@xiuchang-midscene/android 2.0.0 → 2.0.2

package/dist/types/mcp-server.d.ts

@@ -1,22 +1,22 @@
-import { AbstractInterface } from '@midscene/core/device';
-import type { ActionParam } from '@midscene/core';
-import type { ActionReturn } from '@midscene/core';
+import { AbstractInterface } from '@xiuchang-midscene/core/device';
+import type { ActionParam } from '@xiuchang-midscene/core';
+import type { ActionReturn } from '@xiuchang-midscene/core';
 import { ADB } from 'appium-adb';
-import { Agent } from '@midscene/core/agent';
-import { AgentOpt } from '@midscene/core/agent';
-import { AndroidDeviceInputOpt } from '@midscene/core/device';
-import { AndroidDeviceOpt } from '@midscene/core/device';
-import { BaseMCPServer } from '@midscene/shared/mcp';
-import { BaseMidsceneTools } from '@midscene/shared/mcp';
-import { DeviceAction } from '@midscene/core';
-import type { ElementInfo } from '@midscene/shared/extractor';
-import { InterfaceType } from '@midscene/core';
-import { LaunchMCPServerOptions } from '@midscene/shared/mcp';
-import { LaunchMCPServerResult } from '@midscene/shared/mcp';
-import { Point } from '@midscene/core';
-import { Size } from '@midscene/core';
-import { Tool } from '@midscene/shared/mcp';
-import { ToolDefinition } from '@midscene/shared/mcp';
+import { Agent } from '@xiuchang-midscene/core/agent';
+import { AgentOpt } from '@xiuchang-midscene/core/agent';
+import { AndroidDeviceInputOpt } from '@xiuchang-midscene/core/device';
+import { AndroidDeviceOpt } from '@xiuchang-midscene/core/device';
+import { BaseMCPServer } from '@xiuchang-midscene/shared/mcp';
+import { BaseMidsceneTools } from '@xiuchang-midscene/shared/mcp';
+import { DeviceAction } from '@xiuchang-midscene/core';
+import type { ElementInfo } from '@xiuchang-midscene/shared/extractor';
+import { InterfaceType } from '@xiuchang-midscene/core';
+import { LaunchMCPServerOptions } from '@xiuchang-midscene/shared/mcp';
+import { LaunchMCPServerResult } from '@xiuchang-midscene/shared/mcp';
+import { Point } from '@xiuchang-midscene/core';
+import { Size } from '@xiuchang-midscene/core';
+import { Tool } from '@xiuchang-midscene/shared/mcp';
+import { ToolDefinition } from '@xiuchang-midscene/shared/mcp';
 
 declare type ActionArgs<T extends DeviceAction> = [ActionParam<T>] extends [undefined] ? [] : [ActionParam<T>];
 
@@ -37,6 +37,10 @@ declare class AndroidAgent extends Agent<AndroidDevice> {
      * User-provided app name to package name mapping
      */
     private appNameMapping;
+    /**
+     * Cache the last loaded knowledge package name to avoid redundant setAIActContext calls
+     */
+    private lastKnowledgePackageName;
     constructor(device: AndroidDevice, opts?: AndroidAgentOpt);
     /**
      * Launch an Android app or URL
@@ -48,6 +52,29 @@ declare class AndroidAgent extends Agent<AndroidDevice> {
      * @param command - ADB shell command to execute
      */
     runAdbShell(command: string): Promise<string>;
+    /**
+     * Load app-specific business knowledge for the given package name.
+     * If knowledge is found, it will be injected via setAIActContext so that
+     * all AI methods (aiAct, aiAssert, aiQuery, aiBoolean, etc.) can use it.
+     * Also sets up screenshot knowledge provider for planning-time injection.
+     * Skips if the same package knowledge is already loaded.
+     * @param packageName - The Android package name (e.g. "com.taobao.trip")
+     */
+    loadAppKnowledge(packageName: string): void;
+    /**
+     * Resolve annotated reference screenshots for a given page to base64 data URLs.
+     * @param pageId - The page identifier
+     * @param packageName - The Android package name
+     * @returns Array of base64 image data objects, empty if no screenshots found
+     */
+    private resolveReferenceScreenshots;
+    /**
+     * Detect the foreground Android app and automatically load its business knowledge.
+     * Uses ADB to get the currently resumed activity and extract the package name,
+     * then calls loadAppKnowledge() with the detected package name.
+     * This is safe to call repeatedly — loadAppKnowledge() skips if already loaded.
+     */
+    detectAndLoadAppKnowledge(): Promise<void>;
     private createActionWrapper;
 }
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@xiuchang-midscene/android",
-  "version": "2.0.0",
+  "version": "2.0.2",
   "description": "Android automation library for Midscene",
   "keywords": [
     "Android UI automation",
@@ -18,6 +18,7 @@
   "files": [
     "bin",
     "dist",
+    "src/app-knowledge/screenshots",
     "README.md"
   ],
   "exports": {
@@ -41,8 +42,8 @@
     "@yume-chan/stream-extra": "2.1.0",
     "appium-adb": "12.12.1",
     "sharp": "^0.34.3",
-    "@midscene/core": "1.6.0",
-    "@midscene/shared": "1.6.0"
+    "@xiuchang-midscene/shared": "^2.0.2",
+    "@xiuchang-midscene/core": "^2.0.2"
   },
   "optionalDependencies": {
     "@ffmpeg-installer/ffmpeg": "^1.1.0"
@@ -55,8 +56,7 @@
     "typescript": "^5.8.3",
     "tsx": "^4.19.2",
     "vitest": "3.0.5",
-    "zod": "3.24.3",
-    "@midscene/playground": "1.6.0"
+    "zod": "3.24.3"
   },
   "license": "MIT",
   "scripts": {