@midscene/core 0.24.2-beta-20250801111909.0 → 0.24.2-beta-20250805024613.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/es/ai-model.d.ts +8 -4
- package/dist/es/ai-model.js +3 -1
- package/dist/es/{chunk-KFA65L55.js → chunk-JS4CT3XV.js} +22 -17
- package/dist/es/chunk-JS4CT3XV.js.map +1 -0
- package/dist/es/{chunk-2RCMQS5O.js → chunk-NZFWY3M5.js} +3 -3
- package/dist/es/index.d.ts +4 -4
- package/dist/es/index.js +36 -12
- package/dist/es/index.js.map +1 -1
- package/dist/es/{llm-planning-4c782a8d.d.ts → llm-planning-877248da.d.ts} +2 -2
- package/dist/es/{types-7b64b80b.d.ts → types-512d3687.d.ts} +2 -3
- package/dist/es/utils.d.ts +1 -1
- package/dist/es/utils.js +1 -1
- package/dist/lib/ai-model.d.ts +8 -4
- package/dist/lib/ai-model.js +4 -2
- package/dist/lib/{chunk-KFA65L55.js → chunk-JS4CT3XV.js} +22 -17
- package/dist/lib/chunk-JS4CT3XV.js.map +1 -0
- package/dist/lib/{chunk-2RCMQS5O.js → chunk-NZFWY3M5.js} +3 -3
- package/dist/lib/index.d.ts +4 -4
- package/dist/lib/index.js +47 -23
- package/dist/lib/index.js.map +1 -1
- package/dist/lib/{llm-planning-4c782a8d.d.ts → llm-planning-877248da.d.ts} +2 -2
- package/dist/{types/types-7b64b80b.d.ts → lib/types-512d3687.d.ts} +2 -3
- package/dist/lib/utils.d.ts +1 -1
- package/dist/lib/utils.js +2 -2
- package/dist/types/ai-model.d.ts +8 -4
- package/dist/types/index.d.ts +4 -4
- package/dist/types/{llm-planning-4c782a8d.d.ts → llm-planning-877248da.d.ts} +2 -2
- package/dist/{lib/types-7b64b80b.d.ts → types/types-512d3687.d.ts} +2 -3
- package/dist/types/utils.d.ts +1 -1
- package/package.json +3 -3
- package/dist/es/chunk-KFA65L55.js.map +0 -1
- package/dist/lib/chunk-KFA65L55.js.map +0 -1
- /package/dist/es/{chunk-2RCMQS5O.js.map → chunk-NZFWY3M5.js.map} +0 -0
- /package/dist/lib/{chunk-2RCMQS5O.js.map → chunk-NZFWY3M5.js.map} +0 -0
package/dist/es/index.d.ts
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
|
-
import { E as ExecutionTask, a as ExecutionTaskProgressOptions, b as ExecutionTaskApply, c as ExecutionDump, U as UIContext, I as InsightAction, D as DumpSubscriber, d as InsightTaskInfo, e as InsightOptions, f as DetailedLocateParam, L as LocateResult, g as InsightExtractParam, h as InsightExtractOption, T as TMultimodalPrompt, i as TUserPrompt, j as InsightAssertionResponse, A as AIDescribeElementResponse } from './types-
|
|
2
|
-
export {
|
|
3
|
-
import { c as callAiFn } from './llm-planning-
|
|
4
|
-
export { a as AiAssert, A as AiLocateElement, d as describeUserPage, p as plan } from './llm-planning-
|
|
1
|
+
import { E as ExecutionTask, a as ExecutionTaskProgressOptions, b as ExecutionTaskApply, c as ExecutionDump, U as UIContext, I as InsightAction, D as DumpSubscriber, d as InsightTaskInfo, e as InsightOptions, f as DetailedLocateParam, L as LocateResult, g as InsightExtractParam, h as InsightExtractOption, T as TMultimodalPrompt, i as TUserPrompt, j as InsightAssertionResponse, A as AIDescribeElementResponse } from './types-512d3687.js';
|
|
2
|
+
export { y as AIAssertionResponse, w as AIDataExtractionResponse, u as AIElementCoordinatesResponse, t as AIElementLocatorResponse, v as AIElementResponse, p as AIResponseFormat, x as AISectionLocatorResponse, s as AISingleElementResponse, q as AISingleElementResponseById, r as AISingleElementResponseByPosition, o as AIUsageInfo, V as AgentAssertOpt, C as AgentDescribeElementAtPointResult, S as AgentWaitForOpt, a9 as BaseAgentParserOpt, F as CallAIFn, aD as CodeGenerationChunk, a8 as Color, J as DumpMeta, O as ElementById, G as EnsureObject, ac as ExecutionRecorderItem, au as ExecutionTaskAction, at as ExecutionTaskActionApply, af as ExecutionTaskHitBy, as as ExecutionTaskInsightAssertion, ar as ExecutionTaskInsightAssertionApply, aq as ExecutionTaskInsightAssertionParam, aj as ExecutionTaskInsightDumpLog, al as ExecutionTaskInsightLocate, ak as ExecutionTaskInsightLocateApply, ai as ExecutionTaskInsightLocateOutput, ah as ExecutionTaskInsightLocateParam, ap as ExecutionTaskInsightQuery, ao as ExecutionTaskInsightQueryApply, an as ExecutionTaskInsightQueryOutput, am as ExecutionTaskInsightQueryParam, aw as ExecutionTaskLog, av as ExecutionTaskLogApply, ay as ExecutionTaskPlanning, ax as ExecutionTaskPlanningApply, ag as ExecutionTaskReturn, ad as ExecutionTaskType, ae as ExecutorContext, b1 as FreeFn, az as GroupedActionDump, K as InsightDump, N as LiteUISection, aF as LocateOption, H as LocateResultElement, B as LocateValidatorResult, z as LocatorValidatorOption, b4 as MidsceneYamlConfig, b5 as MidsceneYamlConfigOutput, n as MidsceneYamlConfigResult, l as MidsceneYamlFlowItem, aM as MidsceneYamlFlowItemAIAction, aR as MidsceneYamlFlowItemAIAsk, aN as MidsceneYamlFlowItemAIAssert, aS as MidsceneYamlFlowItemAIBoolean, aW as MidsceneYamlFlowItemAIHover, aX as MidsceneYamlFlowItemAIInput, aY as MidsceneYamlFlowItemAIKeyboardPress, aT as MidsceneYamlFlowItemAILocate, aP as MidsceneYamlFlowItemAINumber, aO as MidsceneYamlFlowItemAIQuery, m as MidsceneYamlFlowItemAIRightClick, aZ as MidsceneYamlFlowItemAIScroll, aQ as MidsceneYamlFlowItemAIString, aV as MidsceneYamlFlowItemAITap, aU as MidsceneYamlFlowItemAIWaitFor, a_ as MidsceneYamlFlowItemEvaluateJavaScript, b0 as MidsceneYamlFlowItemLogScreenshot, a$ as MidsceneYamlFlowItemSleep, M as MidsceneYamlScript, aK as MidsceneYamlScriptAndroidEnv, aL as MidsceneYamlScriptEnv, aI as MidsceneYamlScriptEnvBase, aJ as MidsceneYamlScriptWebEnv, k as MidsceneYamlTask, Q as OnTaskStartTip, aA as PageType, P as PartialInsightDumpFromSDK, Y as PlanningAIResponse, X as PlanningAction, a6 as PlanningActionParamAndroidLongPress, a7 as PlanningActionParamAndroidPull, a2 as PlanningActionParamAssert, a4 as PlanningActionParamError, _ as PlanningActionParamHover, a0 as PlanningActionParamInputOrKeyPress, $ as PlanningActionParamRightClick, a1 as PlanningActionParamScroll, a3 as PlanningActionParamSleep, Z as PlanningActionParamTap, a5 as PlanningActionParamWaitFor, W as PlanningLocateParam, ab as PlaywrightParserOpt, aa as PuppeteerParserOpt, aG as ReferenceImage, R as ReportDumpWithAttributes, b3 as ScriptPlayerStatusValue, b2 as ScriptPlayerTaskStatus, aE as StreamingAIResponse, aC as StreamingCallback, aB as StreamingCodeGenerationOptions, aH as scrollParam } from './types-512d3687.js';
|
|
3
|
+
import { c as callAiFn } from './llm-planning-877248da.js';
|
|
4
|
+
export { a as AiAssert, A as AiLocateElement, d as describeUserPage, p as plan } from './llm-planning-877248da.js';
|
|
5
5
|
import { BaseElement, Rect } from '@midscene/shared/types';
|
|
6
6
|
export { BaseElement, ElementTreeNode, Point, Rect, Size } from '@midscene/shared/types';
|
|
7
7
|
export { getVersion } from './utils.js';
|
package/dist/es/index.js
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
import {
|
|
2
2
|
getVersion
|
|
3
|
-
} from "./chunk-
|
|
3
|
+
} from "./chunk-NZFWY3M5.js";
|
|
4
4
|
import {
|
|
5
5
|
AiAssert,
|
|
6
6
|
AiExtractElementInfo,
|
|
@@ -11,7 +11,7 @@ import {
|
|
|
11
11
|
describeUserPage,
|
|
12
12
|
expandSearchArea,
|
|
13
13
|
plan
|
|
14
|
-
} from "./chunk-
|
|
14
|
+
} from "./chunk-JS4CT3XV.js";
|
|
15
15
|
|
|
16
16
|
// src/ai-model/action-executor.ts
|
|
17
17
|
import {
|
|
@@ -186,20 +186,44 @@ ${this.latestErrorTask()?.errorStack}`
|
|
|
186
186
|
// src/ai-model/prompt/describe.ts
|
|
187
187
|
import { getPreferredLanguage } from "@midscene/shared/env";
|
|
188
188
|
var elementDescriberInstruction = () => {
|
|
189
|
-
return `
|
|
189
|
+
return `
|
|
190
|
+
Describe the element in the red rectangle for precise identification. Use ${getPreferredLanguage()}.
|
|
190
191
|
|
|
191
|
-
|
|
192
|
-
1.
|
|
193
|
-
2.
|
|
194
|
-
|
|
195
|
-
|
|
196
|
-
|
|
197
|
-
|
|
198
|
-
|
|
192
|
+
CRITICAL REQUIREMENTS:
|
|
193
|
+
1. UNIQUENESS: The description must uniquely identify this element on the current page
|
|
194
|
+
2. UNIVERSALITY: Use generic, reusable selectors that work across different contexts
|
|
195
|
+
3. PRECISION: Be specific enough to distinguish from similar elements
|
|
196
|
+
|
|
197
|
+
DESCRIPTION STRUCTURE:
