically applied.\n *\n * Locator fields are special business logic fields with complex validation requirements,\n * so they are intentionally excluded from Zod parsing and use existing validation logic.\n *\n * When shrunkShotToLogicalRatio is provided and !== 1, coordinates in locate fields\n * are transformed from screenshot space to logical space.\n */\nexport const parseActionParam = (\n rawParam: Record<string, any> | undefined,\n zodSchema?: z.ZodType<any>,\n options?: { shrunkShotToLogicalRatio?: number },\n): Record<string, any> | undefined => {\n // If no schema is provided, return undefined (action takes no parameters)\n if (!zodSchema) {\n return undefined;\n }\n\n // Handle undefined or null rawParam by providing an empty object\n const param = rawParam ?? {};\n\n // Find all locate fields in the schema\n const locateFields = findAllMidsceneLocatorField(zodSchema);\n\n // If there are no locate fields, just do normal validation\n if (locateFields.length === 0) {\n return zodSchema.parse(param);\n }\n\n // Extract locate field values to restore later\n const locateFieldValues: Record<string, any> = {};\n for (const fieldName of locateFields) {\n if (fieldName in param) {\n locateFieldValues[fieldName] = param[fieldName];\n }\n }\n\n // Build params for validation - skip locate fields and use dummy values\n const paramsForValidation: Record<string, any> = {};\n for (const key in param) {\n if (locateFields.includes(key)) {\n // Use dummy value to satisfy schema validation\n paramsForValidation[key] = { prompt: '_dummy_' };\n } else {\n paramsForValidation[key] = param[key];\n }\n }\n\n // Validate with dummy locate values\n const validated = zodSchema.parse(paramsForValidation);\n\n // Restore the actual locate field values (unvalidated, as per business requirement),\n // and transform coordinates from screenshot space to logical space if needed\n const ratio = options?.shrunkShotToLogicalRatio;\n for (const fieldName in locateFieldValues) {\n let value = locateFieldValues[fieldName];\n if (\n ratio !== undefined &&\n ratio !== 1 &&\n value &&\n typeof value === 'object' &&\n value.center &&\n value.rect\n ) {\n value = {\n ...value,\n center: [\n Math.round(value.center[0] / ratio),\n Math.round(value.center[1] / ratio),\n ],\n rect: {\n ...value.rect,\n left: Math.round(value.rect.left / ratio),\n top: Math.round(value.rect.top / ratio),\n width: Math.round(value.rect.width / ratio),\n height: Math.round(value.rect.height / ratio),\n },\n };\n }\n validated[fieldName] = value;\n }\n\n return validated;\n};\n\nexport const finalizeActionName = 'Finalize';\n\n/**\n * Get a readable time string for a given timestamp or the current time\n * @param format - Optional format string. Supports: YYYY, MM, DD, HH, mm, ss. Default: 'YYYY-MM-DD HH:mm:ss'\n * @param timestamp - Optional timestamp in milliseconds. If not provided, uses current system time.\n * @returns A formatted time string with format label\n */\nexport const getReadableTimeString = (\n format = 'YYYY-MM-DD HH:mm:ss',\n timestamp?: number,\n): string => {\n const now = timestamp !== undefined ? new Date(timestamp) : new Date();\n const year = now.getFullYear();\n const month = String(now.getMonth() + 1).padStart(2, '0');\n const day = String(now.getDate()).padStart(2, '0');\n const hours = String(now.getHours()).padStart(2, '0');\n const minutes = String(now.getMinutes()).padStart(2, '0');\n const seconds = String(now.getSeconds()).padStart(2, '0');\n\n const timeString = format\n .replace('YYYY', String(year))\n .replace('MM', month)\n .replace('DD', day)\n .replace('HH', hours)\n .replace('mm', minutes)\n .replace('ss', seconds);\n\n return `${timeString} (${format})`;\n};\n"],"names":["findActionInActionSpaceOrThrow","planType","actionSpace","action","item","available","Error","buildYamlFlowFromPlans","plans","flow","plan","verb","flowKey","flowParam","dumpActionParam","shortcutField","undefined","shortcutKeys","Object","canInlineShortcut","flowItem","PointSchema","z","SizeSchema","RectSchema","TMultimodalPromptSchema","TUserPromptSchema","userPromptToString","prompt","userPromptToMultimodalPrompt","MidsceneLocationInput","getMidsceneLocationSchema","ifMidsceneLocatorField","field","actualField","shape","formatPromptWithImages","promptObj","promptString","Array","imageCount","findAllMidsceneLocatorField","zodType","requiredOnly","zodObject","keys","key","jsonObject","zodSchema","isPlainObject","locatorFields","result","fieldName","fieldValue","parseActionParam","rawParam","options","param","locateFields","locateFieldValues","paramsForValidation","validated","ratio","value","Math","finalizeActionName","getReadableTimeString","format","timestamp","now","Date","year","month","String","day","hours","minutes","seconds","timeString"],"mappings":";;AAQO,SAASA,+BACdC,QAAgB,EAChBC,WAAgC;IAEhC,MAAMC,SAASD,YAAY,IAAI,CAAC,CAACE,OAASA,KAAK,IAAI,KAAKH;IACxD,IAAI,CAACE,QAAQ;QACX,MAAME,YAAYH,YAAY,GAAG,CAAC,CAACE,OAASA,KAAK,IAAI,EAAE,IAAI,CAAC;QAC5D,MAAM,IAAIE,MACR,CAAC,aAAa,EAAEL,SAAS,yDAAyD,EAAEI,aAAa,UAAU;IAE/G;IACA,OAAOF;AACT;AAEO,SAASI,uBACdC,KAAuB,EACvBN,WAAgC;IAEhC,MAAMO,OAA+B,EAAE;IAEvC,KAAK,MAAMC,QAAQF,MAAO;QACxB,MAAMG,OAAOD,KAAK,IAAI;QAEtB,MAAMP,SAASH,+BAA+BW,MAAMT;QAEpD,MAAMU,UAAUT,OAAO,cAAc,IAAIQ;QACzC,MAAME,YAAYV,OAAO,WAAW,GAChCW,gBAAgBJ,KAAK,KAAK,IAAI,CAAC,GAAGP,OAAO,WAAW,IACpD,CAAC;QAOL,MAAMY,gBACJZ,AAAgB,aAAhBA,OAAO,IAAI,IAAiBA,AAA0B,aAA1BA,OAAO,cAAc,GAC7C,QACAA,AAAgB,gBAAhBA,OAAO,IAAI,IAAoBA,AAA0B,gBAA1BA,OAAO,cAAc,GAClD,QACAA,AAAgB,kBAAhBA,OAAO,IAAI,IACTA,AAA0B,kBAA1BA,OAAO,cAAc,IACrBA,AAAgB,kBAAhBA,OAAO,IAAI,IACXA,AAA0B,kBAA1BA,OAAO,cAAc,GACrB,YACAa;QACV,MAAMC,eAAeF,gBAAgBG,OAAO,IAAI,CAACL,aAAa,EAAE;QAChE,MAAMM,oBACJJ,iBACAE,AAAwB,MAAxBA,aAAa,MAAM,IACnBA,YAAY,CAAC,EAAE,KAAKF,iBACpB,AAAoC,YAApC,OAAOF,SAAS,CAACE,cAAc;QAEjC,MAAMK,WAAiCD,oBACnC;YAAE,CAACP,QAAQ,EAAEC,SAAS,CAACE,cAAwB;QAAC,IAChD;YAAE,CAACH,QAAQ,EAAE;YAAI,GAAGC,SAAS;QAAC;QAElCJ,KAAK,IAAI,CAACW;IACZ;IAEA,OAAOX;AACT;AAGO,MAAMY,cAAcC,EAAE,MAAM,CAAC;IAClC,MAAMA,EAAE,MAAM;IACd,KAAKA,EAAE,MAAM;AACf;AAEO,MAAMC,aAAaD,EAAE,MAAM,CAAC;IACjC,OAAOA,EAAE,MAAM;IACf,QAAQA,EAAE,MAAM;AAClB;AAEO,MAAME,aAAaH,YAAY,GAAG,CAACE,YAAY,GAAG,CACvDD,EAAE,MAAM,CAAC;IACP,MAAMA,EAAE,MAAM,GAAG,QAAQ;AAC3B;AAIK,MAAMG,0BAA0BH,EAAE,MAAM,CAAC;IAC9C,QAAQA,EAAAA,KACA,CACJA,EAAE,MAAM,CAAC;QACP,MAAMA,EAAE,MAAM;QACd,KAAKA,EAAE,MAAM;IACf,IAED,QAAQ;IACX,yBAAyBA,EAAE,OAAO,GAAG,QAAQ;AAC/C;AAGO,MAAMI,oBAAoBJ,EAAE,KAAK,CAAC;IACvCA,EAAE,MAAM;IACRA,EAAAA,MACS,CAAC;QACN,QAAQA,EAAE,MAAM;IAClB,GACC,GAAG,CAACG,wBAAwB,OAAO;CACvC;AAMM,MAAME,qBAAqB,CAACC,SAC1B,AAAkB,YAAlB,OAAOA,SAAsBA,SAASA,OAAO,MAAM;AAGrD,MAAMC,+BAA+B,CAC1CD;IAEA,IAAI,AAAkB,YAAlB,OAAOA,UAAuB,CAACA,OAAO,MAAM,EAC9C;IAEF,OAAO;QACL,QAAQA,OAAO,MAAM;QACrB,yBAAyB,CAAC,CAACA,OAAO,uBAAuB;IAC3D;AACF;AAGA,MAAME,wBAAwBR,EAAAA,MACrB,CAAC;IACN,QAAQI;IACR,YAAYJ,EAAE,OAAO,GAAG,QAAQ;IAChC,WAAWA,EAAAA,OACD,GACP,QAAQ,GACR,QAAQ,CAAC;IACZ,WAAWA,EAAE,OAAO,GAAG,QAAQ;IAC/B,OAAOA,EAAE,KAAK,CAAC;QAACA,EAAE,MAAM;QAAIA,EAAE,OAAO;KAAG,EAAE,QAAQ;AACpD,GACC,WAAW;AAMP,MAAMS,4BAA4B,IAChCD;AAGF,MAAME,yBAAyB,CAACC;IAErC,IAAIC,cAAcD;IAClB,IAAIC,YAAY,IAAI,EAAE,aAAa,eACjCA,cAAcA,YAAY,IAAI,CAAC,SAAS;IAI1C,IAAIA,YAAY,IAAI,EAAE,aAAa,aAAa;QAC9C,MAAMC,QAAQD,YAAY,IAAI,CAAC,KAAK;QAGpC,IAAI,YAAYC,SAASA,MAAM,MAAM,EACnC,OAAO;IAEX;IAEA,OAAO;AACT;AAEA,MAAMC,yBAAyB,CAC7BC;IAEA,IAAIC,eAAeD,UAAU,MAAM;IACnC,IAAIE,MAAM,OAAO,CAACF,UAAU,MAAM,KAAKA,UAAU,MAAM,CAAC,MAAM,GAAG,GAAG;QAClE,MAAMG,aAAaH,UAAU,MAAM,CAAC,MAAM;QAC1CC,gBAAgB,CAAC,OAAO,EAAEE,WAAW,MAAM,EAAEA,aAAa,IAAI,MAAM,GAAG,CAAC,CAAC;IAC3E;IACA,OAAOF;AACT;AAEO,MAAMG,8BAA8B,CACzCC,SACAC;IAEA,IAAI,CAACD,SACH,OAAO,EAAE;IAIX,MAAME,YAAYF;IAClB,IAAIE,UAAU,IAAI,EAAE,aAAa,eAAeA,UAAU,KAAK,EAAE;QAC/D,MAAMC,OAAO3B,OAAO,IAAI,CAAC0B,UAAU,KAAK;QACxC,OAAOC,KAAK,MAAM,CAAC,CAACC;YAClB,MAAMb,QAAQW,UAAU,KAAK,CAACE,IAAI;YAGlC,IAAI,CAACd,uBAAuBC,QAC1B,OAAO;YAIT,IAAIU,cACF,OAAOV,MAAM,IAAI,EAAE,aAAa;YAGlC,OAAO;QACT;IACF;IAGA,OAAO,EAAE;AACX;AAEO,MAAMnB,kBAAkB,CAC7BiC,YACAC;IAGA,IAAI,CAACC,cAAcF,aACjB,OAAO,CAAC;IAGV,MAAMG,gBAAgBT,4BAA4BO;IAClD,MAAMG,SAAS;QAAE,GAAGJ,UAAU;IAAC;IAE/B,KAAK,MAAMK,aAAaF,cAAe;QACrC,MAAMG,aAAaF,MAAM,CAACC,UAAU;QACpC,IAAIC,YAEF;YAAA,IAAI,AAAsB,YAAtB,OAAOA,YACTF,MAAM,CAACC,UAAU,GAAGC;iBACf,IAAI,AAAsB,YAAtB,OAAOA,YAEhB;gBAAA,IAAIA,WAAW,MAAM,EAEnB;oBAAA,IAAI,AAA6B,YAA7B,OAAOA,WAAW,MAAM,EAC1BF,MAAM,CAACC,UAAU,GAAGC,WAAW,MAAM;yBAChC,IACL,AAA6B,YAA7B,OAAOA,WAAW,MAAM,IACxBA,WAAW,MAAM,CAAC,MAAM,EAGxBF,MAAM,CAACC,UAAU,GAAGhB,uBAAuBiB,WAAW,MAAM;gBAC9D;YACF;QACF;IAEJ;IAEA,OAAOF;AACT;AAaO,MAAMG,mBAAmB,CAC9BC,UACAP,WACAQ;IAGA,IAAI,CAACR,WACH;IAIF,MAAMS,QAAQF,YAAY,CAAC;IAG3B,MAAMG,eAAejB,4BAA4BO;IAGjD,IAAIU,AAAwB,MAAxBA,aAAa,MAAM,EACrB,OAAOV,UAAU,KAAK,CAACS;IAIzB,MAAME,oBAAyC,CAAC;IAChD,KAAK,MAAMP,aAAaM,aACtB,IAAIN,aAAaK,OACfE,iBAAiB,CAACP,UAAU,GAAGK,KAAK,CAACL,UAAU;IAKnD,MAAMQ,sBAA2C,CAAC;IAClD,IAAK,MAAMd,OAAOW,MAChB,IAAIC,aAAa,QAAQ,CAACZ,MAExBc,mBAAmB,CAACd,IAAI,GAAG;QAAE,QAAQ;IAAU;SAE/Cc,mBAAmB,CAACd,IAAI,GAAGW,KAAK,CAACX,IAAI;IAKzC,MAAMe,YAAYb,UAAU,KAAK,CAACY;IAIlC,MAAME,QAAQN,SAAS;IACvB,IAAK,MAAMJ,aAAaO,kBAAmB;QACzC,IAAII,QAAQJ,iBAAiB,CAACP,UAAU;QACxC,IACEU,AAAU9C,WAAV8C,SACAA,AAAU,MAAVA,SACAC,SACA,AAAiB,YAAjB,OAAOA,SACPA,MAAM,MAAM,IACZA,MAAM,IAAI,EAEVA,QAAQ;YACN,GAAGA,KAAK;YACR,QAAQ;gBACNC,KAAK,KAAK,CAACD,MAAM,MAAM,CAAC,EAAE,GAAGD;gBAC7BE,KAAK,KAAK,CAACD,MAAM,MAAM,CAAC,EAAE,GAAGD;aAC9B;YACD,MAAM;gBACJ,GAAGC,MAAM,IAAI;gBACb,MAAMC,KAAK,KAAK,CAACD,MAAM,IAAI,CAAC,IAAI,GAAGD;gBACnC,KAAKE,KAAK,KAAK,CAACD,MAAM,IAAI,CAAC,GAAG,GAAGD;gBACjC,OAAOE,KAAK,KAAK,CAACD,MAAM,IAAI,CAAC,KAAK,GAAGD;gBACrC,QAAQE,KAAK,KAAK,CAACD,MAAM,IAAI,CAAC,MAAM,GAAGD;YACzC;QACF;QAEFD,SAAS,CAACT,UAAU,GAAGW;IACzB;IAEA,OAAOF;AACT;AAEO,MAAMI,qBAAqB;AAQ3B,MAAMC,wBAAwB,CACnCC,SAAS,qBAAqB,EAC9BC;IAEA,MAAMC,MAAMD,AAAcpD,WAAdoD,YAA0B,IAAIE,KAAKF,aAAa,IAAIE;IAChE,MAAMC,OAAOF,IAAI,WAAW;IAC5B,MAAMG,QAAQC,OAAOJ,IAAI,QAAQ,KAAK,GAAG,QAAQ,CAAC,GAAG;IACrD,MAAMK,MAAMD,OAAOJ,IAAI,OAAO,IAAI,QAAQ,CAAC,GAAG;IAC9C,MAAMM,QAAQF,OAAOJ,IAAI,QAAQ,IAAI,QAAQ,CAAC,GAAG;IACjD,MAAMO,UAAUH,OAAOJ,IAAI,UAAU,IAAI,QAAQ,CAAC,GAAG;IACrD,MAAMQ,UAAUJ,OAAOJ,IAAI,UAAU,IAAI,QAAQ,CAAC,GAAG;IAErD,MAAMS,aAAaX,OAChB,OAAO,CAAC,QAAQM,OAAOF,OACvB,OAAO,CAAC,MAAMC,OACd,OAAO,CAAC,MAAME,KACd,OAAO,CAAC,MAAMC,OACd,OAAO,CAAC,MAAMC,SACd,OAAO,CAAC,MAAMC;IAEjB,OAAO,GAAGC,WAAW,EAAE,EAAEX,OAAO,CAAC,CAAC;AACpC"}
|
|
1
|
+
{"version":3,"file":"common.mjs","sources":["../../src/common.ts"],"sourcesContent":["import type {\n DeviceAction,\n MidsceneYamlFlowItem,\n PlanningAction,\n} from '@/types';\nimport { isPlainObject } from '@midscene/shared/utils';\nimport { z } from 'zod';\n\nexport function findActionInActionSpaceOrThrow(\n planType: string,\n actionSpace: DeviceAction<any>[],\n): DeviceAction<any> {\n const action = actionSpace.find((item) => item.name === planType);\n if (!action) {\n const available = actionSpace.map((item) => item.name).join(', ');\n throw new Error(\n `Action type '${planType}' is not in the current action space. Available actions: ${available || '(none)'}`,\n );\n }\n return action;\n}\n\nexport function buildYamlFlowFromPlans(\n plans: PlanningAction[],\n actionSpace: DeviceAction<any>[],\n): MidsceneYamlFlowItem[] {\n const flow: MidsceneYamlFlowItem[] = [];\n\n for (const plan of plans) {\n const verb = plan.type;\n\n const action = findActionInActionSpaceOrThrow(verb, actionSpace);\n\n const flowKey = action.interfaceAlias || verb;\n const flowParam = action.paramSchema\n ? dumpActionParam(plan.param || {}, action.paramSchema)\n : {};\n\n // For actions whose param is a single string field (e.g. Launch/Terminate's\n // `uri`, RunAdbShell's `command`), inline the value on the flowKey. Writing\n // `{ terminate: '', uri: '...' }` makes the YAML player treat the empty\n // string as the param and drop the sibling `uri`, so cache replay would\n // call the action with an empty argument.\n const shortcutField =\n action.name === 'Launch' || action.interfaceAlias === 'launch'\n ? 'uri'\n : action.name === 'Terminate' || action.interfaceAlias === 'terminate'\n ? 'uri'\n : action.name === 'RunAdbShell' ||\n action.interfaceAlias === 'runAdbShell' ||\n action.name === 'RunHdcShell' ||\n action.interfaceAlias === 'runHdcShell'\n ? 'command'\n : undefined;\n const shortcutKeys = shortcutField ? Object.keys(flowParam) : [];\n const canInlineShortcut =\n shortcutField &&\n shortcutKeys.length === 1 &&\n shortcutKeys[0] === shortcutField &&\n typeof flowParam[shortcutField] === 'string';\n\n const flowItem: MidsceneYamlFlowItem = canInlineShortcut\n ? { [flowKey]: flowParam[shortcutField as string] }\n : { [flowKey]: '', ...flowParam };\n\n flow.push(flowItem);\n }\n\n return flow;\n}\n\n// Zod schemas for shared types\nexport const PointSchema = z.object({\n left: z.number(),\n top: z.number(),\n});\n\nexport const SizeSchema = z.object({\n width: z.number(),\n height: z.number(),\n});\n\nexport const RectSchema = PointSchema.and(SizeSchema).and(\n z.object({\n zoom: z.number().optional(),\n }),\n);\n\n// Zod schema for TMultimodalPrompt\nexport const TMultimodalPromptSchema = z.object({\n images: z\n .array(\n z.object({\n name: z.string(),\n url: z.string(),\n }),\n )\n .optional(),\n convertHttpImage2Base64: z.boolean().optional(),\n});\n\n// Zod schema for TUserPrompt\nexport const TUserPromptSchema = z.union([\n z.string(),\n z\n .object({\n prompt: z.string(),\n })\n .and(TMultimodalPromptSchema.partial()),\n]);\n\n// Generate TypeScript types from Zod schemas\nexport type TMultimodalPrompt = z.infer<typeof TMultimodalPromptSchema>;\nexport type TUserPrompt = z.infer<typeof TUserPromptSchema>;\n\nexport const userPromptToString = (prompt: TUserPrompt): string => {\n return typeof prompt === 'string' ? prompt : prompt.prompt;\n};\n\nexport const userPromptToMultimodalPrompt = (\n prompt: TUserPrompt,\n): TMultimodalPrompt | undefined => {\n if (typeof prompt === 'string' || !prompt.images) {\n return undefined;\n }\n return {\n images: prompt.images,\n convertHttpImage2Base64: !!prompt.convertHttpImage2Base64,\n };\n};\n\n// Schema for locator field input (when users provide locate parameters)\nconst MidsceneLocationInput = z\n .object({\n prompt: TUserPromptSchema,\n deepLocate: z.boolean().optional(),\n deepThink: z\n .boolean()\n .optional()\n .describe('@deprecated Use `deepLocate` instead.'),\n cacheable: z.boolean().optional(),\n xpath: z.union([z.string(), z.boolean()]).optional(),\n })\n .passthrough();\n\n/**\n * Returns the schema for locator fields.\n * This now returns the input schema which is more permissive and suitable for validation.