@midscene/core 1.10.5 → 1.10.6-beta-20260717061640.0

package/dist/types/ai-model/model-adapter/types.d.ts

@@ -1,6 +1,6 @@
 import type { PixelBbox, PlanningAction } from '../../types';
 import type { AIUsageInfo } from '../../types';
-import type { IModelConfig, TIntent, TModelReasoningEnabled } from '@midscene/shared/env';
+import type { IModelConfig, TIntent, TModelReasoningEnabled, TModelResponseFormat } from '@midscene/shared/env';
 import type OpenAI from 'openai';
 import type { JsonParser, JsonParserContext, JsonParserSource } from '../service-caller/json';
 import type { LocateResultAdapter, LocateResultAdapterDefinition, ResolvedLocateResultCoordinates } from '../shared/model-locate-result/types';
@@ -23,17 +23,28 @@ export interface MidsceneChatCompletionDefaults {
 }
 export interface ChatCompletionCallUserConfig extends ReasoningInput {
     temperature?: number;
+    responseFormat?: TModelResponseFormat;
 }
 export type ChatCompletionUnsupportedUserConfig = keyof ChatCompletionCallUserConfig;
 export interface ChatCompletionCallInput {
     intent?: TIntent;
     userConfig?: ChatCompletionCallUserConfig;
     requiresOriginalImageDetail?: boolean;
+    /**
+     * Whether this call expects a JSON object response.
+     *
+     * This must not be inferred from `intent === 'default'`: intent selects a
+     * model-config slot, while many non-locate calls (for example, browser
+     * extension recording data generation that produces YAML) also use the
+     * default model.
+     */
+    expectedJsonObjectResponse?: boolean;
 }
 export interface ChatCompletionCallContext {
     intent?: TIntent;
     userConfig: ChatCompletionCallUserConfig;
     requiresOriginalImageDetail?: boolean;
+    expectedJsonObjectResponse?: boolean;
     midsceneDefaults: MidsceneChatCompletionDefaults;
 }
 export type ImageDetail = 'auto' | 'low' | 'high' | 'original';

package/dist/types/ai-model/service-caller/index.d.ts

@@ -39,6 +39,7 @@ interface CallAIOptions {
     onChunk?: StreamingCallback;
     abortSignal?: AbortSignal;
     requiresOriginalImageDetail?: boolean;
+    expectedJsonObjectResponse?: boolean;
 }
 export declare function callAI(messages: ChatCompletionMessageParam[], modelRuntime: ModelRuntime, options?: CallAIOptions): Promise<{
     content: string;
@@ -51,6 +52,7 @@ export declare function callAIWithObjectResponse<T>(messages: ChatCompletionMess
     abortSignal?: AbortSignal;
     jsonParserSource?: JsonParserSource;
     retryTimes?: number;
+    retryInterval?: number;
 }): Promise<{
     content: T;
     contentString: string;

package/dist/types/ai-model/service-caller/semantic-retry.d.ts

@@ -7,6 +7,8 @@ export type CallAiAndParseWithRetryOptions<Response, Parsed> = {
     toParseError: (error: unknown, response: Response) => Error;
     /** Number of additional model requests after a parsing failure. Defaults to 0. */
     parseRetryTimes?: number;
+    /** Delay in milliseconds before retrying a parsing failure. Defaults to 0. */
+    parseRetryInterval?: number;
     abortSignal?: AbortSignal;
     /** Runs immediately before a parsing failure triggers another model request. */
     onParseRetry?: (error: unknown, response: Response) => void;
@@ -19,4 +21,4 @@ export type CallAiAndParseWithRetryOptions<Response, Parsed> = {
  * propagated as-is. Only a successful callAi() response that fails the
  * parseResponse() callback triggers another model request.
  */
-export declare function callAiAndParseWithRetry<Response, Parsed>({ callAi, parseResponse, toParseError, parseRetryTimes, abortSignal, onParseRetry, }: CallAiAndParseWithRetryOptions<Response, Parsed>): Promise<Parsed>;
+export declare function callAiAndParseWithRetry<Response, Parsed>({ callAi, parseResponse, toParseError, parseRetryTimes, parseRetryInterval, abortSignal, onParseRetry, }: CallAiAndParseWithRetryOptions<Response, Parsed>): Promise<Parsed>;

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@midscene/core",
   "description": "Automate browser actions, extract data, and perform assertions using AI. It offers JavaScript SDK, Chrome extension, and support for scripting in YAML. See https://midscenejs.com/ for details.",
-  "version": "1.10.5",
+  "version": "1.10.6-beta-20260717061640.0",
   "repository": "https://github.com/web-infra-dev/midscene",
   "homepage": "https://midscenejs.com/",
   "main": "./dist/lib/index.js",
@@ -37,6 +37,11 @@
       "import": "./dist/es/device/index.mjs",
       "require": "./dist/lib/device/index.js"
     },
+    "./dump": {
+      "types": "./dist/types/dump/index.d.ts",
+      "import": "./dist/es/dump/index.mjs",
+      "require": "./dist/lib/dump/index.js"
+    },
     "./dump/task-service-dump": {
       "types": "./dist/types/dump/task-service-dump.d.ts",
       "import": "./dist/es/dump/task-service-dump.mjs",
@@ -80,6 +85,9 @@
       "device": [
         "./dist/types/device.d.ts"
       ],
+      "dump": [
+        "./dist/types/dump/index.d.ts"
+      ],
       "dump/task-service-dump": [
         "./dist/types/dump/task-service-dump.d.ts"
       ],
@@ -100,14 +108,14 @@
     "dotenv": "^16.4.5",
     "fetch-socks": "^1.3.0",
     "js-yaml": "4.1.0",
-    "jsonrepair": "3.14.0",
+    "jsonrepair": "3.15.0",
     "mdast-util-from-markdown": "^2.0.2",
     "mdast-util-to-markdown": "^2.1.2",
     "openai": "6.3.0",
     "semver": "7.5.2",
     "undici": "^6.0.0",
     "zod": "^3.25.1",
-    "@midscene/shared": "1.10.5"
+    "@midscene/shared": "1.10.6-beta-20260717061640.0"
   },
   "devDependencies": {
     "@rslib/core": "^0.18.3",