|
|
198
|
+
1. Element type (button, input, link, div, etc.)
|
|
199
|
+
2. Primary identifier (in order of preference):
|
|
200
|
+
- Unique text content: "with text 'Login'"
|
|
201
|
+
- Unique attribute: "with aria-label 'Search'"
|
|
202
|
+
- Unique class/ID: "with class 'primary-button'"
|
|
203
|
+
- Unique position: "in header navigation"
|
|
204
|
+
3. Secondary identifiers (if needed for uniqueness):
|
|
205
|
+
- Visual features: "blue background", "with icon"
|
|
206
|
+
- Relative position: "below search bar", "in sidebar"
|
|
207
|
+
- Parent context: "in login form", "in main menu"
|
|
208
|
+
|
|
209
|
+
GUIDELINES:
|
|
210
|
+
- Keep description under 25 words
|
|
211
|
+
- Prioritize semantic identifiers over visual ones
|
|
212
|
+
- Use consistent terminology across similar elements
|
|
213
|
+
- Avoid page-specific or temporary content
|
|
214
|
+
- Don't mention the red rectangle or selection box
|
|
215
|
+
- Focus on stable, reusable characteristics
|
|
216
|
+
|
|
217
|
+
EXAMPLES:
|
|
218
|
+
- "Login button with text 'Sign In'"
|
|
219
|
+
- "Search input with placeholder 'Enter keywords'"
|
|
220
|
+
- "Navigation link with text 'Home' in header"
|
|
221
|
+
- "Submit button in contact form"
|
|
222
|
+
- "Menu icon with aria-label 'Open menu'"
|
|
199
223
|
|
|
200
224
|
Return JSON:
|
|
201
225
|
{
|
|
202
|
-
"description": "
|
|
226
|
+
"description": "unique element identifier",
|
|
203
227
|
"error"?: "error message if any"
|
|
204
228
|
}`;
|
|
205
229
|
};
|
package/dist/es/index.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"mappings":";;;;;;;;;;;;;;;;AAUA;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,OACK;AACP,SAAS,cAAc;AAEhB,IAAM,WAAN,MAAe;AAAA,EAUpB,YACE,MACA,SAGA;AACA,SAAK,SACH,SAAS,SAAS,QAAQ,MAAM,SAAS,IAAI,YAAY;AAC3D,SAAK,OAAO;AACZ,SAAK,SAAS,SAAS,SAAS,CAAC,GAAG;AAAA,MAAI,CAAC,SACvC,KAAK,kBAAkB,IAAI;AAAA,IAC7B;AACA,SAAK,cAAc,SAAS;AAAA,EAC9B;AAAA,EAEQ,kBAAkB,MAAyC;AACjE,WAAO;AAAA,MACL,QAAQ;AAAA,MACR,GAAG;AAAA,IACL;AAAA,EACF;AAAA,EAEA,MAAM,OAAO,MAAgE;AAC3E;AAAA,MACE,KAAK,WAAW;AAAA,MAChB;AAAA,QAAyD,KAAK,gBAAgB,GAAG,KAAK;AAAA,EAAK,KAAK,gBAAgB,GAAG,UAAU;AAAA,IAC/H;AACA,QAAI,MAAM,QAAQ,IAAI,GAAG;AACvB,WAAK,MAAM,KAAK,GAAG,KAAK,IAAI,CAAC,SAAS,KAAK,kBAAkB,IAAI,CAAC,CAAC;AAAA,IACrE,OAAO;AACL,WAAK,MAAM,KAAK,KAAK,kBAAkB,IAAI,CAAC;AAAA,IAC9C;AACA,QAAI,KAAK,WAAW,WAAW;AAC7B,WAAK,SAAS;AAAA,IAChB;AAAA,EACF;AAAA,EAEA,MAAM,QAAsB;AAC1B,QAAI,KAAK,WAAW,UAAU,KAAK,MAAM,SAAS,GAAG;AACnD,cAAQ;AAAA,QACN;AAAA,MACF;AAAA,IACF;AAEA,WAAO,KAAK,WAAW,WAAW,6BAA6B;AAC/D,WAAO,KAAK,WAAW,aAAa,+BAA+B;AACnE,WAAO,KAAK,WAAW,SAAS,4BAA4B;AAE5D,UAAM,mBAAmB,KAAK,MAAM;AAAA,MAClC,CAAC,SAAS,KAAK,WAAW;AAAA,IAC5B;AACA,QAAI,mBAAmB,GAAG;AAExB;AAAA,IACF;AAEA,SAAK,SAAS;AACd,QAAI,YAAY;AAChB,QAAI,wBAAwB;AAE5B,QAAI;AAEJ,WAAO,YAAY,KAAK,MAAM,QAAQ;AACpC,YAAM,OAAO,KAAK,MAAM,SAAS;AACjC;AAAA,QACE,KAAK,WAAW;AAAA,QAChB,2CAA2C,KAAK,MAAM;AAAA,MACxD;AACA,WAAK,SAAS;AAAA,QACZ,OAAO,KAAK,IAAI;AAAA,MAClB;AACA,UAAI;AACF,aAAK,SAAS;AACd,YAAI;AACF,cAAI,KAAK,aAAa;AACpB,kBAAM,KAAK,YAAY,IAAI;AAAA,UAC7B;AAAA,QACF,SAAS,GAAG;AACV,kBAAQ,MAAM,wBAAwB,CAAC;AAAA,QACzC;AACA;AAAA,UACE,CAAC,WAAW,UAAU,UAAU,EAAE,QAAQ,KAAK,IAAI,KAAK;AAAA,UACxD,0BAA0B,KAAK,IAAI;AAAA,QACrC;AAEA,cAAM,EAAE,UAAU,MAAM,IAAI;AAC5B,eAAO,UAAU,uCAAuC,KAAK,IAAI,EAAE;AAEnE,YAAI;AACJ,cAAM,kBAAmC;AAAA,UACvC;AAAA,UACA,SAAS,oBAAoB;AAAA,QAC/B;AAEA,YAAI,KAAK,SAAS,WAAW;AAC3B;AAAA,YACE,KAAK,YAAY,YACf,KAAK,YAAY,WACjB,KAAK,YAAY,YACjB,KAAK,YAAY,aACjB,KAAK,YAAY,YACjB,KAAK,YAAY;AAAA,YACnB,gCAAgC,KAAK,OAAO;AAAA,UAC9C;AACA,wBAAc,MAAM,KAAK,SAAS,OAAO,eAAe;AACxD,cAAI,KAAK,YAAY,UAAU;AAC7B,iCACE,aACC;AAAA,UACL;AAAA,QACF,WAAW,KAAK,SAAS,YAAY,KAAK,SAAS,YAAY;AAC7D,wBAAc,MAAM,KAAK,SAAS,OAAO,eAAe;AAAA,QAC1D,OAAO;AACL,kBAAQ;AAAA,YACN,0BAA0B,KAAK,IAAI;AAAA,UACrC;AACA,wBAAc,MAAM,KAAK,SAAS,OAAO,eAAe;AAAA,QAC1D;AAEA,eAAO,OAAO,MAAM,WAAW;AAC/B,aAAK,SAAS;AACd,aAAK,OAAO,MAAM,KAAK,IAAI;AAC3B,aAAK,OAAO,OAAO,KAAK,OAAO,MAAM,KAAK,OAAO;AACjD;AAAA,MACF,SAAS,GAAQ;AACf,gCAAwB;AACxB,aAAK,QAAQ;AACb,aAAK,eACH,GAAG,YAAY,OAAO,MAAM,WAAW,IAAI;AAC7C,aAAK,aAAa,EAAE;AAEpB,aAAK,SAAS;AACd,aAAK,OAAO,MAAM,KAAK,IAAI;AAC3B,aAAK,OAAO,OAAO,KAAK,OAAO,MAAM,KAAK,OAAO;AACjD;AAAA,MACF;AAAA,IACF;AAGA,aAAS,IAAI,YAAY,GAAG,IAAI,KAAK,MAAM,QAAQ,KAAK;AACtD,WAAK,MAAM,CAAC,EAAE,SAAS;AAAA,IACzB;AAEA,QAAI,uBAAuB;AACzB,WAAK,SAAS;AAAA,IAChB,OAAO;AACL,WAAK,SAAS;AAAA,IAChB;AAEA,QAAI,KAAK,MAAM,QAAQ;AAErB,YAAM,cAAc,KAAK,IAAI,WAAW,KAAK,MAAM,SAAS,CAAC;AAC7D,aAAO,KAAK,MAAM,WAAW,EAAE;AAAA,IACjC;AAAA,EACF;AAAA,EAEA,iBAA0B;AACxB,WAAO,KAAK,WAAW;AAAA,EACzB;AAAA,EAEA,kBAAwC;AACtC,QAAI,KAAK,WAAW,SAAS;AAC3B,aAAO;AAAA,IACT;AACA,UAAM,iBAAiB,KAAK,MAAM;AAAA,MAChC,CAAC,SAAS,KAAK,WAAW;AAAA,IAC5B;AACA,QAAI,kBAAkB,GAAG;AACvB,aAAO,KAAK,MAAM,cAAc;AAAA,IAClC;AACA,WAAO;AAAA,EACT;AAAA,EAEA,OAAsB;AACpB,QAAI,mBAAmB;AAEvB,QAAI,aAAa,GAAG;AAClB,YAAM,iBAAiB,mBAAmB;AAC1C,UAAI,gBAAgB;AAClB,2BAAmB,WAAW,cAAc;AAAA,MAC9C,OAAO;AACL,2BAAmB,GAAG,aAAa,CAAC;AAAA,MACtC;AAAA,IACF;AACA,UAAM,WAA0B;AAAA,MAC9B,YAAY,WAAW;AAAA,MACvB,YAAY,YAAY,mBAAmB,KAAK;AAAA,MAChD,mBAAmB;AAAA,MACnB,SAAS,KAAK,IAAI;AAAA,MAClB,MAAM,KAAK;AAAA,MACX,OAAO,KAAK;AAAA,IACd;AACA,WAAO;AAAA,EACT;AACF;;;AC9NA,SAAS,4BAA4B;AAE9B,IAAM,8BAA8B,MAAM;AAC/C,SAAO,6EAA6E,qBAAqB,CAAC;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAgB5G;;;ACcA;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,EACA,gBAAAA;AAAA,OACK;AACP,SAAS,yBAAyB,kBAAkB;AACpD,SAAS,gBAAgB;AACzB,SAAS,UAAAC,eAAc;;;AClCvB,SAAS,uBAAAC,sBAAqB,eAAAC,oBAAmB;AACjD,SAAS,YAAY;AAEd,SAAS,gBACd,MACA,gBACA;AACA,QAAM,WAAqB;AAAA,IACzB,YAAY,WAAW;AAAA,IACvB,SAAS,KAAK,IAAI;AAAA,IAClB,YAAYA,aAAYD,oBAAmB,KAAK;AAAA,EAClD;AACA,QAAM,YAAyB;AAAA,IAC7B,OAAO,KAAK;AAAA,IACZ,GAAG;AAAA,IACH,GAAG;AAAA,EACL;AAEA,mBAAiB,SAAS;AAC5B;;;AD2BA,IAAM,QAAQ,SAAS,YAAY;AACnC,IAAqB,UAArB,MAGE;AAAA,EAWA,YACE,SAGA,KACA;AAXF,sBAAoD;AAYlD,IAAAD,QAAO,SAAS,iCAAiC;AACjD,QAAI,OAAO,YAAY,YAAY;AACjC,WAAK,qBAAqB;AAAA,IAC5B,OAAO;AACL,WAAK,qBAAqB,MAAM,QAAQ,QAAQ,OAAO;AAAA,IACzD;AAEA,QAAI,OAAO,KAAK,eAAe,aAAa;AAC1C,WAAK,aAAa,IAAI;AAAA,IACxB;AACA,QAAI,OAAO,KAAK,aAAa,aAAa;AACxC,WAAK,WAAW,IAAI;AAAA,IACtB;AAAA,EACF;AAAA,EAEA,MAAM,OACJ,OACA,KACuB;AACvB,UAAM,EAAE,OAAO,IAAI,OAAO,CAAC;AAC3B,UAAM,cAAc,OAAO,UAAU,WAAW,QAAQ,MAAM;AAC9D,IAAAA,QAAO,aAAa,8BAA8B;AAClD,UAAM,iBAAiB,KAAK;AAC5B,SAAK,oBAAoB;AAEzB,IAAAA,QAAO,OAAO,UAAU,UAAU,sCAAsC;AAExE,UAAM,wBAAwB;AAAA,MAC5B;AAAA,IACF;AACA,QAAI,uBAAuB;AACzB,YAAM,yBAAyB,qBAAqB;AAAA,IACtD;AACA,QAAI;AACJ,QAAI,MAAM,aAAa,uBAAuB;AAC5C,yBAAmB,MAAM;AAAA,IAC3B;AAEA,QAAI,oBAAoB,CAACD,cAAa,GAAG;AACvC,cAAQ;AAAA,QACN;AAAA,MACF;AACA,yBAAmB;AAAA,IACrB;AAEA,UAAM,UAAU,KAAK,WAAY,MAAM,KAAK,mBAAmB,QAAQ;AAEvE,QAAI,aAA+B;AACnC,QAAI,wBAA4C;AAChD,QAAI,kBAA2C;AAC/C,QAAI,qBAEY;AAChB,QAAI,kBAAkB;AACpB,2BAAqB,MAAM,gBAAgB;AAAA,QACzC;AAAA,QACA,oBAAoB;AAAA,MACtB,CAAC;AACD,MAAAC;AAAA,QACE,mBAAmB;AAAA,QACnB,gCAAgC,gBAAgB,IAC9C,mBAAmB,QAAQ,KAAK,mBAAmB,KAAK,KAAK,EAC/D;AAAA,MACF;AACA,8BAAwB,mBAAmB;AAC3C,wBAAkB,mBAAmB;AACrC,mBAAa,mBAAmB;AAAA,IAClC;AAEA,UAAM,YAAY,KAAK,IAAI;AAC3B,UAAM;AAAA,MACJ;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IACF,IAAI,MAAM,gBAAgB;AAAA,MACxB,QAAQ,UAAU,KAAK;AAAA,MACvB;AAAA,MACA,0BAA0B;AAAA,MAC1B,cAAc;AAAA,IAChB,CAAC;AAED,UAAM,WAAW,KAAK,IAAI,IAAI;AAC9B,UAAM,WAA4B;AAAA,MAChC,GAAI,KAAK,WAAW,KAAK,WAAW,CAAC;AAAA,MACrC,YAAY;AAAA,MACZ,aAAa,KAAK,UAAU,WAAW;AAAA,MACvC,gBAAgB,KAAK,UAAU,WAAW;AAAA,MAC1C;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IACF;AAEA,QAAI;AACJ,QAAI,YAAY,QAAQ,QAAQ;AAC9B,iBAAW;AAAA,EAAgC,YAAY,OAAO,KAAK,IAAI,CAAC;AAAA,IAC1E;AAEA,UAAM,WAAsC;AAAA,MAC1C,MAAM;AAAA,MACN,WAAW;AAAA,QACT,SAAS;AAAA,MACX;AAAA,MACA,gBAAgB,CAAC;AAAA,MACjB,aAAa;AAAA,MACb,MAAM;AAAA,MACN;AAAA,MACA,WAAW,CAAC,CAAC;AAAA,MACb,OAAO;AAAA,IACT;AAEA,UAAM,WAA0B,CAAC;AACjC,KAAC,YAAY,YAAY,CAAC,GAAG,QAAQ,CAAC,SAAS;AAC7C,UAAI,QAAQ,MAAM;AAChB,cAAM,UAAU,YAAY,MAAM,EAAE;AAEpC,YAAI,CAAC,SAAS;AACZ,kBAAQ;AAAA,YACN,kCAAkC,KAAK,EAAE;AAAA,UAC3C;AACA;AAAA,QACF;AACA,iBAAS,KAAK,OAAO;AAAA,MACvB;AAAA,IACF,CAAC;AAED;AAAA,MACE;AAAA,QACE,GAAG;AAAA,QACH,gBAAgB;AAAA,MAClB;AAAA,MACA;AAAA,IACF;AAEA,QAAI,UAAU;AACZ,YAAM,IAAI,MAAM,QAAQ;AAAA,IAC1B;AAEA,IAAAA;AAAA,MACE,SAAS,UAAU;AAAA,MACnB,6CAA6C,SAAS,MAAM;AAAA,IAC9D;AAEA,QAAI,SAAS,WAAW,GAAG;AACzB,aAAO;AAAA,QACL,SAAS;AAAA,UACP,IAAI,SAAS,CAAC,EAAG;AAAA,UACjB,SAAS,SAAS,CAAC,EAAG;AAAA,UACtB,QAAQ,SAAS,CAAC,EAAG;AAAA,UACrB,MAAM,SAAS,CAAC,EAAG;AAAA,UACnB,QAAQ,SAAS,CAAC,EAAG,UAAU,CAAC;AAAA,UAChC,YAAY,SAAS,CAAC,EAAG;AAAA,UACzB;AAAA,QACF;AAAA,QACA;AAAA,MACF;AAAA,IACF;AACA,WAAO;AAAA,MACL,SAAS;AAAA,MACT;AAAA,IACF;AAAA,EACF;AAAA,EAEA,MAAM,QACJ,YACA,KACA,kBACc;AACd,IAAAA;AAAA,MACE,OAAO,eAAe,YAAY,OAAO,eAAe;AAAA,MACxD,kDAAkD,OAAO,UAAU;AAAA,IACrE;AACA,UAAM,iBAAiB,KAAK;AAC5B,SAAK,oBAAoB;AAEzB,UAAM,UAAU,MAAM,KAAK,mBAAmB,SAAS;AAEvD,UAAM,YAAY,KAAK,IAAI;AAC3B,UAAM,EAAE,aAAa,MAAM,IAAI,MAAM,qBAAwB;AAAA,MAC3D;AAAA,MACA,WAAW;AAAA,MACX;AAAA,MACA,eAAe;AAAA,IACjB,CAAC;AAED,UAAM,WAAW,KAAK,IAAI,IAAI;AAC9B,UAAM,WAA4B;AAAA,MAChC,GAAI,KAAK,WAAW,KAAK,WAAW,CAAC;AAAA,MACrC,YAAY;AAAA,MACZ,aAAa,KAAK,UAAU,WAAW;AAAA,IACzC;AAEA,QAAI;AACJ,QAAI,YAAY,QAAQ,QAAQ;AAC9B,iBAAW;AAAA,EAAwB,YAAY,OAAO,KAAK,IAAI,CAAC;AAAA,IAClE;AAEA,UAAM,WAAsC;AAAA,MAC1C,MAAM;AAAA,MACN,WAAW;AAAA,QACT;AAAA,MACF;AAAA,MACA,gBAAgB,CAAC;AAAA,MACjB,MAAM;AAAA,MACN;AAAA,MACA,OAAO;AAAA,IACT;AAEA,UAAM,EAAE,KAAK,IAAI,eAAe,CAAC;AAGjC;AAAA,MACE;AAAA,QACE,GAAG;AAAA,QACH;AAAA,MACF;AAAA,MACA;AAAA,IACF;AAEA,QAAI,YAAY,CAAC,MAAM;AACrB,YAAM,IAAI,MAAM,QAAQ;AAAA,IAC1B;AAEA,WAAO;AAAA,MACL;AAAA,MACA;AAAA,IACF;AAAA,EACF;AAAA,EAEA,MAAM,OAAO,WAA2D;AACtE,UAAM,iBAAiB,KAAK;AAC5B,SAAK,oBAAoB;AAEzB,UAAM,UAAU,MAAM,KAAK,mBAAmB,QAAQ;AACtD,UAAM,YAAY,KAAK,IAAI;AAC3B,UAAM,eAAe,MAAM,SAAS;AAAA,MAClC;AAAA,MACA;AAAA,IACF,CAAC;AAED,UAAM,WAAW,KAAK,IAAI,IAAI;AAC9B,UAAM,WAA4B;AAAA,MAChC,GAAI,KAAK,WAAW,KAAK,WAAW,CAAC;AAAA,MACrC,YAAY;AAAA,MACZ,aAAa,KAAK,UAAU,aAAa,OAAO;AAAA,IAClD;AAEA,UAAM,EAAE,SAAS,KAAK,IAAI,aAAa;AACvC,UAAM,WAAsC;AAAA,MAC1C,MAAM;AAAA,MACN,WAAW;AAAA,QACT;AAAA,MACF;AAAA,MACA,gBAAgB,CAAC;AAAA,MACjB,MAAM;AAAA,MACN;AAAA,MACA,eAAe;AAAA,MACf,kBAAkB;AAAA,MAClB,OAAO,OAAO,SAAY;AAAA,IAC5B;AACA,oBAAgB,UAAU,cAAc;AAExC,WAAO;AAAA,MACL;AAAA,MACA;AAAA,MACA,OAAO,aAAa;AAAA,IACtB;AAAA,EACF;AAAA,EACA,MAAM,SACJ,QACA,KAGyD;AACzD,IAAAA,QAAO,QAAQ,yCAAyC;AACxD,UAAM,UAAU,MAAM,KAAK,mBAAmB,UAAU;AACxD,UAAM,EAAE,kBAAkB,KAAK,IAAI;AACnC,IAAAA,QAAO,kBAAkB,6CAA6C;AAEtE,UAAM,eAAe,4BAA4B;AAGjD,UAAM,kBAAkB;AACxB,UAAM,aAAmB,MAAM,QAAQ,MAAM,IACzC;AAAA,MACE,MAAM,KAAK,MAAM,OAAO,CAAC,IAAI,kBAAkB,CAAC;AAAA,MAChD,KAAK,KAAK,MAAM,OAAO,CAAC,IAAI,kBAAkB,CAAC;AAAA,MAC/C,OAAO;AAAA,MACP,QAAQ;AAAA,IACV,IACA;AAEJ,QAAI,eAAe,MAAM,wBAAwB;AAAA,MAC/C,gBAAgB;AAAA,MAChB;AAAA,MACA,sBAAsB;AAAA,QACpB;AAAA,UACE,MAAM;AAAA,QACR;AAAA,MACF;AAAA,MACA,iBAAiB;AAAA,IACnB,CAAC;AAED,QAAI,KAAK,WAAW;AAClB,YAAM,aAAa,iBAAiB,YAAY,QAAQ,IAAI;AAC5D,YAAM,4BAA4B,UAAU;AAC5C,qBAAe,MAAM;AAAA,QACnB;AAAA,QACA;AAAA,QACA,qBAAqB,oBAAoB;AAAA,MAC3C;AAAA,IACF;AAEA,UAAM,OAAe;AAAA,MACnB,EAAE,MAAM,UAAU,SAAS,aAAa;AAAA,MACxC;AAAA,QACE,MAAM;AAAA,QACN,SAAS;AAAA,UACP;AAAA,YACE,MAAM;AAAA,YACN,WAAW;AAAA,cACT,KAAK;AAAA,cACL,QAAQ;AAAA,YACV;AAAA,UACF;AAAA,QACF;AAAA,MACF;AAAA,IACF;AAEA,UAAM,WACJ,KAAK,cAAc;AAErB,UAAM,MAAM,MAAM,SAAS,8BAAmC;AAE9D,UAAM,EAAE,QAAQ,IAAI;AACpB,IAAAA,QAAO,CAAC,QAAQ,OAAO,oBAAoB,QAAQ,KAAK,EAAE;AAC1D,IAAAA,QAAO,QAAQ,aAAa,gCAAgC;AAC5D,WAAO;AAAA,EACT;AACF;;;AErZA,SAAS,eAAAE,cAAa,uBAAAD,4BAA2B;AAGjD,IAAO,cAAQ","names":["vlLocateMode","assert","MIDSCENE_MODEL_NAME","getAIConfig"],"ignoreList":[],"sources":["../../src/ai-model/action-executor.ts","../../src/ai-model/prompt/describe.ts","../../src/insight/index.ts","../../src/insight/utils.ts","../../src/index.ts"],"sourcesContent":["import type {\n ExecutionDump,\n ExecutionTask,\n ExecutionTaskApply,\n ExecutionTaskInsightLocateOutput,\n ExecutionTaskProgressOptions,\n ExecutionTaskReturn,\n ExecutorContext,\n} from '@/types';\nimport { getVersion } from '@/utils';\nimport {\n MIDSCENE_MODEL_NAME,\n getAIConfig,\n uiTarsModelVersion,\n vlLocateMode,\n} from '@midscene/shared/env';\nimport { assert } from '@midscene/shared/utils';\n\nexport class Executor {\n name: string;\n\n tasks: ExecutionTask[];\n\n // status of executor\n status: 'init' | 'pending' | 'running' | 'completed' | 'error';\n\n onTaskStart?: ExecutionTaskProgressOptions['onTaskStart'];\n\n constructor(\n name: string,\n options?: ExecutionTaskProgressOptions & {\n tasks?: ExecutionTaskApply[];\n },\n ) {\n this.status =\n options?.tasks && options.tasks.length > 0 ? 