\n */\nexport const getMidsceneLocationSchema = () => {\n return MidsceneLocationInput;\n};\n\nexport const ifMidsceneLocatorField = (field: any): boolean => {\n // Handle optional fields by getting the inner type\n let actualField = field;\n if (actualField._def?.typeName === 'ZodOptional') {\n actualField = actualField._def.innerType;\n }\n\n // Check if this is a ZodObject\n if (actualField._def?.typeName === 'ZodObject') {\n const shape = actualField._def.shape();\n\n // Input schema has 'prompt' as a required field\n if ('prompt' in shape && shape.prompt) {\n return true;\n }\n }\n\n return false;\n};\n\nconst formatPromptWithImages = (\n promptObj: Exclude<TUserPrompt, string>,\n): string => {\n let promptString = promptObj.prompt;\n if (Array.isArray(promptObj.images) && promptObj.images.length > 0) {\n const imageCount = promptObj.images.length;\n promptString += ` (with ${imageCount} image${imageCount > 1 ? 's' : ''})`;\n }\n return promptString;\n};\n\nexport const findAllMidsceneLocatorField = (\n zodType?: z.ZodType<any>,\n requiredOnly?: boolean,\n): string[] => {\n if (!zodType) {\n return [];\n }\n\n // Check if this is a ZodObject by checking if it has a shape property\n const zodObject = zodType as any;\n if (zodObject._def?.typeName === 'ZodObject' && zodObject.shape) {\n const keys = Object.keys(zodObject.shape);\n return keys.filter((key) => {\n const field = zodObject.shape[key];\n // TODO: A similar locator-field check exists as isMidsceneLocatorField in\n // @midscene/shared/zod-schema-utils.\n if (!ifMidsceneLocatorField(field)) {\n return false;\n }\n\n // If requiredOnly is true, filter out optional fields\n if (requiredOnly) {\n return field._def?.typeName !== 'ZodOptional';\n }\n\n return true;\n });\n }\n\n // For other ZodType instances, we can't extract field names\n return [];\n};\n\nexport const dumpActionParam = (\n jsonObject: Record<string, any>,\n zodSchema: z.ZodType<any>,\n): Record<string, any> => {\n // Prevent spreading strings into {0: 'c', 1: 'o', ...}\n if (!isPlainObject(jsonObject)) {\n return {};\n }\n\n const locatorFields = findAllMidsceneLocatorField(zodSchema);\n const result = { ...jsonObject };\n\n for (const fieldName of locatorFields) {\n const fieldValue = result[fieldName];\n if (fieldValue) {\n // If it's already a string, keep it as is\n if (typeof fieldValue === 'string') {\n result[fieldName] = fieldValue;\n } else if (typeof fieldValue === 'object') {\n // Check if this field is actually a MidsceneLocationType object\n if (fieldValue.prompt) {\n // If prompt is a string, use it directly\n if (typeof fieldValue.prompt === 'string') {\n result[fieldName] = fieldValue.prompt;\n } else if (\n typeof fieldValue.prompt === 'object' &&\n fieldValue.prompt.prompt\n ) {\n // If prompt is a TUserPrompt object, extract the prompt string\n result[fieldName] = formatPromptWithImages(fieldValue.prompt);\n }\n }\n }\n }\n }\n\n return result;\n};\n\n/**\n * Parse and validate action parameters using Zod schema.\n * All fields are validated through Zod, EXCEPT locator fields which are skipped.\n * Default values defined in the schema are automatically applied.\n *\n * Locator fields are special business logic fields with complex validation requirements,\n * so they are intentionally excluded from Zod parsing and use existing validation logic.\n *\n * When shrunkShotToLogicalRatio is provided and !== 1, coordinates in locate fields\n * are transformed from screenshot space to logical space.\n */\nexport const parseActionParam = (\n rawParam: Record<string, any> | undefined,\n zodSchema?: z.ZodType<any>,\n options?: { shrunkShotToLogicalRatio?: number },\n): Record<string, any> | undefined => {\n // If no schema is provided, return undefined (action takes no parameters)\n if (!zodSchema) {\n return undefined;\n }\n\n // Handle undefined or null rawParam by providing an empty object\n const param = rawParam ?? {};\n\n // Find all locate fields in the schema\n const locateFields = findAllMidsceneLocatorField(zodSchema);\n\n // If there are no locate fields, just do normal validation\n if (locateFields.length === 0) {\n return zodSchema.parse(param);\n }\n\n // Extract locate field values to restore later\n const locateFieldValues: Record<string, any> = {};\n for (const fieldName of locateFields) {\n if (fieldName in param) {\n locateFieldValues[fieldName] = param[fieldName];\n }\n }\n\n // Build params for validation - skip locate fields and use dummy values\n const paramsForValidation: Record<string, any> = {};\n for (const key in param) {\n if (locateFields.includes(key)) {\n // Use dummy value to satisfy schema validation\n paramsForValidation[key] = { prompt: '_dummy_' };\n } else {\n paramsForValidation[key] = param[key];\n }\n }\n\n // Validate with dummy locate values\n const validated = zodSchema.parse(paramsForValidation);\n\n // Restore the actual locate field values (unvalidated, as per business requirement),\n // and transform coordinates from screenshot space to logical space if needed\n const ratio = options?.shrunkShotToLogicalRatio;\n for (const fieldName in locateFieldValues) {\n let value = locateFieldValues[fieldName];\n if (\n ratio !== undefined &&\n value &&\n typeof value === 'object' &&\n value.center\n ) {\n value = {\n ...value,\n center: [\n Math.round(value.center[0] / ratio),\n Math.round(value.center[1] / ratio),\n ],\n };\n }\n validated[fieldName] = value;\n }\n\n return validated;\n};\n\nexport const finalizeActionName = 'Finalize';\n\n/**\n * Get a readable time string for a given timestamp or the current time\n * @param format - Optional format string. Supports: YYYY, MM, DD, HH, mm, ss. Default: 'YYYY-MM-DD HH:mm:ss'\n * @param timestamp - Optional timestamp in milliseconds. If not provided, uses current system time.\n * @returns A formatted time string with format label\n */\nexport const getReadableTimeString = (\n format = 'YYYY-MM-DD HH:mm:ss',\n timestamp?: number,\n): string => {\n const now = timestamp !== undefined ? new Date(timestamp) : new Date();\n const year = now.getFullYear();\n const month = String(now.getMonth() + 1).padStart(2, '0');\n const day = String(now.getDate()).padStart(2, '0');\n const hours = String(now.getHours()).padStart(2, '0');\n const minutes = String(now.getMinutes()).padStart(2, '0');\n const seconds = String(now.getSeconds()).padStart(2, '0');\n\n const timeString = format\n .replace('YYYY', String(year))\n .replace('MM', month)\n .replace('DD', day)\n .replace('HH', hours)\n .replace('mm', minutes)\n .replace('ss', seconds);\n\n return `${timeString} (${format})`;\n};\n"],"names":["findActionInActionSpaceOrThrow","planType","actionSpace","action","item","available","Error","buildYamlFlowFromPlans","plans","flow","plan","verb","flowKey","flowParam","dumpActionParam","shortcutField","undefined","shortcutKeys","Object","canInlineShortcut","flowItem","PointSchema","z","SizeSchema","RectSchema","TMultimodalPromptSchema","TUserPromptSchema","userPromptToString","prompt","userPromptToMultimodalPrompt","MidsceneLocationInput","getMidsceneLocationSchema","ifMidsceneLocatorField","field","actualField","shape","formatPromptWithImages","promptObj","promptString","Array","imageCount","findAllMidsceneLocatorField","zodType","requiredOnly","zodObject","keys","key","jsonObject","zodSchema","isPlainObject","locatorFields","result","fieldName","fieldValue","parseActionParam","rawParam","options","param","locateFields","locateFieldValues","paramsForValidation","validated","ratio","value","Math","finalizeActionName","getReadableTimeString","format","timestamp","now","Date","year","month","String","day","hours","minutes","seconds","timeString"],"mappings":";;AAQO,SAASA,+BACdC,QAAgB,EAChBC,WAAgC;IAEhC,MAAMC,SAASD,YAAY,IAAI,CAAC,CAACE,OAASA,KAAK,IAAI,KAAKH;IACxD,IAAI,CAACE,QAAQ;QACX,MAAME,YAAYH,YAAY,GAAG,CAAC,CAACE,OAASA,KAAK,IAAI,EAAE,IAAI,CAAC;QAC5D,MAAM,IAAIE,MACR,CAAC,aAAa,EAAEL,SAAS,yDAAyD,EAAEI,aAAa,UAAU;IAE/G;IACA,OAAOF;AACT;AAEO,SAASI,uBACdC,KAAuB,EACvBN,WAAgC;IAEhC,MAAMO,OAA+B,EAAE;IAEvC,KAAK,MAAMC,QAAQF,MAAO;QACxB,MAAMG,OAAOD,KAAK,IAAI;QAEtB,MAAMP,SAASH,+BAA+BW,MAAMT;QAEpD,MAAMU,UAAUT,OAAO,cAAc,IAAIQ;QACzC,MAAME,YAAYV,OAAO,WAAW,GAChCW,gBAAgBJ,KAAK,KAAK,IAAI,CAAC,GAAGP,OAAO,WAAW,IACpD,CAAC;QAOL,MAAMY,gBACJZ,AAAgB,aAAhBA,OAAO,IAAI,IAAiBA,AAA0B,aAA1BA,OAAO,cAAc,GAC7C,QACAA,AAAgB,gBAAhBA,OAAO,IAAI,IAAoBA,AAA0B,gBAA1BA,OAAO,cAAc,GAClD,QACAA,AAAgB,kBAAhBA,OAAO,IAAI,IACTA,AAA0B,kBAA1BA,OAAO,cAAc,IACrBA,AAAgB,kBAAhBA,OAAO,IAAI,IACXA,AAA0B,kBAA1BA,OAAO,cAAc,GACrB,YACAa;QACV,MAAMC,eAAeF,gBAAgBG,OAAO,IAAI,CAACL,aAAa,EAAE;QAChE,MAAMM,oBACJJ,iBACAE,AAAwB,MAAxBA,aAAa,MAAM,IACnBA,YAAY,CAAC,EAAE,KAAKF,iBACpB,AAAoC,YAApC,OAAOF,SAAS,CAACE,cAAc;QAEjC,MAAMK,WAAiCD,oBACnC;YAAE,CAACP,QAAQ,EAAEC,SAAS,CAACE,cAAwB;QAAC,IAChD;YAAE,CAACH,QAAQ,EAAE;YAAI,GAAGC,SAAS;QAAC;QAElCJ,KAAK,IAAI,CAACW;IACZ;IAEA,OAAOX;AACT;AAGO,MAAMY,cAAcC,EAAE,MAAM,CAAC;IAClC,MAAMA,EAAE,MAAM;IACd,KAAKA,EAAE,MAAM;AACf;AAEO,MAAMC,aAAaD,EAAE,MAAM,CAAC;IACjC,OAAOA,EAAE,MAAM;IACf,QAAQA,EAAE,MAAM;AAClB;AAEO,MAAME,aAAaH,YAAY,GAAG,CAACE,YAAY,GAAG,CACvDD,EAAE,MAAM,CAAC;IACP,MAAMA,EAAE,MAAM,GAAG,QAAQ;AAC3B;AAIK,MAAMG,0BAA0BH,EAAE,MAAM,CAAC;IAC9C,QAAQA,EAAAA,KACA,CACJA,EAAE,MAAM,CAAC;QACP,MAAMA,EAAE,MAAM;QACd,KAAKA,EAAE,MAAM;IACf,IAED,QAAQ;IACX,yBAAyBA,EAAE,OAAO,GAAG,QAAQ;AAC/C;AAGO,MAAMI,oBAAoBJ,EAAE,KAAK,CAAC;IACvCA,EAAE,MAAM;IACRA,EAAAA,MACS,CAAC;QACN,QAAQA,EAAE,MAAM;IAClB,GACC,GAAG,CAACG,wBAAwB,OAAO;CACvC;AAMM,MAAME,qBAAqB,CAACC,SAC1B,AAAkB,YAAlB,OAAOA,SAAsBA,SAASA,OAAO,MAAM;AAGrD,MAAMC,+BAA+B,CAC1CD;IAEA,IAAI,AAAkB,YAAlB,OAAOA,UAAuB,CAACA,OAAO,MAAM,EAC9C;IAEF,OAAO;QACL,QAAQA,OAAO,MAAM;QACrB,yBAAyB,CAAC,CAACA,OAAO,uBAAuB;IAC3D;AACF;AAGA,MAAME,wBAAwBR,EAAAA,MACrB,CAAC;IACN,QAAQI;IACR,YAAYJ,EAAE,OAAO,GAAG,QAAQ;IAChC,WAAWA,EAAAA,OACD,GACP,QAAQ,GACR,QAAQ,CAAC;IACZ,WAAWA,EAAE,OAAO,GAAG,QAAQ;IAC/B,OAAOA,EAAE,KAAK,CAAC;QAACA,EAAE,MAAM;QAAIA,EAAE,OAAO;KAAG,EAAE,QAAQ;AACpD,GACC,WAAW;AAMP,MAAMS,4BAA4B,IAChCD;AAGF,MAAME,yBAAyB,CAACC;IAErC,IAAIC,cAAcD;IAClB,IAAIC,YAAY,IAAI,EAAE,aAAa,eACjCA,cAAcA,YAAY,IAAI,CAAC,SAAS;IAI1C,IAAIA,YAAY,IAAI,EAAE,aAAa,aAAa;QAC9C,MAAMC,QAAQD,YAAY,IAAI,CAAC,KAAK;QAGpC,IAAI,YAAYC,SAASA,MAAM,MAAM,EACnC,OAAO;IAEX;IAEA,OAAO;AACT;AAEA,MAAMC,yBAAyB,CAC7BC;IAEA,IAAIC,eAAeD,UAAU,MAAM;IACnC,IAAIE,MAAM,OAAO,CAACF,UAAU,MAAM,KAAKA,UAAU,MAAM,CAAC,MAAM,GAAG,GAAG;QAClE,MAAMG,aAAaH,UAAU,MAAM,CAAC,MAAM;QAC1CC,gBAAgB,CAAC,OAAO,EAAEE,WAAW,MAAM,EAAEA,aAAa,IAAI,MAAM,GAAG,CAAC,CAAC;IAC3E;IACA,OAAOF;AACT;AAEO,MAAMG,8BAA8B,CACzCC,SACAC;IAEA,IAAI,CAACD,SACH,OAAO,EAAE;IAIX,MAAME,YAAYF;IAClB,IAAIE,UAAU,IAAI,EAAE,aAAa,eAAeA,UAAU,KAAK,EAAE;QAC/D,MAAMC,OAAO3B,OAAO,IAAI,CAAC0B,UAAU,KAAK;QACxC,OAAOC,KAAK,MAAM,CAAC,CAACC;YAClB,MAAMb,QAAQW,UAAU,KAAK,CAACE,IAAI;YAGlC,IAAI,CAACd,uBAAuBC,QAC1B,OAAO;YAIT,IAAIU,cACF,OAAOV,MAAM,IAAI,EAAE,aAAa;YAGlC,OAAO;QACT;IACF;IAGA,OAAO,EAAE;AACX;AAEO,MAAMnB,kBAAkB,CAC7BiC,YACAC;IAGA,IAAI,CAACC,cAAcF,aACjB,OAAO,CAAC;IAGV,MAAMG,gBAAgBT,4BAA4BO;IAClD,MAAMG,SAAS;QAAE,GAAGJ,UAAU;IAAC;IAE/B,KAAK,MAAMK,aAAaF,cAAe;QACrC,MAAMG,aAAaF,MAAM,CAACC,UAAU;QACpC,IAAIC,YAEF;YAAA,IAAI,AAAsB,YAAtB,OAAOA,YACTF,MAAM,CAACC,UAAU,GAAGC;iBACf,IAAI,AAAsB,YAAtB,OAAOA,YAEhB;gBAAA,IAAIA,WAAW,MAAM,EAEnB;oBAAA,IAAI,AAA6B,YAA7B,OAAOA,WAAW,MAAM,EAC1BF,MAAM,CAACC,UAAU,GAAGC,WAAW,MAAM;yBAChC,IACL,AAA6B,YAA7B,OAAOA,WAAW,MAAM,IACxBA,WAAW,MAAM,CAAC,MAAM,EAGxBF,MAAM,CAACC,UAAU,GAAGhB,uBAAuBiB,WAAW,MAAM;gBAC9D;YACF;QACF;IAEJ;IAEA,OAAOF;AACT;AAaO,MAAMG,mBAAmB,CAC9BC,UACAP,WACAQ;IAGA,IAAI,CAACR,WACH;IAIF,MAAMS,QAAQF,YAAY,CAAC;IAG3B,MAAMG,eAAejB,4BAA4BO;IAGjD,IAAIU,AAAwB,MAAxBA,aAAa,MAAM,EACrB,OAAOV,UAAU,KAAK,CAACS;IAIzB,MAAME,oBAAyC,CAAC;IAChD,KAAK,MAAMP,aAAaM,aACtB,IAAIN,aAAaK,OACfE,iBAAiB,CAACP,UAAU,GAAGK,KAAK,CAACL,UAAU;IAKnD,MAAMQ,sBAA2C,CAAC;IAClD,IAAK,MAAMd,OAAOW,MAChB,IAAIC,aAAa,QAAQ,CAACZ,MAExBc,mBAAmB,CAACd,IAAI,GAAG;QAAE,QAAQ;IAAU;SAE/Cc,mBAAmB,CAACd,IAAI,GAAGW,KAAK,CAACX,IAAI;IAKzC,MAAMe,YAAYb,UAAU,KAAK,CAACY;IAIlC,MAAME,QAAQN,SAAS;IACvB,IAAK,MAAMJ,aAAaO,kBAAmB;QACzC,IAAII,QAAQJ,iBAAiB,CAACP,UAAU;QACxC,IACEU,AAAU9C,WAAV8C,SACAC,SACA,AAAiB,YAAjB,OAAOA,SACPA,MAAM,MAAM,EAEZA,QAAQ;YACN,GAAGA,KAAK;YACR,QAAQ;gBACNC,KAAK,KAAK,CAACD,MAAM,MAAM,CAAC,EAAE,GAAGD;gBAC7BE,KAAK,KAAK,CAACD,MAAM,MAAM,CAAC,EAAE,GAAGD;aAC9B;QACH;QAEFD,SAAS,CAACT,UAAU,GAAGW;IACzB;IAEA,OAAOF;AACT;AAEO,MAAMI,qBAAqB;AAQ3B,MAAMC,wBAAwB,CACnCC,SAAS,qBAAqB,EAC9BC;IAEA,MAAMC,MAAMD,AAAcpD,WAAdoD,YAA0B,IAAIE,KAAKF,aAAa,IAAIE;IAChE,MAAMC,OAAOF,IAAI,WAAW;IAC5B,MAAMG,QAAQC,OAAOJ,IAAI,QAAQ,KAAK,GAAG,QAAQ,CAAC,GAAG;IACrD,MAAMK,MAAMD,OAAOJ,IAAI,OAAO,IAAI,QAAQ,CAAC,GAAG;IAC9C,MAAMM,QAAQF,OAAOJ,IAAI,QAAQ,IAAI,QAAQ,CAAC,GAAG;IACjD,MAAMO,UAAUH,OAAOJ,IAAI,UAAU,IAAI,QAAQ,CAAC,GAAG;IACrD,MAAMQ,UAAUJ,OAAOJ,IAAI,UAAU,IAAI,QAAQ,CAAC,GAAG;IAErD,MAAMS,aAAaX,OAChB,OAAO,CAAC,QAAQM,OAAOF,OACvB,OAAO,CAAC,MAAMC,OACd,OAAO,CAAC,MAAME,KACd,OAAO,CAAC,MAAMC,OACd,OAAO,CAAC,MAAMC,SACd,OAAO,CAAC,MAAMC;IAEjB,OAAO,GAAGC,WAAW,EAAE,EAAEX,OAAO,CAAC,CAAC;AACpC"}
|
|
@@ -13,8 +13,8 @@ const includedInRect = (point, rect)=>{
|
|
|
13
13
|
};
|
|
14
14
|
const buildLocateValidatorResult = (expectCenter, located, verifyLocateOption)=>{
|
|
15
15
|
const distance = distanceOfTwoPoints(expectCenter, located.center);
|
|
16
|
-
const included = includedInRect(expectCenter, located.rect);
|
|
17
|
-
const pass = distance <= (verifyLocateOption?.centerDistanceThreshold || 20) || included;
|
|
16
|
+
const included = located.rect ? includedInRect(expectCenter, located.rect) : void 0;
|
|
17
|
+
const pass = distance <= (verifyLocateOption?.centerDistanceThreshold || 20) || true === included;
|
|
18
18
|
return {
|
|
19
19
|
pass,
|
|
20
20
|
rect: located.rect,
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"element-describer.mjs","sources":["../../src/element-describer.ts"],"sourcesContent":["import type { ModelRuntime } from '@/ai-model/models';\nimport { assert } from '@midscene/shared/utils';\nimport { createScreenshotBoundUIContext } from './agent/utils';\nimport type Service from './service';\nimport type {\n AgentDescribeElementAtPointResult,\n LocateOption,\n LocateResultElement,\n LocateValidatorResult,\n LocatorValidatorOption,\n Rect,\n Size,\n UIContext,\n} from './types';\nimport { buildDetailedLocateParam } from './yaml/utils';\n\nexport type DescribeElementCoordinateSpace = 'screenshot' | 'logical';\n\nexport type LocatorVerifyFn = (input: {\n prompt: string;\n expectCenter: [number, number];\n retryCount: number;\n verifyResult: LocateValidatorResult;\n}) => LocateValidatorResult | boolean;\n\nexport type DescribeElementAtPointOptions = {\n verifyPrompt?: boolean;\n retryLimit?: number;\n deepDescribe?: boolean;\n deepLocate?: boolean;\n locatorVerifyFn?: LocatorVerifyFn;\n screenshotBase64?: string;\n screenshotSize?: Size;\n coordinateSpace?: DescribeElementCoordinateSpace;\n logicalSize?: Size;\n onProgress?: (progress: {\n prompt?: string;\n deepDescribe?: boolean;\n deepLocate?: boolean;\n verifyResult?: LocateValidatorResult;\n }) => void;\n} & LocatorValidatorOption;\n\ntype ScreenshotBoundContextOptions = {\n screenshotBase64?: string;\n screenshotSize?: Size;\n coordinateSpace?: DescribeElementCoordinateSpace;\n logicalSize?: Size;\n};\n\nexport type VerifyElementDescriptionAtPointOptions =\n ScreenshotBoundContextOptions & LocatorValidatorOption;\n\ntype LocateAndVerifyOptions = VerifyElementDescriptionAtPointOptions &\n Pick<LocateOption, 'cacheable' | 'deepLocate' | 'xpath'> & {\n abortSignal?: AbortSignal;\n };\n\nexport type ElementDescriberRuntime = {\n service: Pick<Service, 'describe' | 'locate'>;\n describeModelRuntime: ModelRuntime;\n locateModelRuntime: ModelRuntime;\n};\n\ntype ServiceDescribeOptions = NonNullable<\n Parameters<ElementDescriberRuntime['service']['describe']>[2]\n>;\n\nconst distanceOfTwoPoints = (p1: [number, number], p2: [number, number]) => {\n const [x1, y1] = p1;\n const [x2, y2] = p2;\n return Math.round(Math.sqrt((x1 - x2) ** 2 + (y1 - y2) ** 2));\n};\n\nconst includedInRect = (point: [number, number], rect: Rect) => {\n const [x, y] = point;\n const { left, top, width, height } = rect;\n return x >= left && x <= left + width && y >= top && y <= top + height;\n};\n\nconst buildLocateValidatorResult = (\n expectCenter: [number, number],\n located: Pick<LocateResultElement, 'center' | 'rect'>,\n verifyLocateOption?: LocatorValidatorOption,\n): LocateValidatorResult => {\n const distance = distanceOfTwoPoints(expectCenter, located.center);\n const included = includedInRect(expectCenter, located.rect);\n const pass =\n distance <= (verifyLocateOption?.centerDistanceThreshold || 20) || included;\n return {\n pass,\n rect: located.rect,\n center: located.center,\n centerDistance: distance,\n includedInRect: included,\n };\n};\n\nfunction assertPositiveSize(\n size: Size | undefined,\n label: string,\n): asserts size is Size {\n assert(\n size &&\n Number.isFinite(size.width) &&\n Number.isFinite(size.height) &&\n size.width > 0 &&\n size.height > 0,\n `${label} must include positive width and height`,\n );\n}\n\nconst mapPointToScreenshotSpace = (\n center: [number, number],\n screenshotSize: Size,\n opt: ScreenshotBoundContextOptions,\n): [number, number] => {\n const coordinateSpace = opt.coordinateSpace || 'screenshot';\n if (coordinateSpace === 'screenshot') {\n return center;\n }\n\n assertPositiveSize(\n opt.logicalSize,\n 'logicalSize is required when coordinateSpace is logical',\n );\n return [\n (center[0] * screenshotSize.width) / opt.logicalSize.width,\n (center[1] * screenshotSize.height) / opt.logicalSize.height,\n ];\n};\n\nconst createScreenshotBoundLocatorContext = async (\n center: [number, number],\n opt?: ScreenshotBoundContextOptions,\n): Promise<{\n screenshotContext?: UIContext;\n locateOpt?: LocateOption;\n targetCenter: [number, number];\n}> => {\n const screenshotContext = opt?.screenshotBase64\n ? await createScreenshotBoundUIContext(opt.screenshotBase64, opt)\n : undefined;\n const targetCenter = screenshotContext\n ? mapPointToScreenshotSpace(center, screenshotContext.shotSize, opt || {})\n : center;\n return {\n screenshotContext,\n locateOpt: screenshotContext ? { uiContext: screenshotContext } : undefined,\n targetCenter,\n };\n};\n\nexport async function verifyLocator(\n runtime: Pick<ElementDescriberRuntime, 'service' | 'locateModelRuntime'>,\n prompt: string,\n locateOpt: LocateOption | undefined,\n expectCenter: [number, number],\n verifyLocateOption?: LocatorValidatorOption &\n Pick<LocateOption, 'deepLocate'>,\n): Promise<LocateValidatorResult> {\n return locateAndVerify(runtime, prompt, expectCenter, {\n centerDistanceThreshold: verifyLocateOption?.centerDistanceThreshold,\n deepLocate: verifyLocateOption?.deepLocate,\n uiContext: locateOpt?.uiContext,\n });\n}\n\nfunction applyLocatorVerifyFn(\n locatorVerifyFn: LocatorVerifyFn | undefined,\n input: Parameters<LocatorVerifyFn>[0],\n): LocateValidatorResult {\n if (!locatorVerifyFn) {\n return input.verifyResult;\n }\n const customResult = locatorVerifyFn(input);\n if (typeof customResult === 'boolean') {\n return {\n ...input.verifyResult,\n pass: customResult,\n };\n }\n return customResult;\n}\n\nfunction errorMessage(error: unknown): string {\n return error instanceof Error ? error.message : String(error);\n}\n\nexport async function describeElementAtPoint(\n runtime: ElementDescriberRuntime,\n center: [number, number],\n opt?: DescribeElementAtPointOptions,\n): Promise<AgentDescribeElementAtPointResult> {\n const { verifyPrompt = true, retryLimit = 3 } = opt || {};\n const { screenshotContext, locateOpt, targetCenter } =\n await createScreenshotBoundLocatorContext(center, opt);\n\n let success = false;\n let retryCount = 0;\n let resultPrompt = '';\n const autoRetryDeepDescribe = opt?.deepDescribe === undefined;\n const autoRetryDeepLocate = opt?.deepLocate === undefined;\n let deepDescribe = opt?.deepDescribe || false;\n let deepLocate = opt?.deepLocate || false;\n let verifyResult: LocateValidatorResult | undefined;\n let lastError: string | undefined;\n let failureStage: AgentDescribeElementAtPointResult['failureStage'];\n\n while (!success && retryCount < retryLimit) {\n if (retryCount >= 1 && autoRetryDeepDescribe) {\n deepDescribe = true;\n }\n if (retryCount >= 1 && autoRetryDeepLocate) {\n deepLocate = true;\n }\n const describeModelRuntime = runtime.describeModelRuntime;\n const locateModelRuntime = runtime.locateModelRuntime;\n const retryRuntime: ElementDescriberRuntime = {\n ...runtime,\n describeModelRuntime,\n locateModelRuntime,\n };\n const describeOpt: ServiceDescribeOptions = screenshotContext\n ? {\n deepDescribe,\n context: screenshotContext,\n }\n : {\n deepDescribe,\n };\n let text: Awaited<\n ReturnType<ElementDescriberRuntime['service']['describe']>\n >;\n try {\n text = await retryRuntime.service.describe(\n targetCenter,\n retryRuntime.describeModelRuntime,\n describeOpt,\n );\n } catch (error) {\n return {\n prompt: resultPrompt,\n deepLocate,\n deepDescribe,\n verifyResult,\n success: false,\n error: errorMessage(error),\n failureStage: 'describe',\n };\n }\n if (!text.description) {\n return {\n prompt: resultPrompt,\n deepLocate,\n deepDescribe,\n verifyResult,\n success: false,\n error: `failed to describe element at [${targetCenter}]`,\n failureStage: 'describe',\n };\n }\n resultPrompt = text.description;\n if (!verifyPrompt) {\n opt?.onProgress?.({\n prompt: resultPrompt,\n deepDescribe,\n deepLocate,\n });\n success = true;\n break;\n }\n\n try {\n const candidateVerifyResult = await verifyLocator(\n retryRuntime,\n resultPrompt,\n locateOpt,\n targetCenter,\n {\n ...opt,\n deepLocate,\n },\n );\n verifyResult = applyLocatorVerifyFn(opt?.locatorVerifyFn, {\n prompt: resultPrompt,\n expectCenter: targetCenter,\n retryCount,\n verifyResult: candidateVerifyResult,\n });\n opt?.onProgress?.({\n prompt: resultPrompt,\n deepDescribe,\n deepLocate,\n verifyResult,\n });\n if (verifyResult.pass) {\n success = true;\n break;\n }\n lastError = undefined;\n failureStage = 'verify';\n } catch (error) {\n lastError = errorMessage(error);\n failureStage = 'verify';\n opt?.onProgress?.