'pending' : 'init';\n this.name = name;\n this.tasks = (options?.tasks || []).map((item) =>\n this.markTaskAsPending(item),\n );\n this.onTaskStart = options?.onTaskStart;\n }\n\n private markTaskAsPending(task: ExecutionTaskApply): ExecutionTask {\n return {\n status: 'pending',\n ...task,\n };\n }\n\n async append(task: ExecutionTaskApply[] | ExecutionTaskApply): Promise<void> {\n assert(\n this.status !== 'error',\n `executor is in error state, cannot append task\\nerror=${this.latestErrorTask()?.error}\\n${this.latestErrorTask()?.errorStack}`,\n );\n if (Array.isArray(task)) {\n this.tasks.push(...task.map((item) => this.markTaskAsPending(item)));\n } else {\n this.tasks.push(this.markTaskAsPending(task));\n }\n if (this.status !== 'running') {\n this.status = 'pending';\n }\n }\n\n async flush(): Promise<any> {\n if (this.status === 'init' && this.tasks.length > 0) {\n console.warn(\n 'illegal state for executor, status is init but tasks are not empty',\n );\n }\n\n assert(this.status !== 'running', 'executor is already running');\n assert(this.status !== 'completed', 'executor is already completed');\n assert(this.status !== 'error', 'executor is in error state');\n\n const nextPendingIndex = this.tasks.findIndex(\n (task) => task.status === 'pending',\n );\n if (nextPendingIndex < 0) {\n // all tasks are completed\n return;\n }\n\n this.status = 'running';\n let taskIndex = nextPendingIndex;\n let successfullyCompleted = true;\n\n let previousFindOutput: ExecutionTaskInsightLocateOutput | undefined;\n\n while (taskIndex < this.tasks.length) {\n const task = this.tasks[taskIndex];\n assert(\n task.status === 'pending',\n `task status should be pending, but got: ${task.status}`,\n );\n task.timing = {\n start: Date.now(),\n };\n try {\n task.status = 'running';\n try {\n if (this.onTaskStart) {\n await this.onTaskStart(task);\n }\n } catch (e) {\n console.error('error in onTaskStart', e);\n }\n assert(\n ['Insight', 'Action', 'Planning'].indexOf(task.type) >= 0,\n `unsupported task type: ${task.type}`,\n );\n\n const { executor, param } = task;\n assert(executor, `executor is required for task type: ${task.type}`);\n\n let returnValue;\n const executorContext: ExecutorContext = {\n task,\n element: previousFindOutput?.element,\n };\n\n if (task.type === 'Insight') {\n assert(\n task.subType === 'Locate' ||\n task.subType === 'Query' ||\n task.subType === 'Assert' ||\n task.subType === 'Boolean' ||\n task.subType === 'Number' ||\n task.subType === 'String',\n `unsupported insight subType: ${task.subType}`,\n );\n returnValue = await task.executor(param, executorContext);\n if (task.subType === 'Locate') {\n previousFindOutput = (\n returnValue as ExecutionTaskReturn<ExecutionTaskInsightLocateOutput>\n )?.output;\n }\n } else if (task.type === 'Action' || task.type === 'Planning') {\n returnValue = await task.executor(param, executorContext);\n } else {\n console.warn(\n `unsupported task type: ${task.type}, will try to execute it directly`,\n );\n returnValue = await task.executor(param, executorContext);\n }\n\n Object.assign(task, returnValue);\n task.status = 'finished';\n task.timing.end = Date.now();\n task.timing.cost = task.timing.end - task.timing.start;\n taskIndex++;\n } catch (e: any) {\n successfullyCompleted = false;\n task.error = e;\n task.errorMessage =\n e?.message || (typeof e === 'string' ? e : 'error-without-message');\n task.errorStack = e.stack;\n\n task.status = 'failed';\n task.timing.end = Date.now();\n task.timing.cost = task.timing.end - task.timing.start;\n break;\n }\n }\n\n // set all remaining tasks as cancelled\n for (let i = taskIndex + 1; i < this.tasks.length; i++) {\n this.tasks[i].status = 'cancelled';\n }\n\n if (successfullyCompleted) {\n this.status = 'completed';\n } else {\n this.status = 'error';\n }\n\n if (this.tasks.length) {\n // return the last output\n const outputIndex = Math.min(taskIndex, this.tasks.length - 1);\n return this.tasks[outputIndex].output;\n }\n }\n\n isInErrorState(): boolean {\n return this.status === 'error';\n }\n\n latestErrorTask(): ExecutionTask | null {\n if (this.status !== 'error') {\n return null;\n }\n const errorTaskIndex = this.tasks.findIndex(\n (task) => task.status === 'failed',\n );\n if (errorTaskIndex >= 0) {\n return this.tasks[errorTaskIndex];\n }\n return null;\n }\n\n dump(): ExecutionDump {\n let modelDescription = '';\n\n if (vlLocateMode()) {\n const uiTarsModelVer = uiTarsModelVersion();\n if (uiTarsModelVer) {\n modelDescription = `UI-TARS=${uiTarsModelVer}`;\n } else {\n modelDescription = `${vlLocateMode()} mode`;\n }\n }\n const dumpData: ExecutionDump = {\n sdkVersion: getVersion(),\n model_name: getAIConfig(MIDSCENE_MODEL_NAME) || '',\n model_description: modelDescription,\n logTime: Date.now(),\n name: this.name,\n tasks: this.tasks,\n };\n return dumpData;\n }\n}\n","import { getPreferredLanguage } from '@midscene/shared/env';\n\nexport const elementDescriberInstruction = () => {\n return `Describe the element in the red rectangle for precise identification. Use ${getPreferredLanguage()}.\n\nRules:\n1. Start with element type (button, input, link, etc.) \n2. Include key identifiers:\n - Text content: \"with text 'Submit'\"\n - Visual features: \"blue background\", \"icon only\"\n - Position: \"top-right\", \"below search bar\"\n3. Keep description under 20 words\n4. Don't mention the red rectangle\n\nReturn JSON:\n{\n \"description\": \"brief element type with key identifiers\",\n \"error\"?: \"error message if any\"\n}`;\n};\n","import {\n AIActionType,\n type AIArgs,\n callAiFn,\n expandSearchArea,\n} from '@/ai-model/common';\nimport {\n AiExtractElementInfo,\n AiLocateElement,\n callToGetJSONObject,\n} from '@/ai-model/index';\nimport { AiAssert, AiLocateSection } from '@/ai-model/inspect';\nimport { elementDescriberInstruction } from '@/ai-model/prompt/describe';\nimport type {\n AIDescribeElementResponse,\n AIElementResponse,\n AIUsageInfo,\n BaseElement,\n DetailedLocateParam,\n DumpSubscriber,\n InsightAction,\n InsightAssertionResponse,\n InsightExtractOption,\n InsightExtractParam,\n InsightOptions,\n InsightTaskInfo,\n LocateResult,\n PartialInsightDumpFromSDK,\n Rect,\n TMultimodalPrompt,\n TUserPrompt,\n UIContext,\n} from '@/types';\nimport {\n MIDSCENE_FORCE_DEEP_THINK,\n MIDSCENE_USE_QWEN_VL,\n getAIConfigInBoolean,\n vlLocateMode,\n} from '@midscene/shared/env';\nimport { compositeElementInfoImg, cropByRect } from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport { emitInsightDump } from './utils';\n\nexport interface LocateOpts {\n context?: UIContext<BaseElement>;\n callAI?: typeof callAiFn<AIElementResponse>;\n}\n\nexport type AnyValue<T> = {\n [K in keyof T]: unknown extends T[K] ? any : T[K];\n};\n\nconst debug = getDebug('ai:insight');\nexport default class Insight<\n ElementType extends BaseElement = BaseElement,\n ContextType extends UIContext<ElementType> = UIContext<ElementType>,\n> {\n contextRetrieverFn: (\n action: InsightAction,\n ) => Promise<ContextType> | ContextType;\n\n aiVendorFn: (...args: Array<any>) => Promise<any> = callAiFn;\n\n onceDumpUpdatedFn?: DumpSubscriber;\n\n taskInfo?: Omit<InsightTaskInfo, 'durationMs'>;\n\n constructor(\n context:\n | ContextType\n | ((action: InsightAction) => Promise<ContextType> | ContextType),\n opt?: InsightOptions,\n ) {\n assert(context, 'context is required for Insight');\n if (typeof context === 'function') {\n this.contextRetrieverFn = context;\n } else {\n this.contextRetrieverFn = () => Promise.resolve(context);\n }\n\n if (typeof opt?.aiVendorFn !== 'undefined') {\n this.aiVendorFn = opt.aiVendorFn;\n }\n if (typeof opt?.taskInfo !== 'undefined') {\n this.taskInfo = opt.taskInfo;\n }\n }\n\n async locate(\n query: DetailedLocateParam,\n opt?: LocateOpts,\n ): Promise<LocateResult> {\n const { callAI } = opt || {};\n const queryPrompt = typeof query === 'string' ? query : query.prompt;\n assert(queryPrompt, 'query is required for locate');\n const dumpSubscriber = this.onceDumpUpdatedFn;\n this.onceDumpUpdatedFn = undefined;\n\n assert(typeof query === 'object', 'query should be an object for locate');\n\n const globalDeepThinkSwitch = getAIConfigInBoolean(\n MIDSCENE_FORCE_DEEP_THINK,\n );\n if (globalDeepThinkSwitch) {\n debug('globalDeepThinkSwitch', globalDeepThinkSwitch);\n }\n let searchAreaPrompt;\n if (query.deepThink || globalDeepThinkSwitch) {\n searchAreaPrompt = query.prompt;\n }\n\n if (searchAreaPrompt && !vlLocateMode()) {\n console.warn(\n 'The \"deepThink\" feature is not supported with multimodal LLM. Please config VL model for Midscene. https://midscenejs.com/choose-a-model',\n );\n searchAreaPrompt = undefined;\n }\n\n const context = opt?.context || (await this.contextRetrieverFn('locate'));\n\n let searchArea: Rect | undefined = undefined;\n let searchAreaRawResponse: string | undefined = undefined;\n let searchAreaUsage: AIUsageInfo | undefined = undefined;\n let searchAreaResponse:\n | Awaited<ReturnType<typeof AiLocateSection>>\n | undefined = undefined;\n if (searchAreaPrompt) {\n searchAreaResponse = await AiLocateSection({\n context,\n sectionDescription: searchAreaPrompt,\n });\n assert(\n searchAreaResponse.rect,\n `cannot find search area for \"${searchAreaPrompt}\"${\n searchAreaResponse.error ? `: ${searchAreaResponse.error}` : ''\n }`,\n );\n searchAreaRawResponse = searchAreaResponse.rawResponse;\n searchAreaUsage = searchAreaResponse.usage;\n searchArea = searchAreaResponse.rect;\n }\n\n const startTime = Date.now();\n const {\n parseResult,\n rect,\n elementById,\n rawResponse,\n usage,\n isOrderSensitive,\n } = await AiLocateElement({\n callAI: callAI || this.aiVendorFn,\n context,\n targetElementDescription: queryPrompt,\n searchConfig: searchAreaResponse,\n });\n\n const timeCost = Date.now() - startTime;\n const taskInfo: InsightTaskInfo = {\n ...(this.taskInfo ? this.taskInfo : {}),\n durationMs: timeCost,\n rawResponse: JSON.stringify(rawResponse),\n formatResponse: JSON.stringify(parseResult),\n usage,\n searchArea,\n searchAreaRawResponse,\n searchAreaUsage,\n };\n\n let errorLog: string | undefined;\n if (parseResult.errors?.length) {\n errorLog = `AI model failed to locate: \\n${parseResult.errors.join('\\n')}`;\n }\n\n const dumpData: PartialInsightDumpFromSDK = {\n type: 'locate',\n userQuery: {\n element: queryPrompt,\n },\n matchedElement: [],\n matchedRect: rect,\n data: null,\n taskInfo,\n deepThink: !!searchArea,\n error: errorLog,\n };\n\n const elements: BaseElement[] = [];\n (parseResult.elements || []).forEach((item) => {\n if ('id' in item) {\n const element = elementById(item?.id);\n\n if (!element) {\n console.warn(\n `locate: cannot find element id=${item.id}. Maybe an unstable response from AI model`,\n );\n return;\n }\n elements.push(element);\n }\n });\n\n emitInsightDump(\n {\n ...dumpData,\n matchedElement: elements,\n },\n dumpSubscriber,\n );\n\n if (errorLog) {\n throw new Error(errorLog);\n }\n\n assert(\n elements.length <= 1,\n `locate: multiple elements found, length = ${elements.length}`,\n );\n\n if (elements.length === 1) {\n return {\n element: {\n id: elements[0]!.id,\n indexId: elements[0]!.indexId,\n center: elements[0]!.center,\n rect: elements[0]!.rect,\n xpaths: elements[0]!.xpaths || [],\n attributes: elements[0]!.attributes,\n isOrderSensitive,\n },\n rect,\n };\n }\n return {\n element: null,\n rect,\n };\n }\n\n async extract<T>(\n dataDemand: InsightExtractParam,\n opt?: InsightExtractOption,\n multimodalPrompt?: TMultimodalPrompt,\n ): Promise<any> {\n assert(\n typeof dataDemand === 'object' || typeof dataDemand === 'string',\n `dataDemand should be object or string, but get ${typeof dataDemand}`,\n );\n const dumpSubscriber = this.onceDumpUpdatedFn;\n this.onceDumpUpdatedFn = undefined;\n\n const context = await this.contextRetrieverFn('extract');\n\n const startTime = Date.now();\n const { parseResult, usage } = await AiExtractElementInfo<T>({\n context,\n dataQuery: dataDemand,\n multimodalPrompt,\n extractOption: opt,\n });\n\n const timeCost = Date.now() - startTime;\n const taskInfo: InsightTaskInfo = {\n ...(this.taskInfo ? this.taskInfo : {}),\n durationMs: timeCost,\n rawResponse: JSON.stringify(parseResult),\n };\n\n let errorLog: string | undefined;\n if (parseResult.errors?.length) {\n errorLog = `AI response error: \\n${parseResult.errors.join('\\n')}`;\n }\n\n const dumpData: PartialInsightDumpFromSDK = {\n type: 'extract',\n userQuery: {\n dataDemand,\n },\n matchedElement: [],\n data: null,\n taskInfo,\n error: errorLog,\n };\n\n const { data } = parseResult || {};\n\n // 4\n emitInsightDump(\n {\n ...dumpData,\n data,\n },\n dumpSubscriber,\n );\n\n if (errorLog && !data) {\n throw new Error(errorLog);\n }\n\n return {\n data,\n usage,\n };\n }\n\n async assert(assertion: TUserPrompt): Promise<InsightAssertionResponse> {\n const dumpSubscriber = this.onceDumpUpdatedFn;\n this.onceDumpUpdatedFn = undefined;\n\n const context = await this.contextRetrieverFn('assert');\n const startTime = Date.now();\n const assertResult = await AiAssert({\n assertion,\n context,\n });\n\n const timeCost = Date.now() - startTime;\n const taskInfo: InsightTaskInfo = {\n ...(this.taskInfo ? this.taskInfo : {}),\n durationMs: timeCost,\n rawResponse: JSON.stringify(assertResult.content),\n };\n\n const { thought, pass } = assertResult.content;\n const dumpData: PartialInsightDumpFromSDK = {\n type: 'assert',\n userQuery: {\n assertion,\n },\n matchedElement: [],\n data: null,\n taskInfo,\n assertionPass: pass,\n assertionThought: thought,\n error: pass ? undefined : thought,\n };\n emitInsightDump(dumpData, dumpSubscriber);\n\n return {\n pass,\n thought,\n usage: assertResult.usage,\n };\n }\n async describe(\n target: Rect | [number, number],\n opt?: {\n deepThink?: boolean;\n },\n ): Promise<Pick<AIDescribeElementResponse, 'description'>> {\n assert(target, 'target is required for insight.describe');\n const context = await this.contextRetrieverFn('describe');\n const { screenshotBase64, size } = context;\n assert(screenshotBase64, 'screenshot is required for insight.describe');\n\n const systemPrompt = elementDescriberInstruction();\n\n // Convert [x,y] center point to Rect if needed\n const defaultRectSize = 30;\n const targetRect: Rect = Array.isArray(target)\n ? {\n left: Math.floor(target[0] - defaultRectSize / 2),\n top: Math.floor(target[1] - defaultRectSize / 2),\n width: defaultRectSize,\n height: defaultRectSize,\n }\n : target;\n\n let imagePayload = await compositeElementInfoImg({\n inputImgBase64: screenshotBase64,\n size,\n elementsPositionInfo: [\n {\n rect: targetRect,\n },\n ],\n borderThickness: 3,\n });\n\n if (opt?.deepThink) {\n const searchArea = expandSearchArea(targetRect, context.size);\n debug('describe: set searchArea', searchArea);\n imagePayload = await cropByRect(\n imagePayload,\n searchArea,\n getAIConfigInBoolean(MIDSCENE_USE_QWEN_VL),\n );\n }\n\n const msgs: AIArgs = [\n { role: 'system', content: systemPrompt },\n {\n role: 'user',\n content: [\n {\n type: 'image_url',\n image_url: {\n url: imagePayload,\n detail: 'high',\n },\n },\n ],\n },\n ];\n\n const callAIFn =\n this.aiVendorFn || callToGetJSONObject<AIDescribeElementResponse>;\n\n const res = await callAIFn(msgs, AIActionType.DESCRIBE_ELEMENT);\n\n const { content } = res;\n assert(!content.error, `describe failed: ${content.error}`);\n assert(content.description, 'failed to describe the element');\n return content;\n }\n}\n","import type {\n DumpMeta,\n DumpSubscriber,\n InsightDump,\n PartialInsightDumpFromSDK,\n} from '@/types';\nimport { getVersion } from '@/utils';\nimport { MIDSCENE_MODEL_NAME, getAIConfig } from '@midscene/shared/env';\nimport { uuid } from '@midscene/shared/utils';\n\nexport function emitInsightDump(\n data: PartialInsightDumpFromSDK,\n dumpSubscriber?: DumpSubscriber,\n) {\n const baseData: DumpMeta = {\n sdkVersion: getVersion(),\n logTime: Date.now(),\n model_name: getAIConfig(MIDSCENE_MODEL_NAME) || '',\n };\n const finalData: InsightDump = {\n logId: uuid(),\n ...baseData,\n ...data,\n };\n\n dumpSubscriber?.(finalData);\n}\n","import { Executor } from './ai-model/action-executor';\nimport Insight from './insight/index';\nimport { getVersion } from './utils';\n\nexport {\n plan,\n describeUserPage,\n AiLocateElement,\n AiAssert,\n} from './ai-model/index';\n\nexport { getAIConfig, MIDSCENE_MODEL_NAME } from '@midscene/shared/env';\n\nexport type * from './types';\nexport default Insight;\nexport { Executor, Insight, getVersion };\n\nexport type {\n MidsceneYamlScript,\n MidsceneYamlTask,\n MidsceneYamlFlowItem,\n MidsceneYamlFlowItemAIRightClick,\n MidsceneYamlConfigResult,\n LocateOption,\n DetailedLocateParam,\n} from './yaml';\n"]}
|
|
1
|
+
{"version":3,"mappings":";;;;;;;;;;;;;;;;AAUA;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,OACK;AACP,SAAS,cAAc;AAEhB,IAAM,WAAN,MAAe;AAAA,EAUpB,YACE,MACA,SAGA;AACA,SAAK,SACH,SAAS,SAAS,QAAQ,MAAM,SAAS,IAAI,YAAY;AAC3D,SAAK,OAAO;AACZ,SAAK,SAAS,SAAS,SAAS,CAAC,GAAG;AAAA,MAAI,CAAC,SACvC,KAAK,kBAAkB,IAAI;AAAA,IAC7B;AACA,SAAK,cAAc,SAAS;AAAA,EAC9B;AAAA,EAEQ,kBAAkB,MAAyC;AACjE,WAAO;AAAA,MACL,QAAQ;AAAA,MACR,GAAG;AAAA,IACL;AAAA,EACF;AAAA,EAEA,MAAM,OAAO,MAAgE;AAC3E;AAAA,MACE,KAAK,WAAW;AAAA,MAChB;AAAA,QAAyD,KAAK,gBAAgB,GAAG,KAAK;AAAA,EAAK,KAAK,gBAAgB,GAAG,UAAU;AAAA,IAC/H;AACA,QAAI,MAAM,QAAQ,IAAI,GAAG;AACvB,WAAK,MAAM,KAAK,GAAG,KAAK,IAAI,CAAC,SAAS,KAAK,kBAAkB,IAAI,CAAC,CAAC;AAAA,IACrE,OAAO;AACL,WAAK,MAAM,KAAK,KAAK,kBAAkB,IAAI,CAAC;AAAA,IAC9C;AACA,QAAI,KAAK,WAAW,WAAW;AAC7B,WAAK,SAAS;AAAA,IAChB;AAAA,EACF;AAAA,EAEA,MAAM,QAAsB;AAC1B,QAAI,KAAK,WAAW,UAAU,KAAK,MAAM,SAAS,GAAG;AACnD,cAAQ;AAAA,QACN;AAAA,MACF;AAAA,IACF;AAEA,WAAO,KAAK,WAAW,WAAW,6BAA6B;AAC/D,WAAO,KAAK,WAAW,aAAa,+BAA+B;AACnE,WAAO,KAAK,WAAW,SAAS,4BAA4B;AAE5D,UAAM,mBAAmB,KAAK,MAAM;AAAA,MAClC,CAAC,SAAS,KAAK,WAAW;AAAA,IAC5B;AACA,QAAI,mBAAmB,GAAG;AAExB;AAAA,IACF;AAEA,SAAK,SAAS;AACd,QAAI,YAAY;AAChB,QAAI,wBAAwB;AAE5B,QAAI;AAEJ,WAAO,YAAY,KAAK,MAAM,QAAQ;AACpC,YAAM,OAAO,KAAK,MAAM,SAAS;AACjC;AAAA,QACE,KAAK,WAAW;AAAA,QAChB,2CAA2C,KAAK,MAAM;AAAA,MACxD;AACA,WAAK,SAAS;AAAA,QACZ,OAAO,KAAK,IAAI;AAAA,MAClB;AACA,UAAI;AACF,aAAK,SAAS;AACd,YAAI;AACF,cAAI,KAAK,aAAa;AACpB,kBAAM,KAAK,YAAY,IAAI;AAAA,UAC7B;AAAA,QACF,SAAS,GAAG;AACV,kBAAQ,MAAM,wBAAwB,CAAC;AAAA,QACzC;AACA;AAAA,UACE,CAAC,WAAW,UAAU,UAAU,EAAE,QAAQ,KAAK,IAAI,KAAK;AAAA,UACxD,0BAA0B,KAAK,IAAI;AAAA,QACrC;AAEA,cAAM,EAAE,UAAU,MAAM,IAAI;AAC5B,eAAO,UAAU,uCAAuC,KAAK,IAAI,EAAE;AAEnE,YAAI;AACJ,cAAM,kBAAmC;AAAA,UACvC;AAAA,UACA,SAAS,oBAAoB;AAAA,QAC/B;AAEA,YAAI,KAAK,SAAS,WAAW;AAC3B;AAAA,YACE,KAAK,YAAY,YACf,KAAK,YAAY,WACjB,KAAK,YAAY,YACjB,KAAK,YAAY,aACjB,KAAK,YAAY,YACjB,KAAK,YAAY;AAAA,YACnB,gCAAgC,KAAK,OAAO;AAAA,UAC9C;AACA,wBAAc,MAAM,KAAK,SAAS,OAAO,eAAe;AACxD,cAAI,KAAK,YAAY,UAAU;AAC7B,iCACE,aACC;AAAA,UACL;AAAA,QACF,WAAW,KAAK,SAAS,YAAY,KAAK,SAAS,YAAY;AAC7D,wBAAc,MAAM,KAAK,SAAS,OAAO,eAAe;AAAA,QAC1D,OAAO;AACL,kBAAQ;AAAA,YACN,0BAA0B,KAAK,IAAI;AAAA,UACrC;AACA,wBAAc,MAAM,KAAK,SAAS,OAAO,eAAe;AAAA,QAC1D;AAEA,eAAO,OAAO,MAAM,WAAW;AAC/B,aAAK,SAAS;AACd,aAAK,OAAO,MAAM,KAAK,IAAI;AAC3B,aAAK,OAAO,OAAO,KAAK,OAAO,MAAM,KAAK,OAAO;AACjD;AAAA,MACF,SAAS,GAAQ;AACf,gCAAwB;AACxB,aAAK,QAAQ;AACb,aAAK,eACH,GAAG,YAAY,OAAO,MAAM,WAAW,IAAI;AAC7C,aAAK,aAAa,EAAE;AAEpB,aAAK,SAAS;AACd,aAAK,OAAO,MAAM,KAAK,IAAI;AAC3B,aAAK,OAAO,OAAO,KAAK,OAAO,MAAM,KAAK,OAAO;AACjD;AAAA,MACF;AAAA,IACF;AAGA,aAAS,IAAI,YAAY,GAAG,IAAI,KAAK,MAAM,QAAQ,KAAK;AACtD,WAAK,MAAM,CAAC,EAAE,SAAS;AAAA,IACzB;AAEA,QAAI,uBAAuB;AACzB,WAAK,SAAS;AAAA,IAChB,OAAO;AACL,WAAK,SAAS;AAAA,IAChB;AAEA,QAAI,KAAK,MAAM,QAAQ;AAErB,YAAM,cAAc,KAAK,IAAI,WAAW,KAAK,MAAM,SAAS,CAAC;AAC7D,aAAO,KAAK,MAAM,WAAW,EAAE;AAAA,IACjC;AAAA,EACF;AAAA,EAEA,iBAA0B;AACxB,WAAO,KAAK,WAAW;AAAA,EACzB;AAAA,EAEA,kBAAwC;AACtC,QAAI,KAAK,WAAW,SAAS;AAC3B,aAAO;AAAA,IACT;AACA,UAAM,iBAAiB,KAAK,MAAM;AAAA,MAChC,CAAC,SAAS,KAAK,WAAW;AAAA,IAC5B;AACA,QAAI,kBAAkB,GAAG;AACvB,aAAO,KAAK,MAAM,cAAc;AAAA,IAClC;AACA,WAAO;AAAA,EACT;AAAA,EAEA,OAAsB;AACpB,QAAI,mBAAmB;AAEvB,QAAI,aAAa,GAAG;AAClB,YAAM,iBAAiB,mBAAmB;AAC1C,UAAI,gBAAgB;AAClB,2BAAmB,WAAW,cAAc;AAAA,MAC9C,OAAO;AACL,2BAAmB,GAAG,aAAa,CAAC;AAAA,MACtC;AAAA,IACF;AACA,UAAM,WAA0B;AAAA,MAC9B,YAAY,WAAW;AAAA,MACvB,YAAY,YAAY,mBAAmB,KAAK;AAAA,MAChD,mBAAmB;AAAA,MACnB,SAAS,KAAK,IAAI;AAAA,MAClB,MAAM,KAAK;AAAA,MACX,OAAO,KAAK;AAAA,IACd;AACA,WAAO;AAAA,EACT;AACF;;;AC9NA,SAAS,4BAA4B;AAE9B,IAAM,8BAA8B,MAAM;AAC/C,SAAO;AAAA,4EACmE,qBAAqB,CAAC;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAuClG;;;ACVA;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,EACA,gBAAAA;AAAA,OACK;AACP,SAAS,yBAAyB,kBAAkB;AACpD,SAAS,gBAAgB;AACzB,SAAS,UAAAC,eAAc;;;AClCvB,SAAS,uBAAAC,sBAAqB,eAAAC,oBAAmB;AACjD,SAAS,YAAY;AAEd,SAAS,gBACd,MACA,gBACA;AACA,QAAM,WAAqB;AAAA,IACzB,YAAY,WAAW;AAAA,IACvB,SAAS,KAAK,IAAI;AAAA,IAClB,YAAYA,aAAYD,oBAAmB,KAAK;AAAA,EAClD;AACA,QAAM,YAAyB;AAAA,IAC7B,OAAO,KAAK;AAAA,IACZ,GAAG;AAAA,IACH,GAAG;AAAA,EACL;AAEA,mBAAiB,SAAS;AAC5B;;;AD2BA,IAAM,QAAQ,SAAS,YAAY;AACnC,IAAqB,UAArB,MAGE;AAAA,EAWA,YACE,SAGA,KACA;AAXF,sBAAoD;AAYlD,IAAAD,QAAO,SAAS,iCAAiC;AACjD,QAAI,OAAO,YAAY,YAAY;AACjC,WAAK,qBAAqB;AAAA,IAC5B,OAAO;AACL,WAAK,qBAAqB,MAAM,QAAQ,QAAQ,OAAO;AAAA,IACzD;AAEA,QAAI,OAAO,KAAK,eAAe,aAAa;AAC1C,WAAK,aAAa,IAAI;AAAA,IACxB;AACA,QAAI,OAAO,KAAK,aAAa,aAAa;AACxC,WAAK,WAAW,IAAI;AAAA,IACtB;AAAA,EACF;AAAA,EAEA,MAAM,OACJ,OACA,KACuB;AACvB,UAAM,EAAE,OAAO,IAAI,OAAO,CAAC;AAC3B,UAAM,cAAc,OAAO,UAAU,WAAW,QAAQ,MAAM;AAC9D,IAAAA,QAAO,aAAa,8BAA8B;AAClD,UAAM,iBAAiB,KAAK;AAC5B,SAAK,oBAAoB;AAEzB,IAAAA,QAAO,OAAO,UAAU,UAAU,sCAAsC;AAExE,UAAM,wBAAwB;AAAA,MAC5B;AAAA,IACF;AACA,QAAI,uBAAuB;AACzB,YAAM,yBAAyB,qBAAqB;AAAA,IACtD;AACA,QAAI;AACJ,QAAI,MAAM,aAAa,uBAAuB;AAC5C,yBAAmB,MAAM;AAAA,IAC3B;AAEA,QAAI,oBAAoB,CAACD,cAAa,GAAG;AACvC,cAAQ;AAAA,QACN;AAAA,MACF;AACA,yBAAmB;AAAA,IACrB;AAEA,UAAM,UAAU,KAAK,WAAY,MAAM,KAAK,mBAAmB,QAAQ;AAEvE,QAAI,aAA+B;AACnC,QAAI,wBAA4C;AAChD,QAAI,kBAA2C;AAC/C,QAAI,qBAEY;AAChB,QAAI,kBAAkB;AACpB,2BAAqB,MAAM,gBAAgB;AAAA,QACzC;AAAA,QACA,oBAAoB;AAAA,MACtB,CAAC;AACD,MAAAC;AAAA,QACE,mBAAmB;AAAA,QACnB,gCAAgC,gBAAgB,IAC9C,mBAAmB,QAAQ,KAAK,mBAAmB,KAAK,KAAK,EAC/D;AAAA,MACF;AACA,8BAAwB,mBAAmB;AAC3C,wBAAkB,mBAAmB;AACrC,mBAAa,mBAAmB;AAAA,IAClC;AAEA,UAAM,YAAY,KAAK,IAAI;AAC3B,UAAM;AAAA,MACJ;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IACF,IAAI,MAAM,gBAAgB;AAAA,MACxB,QAAQ,UAAU,KAAK;AAAA,MACvB;AAAA,MACA,0BAA0B;AAAA,MAC1B,cAAc;AAAA,IAChB,CAAC;AAED,UAAM,WAAW,KAAK,IAAI,IAAI;AAC9B,UAAM,WAA4B;AAAA,MAChC,GAAI,KAAK,WAAW,KAAK,WAAW,CAAC;AAAA,MACrC,YAAY;AAAA,MACZ,aAAa,KAAK,UAAU,WAAW;AAAA,MACvC,gBAAgB,KAAK,UAAU,WAAW;AAAA,MAC1C;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IACF;AAEA,QAAI;AACJ,QAAI,YAAY,QAAQ,QAAQ;AAC9B,iBAAW;AAAA,EAAgC,YAAY,OAAO,KAAK,IAAI,CAAC;AAAA,IAC1E;AAEA,UAAM,WAAsC;AAAA,MAC1C,MAAM;AAAA,MACN,WAAW;AAAA,QACT,SAAS;AAAA,MACX;AAAA,MACA,gBAAgB,CAAC;AAAA,MACjB,aAAa;AAAA,MACb,MAAM;AAAA,MACN;AAAA,MACA,WAAW,CAAC,CAAC;AAAA,MACb,OAAO;AAAA,IACT;AAEA,UAAM,WAA0B,CAAC;AACjC,KAAC,YAAY,YAAY,CAAC,GAAG,QAAQ,CAAC,SAAS;AAC7C,UAAI,QAAQ,MAAM;AAChB,cAAM,UAAU,YAAY,MAAM,EAAE;AAEpC,YAAI,CAAC,SAAS;AACZ,kBAAQ;AAAA,YACN,kCAAkC,KAAK,EAAE;AAAA,UAC3C;AACA;AAAA,QACF;AACA,iBAAS,KAAK,OAAO;AAAA,MACvB;AAAA,IACF,CAAC;AAED;AAAA,MACE;AAAA,QACE,GAAG;AAAA,QACH,gBAAgB;AAAA,MAClB;AAAA,MACA;AAAA,IACF;AAEA,QAAI,UAAU;AACZ,YAAM,IAAI,MAAM,QAAQ;AAAA,IAC1B;AAEA,IAAAA;AAAA,MACE,SAAS,UAAU;AAAA,MACnB,6CAA6C,SAAS,MAAM;AAAA,IAC9D;AAEA,QAAI,SAAS,WAAW,GAAG;AACzB,aAAO;AAAA,QACL,SAAS;AAAA,UACP,IAAI,SAAS,CAAC,EAAG;AAAA,UACjB,SAAS,SAAS,CAAC,EAAG;AAAA,UACtB,QAAQ,SAAS,CAAC,EAAG;AAAA,UACrB,MAAM,SAAS,CAAC,EAAG;AAAA,UACnB,QAAQ,SAAS,CAAC,EAAG,UAAU,CAAC;AAAA,UAChC,YAAY,SAAS,CAAC,EAAG;AAAA,UACzB;AAAA,QACF;AAAA,QACA;AAAA,MACF;AAAA,IACF;AACA,WAAO;AAAA,MACL,SAAS;AAAA,MACT;AAAA,IACF;AAAA,EACF;AAAA,EAEA,MAAM,QACJ,YACA,KACA,kBACc;AACd,IAAAA;AAAA,MACE,OAAO,eAAe,YAAY,OAAO,eAAe;AAAA,MACxD,kDAAkD,OAAO,UAAU;AAAA,IACrE;AACA,UAAM,iBAAiB,KAAK;AAC5B,SAAK,oBAAoB;AAEzB,UAAM,UAAU,MAAM,KAAK,mBAAmB,SAAS;AAEvD,UAAM,YAAY,KAAK,IAAI;AAC3B,UAAM,EAAE,aAAa,MAAM,IAAI,MAAM,qBAAwB;AAAA,MAC3D;AAAA,MACA,WAAW;AAAA,MACX;AAAA,MACA,eAAe;AAAA,IACjB,CAAC;AAED,UAAM,WAAW,KAAK,IAAI,IAAI;AAC9B,UAAM,WAA4B;AAAA,MAChC,GAAI,KAAK,WAAW,KAAK,WAAW,CAAC;AAAA,MACrC,YAAY;AAAA,MACZ,aAAa,KAAK,UAAU,WAAW;AAAA,IACzC;AAEA,QAAI;AACJ,QAAI,YAAY,QAAQ,QAAQ;AAC9B,iBAAW;AAAA,EAAwB,YAAY,OAAO,KAAK,IAAI,CAAC;AAAA,IAClE;AAEA,UAAM,WAAsC;AAAA,MAC1C,MAAM;AAAA,MACN,WAAW;AAAA,QACT;AAAA,MACF;AAAA,MACA,gBAAgB,CAAC;AAAA,MACjB,MAAM;AAAA,MACN;AAAA,MACA,OAAO;AAAA,IACT;AAEA,UAAM,EAAE,KAAK,IAAI,eAAe,CAAC;AAGjC;AAAA,MACE;AAAA,QACE,GAAG;AAAA,QACH;AAAA,MACF;AAAA,MACA;AAAA,IACF;AAEA,QAAI,YAAY,CAAC,MAAM;AACrB,YAAM,IAAI,MAAM,QAAQ;AAAA,IAC1B;AAEA,WAAO;AAAA,MACL;AAAA,MACA;AAAA,IACF;AAAA,EACF;AAAA,EAEA,MAAM,OAAO,WAA2D;AACtE,UAAM,iBAAiB,KAAK;AAC5B,SAAK,oBAAoB;AAEzB,UAAM,UAAU,MAAM,KAAK,mBAAmB,QAAQ;AACtD,UAAM,YAAY,KAAK,IAAI;AAC3B,UAAM,eAAe,MAAM,SAAS;AAAA,MAClC;AAAA,MACA;AAAA,IACF,CAAC;AAED,UAAM,WAAW,KAAK,IAAI,IAAI;AAC9B,UAAM,WAA4B;AAAA,MAChC,GAAI,KAAK,WAAW,KAAK,WAAW,CAAC;AAAA,MACrC,YAAY;AAAA,MACZ,aAAa,KAAK,UAAU,aAAa,OAAO;AAAA,IAClD;AAEA,UAAM,EAAE,SAAS,KAAK,IAAI,aAAa;AACvC,UAAM,WAAsC;AAAA,MAC1C,MAAM;AAAA,MACN,WAAW;AAAA,QACT;AAAA,MACF;AAAA,MACA,gBAAgB,CAAC;AAAA,MACjB,MAAM;AAAA,MACN;AAAA,MACA,eAAe;AAAA,MACf,kBAAkB;AAAA,MAClB,OAAO,OAAO,SAAY;AAAA,IAC5B;AACA,oBAAgB,UAAU,cAAc;AAExC,WAAO;AAAA,MACL;AAAA,MACA;AAAA,MACA,OAAO,aAAa;AAAA,IACtB;AAAA,EACF;AAAA,EACA,MAAM,SACJ,QACA,KAGyD;AACzD,IAAAA,QAAO,QAAQ,yCAAyC;AACxD,UAAM,UAAU,MAAM,KAAK,mBAAmB,UAAU;AACxD,UAAM,EAAE,kBAAkB,KAAK,IAAI;AACnC,IAAAA,QAAO,kBAAkB,6CAA6C;AAEtE,UAAM,eAAe,4BAA4B;AAGjD,UAAM,kBAAkB;AACxB,UAAM,aAAmB,MAAM,QAAQ,MAAM,IACzC;AAAA,MACE,MAAM,KAAK,MAAM,OAAO,CAAC,IAAI,kBAAkB,CAAC;AAAA,MAChD,KAAK,KAAK,MAAM,OAAO,CAAC,IAAI,kBAAkB,CAAC;AAAA,MAC/C,OAAO;AAAA,MACP,QAAQ;AAAA,IACV,IACA;AAEJ,QAAI,eAAe,MAAM,wBAAwB;AAAA,MAC/C,gBAAgB;AAAA,MAChB;AAAA,MACA,sBAAsB;AAAA,QACpB;AAAA,UACE,MAAM;AAAA,QACR;AAAA,MACF;AAAA,MACA,iBAAiB;AAAA,IACnB,CAAC;AAED,QAAI,KAAK,WAAW;AAClB,YAAM,aAAa,iBAAiB,YAAY,QAAQ,IAAI;AAC5D,YAAM,4BAA4B,UAAU;AAC5C,qBAAe,MAAM;AAAA,QACnB;AAAA,QACA;AAAA,QACA,qBAAqB,oBAAoB;AAAA,MAC3C;AAAA,IACF;AAEA,UAAM,OAAe;AAAA,MACnB,EAAE,MAAM,UAAU,SAAS,aAAa;AAAA,MACxC;AAAA,QACE,MAAM;AAAA,QACN,SAAS;AAAA,UACP;AAAA,YACE,MAAM;AAAA,YACN,WAAW;AAAA,cACT,KAAK;AAAA,cACL,QAAQ;AAAA,YACV;AAAA,UACF;AAAA,QACF;AAAA,MACF;AAAA,IACF;AAEA,UAAM,WACJ,KAAK,cAAc;AAErB,UAAM,MAAM,MAAM,SAAS,8BAAmC;AAE9D,UAAM,EAAE,QAAQ,IAAI;AACpB,IAAAA,QAAO,CAAC,QAAQ,OAAO,oBAAoB,QAAQ,KAAK,EAAE;AAC1D,IAAAA,QAAO,QAAQ,aAAa,gCAAgC;AAC5D,WAAO;AAAA,EACT;AACF;;;AErZA,SAAS,eAAAE,cAAa,uBAAAD,4BAA2B;AAGjD,IAAO,cAAQ","names":["vlLocateMode","assert","MIDSCENE_MODEL_NAME","getAIConfig"],"ignoreList":[],"sources":["../../src/ai-model/action-executor.ts","../../src/ai-model/prompt/describe.ts","../../src/insight/index.ts","../../src/insight/utils.ts","../../src/index.ts"],"sourcesContent":["import type {\n ExecutionDump,\n ExecutionTask,\n ExecutionTaskApply,\n ExecutionTaskInsightLocateOutput,\n ExecutionTaskProgressOptions,\n ExecutionTaskReturn,\n ExecutorContext,\n} from '@/types';\nimport { getVersion } from '@/utils';\nimport {\n MIDSCENE_MODEL_NAME,\n getAIConfig,\n uiTarsModelVersion,\n vlLocateMode,\n} from '@midscene/shared/env';\nimport { assert } from '@midscene/shared/utils';\n\nexport class Executor {\n name: string;\n\n tasks: ExecutionTask[];\n\n // status of executor\n status: 'init' | 'pending' | 'running' | 'completed' | 'error';\n\n onTaskStart?: ExecutionTaskProgressOptions['onTaskStart'];\n\n constructor(\n name: string,\n options?: ExecutionTaskProgressOptions & {\n tasks?: ExecutionTaskApply[];\n },\n ) {\n this.status =\n options?.tasks && options.tasks.length > 0 ? 'pending' : 'init';\n this.name = name;\n this.tasks = (options?.tasks || []).map((item) =>\n this.markTaskAsPending(item),\n );\n this.onTaskStart = options?.onTaskStart;\n }\n\n private markTaskAsPending(task: ExecutionTaskApply): ExecutionTask {\n return {\n status: 'pending',\n ...task,\n };\n }\n\n async append(task: ExecutionTaskApply[] | ExecutionTaskApply): Promise<void> {\n assert(\n this.status !== 'error',\n `executor is in error state, cannot append task\\nerror=${this.latestErrorTask()?.error}\\n${this.latestErrorTask()?.errorStack}`,\n );\n if (Array.isArray(task)) {\n this.tasks.push(...task.map((item) => this.markTaskAsPending(item)));\n } else {\n this.tasks.push(this.markTaskAsPending(task));\n }\n if (this.status !== 'running') {\n this.status = 'pending';\n }\n }\n\n async flush(): Promise<any> {\n if (this.status === 'init' && this.tasks.length > 0) {\n console.warn(\n 'illegal state for executor, status is init but tasks are not empty',\n );\n }\n\n assert(this.status !== 'running', 'executor is already running');\n assert(this.status !== 'completed', 'executor is already completed');\n assert(this.status !== 'error', 'executor is in error state');\n\n const nextPendingIndex = this.tasks.findIndex(\n (task) => task.status === 'pending',\n );\n if (nextPendingIndex < 0) {\n // all tasks are completed\n return;\n }\n\n this.status = 'running';\n let taskIndex = nextPendingIndex;\n let successfullyCompleted = true;\n\n let previousFindOutput: ExecutionTaskInsightLocateOutput | undefined;\n\n while (taskIndex < this.tasks.length) {\n const task = this.tasks[taskIndex];\n assert(\n task.status === 'pending',\n `task status should be pending, but got: ${task.status}`,\n );\n task.timing = {\n start: Date.now(),\n };\n try {\n task.status = 'running';\n try {\n if (this.onTaskStart) {\n await this.onTaskStart(task);\n }\n } catch (e) {\n console.error('error in onTaskStart', e);\n }\n assert(\n ['Insight', 'Action', 'Planning'].indexOf(task.type) >= 0,\n `unsupported task type: ${task.type}`,\n );\n\n const { executor, param } = task;\n assert(executor, `executor is required for task type: ${task.type}`);\n\n let returnValue;\n const executorContext: ExecutorContext = {\n task,\n element: previousFindOutput?.element,\n };\n\n if (task.type === 'Insight') {\n assert(\n task.subType === 'Locate' ||\n task.subType === 'Query' ||\n task.subType === 'Assert' ||\n task.subType === 'Boolean' ||\n task.subType === 'Number' ||\n task.subType === 'String',\n `unsupported insight subType: ${task.subType}`,\n );\n returnValue = await task.executor(param, executorContext);\n if (task.subType === 'Locate') {\n previousFindOutput = (\n returnValue as ExecutionTaskReturn<ExecutionTaskInsightLocateOutput>\n )?.output;\n }\n } else if (task.type === 'Action' || task.type === 'Planning') {\n returnValue = await task.executor(param, executorContext);\n } else {\n console.warn(\n `unsupported task type: ${task.type}, will try to execute it directly`,\n );\n returnValue = await task.executor(param, executorContext);\n }\n\n Object.assign(task, returnValue);\n task.status = 'finished';\n task.timing.end = Date.now();\n task.timing.cost = task.timing.end - task.timing.start;\n taskIndex++;\n } catch (e: any) {\n successfullyCompleted = false;\n task.error = e;\n task.errorMessage =\n e?.message || (typeof e === 'string' ? e : 'error-without-message');\n task.errorStack = e.stack;\n\n task.status = 'failed';\n task.timing.end = Date.now();\n task.timing.cost = task.timing.end - task.timing.start;\n break;\n }\n }\n\n // set all remaining tasks as cancelled\n for (let i = taskIndex + 1; i < this.tasks.length; i++) {\n this.tasks[i].status = 'cancelled';\n }\n\n if (successfullyCompleted) {\n this.status = 'completed';\n } else {\n this.status = 'error';\n }\n\n if (this.tasks.length) {\n // return the last output\n const outputIndex = Math.min(taskIndex, this.tasks.length - 1);\n return this.tasks[outputIndex].output;\n }\n }\n\n isInErrorState(): boolean {\n return this.status === 'error';\n }\n\n latestErrorTask(): ExecutionTask | null {\n if (this.status !