({\n prompt: resultPrompt,\n deepDescribe,\n deepLocate,\n });\n }\n retryCount++;\n }\n\n return {\n prompt: resultPrompt,\n deepLocate,\n deepDescribe,\n verifyResult,\n success,\n error:\n success || !verifyPrompt\n ? undefined\n : lastError || 'describeElementAtPoint verify failed',\n failureStage: success ? undefined : failureStage,\n };\n}\n\nexport async function verifyElementDescriptionAtPoint(\n runtime: ElementDescriberRuntime,\n description: string,\n center: [number, number],\n opt?: VerifyElementDescriptionAtPointOptions,\n): Promise<LocateValidatorResult> {\n assert(description?.trim(), 'description must not be empty');\n const { locateOpt, targetCenter } = await createScreenshotBoundLocatorContext(\n center,\n opt,\n );\n\n return verifyLocator(runtime, description, locateOpt, targetCenter, opt);\n}\n\nasync function locateAndVerify(\n runtime: Pick<ElementDescriberRuntime, 'service' | 'locateModelRuntime'>,\n description: string,\n center: [number, number],\n opt?: LocateAndVerifyOptions & {\n uiContext?: UIContext;\n },\n): Promise<LocateValidatorResult> {\n assert(description?.trim(), 'description must not be empty');\n const { screenshotContext, targetCenter } =\n await createScreenshotBoundLocatorContext(center, opt);\n const context = opt?.uiContext || screenshotContext;\n const locateParam = buildDetailedLocateParam(description, {\n cacheable: opt?.cacheable,\n deepLocate: opt?.deepLocate,\n xpath: opt?.xpath,\n });\n assert(locateParam, 'cannot get locate param for service locate');\n\n const locateResult = await runtime.service.locate(\n locateParam,\n context ? { context } : {},\n runtime.locateModelRuntime,\n opt?.abortSignal,\n );\n assert(locateResult.element, `Element not found: ${description}`);\n const verifyResult = buildLocateValidatorResult(\n targetCenter,\n locateResult.element,\n opt,\n );\n return verifyResult;\n}\n"],"names":["distanceOfTwoPoints","p1","p2","x1","y1","x2","y2","Math","includedInRect","point","rect","x","y","left","top","width","height","buildLocateValidatorResult","expectCenter","located","verifyLocateOption","distance","included","pass","assertPositiveSize","size","label","assert","Number","mapPointToScreenshotSpace","center","screenshotSize","opt","coordinateSpace","createScreenshotBoundLocatorContext","screenshotContext","createScreenshotBoundUIContext","undefined","targetCenter","verifyLocator","runtime","prompt","locateOpt","locateAndVerify","applyLocatorVerifyFn","locatorVerifyFn","input","customResult","errorMessage","error","Error","String","describeElementAtPoint","verifyPrompt","retryLimit","success","retryCount","resultPrompt","autoRetryDeepDescribe","autoRetryDeepLocate","deepDescribe","deepLocate","verifyResult","lastError","failureStage","describeModelRuntime","locateModelRuntime","retryRuntime","describeOpt","text","candidateVerifyResult","verifyElementDescriptionAtPoint","description","context","locateParam","buildDetailedLocateParam","locateResult"],"mappings":";;;AAoEA,MAAMA,sBAAsB,CAACC,IAAsBC;IACjD,MAAM,CAACC,IAAIC,GAAG,GAAGH;IACjB,MAAM,CAACI,IAAIC,GAAG,GAAGJ;IACjB,OAAOK,KAAK,KAAK,CAACA,KAAK,IAAI,CAAEJ,AAAAA,CAAAA,KAAKE,EAAC,KAAM,IAAKD,AAAAA,CAAAA,KAAKE,EAAC,KAAM;AAC5D;AAEA,MAAME,iBAAiB,CAACC,OAAyBC;IAC/C,MAAM,CAACC,GAAGC,EAAE,GAAGH;IACf,MAAM,EAAEI,IAAI,EAAEC,GAAG,EAAEC,KAAK,EAAEC,MAAM,EAAE,GAAGN;IACrC,OAAOC,KAAKE,QAAQF,KAAKE,OAAOE,SAASH,KAAKE,OAAOF,KAAKE,MAAME;AAClE;AAEA,MAAMC,6BAA6B,CACjCC,cACAC,SACAC;IAEA,MAAMC,WAAWrB,oBAAoBkB,cAAcC,QAAQ,MAAM;IACjE,MAAMG,WAAWd,eAAeU,cAAcC,QAAQ,IAAI;IAC1D,MAAMI,OACJF,YAAaD,CAAAA,oBAAoB,2BAA2B,EAAC,KAAME;IACrE,OAAO;QACLC;QACA,MAAMJ,QAAQ,IAAI;QAClB,QAAQA,QAAQ,MAAM;QACtB,gBAAgBE;QAChB,gBAAgBC;IAClB;AACF;AAEA,SAASE,mBACPC,IAAsB,EACtBC,KAAa;IAEbC,OACEF,QACEG,OAAO,QAAQ,CAACH,KAAK,KAAK,KAC1BG,OAAO,QAAQ,CAACH,KAAK,MAAM,KAC3BA,KAAK,KAAK,GAAG,KACbA,KAAK,MAAM,GAAG,GAChB,GAAGC,MAAM,uCAAuC,CAAC;AAErD;AAEA,MAAMG,4BAA4B,CAChCC,QACAC,gBACAC;IAEA,MAAMC,kBAAkBD,IAAI,eAAe,IAAI;IAC/C,IAAIC,AAAoB,iBAApBA,iBACF,OAAOH;IAGTN,mBACEQ,IAAI,WAAW,EACf;IAEF,OAAO;QACJF,MAAM,CAAC,EAAE,GAAGC,eAAe,KAAK,GAAIC,IAAI,WAAW,CAAC,KAAK;QACzDF,MAAM,CAAC,EAAE,GAAGC,eAAe,MAAM,GAAIC,IAAI,WAAW,CAAC,MAAM;KAC7D;AACH;AAEA,MAAME,sCAAsC,OAC1CJ,QACAE;IAMA,MAAMG,oBAAoBH,KAAK,mBAC3B,MAAMI,+BAA+BJ,IAAI,gBAAgB,EAAEA,OAC3DK;IACJ,MAAMC,eAAeH,oBACjBN,0BAA0BC,QAAQK,kBAAkB,QAAQ,EAAEH,OAAO,CAAC,KACtEF;IACJ,OAAO;QACLK;QACA,WAAWA,oBAAoB;YAAE,WAAWA;QAAkB,IAAIE;QAClEC;IACF;AACF;AAEO,eAAeC,cACpBC,OAAwE,EACxEC,MAAc,EACdC,SAAmC,EACnCxB,YAA8B,EAC9BE,kBACkC;IAElC,OAAOuB,gBAAgBH,SAASC,QAAQvB,cAAc;QACpD,yBAAyBE,oBAAoB;QAC7C,YAAYA,oBAAoB;QAChC,WAAWsB,WAAW;IACxB;AACF;AAEA,SAASE,qBACPC,eAA4C,EAC5CC,KAAqC;IAErC,IAAI,CAACD,iBACH,OAAOC,MAAM,YAAY;IAE3B,MAAMC,eAAeF,gBAAgBC;IACrC,IAAI,AAAwB,aAAxB,OAAOC,cACT,OAAO;QACL,GAAGD,MAAM,YAAY;QACrB,MAAMC;IACR;IAEF,OAAOA;AACT;AAEA,SAASC,aAAaC,KAAc;IAClC,OAAOA,iBAAiBC,QAAQD,MAAM,OAAO,GAAGE,OAAOF;AACzD;AAEO,eAAeG,uBACpBZ,OAAgC,EAChCV,MAAwB,EACxBE,GAAmC;IAEnC,MAAM,EAAEqB,eAAe,IAAI,EAAEC,aAAa,CAAC,EAAE,GAAGtB,OAAO,CAAC;IACxD,MAAM,EAAEG,iBAAiB,EAAEO,SAAS,EAAEJ,YAAY,EAAE,GAClD,MAAMJ,oCAAoCJ,QAAQE;IAEpD,IAAIuB,UAAU;IACd,IAAIC,aAAa;IACjB,IAAIC,eAAe;IACnB,MAAMC,wBAAwB1B,KAAK,iBAAiBK;IACpD,MAAMsB,sBAAsB3B,KAAK,eAAeK;IAChD,IAAIuB,eAAe5B,KAAK,gBAAgB;IACxC,IAAI6B,aAAa7B,KAAK,cAAc;IACpC,IAAI8B;IACJ,IAAIC;IACJ,IAAIC;IAEJ,MAAO,CAACT,WAAWC,aAAaF,WAAY;QAC1C,IAAIE,cAAc,KAAKE,uBACrBE,eAAe;QAEjB,IAAIJ,cAAc,KAAKG,qBACrBE,aAAa;QAEf,MAAMI,uBAAuBzB,QAAQ,oBAAoB;QACzD,MAAM0B,qBAAqB1B,QAAQ,kBAAkB;QACrD,MAAM2B,eAAwC;YAC5C,GAAG3B,OAAO;YACVyB;YACAC;QACF;QACA,MAAME,cAAsCjC,oBACxC;YACEyB;YACA,SAASzB;QACX,IACA;YACEyB;QACF;QACJ,IAAIS;QAGJ,IAAI;YACFA,OAAO,MAAMF,aAAa,OAAO,CAAC,QAAQ,CACxC7B,cACA6B,aAAa,oBAAoB,EACjCC;QAEJ,EAAE,OAAOnB,OAAO;YACd,OAAO;gBACL,QAAQQ;gBACRI;gBACAD;gBACAE;gBACA,SAAS;gBACT,OAAOd,aAAaC;gBACpB,cAAc;YAChB;QACF;QACA,IAAI,CAACoB,KAAK,WAAW,EACnB,OAAO;YACL,QAAQZ;YACRI;YACAD;YACAE;YACA,SAAS;YACT,OAAO,CAAC,+BAA+B,EAAExB,aAAa,CAAC,CAAC;YACxD,cAAc;QAChB;QAEFmB,eAAeY,KAAK,WAAW;QAC/B,IAAI,CAAChB,cAAc;YACjBrB,KAAK,aAAa;gBAChB,QAAQyB;gBACRG;gBACAC;YACF;YACAN,UAAU;YACV;QACF;QAEA,IAAI;YACF,MAAMe,wBAAwB,MAAM/B,cAClC4B,cACAV,cACAf,WACAJ,cACA;gBACE,GAAGN,GAAG;gBACN6B;YACF;YAEFC,eAAelB,qBAAqBZ,KAAK,iBAAiB;gBACxD,QAAQyB;gBACR,cAAcnB;gBACdkB;gBACA,cAAcc;YAChB;YACAtC,KAAK,aAAa;gBAChB,QAAQyB;gBACRG;gBACAC;gBACAC;YACF;YACA,IAAIA,aAAa,IAAI,EAAE;gBACrBP,UAAU;gBACV;YACF;YACAQ,YAAY1B;YACZ2B,eAAe;QACjB,EAAE,OAAOf,OAAO;YACdc,YAAYf,aAAaC;YACzBe,eAAe;YACfhC,KAAK,aAAa;gBAChB,QAAQyB;gBACRG;gBACAC;YACF;QACF;QACAL;IACF;IAEA,OAAO;QACL,QAAQC;QACRI;QACAD;QACAE;QACAP;QACA,OACEA,WAAW,CAACF,eACRhB,SACA0B,aAAa;QACnB,cAAcR,UAAUlB,SAAY2B;IACtC;AACF;AAEO,eAAeO,gCACpB/B,OAAgC,EAChCgC,WAAmB,EACnB1C,MAAwB,EACxBE,GAA4C;IAE5CL,OAAO6C,aAAa,QAAQ;IAC5B,MAAM,EAAE9B,SAAS,EAAEJ,YAAY,EAAE,GAAG,MAAMJ,oCACxCJ,QACAE;IAGF,OAAOO,cAAcC,SAASgC,aAAa9B,WAAWJ,cAAcN;AACtE;AAEA,eAAeW,gBACbH,OAAwE,EACxEgC,WAAmB,EACnB1C,MAAwB,EACxBE,GAEC;IAEDL,OAAO6C,aAAa,QAAQ;IAC5B,MAAM,EAAErC,iBAAiB,EAAEG,YAAY,EAAE,GACvC,MAAMJ,oCAAoCJ,QAAQE;IACpD,MAAMyC,UAAUzC,KAAK,aAAaG;IAClC,MAAMuC,cAAcC,yBAAyBH,aAAa;QACxD,WAAWxC,KAAK;QAChB,YAAYA,KAAK;QACjB,OAAOA,KAAK;IACd;IACAL,OAAO+C,aAAa;IAEpB,MAAME,eAAe,MAAMpC,QAAQ,OAAO,CAAC,MAAM,CAC/CkC,aACAD,UAAU;QAAEA;IAAQ,IAAI,CAAC,GACzBjC,QAAQ,kBAAkB,EAC1BR,KAAK;IAEPL,OAAOiD,aAAa,OAAO,EAAE,CAAC,mBAAmB,EAAEJ,aAAa;IAChE,MAAMV,eAAe7C,2BACnBqB,cACAsC,aAAa,OAAO,EACpB5C;IAEF,OAAO8B;AACT"}
|
|
1
|
+