== 'error') {\n return null;\n }\n const errorTaskIndex = this.tasks.findIndex(\n (task) => task.status === 'failed',\n );\n if (errorTaskIndex >= 0) {\n return this.tasks[errorTaskIndex];\n }\n return null;\n }\n\n dump(): ExecutionDump {\n let modelDescription = '';\n\n if (vlLocateMode()) {\n const uiTarsModelVer = uiTarsModelVersion();\n if (uiTarsModelVer) {\n modelDescription = `UI-TARS=${uiTarsModelVer}`;\n } else {\n modelDescription = `${vlLocateMode()} mode`;\n }\n }\n const dumpData: ExecutionDump = {\n sdkVersion: getVersion(),\n model_name: getAIConfig(MIDSCENE_MODEL_NAME) || '',\n model_description: modelDescription,\n logTime: Date.now(),\n name: this.name,\n tasks: this.tasks,\n };\n return dumpData;\n }\n}\n","import { getPreferredLanguage } from '@midscene/shared/env';\n\nexport const elementDescriberInstruction = () => {\n return `\nDescribe the element in the red rectangle for precise identification. Use ${getPreferredLanguage()}.\n\nCRITICAL REQUIREMENTS:\n1. UNIQUENESS: The description must uniquely identify this element on the current page\n2. UNIVERSALITY: Use generic, reusable selectors that work across different contexts\n3. PRECISION: Be specific enough to distinguish from similar elements\n\nDESCRIPTION STRUCTURE:\n1. Element type (button, input, link, div, etc.)\n2. Primary identifier (in order of preference):\n - Unique text content: \"with text 'Login'\"\n - Unique attribute: \"with aria-label 'Search'\"\n - Unique class/ID: \"with class 'primary-button'\"\n - Unique position: \"in header navigation\"\n3. Secondary identifiers (if needed for uniqueness):\n - Visual features: \"blue background\", \"with icon\"\n - Relative position: \"below search bar\", \"in sidebar\"\n - Parent context: \"in login form\", \"in main menu\"\n\nGUIDELINES:\n- Keep description under 25 words\n- Prioritize semantic identifiers over visual ones\n- Use consistent terminology across similar elements\n- Avoid page-specific or temporary content\n- Don't mention the red rectangle or selection box\n- Focus on stable, reusable characteristics\n\nEXAMPLES:\n- \"Login button with text 'Sign In'\"\n- \"Search input with placeholder 'Enter keywords'\"\n- \"Navigation link with text 'Home' in header\"\n- \"Submit button in contact form\"\n- \"Menu icon with aria-label 'Open menu'\"\n\nReturn JSON:\n{\n \"description\": \"unique element identifier\",\n \"error\"?: \"error message if any\"\n}`;\n};\n","import {\n AIActionType,\n type AIArgs,\n callAiFn,\n expandSearchArea,\n} from '@/ai-model/common';\nimport {\n AiExtractElementInfo,\n AiLocateElement,\n callToGetJSONObject,\n} from '@/ai-model/index';\nimport { AiAssert, AiLocateSection } from '@/ai-model/inspect';\nimport { elementDescriberInstruction } from '@/ai-model/prompt/describe';\nimport type {\n AIDescribeElementResponse,\n AIElementResponse,\n AIUsageInfo,\n BaseElement,\n DetailedLocateParam,\n DumpSubscriber,\n InsightAction,\n InsightAssertionResponse,\n InsightExtractOption,\n InsightExtractParam,\n InsightOptions,\n InsightTaskInfo,\n LocateResult,\n PartialInsightDumpFromSDK,\n Rect,\n TMultimodalPrompt,\n TUserPrompt,\n UIContext,\n} from '@/types';\nimport {\n MIDSCENE_FORCE_DEEP_THINK,\n MIDSCENE_USE_QWEN_VL,\n getAIConfigInBoolean,\n vlLocateMode,\n} from '@midscene/shared/env';\nimport { compositeElementInfoImg, cropByRect } from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport { emitInsightDump } from './utils';\n\nexport interface LocateOpts {\n context?: UIContext<BaseElement>;\n callAI?: typeof callAiFn<AIElementResponse>;\n}\n\nexport type AnyValue<T> = {\n [K in keyof T]: unknown extends T[K] ? any : T[K];\n};\n\nconst debug = getDebug('ai:insight');\nexport default class Insight<\n ElementType extends BaseElement = BaseElement,\n ContextType extends UIContext<ElementType> = UIContext<ElementType>,\n> {\n contextRetrieverFn: (\n action: InsightAction,\n ) => Promise<ContextType> | ContextType;\n\n aiVendorFn: (...args: Array<any>) => Promise<any> = callAiFn;\n\n onceDumpUpdatedFn?: DumpSubscriber;\n\n taskInfo?: Omit<InsightTaskInfo, 'durationMs'>;\n\n constructor(\n context:\n | ContextType\n | ((action: InsightAction) => Promise<ContextType> | ContextType),\n opt?: InsightOptions,\n ) {\n assert(context, 'context is required for Insight');\n if (typeof context === 'function') {\n this.contextRetrieverFn = context;\n } else {\n this.contextRetrieverFn = () => Promise.resolve(context);\n }\n\n if (typeof opt?.aiVendorFn !== 'undefined') {\n this.aiVendorFn = opt.aiVendorFn;\n }\n if (typeof opt?.taskInfo !== 'undefined') {\n this.taskInfo = opt.taskInfo;\n }\n }\n\n async locate(\n query: DetailedLocateParam,\n opt?: LocateOpts,\n ): Promise<LocateResult> {\n const { callAI } = opt || {};\n const queryPrompt = typeof query === 'string' ? query : query.prompt;\n assert(queryPrompt, 'query is required for locate');\n const dumpSubscriber = this.onceDumpUpdatedFn;\n this.onceDumpUpdatedFn = undefined;\n\n assert(typeof query === 'object', 'query should be an object for locate');\n\n const globalDeepThinkSwitch = getAIConfigInBoolean(\n MIDSCENE_FORCE_DEEP_THINK,\n );\n if (globalDeepThinkSwitch) {\n debug('globalDeepThinkSwitch', globalDeepThinkSwitch);\n }\n let searchAreaPrompt;\n if (query.deepThink || globalDeepThinkSwitch) {\n searchAreaPrompt = query.prompt;\n }\n\n if (searchAreaPrompt && !vlLocateMode()) {\n console.warn(\n 'The \"deepThink\" feature is not supported with multimodal LLM. Please config VL model for Midscene. https://midscenejs.com/choose-a-model',\n );\n searchAreaPrompt = undefined;\n }\n\n const context = opt?.context || (await this.contextRetrieverFn('locate'));\n\n let searchArea: Rect | undefined = undefined;\n let searchAreaRawResponse: string | undefined = undefined;\n let searchAreaUsage: AIUsageInfo | undefined = undefined;\n let searchAreaResponse:\n | Awaited<ReturnType<typeof AiLocateSection>>\n | undefined = undefined;\n if (searchAreaPrompt) {\n searchAreaResponse = await AiLocateSection({\n context,\n sectionDescription: searchAreaPrompt,\n });\n assert(\n searchAreaResponse.rect,\n `cannot find search area for \"${searchAreaPrompt}\"${\n searchAreaResponse.error ? `: ${searchAreaResponse.error}` : ''\n }`,\n );\n searchAreaRawResponse = searchAreaResponse.rawResponse;\n searchAreaUsage = searchAreaResponse.usage;\n searchArea = searchAreaResponse.rect;\n }\n\n const startTime = Date.now();\n const {\n parseResult,\n rect,\n elementById,\n rawResponse,\n usage,\n isOrderSensitive,\n } = await AiLocateElement({\n callAI: callAI || this.aiVendorFn,\n context,\n targetElementDescription: queryPrompt,\n searchConfig: searchAreaResponse,\n });\n\n const timeCost = Date.now() - startTime;\n const taskInfo: InsightTaskInfo = {\n ...(this.taskInfo ? this.taskInfo : {}),\n durationMs: timeCost,\n rawResponse: JSON.stringify(rawResponse),\n formatResponse: JSON.stringify(parseResult),\n usage,\n searchArea,\n searchAreaRawResponse,\n searchAreaUsage,\n };\n\n let errorLog: string | undefined;\n if (parseResult.errors?.length) {\n errorLog = `AI model failed to locate: \\n${parseResult.errors.join('\\n')}`;\n }\n\n const dumpData: PartialInsightDumpFromSDK = {\n type: 'locate',\n userQuery: {\n element: queryPrompt,\n },\n matchedElement: [],\n matchedRect: rect,\n data: null,\n taskInfo,\n deepThink: !!searchArea,\n error: errorLog,\n };\n\n const elements: BaseElement[] = [];\n (parseResult.elements || []).forEach((item) => {\n if ('id' in item) {\n const element = elementById(item?.id);\n\n if (!element) {\n console.warn(\n `locate: cannot find element id=${item.id}. Maybe an unstable response from AI model`,\n );\n return;\n }\n elements.push(element);\n }\n });\n\n emitInsightDump(\n {\n ...dumpData,\n matchedElement: elements,\n },\n dumpSubscriber,\n );\n\n if (errorLog) {\n throw new Error(errorLog);\n }\n\n assert(\n elements.length <= 1,\n `locate: multiple elements found, length = ${elements.length}`,\n );\n\n if (elements.length === 1) {\n return {\n element: {\n id: elements[0]!.id,\n indexId: elements[0]!.indexId,\n center: elements[0]!.center,\n rect: elements[0]!.rect,\n xpaths: elements[0]!.xpaths || [],\n attributes: elements[0]!.attributes,\n isOrderSensitive,\n },\n rect,\n };\n }\n return {\n element: null,\n rect,\n };\n }\n\n async extract<T>(\n dataDemand: InsightExtractParam,\n opt?: InsightExtractOption,\n multimodalPrompt?: TMultimodalPrompt,\n ): Promise<any> {\n assert(\n typeof dataDemand === 'object' || typeof dataDemand === 'string',\n `dataDemand should be object or string, but get ${typeof dataDemand}`,\n );\n const dumpSubscriber = this.onceDumpUpdatedFn;\n this.onceDumpUpdatedFn = undefined;\n\n const context = await this.contextRetrieverFn('extract');\n\n const startTime = Date.now();\n const { parseResult, usage } = await AiExtractElementInfo<T>({\n context,\n dataQuery: dataDemand,\n multimodalPrompt,\n extractOption: opt,\n });\n\n const timeCost = Date.now() - startTime;\n const taskInfo: InsightTaskInfo = {\n ...(this.taskInfo ? this.taskInfo : {}),\n durationMs: timeCost,\n rawResponse: JSON.stringify(parseResult),\n };\n\n let errorLog: string | undefined;\n if (parseResult.errors?.length) {\n errorLog = `AI response error: \\n${parseResult.errors.join('\\n')}`;\n }\n\n const dumpData: PartialInsightDumpFromSDK = {\n type: 'extract',\n userQuery: {\n dataDemand,\n },\n matchedElement: [],\n data: null,\n taskInfo,\n error: errorLog,\n };\n\n const { data } = parseResult || {};\n\n // 4\n emitInsightDump(\n {\n ...dumpData,\n data,\n },\n dumpSubscriber,\n );\n\n if (errorLog && !data) {\n throw new Error(errorLog);\n }\n\n return {\n data,\n usage,\n };\n }\n\n async assert(assertion: TUserPrompt): Promise<InsightAssertionResponse> {\n const dumpSubscriber = this.onceDumpUpdatedFn;\n this.onceDumpUpdatedFn = undefined;\n\n const context = await this.contextRetrieverFn('assert');\n const startTime = Date.now();\n const assertResult = await AiAssert({\n assertion,\n context,\n });\n\n const timeCost = Date.now() - startTime;\n const taskInfo: InsightTaskInfo = {\n ...(this.taskInfo ? this.taskInfo : {}),\n durationMs: timeCost,\n rawResponse: JSON.stringify(assertResult.content),\n };\n\n const { thought, pass } = assertResult.content;\n const dumpData: PartialInsightDumpFromSDK = {\n type: 'assert',\n userQuery: {\n assertion,\n },\n matchedElement: [],\n data: null,\n taskInfo,\n assertionPass: pass,\n assertionThought: thought,\n error: pass ? undefined : thought,\n };\n emitInsightDump(dumpData, dumpSubscriber);\n\n return {\n pass,\n thought,\n usage: assertResult.usage,\n };\n }\n async describe(\n target: Rect | [number, number],\n opt?: {\n deepThink?: boolean;\n },\n ): Promise<Pick<AIDescribeElementResponse, 'description'>> {\n assert(target, 'target is required for insight.describe');\n const context = await this.contextRetrieverFn('describe');\n const { screenshotBase64, size } = context;\n assert(screenshotBase64, 'screenshot is required for insight.describe');\n\n const systemPrompt = elementDescriberInstruction();\n\n // Convert [x,y] center point to Rect if needed\n const defaultRectSize = 30;\n const targetRect: Rect = Array.isArray(target)\n ? {\n left: Math.floor(target[0] - defaultRectSize / 2),\n top: Math.floor(target[1] - defaultRectSize / 2),\n width: defaultRectSize,\n height: defaultRectSize,\n }\n : target;\n\n let imagePayload = await compositeElementInfoImg({\n inputImgBase64: screenshotBase64,\n size,\n elementsPositionInfo: [\n {\n rect: targetRect,\n },\n ],\n borderThickness: 3,\n });\n\n if (opt?.deepThink) {\n const searchArea = expandSearchArea(targetRect, context.size);\n debug('describe: set searchArea', searchArea);\n imagePayload = await cropByRect(\n imagePayload,\n searchArea,\n getAIConfigInBoolean(MIDSCENE_USE_QWEN_VL),\n );\n }\n\n const msgs: AIArgs = [\n { role: 'system', content: systemPrompt },\n {\n role: 'user',\n content: [\n {\n type: 'image_url',\n image_url: {\n url: imagePayload,\n detail: 'high',\n },\n },\n ],\n },\n ];\n\n const callAIFn =\n this.aiVendorFn || callToGetJSONObject<AIDescribeElementResponse>;\n\n const res = await callAIFn(msgs, AIActionType.DESCRIBE_ELEMENT);\n\n const { content } = res;\n assert(!content.error, `describe failed: ${content.error}`);\n assert(content.description, 'failed to describe the element');\n return content;\n }\n}\n","import type {\n DumpMeta,\n DumpSubscriber,\n InsightDump,\n PartialInsightDumpFromSDK,\n} from '@/types';\nimport { getVersion } from '@/utils';\nimport { MIDSCENE_MODEL_NAME, getAIConfig } from '@midscene/shared/env';\nimport { uuid } from '@midscene/shared/utils';\n\nexport function emitInsightDump(\n data: PartialInsightDumpFromSDK,\n dumpSubscriber?: DumpSubscriber,\n) {\n const baseData: DumpMeta = {\n sdkVersion: getVersion(),\n logTime: Date.now(),\n model_name: getAIConfig(MIDSCENE_MODEL_NAME) || '',\n };\n const finalData: InsightDump = {\n logId: uuid(),\n ...baseData,\n ...data,\n };\n\n dumpSubscriber?.(finalData);\n}\n","import { Executor } from './ai-model/action-executor';\nimport Insight from './insight/index';\nimport { getVersion } from './utils';\n\nexport {\n plan,\n describeUserPage,\n AiLocateElement,\n AiAssert,\n} from './ai-model/index';\n\nexport { getAIConfig, MIDSCENE_MODEL_NAME } from '@midscene/shared/env';\n\nexport type * from './types';\nexport default Insight;\nexport { Executor, Insight, getVersion };\n\nexport type {\n MidsceneYamlScript,\n MidsceneYamlTask,\n MidsceneYamlFlowItem,\n MidsceneYamlFlowItemAIRightClick,\n MidsceneYamlConfigResult,\n} from './yaml';\n"]}
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import {
|
|
1
|
+
import { o as AIUsageInfo, U as UIContext, i as TUserPrompt, aG as ReferenceImage, t as AIElementLocatorResponse, O as ElementById, T as TMultimodalPrompt, h as InsightExtractOption, w as AIDataExtractionResponse, y as AIAssertionResponse, aA as PageType, Y as PlanningAIResponse } from './types-512d3687.js';
|
|
2
2
|
import { Rect, ElementTreeNode, BaseElement } from '@midscene/shared/types';
|
|
3
3
|
import { ChatCompletionSystemMessageParam, ChatCompletionUserMessageParam } from 'openai/resources';
|
|
4
4
|
|
|
@@ -102,4 +102,4 @@ declare function plan(userInstruction: string, opts: {
|
|
|
102
102
|
actionContext?: string;
|
|
103
103
|
}): Promise<PlanningAIResponse>;
|
|
104
104
|
|
|
105
|
-
export { AiLocateElement as A, AiAssert as a, AIActionType as b, callAiFn as c, describeUserPage as d,
|
|
105
|
+
export { AiLocateElement as A, AiAssert as a, AIActionType as b, callAiFn as c, describeUserPage as d, type AIArgs as e, elementByPositionWithElementInfo as f, AiExtractElementInfo as g, AiLocateSection as h, adaptBboxToRect as i, plan as p };
|
|
@@ -1,12 +1,11 @@
|
|
|
1
1
|
import { NodeType } from '@midscene/shared/constants';
|
|
2
|
-
import {
|
|
2
|
+
import { Rect, BaseElement, ElementTreeNode, Size } from '@midscene/shared/types';
|
|
3
3
|
import { ChatCompletionMessageParam } from 'openai/resources';
|
|
4
4
|
|
|
5
5
|
interface LocateOption {
|
|
6
6
|
deepThink?: boolean;
|
|
7
7
|
cacheable?: boolean;
|
|
8
8
|
xpath?: string;
|
|
9
|
-
_forceContext?: UIContext<BaseElement>;
|
|
10
9
|
}
|
|
11
10
|
interface InsightExtractOption {
|
|
12
11
|
domIncluded?: boolean | 'visible-only';
|
|
@@ -563,4 +562,4 @@ type TUserPrompt = string | ({
|
|
|
563
562
|
prompt: string;
|
|
564
563
|
} & Partial<TMultimodalPrompt>);
|
|
565
564
|
|
|
566
|
-
export { type
|
|
565
|
+
export { type PlanningActionParamRightClick as $, type AIDescribeElementResponse as A, type LocateValidatorResult as B, type AgentDescribeElementAtPointResult as C, type DumpSubscriber as D, type ExecutionTask as E, type CallAIFn as F, type EnsureObject as G, type LocateResultElement as H, type InsightAction as I, type DumpMeta as J, type InsightDump as K, type LocateResult as L, type MidsceneYamlScript as M, type LiteUISection as N, type ElementById as O, type PartialInsightDumpFromSDK as P, type OnTaskStartTip as Q, type ReportDumpWithAttributes as R, type AgentWaitForOpt as S, type TMultimodalPrompt as T, UIContext as U, type AgentAssertOpt as V, type PlanningLocateParam as W, type PlanningAction as X, type PlanningAIResponse as Y, type PlanningActionParamTap as Z, type PlanningActionParamHover as _, type ExecutionTaskProgressOptions as a, type MidsceneYamlFlowItemSleep as a$, type PlanningActionParamInputOrKeyPress as a0, type PlanningActionParamScroll as a1, type PlanningActionParamAssert as a2, type PlanningActionParamSleep as a3, type PlanningActionParamError as a4, type PlanningActionParamWaitFor as a5, type PlanningActionParamAndroidLongPress as a6, type PlanningActionParamAndroidPull as a7, type Color as a8, type BaseAgentParserOpt as a9, type PageType as aA, type StreamingCodeGenerationOptions as aB, type StreamingCallback as aC, type CodeGenerationChunk as aD, type StreamingAIResponse as aE, type LocateOption as aF, type ReferenceImage as aG, type scrollParam as aH, type MidsceneYamlScriptEnvBase as aI, type MidsceneYamlScriptWebEnv as aJ, type MidsceneYamlScriptAndroidEnv as aK, type MidsceneYamlScriptEnv as aL, type MidsceneYamlFlowItemAIAction as aM, type MidsceneYamlFlowItemAIAssert as aN, type MidsceneYamlFlowItemAIQuery as aO, type MidsceneYamlFlowItemAINumber as aP, type MidsceneYamlFlowItemAIString as aQ, type MidsceneYamlFlowItemAIAsk as aR, type MidsceneYamlFlowItemAIBoolean as aS, type MidsceneYamlFlowItemAILocate as aT, type MidsceneYamlFlowItemAIWaitFor as aU, type MidsceneYamlFlowItemAITap as aV, type MidsceneYamlFlowItemAIHover as aW, type MidsceneYamlFlowItemAIInput as aX, type MidsceneYamlFlowItemAIKeyboardPress as aY, type MidsceneYamlFlowItemAIScroll as aZ, type MidsceneYamlFlowItemEvaluateJavaScript as a_, type PuppeteerParserOpt as aa, type PlaywrightParserOpt as ab, type ExecutionRecorderItem as ac, type ExecutionTaskType as ad, type ExecutorContext as ae, type ExecutionTaskHitBy as af, type ExecutionTaskReturn as ag, type ExecutionTaskInsightLocateParam as ah, type ExecutionTaskInsightLocateOutput as ai, type ExecutionTaskInsightDumpLog as aj, type ExecutionTaskInsightLocateApply as ak, type ExecutionTaskInsightLocate as al, type ExecutionTaskInsightQueryParam as am, type ExecutionTaskInsightQueryOutput as an, type ExecutionTaskInsightQueryApply as ao, type ExecutionTaskInsightQuery as ap, type ExecutionTaskInsightAssertionParam as aq, type ExecutionTaskInsightAssertionApply as ar, type ExecutionTaskInsightAssertion as as, type ExecutionTaskActionApply as at, type ExecutionTaskAction as au, type ExecutionTaskLogApply as av, type ExecutionTaskLog as aw, type ExecutionTaskPlanningApply as ax, type ExecutionTaskPlanning as ay, type GroupedActionDump as az, type ExecutionTaskApply as b, type MidsceneYamlFlowItemLogScreenshot as b0, type FreeFn as b1, type ScriptPlayerTaskStatus as b2, type ScriptPlayerStatusValue as b3, type MidsceneYamlConfig as b4, type MidsceneYamlConfigOutput as b5, type ExecutionDump as c, type InsightTaskInfo as d, type InsightOptions as e, type DetailedLocateParam as f, type InsightExtractParam as g, type InsightExtractOption as h, type TUserPrompt as i, type InsightAssertionResponse as j, type MidsceneYamlTask as k, type MidsceneYamlFlowItem as l, type MidsceneYamlFlowItemAIRightClick as m, type MidsceneYamlConfigResult as n, type AIUsageInfo as o, AIResponseFormat as p, type AISingleElementResponseById as q, type AISingleElementResponseByPosition as r, type AISingleElementResponse as s, type AIElementLocatorResponse as t, type AIElementCoordinatesResponse as u, type AIElementResponse as v, type AIDataExtractionResponse as w, type AISectionLocatorResponse as x, type AIAssertionResponse as y, type LocatorValidatorOption as z };
|
package/dist/es/utils.d.ts
CHANGED
package/dist/es/utils.js
CHANGED
package/dist/lib/ai-model.d.ts
CHANGED
|
@@ -1,9 +1,9 @@
|
|
|
1
|
-
import {
|
|
1
|
+
import { aC as StreamingCallback, o as AIUsageInfo, aB as StreamingCodeGenerationOptions, aE as StreamingAIResponse, X as PlanningAction, l as MidsceneYamlFlowItem } from './types-512d3687.js';
|
|
2
2
|
import OpenAI from 'openai';
|
|
3
3
|
import { ChatCompletionMessageParam } from 'openai/resources';
|
|
4
4
|
export { ChatCompletionMessageParam } from 'openai/resources';
|
|
5
|
-
import { b as AIActionType } from './llm-planning-
|
|
6
|
-
export { a as AiAssert,
|
|
5
|
+
import { b as AIActionType, e as AIArgs } from './llm-planning-877248da.js';
|
|
6
|
+
export { a as AiAssert, g as AiExtractElementInfo, A as AiLocateElement, h as AiLocateSection, i as adaptBboxToRect, c as callAiFn, d as describeUserPage, f as elementByPositionWithElementInfo, p as plan } from './llm-planning-877248da.js';
|
|
7
7
|
import { vlLocateMode } from '@midscene/shared/env';
|
|
8
8
|