{"version":3,"file":"element-describer.mjs","sources":["../../src/element-describer.ts"],"sourcesContent":["import type { ModelRuntime } from '@/ai-model/models';\nimport { assert } from '@midscene/shared/utils';\nimport { createScreenshotBoundUIContext } from './agent/utils';\nimport type Service from './service';\nimport type {\n AgentDescribeElementAtPointResult,\n LocateOption,\n LocateResultElement,\n LocateValidatorResult,\n LocatorValidatorOption,\n Rect,\n Size,\n UIContext,\n} from './types';\nimport { buildDetailedLocateParam } from './yaml/utils';\n\nexport type DescribeElementCoordinateSpace = 'screenshot' | 'logical';\n\nexport type LocatorVerifyFn = (input: {\n prompt: string;\n expectCenter: [number, number];\n retryCount: number;\n verifyResult: LocateValidatorResult;\n}) => LocateValidatorResult | boolean;\n\nexport type DescribeElementAtPointOptions = {\n verifyPrompt?: boolean;\n retryLimit?: number;\n deepDescribe?: boolean;\n deepLocate?: boolean;\n locatorVerifyFn?: LocatorVerifyFn;\n screenshotBase64?: string;\n screenshotSize?: Size;\n coordinateSpace?: DescribeElementCoordinateSpace;\n logicalSize?: Size;\n onProgress?: (progress: {\n prompt?: string;\n deepDescribe?: boolean;\n deepLocate?: boolean;\n verifyResult?: LocateValidatorResult;\n }) => void;\n} & LocatorValidatorOption;\n\ntype ScreenshotBoundContextOptions = {\n screenshotBase64?: string;\n screenshotSize?: Size;\n coordinateSpace?: DescribeElementCoordinateSpace;\n logicalSize?: Size;\n};\n\nexport type VerifyElementDescriptionAtPointOptions =\n ScreenshotBoundContextOptions & LocatorValidatorOption;\n\ntype LocateAndVerifyOptions = VerifyElementDescriptionAtPointOptions &\n Pick<LocateOption, 'cacheable' | 'deepLocate' | 'xpath'> & {\n abortSignal?: AbortSignal;\n };\n\nexport type ElementDescriberRuntime = {\n service: Pick<Service, 'describe' | 'locate'>;\n describeModelRuntime: ModelRuntime;\n locateModelRuntime: ModelRuntime;\n};\n\ntype ServiceDescribeOptions = NonNullable<\n Parameters<ElementDescriberRuntime['service']['describe']>[2]\n>;\n\nconst distanceOfTwoPoints = (p1: [number, number], p2: [number, number]) => {\n const [x1, y1] = p1;\n const [x2, y2] = p2;\n return Math.round(Math.sqrt((x1 - x2) ** 2 + (y1 - y2) ** 2));\n};\n\nconst includedInRect = (point: [number, number], rect: Rect) => {\n const [x, y] = point;\n const { left, top, width, height } = rect;\n return x >= left && x <= left + width && y >= top && y <= top + height;\n};\n\nconst buildLocateValidatorResult = (\n expectCenter: [number, number],\n located: Pick<LocateResultElement, 'center' | 'rect'>,\n verifyLocateOption?: LocatorValidatorOption,\n): LocateValidatorResult => {\n const distance = distanceOfTwoPoints(expectCenter, located.center);\n const included = located.rect\n ? includedInRect(expectCenter, located.rect)\n : undefined;\n const pass =\n distance <= (verifyLocateOption?.centerDistanceThreshold || 20) ||\n included === true;\n return {\n pass,\n rect: located.rect,\n center: located.center,\n centerDistance: distance,\n includedInRect: included,\n };\n};\n\nfunction assertPositiveSize(\n size: Size | undefined,\n label: string,\n): asserts size is Size {\n assert(\n size &&\n Number.isFinite(size.width) &&\n Number.isFinite(size.height) &&\n size.width > 0 &&\n size.height > 0,\n `${label} must include positive width and height`,\n );\n}\n\nconst mapPointToScreenshotSpace = (\n center: [number, number],\n screenshotSize: Size,\n opt: ScreenshotBoundContextOptions,\n): [number, number] => {\n const coordinateSpace = opt.coordinateSpace || 'screenshot';\n if (coordinateSpace === 'screenshot') {\n return center;\n }\n\n assertPositiveSize(\n opt.logicalSize,\n 'logicalSize is required when coordinateSpace is logical',\n );\n return [\n (center[0] * screenshotSize.width) / opt.logicalSize.width,\n (center[1] * screenshotSize.height) / opt.logicalSize.height,\n ];\n};\n\nconst createScreenshotBoundLocatorContext = async (\n center: [number, number],\n opt?: ScreenshotBoundContextOptions,\n): Promise<{\n screenshotContext?: UIContext;\n locateOpt?: LocateOption;\n targetCenter: [number, number];\n}> => {\n const screenshotContext = opt?.screenshotBase64\n ? await createScreenshotBoundUIContext(opt.screenshotBase64, opt)\n : undefined;\n const targetCenter = screenshotContext\n ? mapPointToScreenshotSpace(center, screenshotContext.shotSize, opt || {})\n : center;\n return {\n screenshotContext,\n locateOpt: screenshotContext ? { uiContext: screenshotContext } : undefined,\n targetCenter,\n };\n};\n\nexport async function verifyLocator(\n runtime: Pick<ElementDescriberRuntime, 'service' | 'locateModelRuntime'>,\n prompt: string,\n locateOpt: LocateOption | undefined,\n expectCenter: [number, number],\n verifyLocateOption?: LocatorValidatorOption &\n Pick<LocateOption, 'deepLocate'>,\n): Promise<LocateValidatorResult> {\n return locateAndVerify(runtime, prompt, expectCenter, {\n centerDistanceThreshold: verifyLocateOption?.centerDistanceThreshold,\n deepLocate: verifyLocateOption?.deepLocate,\n uiContext: locateOpt?.uiContext,\n });\n}\n\nfunction applyLocatorVerifyFn(\n locatorVerifyFn: LocatorVerifyFn | undefined,\n input: Parameters<LocatorVerifyFn>[0],\n): LocateValidatorResult {\n if (!locatorVerifyFn) {\n return input.verifyResult;\n }\n const customResult = locatorVerifyFn(input);\n if (typeof customResult === 'boolean') {\n return {\n ...input.verifyResult,\n pass: customResult,\n };\n }\n return customResult;\n}\n\nfunction errorMessage(error: unknown): string {\n return error instanceof Error ? error.message : String(error);\n}\n\nexport async function describeElementAtPoint(\n runtime: ElementDescriberRuntime,\n center: [number, number],\n opt?: DescribeElementAtPointOptions,\n): Promise<AgentDescribeElementAtPointResult> {\n const { verifyPrompt = true, retryLimit = 3 } = opt || {};\n const { screenshotContext, locateOpt, targetCenter } =\n await createScreenshotBoundLocatorContext(center, opt);\n\n let success = false;\n let retryCount = 0;\n let resultPrompt = '';\n const autoRetryDeepDescribe = opt?.deepDescribe === undefined;\n const autoRetryDeepLocate = opt?.deepLocate === undefined;\n let deepDescribe = opt?.deepDescribe || false;\n let deepLocate = opt?.deepLocate || false;\n let verifyResult: LocateValidatorResult | undefined;\n let lastError: string | undefined;\n let failureStage: AgentDescribeElementAtPointResult['failureStage'];\n\n while (!success && retryCount < retryLimit) {\n if (retryCount >= 1 && autoRetryDeepDescribe) {\n deepDescribe = true;\n }\n if (retryCount >= 1 && autoRetryDeepLocate) {\n deepLocate = true;\n }\n const describeModelRuntime = runtime.describeModelRuntime;\n const locateModelRuntime = runtime.locateModelRuntime;\n const retryRuntime: ElementDescriberRuntime = {\n ...runtime,\n describeModelRuntime,\n locateModelRuntime,\n };\n const describeOpt: ServiceDescribeOptions = screenshotContext\n ? {\n deepDescribe,\n context: screenshotContext,\n }\n : {\n deepDescribe,\n };\n let text: Awaited<\n ReturnType<ElementDescriberRuntime['service']['describe']>\n >;\n try {\n text = await retryRuntime.service.describe(\n targetCenter,\n retryRuntime.describeModelRuntime,\n describeOpt,\n );\n } catch (error) {\n return {\n prompt: resultPrompt,\n deepLocate,\n deepDescribe,\n verifyResult,\n success: false,\n error: errorMessage(error),\n failureStage: 'describe',\n };\n }\n if (!text.description) {\n return {\n prompt: resultPrompt,\n deepLocate,\n deepDescribe,\n verifyResult,\n success: false,\n error: `failed to describe element at [${targetCenter}]`,\n failureStage: 'describe',\n };\n }\n resultPrompt = text.description;\n if (!verifyPrompt) {\n opt?.onProgress?.({\n prompt: resultPrompt,\n deepDescribe,\n deepLocate,\n });\n success = true;\n break;\n }\n\n try {\n const candidateVerifyResult = await verifyLocator(\n retryRuntime,\n resultPrompt,\n locateOpt,\n targetCenter,\n {\n ...opt,\n deepLocate,\n },\n );\n verifyResult = applyLocatorVerifyFn(opt?.locatorVerifyFn, {\n prompt: resultPrompt,\n expectCenter: targetCenter,\n retryCount,\n verifyResult: candidateVerifyResult,\n });\n opt?.onProgress?.({\n prompt: resultPrompt,\n deepDescribe,\n deepLocate,\n verifyResult,\n });\n if (verifyResult.pass) {\n success = true;\n break;\n }\n lastError = undefined;\n failureStage = 'verify';\n } catch (error) {\n lastError = errorMessage(error);\n failureStage = 'verify';\n opt?.onProgress?.({\n prompt: resultPrompt,\n deepDescribe,\n deepLocate,\n });\n }\n retryCount++;\n }\n\n return {\n prompt: resultPrompt,\n deepLocate,\n deepDescribe,\n verifyResult,\n success,\n error:\n success || !verifyPrompt\n ? undefined\n : lastError || 'describeElementAtPoint verify failed',\n failureStage: success ? undefined : failureStage,\n };\n}\n\nexport async function verifyElementDescriptionAtPoint(\n runtime: ElementDescriberRuntime,\n description: string,\n center: [number, number],\n opt?: VerifyElementDescriptionAtPointOptions,\n): Promise<LocateValidatorResult> {\n assert(description?.trim(), 'description must not be empty');\n const { locateOpt, targetCenter } = await createScreenshotBoundLocatorContext(\n center,\n opt,\n );\n\n return verifyLocator(runtime, description, locateOpt, targetCenter, opt);\n}\n\nasync function locateAndVerify(\n runtime: Pick<ElementDescriberRuntime, 'service' | 'locateModelRuntime'>,\n description: string,\n center: [number, number],\n opt?: LocateAndVerifyOptions & {\n uiContext?: UIContext;\n },\n): Promise<LocateValidatorResult> {\n assert(description?.trim(), 'description must not be empty');\n const { screenshotContext, targetCenter } =\n await createScreenshotBoundLocatorContext(center, opt);\n const context = opt?.uiContext || screenshotContext;\n const locateParam = buildDetailedLocateParam(description, {\n cacheable: opt?.cacheable,\n deepLocate: opt?.deepLocate,\n xpath: opt?.xpath,\n });\n assert(locateParam, 'cannot get locate param for service locate');\n\n const locateResult = await runtime.service.locate(\n locateParam,\n context ? { context } : {},\n runtime.locateModelRuntime,\n opt?.abortSignal,\n );\n assert(locateResult.element, `Element not found: ${description}`);\n const verifyResult = buildLocateValidatorResult(\n targetCenter,\n locateResult.element,\n opt,\n );\n return verifyResult;\n}\n"],"names":["distanceOfTwoPoints","p1","p2","x1","y1","x2","y2","Math","includedInRect","point","rect","x","y","left","top","width","height","buildLocateValidatorResult","expectCenter","located","verifyLocateOption","distance","included","undefined","pass","assertPositiveSize","size","label","assert","Number","mapPointToScreenshotSpace","center","screenshotSize","opt","coordinateSpace","createScreenshotBoundLocatorContext","screenshotContext","createScreenshotBoundUIContext","targetCenter","verifyLocator","runtime","prompt","locateOpt","locateAndVerify","applyLocatorVerifyFn","locatorVerifyFn","input","customResult","errorMessage","error","Error","String","describeElementAtPoint","verifyPrompt","retryLimit","success","retryCount","resultPrompt","autoRetryDeepDescribe","autoRetryDeepLocate","deepDescribe","deepLocate","verifyResult","lastError","failureStage","describeModelRuntime","locateModelRuntime","retryRuntime","describeOpt","text","candidateVerifyResult","verifyElementDescriptionAtPoint","description","context","locateParam","buildDetailedLocateParam","locateResult"],"mappings":";;;AAoEA,MAAMA,sBAAsB,CAACC,IAAsBC;IACjD,MAAM,CAACC,IAAIC,GAAG,GAAGH;IACjB,MAAM,CAACI,IAAIC,GAAG,GAAGJ;IACjB,OAAOK,KAAK,KAAK,CAACA,KAAK,IAAI,CAAEJ,AAAAA,CAAAA,KAAKE,EAAC,KAAM,IAAKD,AAAAA,CAAAA,KAAKE,EAAC,KAAM;AAC5D;AAEA,MAAME,iBAAiB,CAACC,OAAyBC;IAC/C,MAAM,CAACC,GAAGC,EAAE,GAAGH;IACf,MAAM,EAAEI,IAAI,EAAEC,GAAG,EAAEC,KAAK,EAAEC,MAAM,EAAE,GAAGN;IACrC,OAAOC,KAAKE,QAAQF,KAAKE,OAAOE,SAASH,KAAKE,OAAOF,KAAKE,MAAME;AAClE;AAEA,MAAMC,6BAA6B,CACjCC,cACAC,SACAC;IAEA,MAAMC,WAAWrB,oBAAoBkB,cAAcC,QAAQ,MAAM;IACjE,MAAMG,WAAWH,QAAQ,IAAI,GACzBX,eAAeU,cAAcC,QAAQ,IAAI,IACzCI;IACJ,MAAMC,OACJH,YAAaD,CAAAA,oBAAoB,2BAA2B,EAAC,KAC7DE,AAAa,SAAbA;IACF,OAAO;QACLE;QACA,MAAML,QAAQ,IAAI;QAClB,QAAQA,QAAQ,MAAM;QACtB,gBAAgBE;QAChB,gBAAgBC;IAClB;AACF;AAEA,SAASG,mBACPC,IAAsB,EACtBC,KAAa;IAEbC,OACEF,QACEG,OAAO,QAAQ,CAACH,KAAK,KAAK,KAC1BG,OAAO,QAAQ,CAACH,KAAK,MAAM,KAC3BA,KAAK,KAAK,GAAG,KACbA,KAAK,MAAM,GAAG,GAChB,GAAGC,MAAM,uCAAuC,CAAC;AAErD;AAEA,MAAMG,4BAA4B,CAChCC,QACAC,gBACAC;IAEA,MAAMC,kBAAkBD,IAAI,eAAe,IAAI;IAC/C,IAAIC,AAAoB,iBAApBA,iBACF,OAAOH;IAGTN,mBACEQ,IAAI,WAAW,EACf;IAEF,OAAO;QACJF,MAAM,CAAC,EAAE,GAAGC,eAAe,KAAK,GAAIC,IAAI,WAAW,CAAC,KAAK;QACzDF,MAAM,CAAC,EAAE,GAAGC,eAAe,MAAM,GAAIC,IAAI,WAAW,CAAC,MAAM;KAC7D;AACH;AAEA,MAAME,sCAAsC,OAC1CJ,QACAE;IAMA,MAAMG,oBAAoBH,KAAK,mBAC3B,MAAMI,+BAA+BJ,IAAI,gBAAgB,EAAEA,OAC3DV;IACJ,MAAMe,eAAeF,oBACjBN,0BAA0BC,QAAQK,kBAAkB,QAAQ,EAAEH,OAAO,CAAC,KACtEF;IACJ,OAAO;QACLK;QACA,WAAWA,oBAAoB;YAAE,WAAWA;QAAkB,IAAIb;QAClEe;IACF;AACF;AAEO,eAAeC,cACpBC,OAAwE,EACxEC,MAAc,EACdC,SAAmC,EACnCxB,YAA8B,EAC9BE,kBACkC;IAElC,OAAOuB,gBAAgBH,SAASC,QAAQvB,cAAc;QACpD,yBAAyBE,oBAAoB;QAC7C,YAAYA,oBAAoB;QAChC,WAAWsB,WAAW;IACxB;AACF;AAEA,SAASE,qBACPC,eAA4C,EAC5CC,KAAqC;IAErC,IAAI,CAACD,iBACH,OAAOC,MAAM,YAAY;IAE3B,MAAMC,eAAeF,gBAAgBC;IACrC,IAAI,AAAwB,aAAxB,OAAOC,cACT,OAAO;QACL,GAAGD,MAAM,YAAY;QACrB,MAAMC;IACR;IAEF,OAAOA;AACT;AAEA,SAASC,aAAaC,KAAc;IAClC,OAAOA,iBAAiBC,QAAQD,MAAM,OAAO,GAAGE,OAAOF;AACzD;AAEO,eAAeG,uBACpBZ,OAAgC,EAChCT,MAAwB,EACxBE,GAAmC;IAEnC,MAAM,EAAEoB,eAAe,IAAI,EAAEC,aAAa,CAAC,EAAE,GAAGrB,OAAO,CAAC;IACxD,MAAM,EAAEG,iBAAiB,EAAEM,SAAS,EAAEJ,YAAY,EAAE,GAClD,MAAMH,oCAAoCJ,QAAQE;IAEpD,IAAIsB,UAAU;IACd,IAAIC,aAAa;IACjB,IAAIC,eAAe;IACnB,MAAMC,wBAAwBzB,KAAK,iBAAiBV;IACpD,MAAMoC,sBAAsB1B,KAAK,eAAeV;IAChD,IAAIqC,eAAe3B,KAAK,gBAAgB;IACxC,IAAI4B,aAAa5B,KAAK,cAAc;IACpC,IAAI6B;IACJ,IAAIC;IACJ,IAAIC;IAEJ,MAAO,CAACT,WAAWC,aAAaF,WAAY;QAC1C,IAAIE,cAAc,KAAKE,uBACrBE,eAAe;QAEjB,IAAIJ,cAAc,KAAKG,qBACrBE,aAAa;QAEf,MAAMI,uBAAuBzB,QAAQ,oBAAoB;QACzD,MAAM0B,qBAAqB1B,QAAQ,kBAAkB;QACrD,MAAM2B,eAAwC;YAC5C,GAAG3B,OAAO;YACVyB;YACAC;QACF;QACA,MAAME,cAAsChC,oBACxC;YACEwB;YACA,SAASxB;QACX,IACA;YACEwB;QACF;QACJ,IAAIS;QAGJ,IAAI;YACFA,OAAO,MAAMF,aAAa,OAAO,CAAC,QAAQ,CACxC7B,cACA6B,aAAa,oBAAoB,EACjCC;QAEJ,EAAE,OAAOnB,OAAO;YACd,OAAO;gBACL,QAAQQ;gBACRI;gBACAD;gBACAE;gBACA,SAAS;gBACT,OAAOd,aAAaC;gBACpB,cAAc;YAChB;QACF;QACA,IAAI,CAACoB,KAAK,WAAW,EACnB,OAAO;YACL,QAAQZ;YACRI;YACAD;YACAE;YACA,SAAS;YACT,OAAO,CAAC,+BAA+B,EAAExB,aAAa,CAAC,CAAC;YACxD,cAAc;QAChB;QAEFmB,eAAeY,KAAK,WAAW;QAC/B,IAAI,CAAChB,cAAc;YACjBpB,KAAK,aAAa;gBAChB,QAAQwB;gBACRG;gBACAC;YACF;YACAN,UAAU;YACV;QACF;QAEA,IAAI;YACF,MAAMe,wBAAwB,MAAM/B,cAClC4B,cACAV,cACAf,WACAJ,cACA;gBACE,GAAGL,GAAG;gBACN4B;YACF;YAEFC,eAAelB,qBAAqBX,KAAK,iBAAiB;gBACxD,QAAQwB;gBACR,cAAcnB;gBACdkB;gBACA,cAAcc;YAChB;YACArC,KAAK,aAAa;gBAChB,QAAQwB;gBACRG;gBACAC;gBACAC;YACF;YACA,IAAIA,aAAa,IAAI,EAAE;gBACrBP,UAAU;gBACV;YACF;YACAQ,YAAYxC;YACZyC,eAAe;QACjB,EAAE,OAAOf,OAAO;YACdc,YAAYf,aAAaC;YACzBe,eAAe;YACf/B,KAAK,aAAa;gBAChB,QAAQwB;gBACRG;gBACAC;YACF;QACF;QACAL;IACF;IAEA,OAAO;QACL,QAAQC;QACRI;QACAD;QACAE;QACAP;QACA,OACEA,WAAW,CAACF,eACR9B,SACAwC,aAAa;QACnB,cAAcR,UAAUhC,SAAYyC;IACtC;AACF;AAEO,eAAeO,gCACpB/B,OAAgC,EAChCgC,WAAmB,EACnBzC,MAAwB,EACxBE,GAA4C;IAE5CL,OAAO4C,aAAa,QAAQ;IAC5B,MAAM,EAAE9B,SAAS,EAAEJ,YAAY,EAAE,GAAG,MAAMH,oCACxCJ,QACAE;IAGF,OAAOM,cAAcC,SAASgC,aAAa9B,WAAWJ,cAAcL;AACtE;AAEA,eAAeU,gBACbH,OAAwE,EACxEgC,WAAmB,EACnBzC,MAAwB,EACxBE,GAEC;IAEDL,OAAO4C,aAAa,QAAQ;IAC5B,MAAM,EAAEpC,iBAAiB,EAAEE,YAAY,EAAE,GACvC,MAAMH,oCAAoCJ,QAAQE;IACpD,MAAMwC,UAAUxC,KAAK,aAAaG;IAClC,MAAMsC,cAAcC,yBAAyBH,aAAa;QACxD,WAAWvC,KAAK;QAChB,YAAYA,KAAK;QACjB,OAAOA,KAAK;IACd;IACAL,OAAO8C,aAAa;IAEpB,MAAME,eAAe,MAAMpC,QAAQ,OAAO,CAAC,MAAM,CAC/CkC,aACAD,UAAU;QAAEA;IAAQ,IAAI,CAAC,GACzBjC,QAAQ,kBAAkB,EAC1BP,KAAK;IAEPL,OAAOgD,aAAa,OAAO,EAAE,CAAC,mBAAmB,EAAEJ,aAAa;IAChE,MAAMV,eAAe7C,2BACnBqB,cACAsC,aAAa,OAAO,EACpB3C;IAEF,OAAO6B;AACT"}
|