import { actionParser } from '@ui-tars/action-parser';
|
|
9
9
|
import { Size } from '@midscene/shared/types';
|
|
@@ -21,6 +21,10 @@ declare function callToGetJSONObject<T>(messages: ChatCompletionMessageParam[],
|
|
|
21
21
|
content: T;
|
|
22
22
|
usage?: AIUsageInfo;
|
|
23
23
|
}>;
|
|
24
|
+
declare function callAiFnWithStringResponse<T>(msgs: AIArgs, AIActionTypeValue: AIActionType): Promise<{
|
|
25
|
+
content: string;
|
|
26
|
+
usage?: AIUsageInfo;
|
|
27
|
+
}>;
|
|
24
28
|
|
|
25
29
|
declare function systemPromptToLocateElement(vlMode: ReturnType<typeof vlLocateMode>): string;
|
|
26
30
|
|
|
@@ -92,4 +96,4 @@ declare function vlmPlanning(options: {
|
|
|
92
96
|
}>;
|
|
93
97
|
declare function resizeImageForUiTars(imageBase64: string, size: Size): Promise<string>;
|
|
94
98
|
|
|
95
|
-
export { AIActionType, call as callAi, callToGetJSONObject, generatePlaywrightTest, generatePlaywrightTestStream, generateYamlTest, generateYamlTestStream, resizeImageForUiTars, systemPromptToLocateElement, vlmPlanning };
|
|
99
|
+
export { AIActionType, AIArgs, call as callAi, callAiFnWithStringResponse, callToGetJSONObject, generatePlaywrightTest, generatePlaywrightTestStream, generateYamlTest, generateYamlTestStream, resizeImageForUiTars, systemPromptToLocateElement, vlmPlanning };
|
package/dist/lib/ai-model.js
CHANGED
|
@@ -18,8 +18,8 @@
|
|
|
18
18
|
|
|
19
19
|
|
|
20
20
|
|
|
21
|
-
var _chunkKFA65L55js = require('./chunk-KFA65L55.js');
|
|
22
21
|
|
|
22
|
+
var _chunkJS4CT3XVjs = require('./chunk-JS4CT3XV.js');
|
|
23
23
|
|
|
24
24
|
|
|
25
25
|
|
|
@@ -39,4 +39,6 @@ var _chunkKFA65L55js = require('./chunk-KFA65L55.js');
|
|
|
39
39
|
|
|
40
40
|
|
|
41
41
|
|
|
42
|
-
|
|
42
|
+
|
|
43
|
+
|
|
44
|
+
exports.AIActionType = _chunkJS4CT3XVjs.AIActionType; exports.AiAssert = _chunkJS4CT3XVjs.AiAssert; exports.AiExtractElementInfo = _chunkJS4CT3XVjs.AiExtractElementInfo; exports.AiLocateElement = _chunkJS4CT3XVjs.AiLocateElement; exports.AiLocateSection = _chunkJS4CT3XVjs.AiLocateSection; exports.adaptBboxToRect = _chunkJS4CT3XVjs.adaptBboxToRect; exports.callAi = _chunkJS4CT3XVjs.call; exports.callAiFn = _chunkJS4CT3XVjs.callAiFn; exports.callAiFnWithStringResponse = _chunkJS4CT3XVjs.callAiFnWithStringResponse; exports.callToGetJSONObject = _chunkJS4CT3XVjs.callToGetJSONObject; exports.describeUserPage = _chunkJS4CT3XVjs.describeUserPage; exports.elementByPositionWithElementInfo = _chunkJS4CT3XVjs.elementByPositionWithElementInfo; exports.generatePlaywrightTest = _chunkJS4CT3XVjs.generatePlaywrightTest; exports.generatePlaywrightTestStream = _chunkJS4CT3XVjs.generatePlaywrightTestStream; exports.generateYamlTest = _chunkJS4CT3XVjs.generateYamlTest; exports.generateYamlTestStream = _chunkJS4CT3XVjs.generateYamlTestStream; exports.plan = _chunkJS4CT3XVjs.plan; exports.resizeImageForUiTars = _chunkJS4CT3XVjs.resizeImageForUiTars; exports.systemPromptToLocateElement = _chunkJS4CT3XVjs.systemPromptToLocateElement; exports.vlmPlanning = _chunkJS4CT3XVjs.vlmPlanning;
|
|
@@ -56,10 +56,6 @@ var AIActionType = /* @__PURE__ */ ((AIActionType2) => {
|
|
|
56
56
|
return AIActionType2;
|
|
57
57
|
})(AIActionType || {});
|
|
58
58
|
async function callAiFn(msgs, AIActionTypeValue) {
|
|
59
|
-
_utils.assert.call(void 0,
|
|
60
|
-
checkAIConfig(),
|
|
61
|
-
"Cannot find config for AI model service. If you are using a self-hosted model without validating the API key, please set `OPENAI_API_KEY` to any non-null value. https://midscenejs.com/model-provider.html"
|
|
62
|
-
);
|
|
63
59
|
const { content, usage } = await callToGetJSONObject(
|
|
64
60
|
msgs,
|
|
65
61
|
AIActionTypeValue
|
|
@@ -1331,7 +1327,11 @@ Please check your config.`
|
|
|
1331
1327
|
}
|
|
1332
1328
|
throw new Error("Openai SDK or Anthropic SDK is not initialized");
|
|
1333
1329
|
}
|
|
1334
|
-
async function
|
|
1330
|
+
async function call2(messages, AIActionTypeValue, responseFormat, options) {
|
|
1331
|
+
_utils.assert.call(void 0,
|
|
1332
|
+
checkAIConfig(),
|
|
1333
|
+
"Cannot find config for AI model service. If you are using a self-hosted model without validating the API key, please set `OPENAI_API_KEY` to any non-null value. https://midscenejs.com/model-provider.html"
|
|
1334
|
+
);
|
|
1335
1335
|
const { completion, style } = await createChatClient({
|
|
1336
1336
|
AIActionTypeValue
|
|
1337
1337
|
});
|
|
@@ -1576,11 +1576,15 @@ async function callToGetJSONObject(messages, AIActionTypeValue) {
|
|
|
1576
1576
|
if (model === "gpt-4o-2024-05-13") {
|
|
1577
1577
|
responseFormat = { type: "json_object" /* JSON */ };
|
|
1578
1578
|
}
|
|
1579
|
-
const response = await
|
|
1579
|
+
const response = await call2(messages, AIActionTypeValue, responseFormat);
|
|
1580
1580
|
_utils.assert.call(void 0, response, "empty response");
|
|
1581
1581
|
const jsonContent = safeParseJson(response.content);
|
|
1582
1582
|
return { content: jsonContent, usage: response.usage };
|
|
1583
1583
|
}
|
|
1584
|
+
async function callAiFnWithStringResponse(msgs, AIActionTypeValue) {
|
|
1585
|
+
const { content, usage } = await call2(msgs, AIActionTypeValue);
|
|
1586
|
+
return { content, usage };
|
|
1587
|
+
}
|
|
1584
1588
|
function extractJSONFromCodeBlock(response) {
|
|
1585
1589
|
try {
|
|
1586
1590
|
const jsonMatch = response.match(/^\s*(\{[\s\S]*\})\s*$/);
|
|
@@ -1795,7 +1799,7 @@ Respond with YAML only, no explanations.`
|
|
|
1795
1799
|
}))
|
|
1796
1800
|
});
|
|
1797
1801
|
}
|
|
1798
|
-
const response = await
|
|
1802
|
+
const response = await call2(prompt, 2 /* EXTRACT_DATA */);
|
|
1799
1803
|
if (_optionalChain([response, 'optionalAccess', _51 => _51.content]) && typeof response.content === "string") {
|
|
1800
1804
|
return response.content;
|
|
1801
1805
|
}
|
|
@@ -1857,12 +1861,12 @@ Respond with YAML only, no explanations.`
|
|
|
1857
1861
|
});
|
|
1858
1862
|
}
|
|
1859
1863
|
if (options.stream && options.onChunk) {
|
|
1860
|
-
return await
|
|
1864
|
+
return await call2(prompt, 2 /* EXTRACT_DATA */, void 0, {
|
|
1861
1865
|
stream: true,
|
|
1862
1866
|
onChunk: options.onChunk
|
|
1863
1867
|
});
|
|
1864
1868
|
} else {
|
|
1865
|
-
const response = await
|
|
1869
|
+
const response = await call2(prompt, 2 /* EXTRACT_DATA */);
|
|
1866
1870
|
if (_optionalChain([response, 'optionalAccess', _52 => _52.content]) && typeof response.content === "string") {
|
|
1867
1871
|
return {
|
|
1868
1872
|
content: response.content,
|
|
@@ -1925,7 +1929,7 @@ ${_constants.PLAYWRIGHT_EXAMPLE_CODE}`;
|
|
|
1925
1929
|
content: messageContent
|
|
1926
1930
|
}
|
|
1927
1931
|
];
|
|
1928
|
-
const response = await
|
|
1932
|
+
const response = await call2(prompt, 2 /* EXTRACT_DATA */);
|
|
1929
1933
|
if (_optionalChain([response, 'optionalAccess', _53 => _53.content]) && typeof response.content === "string") {
|
|
1930
1934
|
return response.content;
|
|
1931
1935
|
}
|
|
@@ -1980,12 +1984,12 @@ ${_constants.PLAYWRIGHT_EXAMPLE_CODE}`;
|
|
|
1980
1984
|
}
|
|
1981
1985
|
];
|
|
1982
1986
|
if (options.stream && options.onChunk) {
|
|
1983
|
-
return await
|
|
1987
|
+
return await call2(prompt, 2 /* EXTRACT_DATA */, void 0, {
|
|
1984
1988
|
stream: true,
|
|
1985
1989
|
onChunk: options.onChunk
|
|
1986
1990
|
});
|
|
1987
1991
|
} else {
|
|
1988
|
-
const response = await
|
|
1992
|
+
const response = await call2(prompt, 2 /* EXTRACT_DATA */);
|
|
1989
1993
|
if (_optionalChain([response, 'optionalAccess', _54 => _54.content]) && typeof response.content === "string") {
|
|
1990
1994
|
return {
|
|
1991
1995
|
content: response.content,
|
|
@@ -2549,8 +2553,8 @@ async function plan(userInstruction, opts) {
|
|
|
2549
2553
|
]
|
|
2550
2554
|
}
|
|
2551
2555
|
];
|
|
2552
|
-
const
|
|
2553
|
-
const { content, usage } = await
|
|
2556
|
+
const call3 = callAI || callAiFn;
|
|
2557
|
+
const { content, usage } = await call3(msgs, 3 /* PLAN */);
|
|
2554
2558
|
const rawResponse = JSON.stringify(content, void 0, 2);
|
|
2555
2559
|
const planFromAI = content;
|
|
2556
2560
|
const actions = (_optionalChain([planFromAI, 'access', _72 => _72.action, 'optionalAccess', _73 => _73.type]) ? [planFromAI.action] : planFromAI.actions) || [];
|
|
@@ -2658,7 +2662,7 @@ var pointToBbox = (point, width, height) => {
|
|
|
2658
2662
|
async function vlmPlanning(options) {
|
|
2659
2663
|
const { conversationHistory, userInstruction, size } = options;
|
|
2660
2664
|
const systemPrompt = getUiTarsPlanningPrompt() + userInstruction;
|
|
2661
|
-
const res = await
|
|
2665
|
+
const res = await call2(
|
|
2662
2666
|
[
|
|
2663
2667
|
{
|
|
2664
2668
|
role: "user",
|
|
@@ -2906,6 +2910,7 @@ async function resizeImageForUiTars(imageBase64, size) {
|
|
|
2906
2910
|
|
|
2907
2911
|
|
|
2908
2912
|
|
|
2909
|
-
exports.systemPromptToLocateElement = systemPromptToLocateElement; exports.elementByPositionWithElementInfo = elementByPositionWithElementInfo; exports.describeUserPage = describeUserPage; exports.call = call; exports.callToGetJSONObject = callToGetJSONObject; exports.AIActionType = AIActionType; exports.callAiFn = callAiFn; exports.adaptBboxToRect = adaptBboxToRect; exports.expandSearchArea = expandSearchArea; exports.generateYamlTest = generateYamlTest; exports.generateYamlTestStream = generateYamlTestStream; exports.generatePlaywrightTest = generatePlaywrightTest; exports.generatePlaywrightTestStream = generatePlaywrightTestStream; exports.AiLocateElement = AiLocateElement; exports.AiLocateSection = AiLocateSection; exports.AiExtractElementInfo = AiExtractElementInfo; exports.AiAssert = AiAssert; exports.plan = plan; exports.vlmPlanning = vlmPlanning; exports.resizeImageForUiTars = resizeImageForUiTars;
|
|
2910
2913
|
|
|
2911
|
-
|
|
2914
|
+
exports.systemPromptToLocateElement = systemPromptToLocateElement; exports.elementByPositionWithElementInfo = elementByPositionWithElementInfo; exports.describeUserPage = describeUserPage; exports.call = call2; exports.callToGetJSONObject = callToGetJSONObject; exports.callAiFnWithStringResponse = callAiFnWithStringResponse; exports.AIActionType = AIActionType; exports.callAiFn = callAiFn; exports.adaptBboxToRect = adaptBboxToRect; exports.expandSearchArea = expandSearchArea; exports.generateYamlTest = generateYamlTest; exports.generateYamlTestStream = generateYamlTestStream; exports.generatePlaywrightTest = generatePlaywrightTest; exports.generatePlaywrightTestStream = generatePlaywrightTestStream; exports.AiLocateElement = AiLocateElement; exports.AiLocateSection = AiLocateSection; exports.AiExtractElementInfo = AiExtractElementInfo; exports.AiAssert = AiAssert; exports.plan = plan; exports.vlmPlanning = vlmPlanning; exports.resizeImageForUiTars = resizeImageForUiTars;
|
|
2915
|
+
|
|
2916
|
+
//# sourceMappingURL=chunk-JS4CT3XV.js.map
|