@donggui/core 1.5.14 → 1.6.1

package/dist/es/agent/utils.mjs

@@ -152,7 +152,7 @@ async function matchElementFromCache(context, cacheEntry, cachePrompt, cacheable
         return;
     }
 }
-const getMidsceneVersion = ()=>"1.5.14";
+const getMidsceneVersion = ()=>"1.6.0";
 const parsePrompt = (prompt)=>{
     if ('string' == typeof prompt) return {
         textPrompt: prompt,

package/dist/es/ai-model/assert.mjs

@@ -0,0 +1,415 @@
+import { getDebug } from "@midscene/shared/logger";
+import { callAIWithObjectResponse } from "./service-caller/index.mjs";
+const debug = getDebug('ai:assert');
+const DEFAULT_SYSTEM_CHECK_PROMPT = `请检查以下系统级问题：
+
+1. **白屏检测**：页面是否完全白屏或大面积空白
+2. **布局遮挡**：重要元素是否存在组件间重叠遮挡
+3. **未加载内容**：页面是否有加载中的占位符、骨架屏
+4. **错误提示**：页面是否显示错误信息、异常提示
+5. **后端错误**：是否有后端请求失败的提示，如"活动太火爆了"、报错 Toast、错误码 等
+
+如果发现以上问题，请在 systemCheckResults 中标注对应字段为 true。`;
+function buildAssertionPrompt(options) {
+    const { assertion, businessContext, systemCheckPrompt } = options;
+    let prompt = `## 用户断言描述
+${assertion}`;
+    if (businessContext) prompt += `
+
+## 业务知识上下文
+${businessContext}`;
+    if (systemCheckPrompt) prompt += `
+
+## 系统校验规则
+${systemCheckPrompt}`;
+    prompt += `
+
+## 输出要求
+请以 JSON 格式输出断言结果：
+{
+  "pass": boolean,    // 断言是否通过
+  "thought": string,  // 思考过程
+  "reason": string,   // 失败原因（如果失败）
+  "systemCheckResults": {  // 系统校验结果（可选）
+    "whiteScreen": boolean,
+    "layoutBlocked": boolean,
+    "loadingContent": boolean,
+    "errorPrompt": boolean,
+    "backendError": boolean
+  }
+}`;
+    return prompt;
+}
+async function AiAssertElement(options) {
+    const { beforeScreenshot, afterScreenshot, assertion, businessContext, enableSystemCheck = true, customSystemCheckRules, modelConfig, abortSignal } = options;
+    const systemPrompt = `你是一个自动化测试断言专家。请根据以下信息判断断言是否通过。
+
+你需要：
+1. 分析操作前后的截图变化（如果提供了操作前截图）
+2. 验证用户的断言描述是否成立
+3. 检查是否存在系统级问题（如果启用）
+4. 给出详细的思考过程和判断结果`;
+    const systemCheckPrompt = enableSystemCheck ? customSystemCheckRules || DEFAULT_SYSTEM_CHECK_PROMPT : '';
+    const userContent = [];
+    if (beforeScreenshot) {
+        userContent.push({
+            type: 'text',
+            text: '## 操作前截图（执行操作前的页面状态）'
+        });
+        userContent.push({
+            type: 'image_url',
+            image_url: {
+                url: beforeScreenshot,
+                detail: 'high'
+            }
+        });
+        userContent.push({
+            type: 'text',
+            text: '## 操作后截图（执行操作后的页面状态）'
+        });
+    } else userContent.push({
+        type: 'text',
+        text: '## 当前页面截图'
+    });
+    userContent.push({
+        type: 'image_url',
+        image_url: {
+            url: afterScreenshot,
+            detail: 'high'
+        }
+    });
+    userContent.push({
+        type: 'text',
+        text: buildAssertionPrompt({
+            assertion,
+            businessContext,
+            systemCheckPrompt
+        })
+    });
+    const msgs = [
+        {
+            role: 'system',
+            content: systemPrompt
+        },
+        {
+            role: 'user',
+            content: userContent
+        }
+    ];
+    debug('calling AI for assertion:', assertion);
+    try {
+        const result = await callAIWithObjectResponse(msgs, modelConfig, {
+            abortSignal
+        });
+        debug('assertion result:', result.content);
+        return {
+            pass: result.content.pass,
+            thought: result.content.thought || '',
+            reason: result.content.reason,
+            usage: result.usage,
+            systemCheckResults: result.content.systemCheckResults,
+            rawResponse: JSON.stringify(result.content)
+        };
+    } catch (error) {
+        debug('assertion error:', error);
+        throw error;
+    }
+}
+const DIFF_SYSTEM_PROMPT = `你是一个自动化测试的图像对比专家。你的任务是对比当前页面截图与基准图片，判断页面样式是否符合预期。
+
+你需要：
+1. 仔细对比两张图片的布局、颜色、元素位置
+2. 分析差异是否在可接受范围内
+3. 识别动态内容（如时间、日期）导致的差异
+4. 给出详细的对比分析和判断结果`;
+function buildDiffPrompt(options) {
+    const { assertion, businessContext, diffThreshold = 0.1, ignoreRegions, ignoreDynamicContent, strictMode } = options;
+    let prompt = `## 图像对比任务
+
+请对比基准图片（预期样式）和当前截图（实际样式），判断页面是否符合预期。
+
+## 用户断言描述
+${assertion}`;
+    if (businessContext) prompt += `
+
+## 业务知识上下文
+${businessContext}`;
+    prompt += `
+
+## 对比要求
+
+请从以下维度进行对比分析：
+
+1. **布局一致性**：页面布局是否与基准图片一致
+2. **颜色一致性**：主要颜色是否与基准图片一致
+3. **元素位置**：关键元素的位置是否与基准图片一致
+4. **文字内容**：文字内容是否与基准图片一致
+5. **图片资源**：图片、图标等资源是否正确加载
+
+## 差异阈值
+允许的差异阈值：${100 * diffThreshold}%`;
+    if (ignoreRegions && ignoreRegions.length > 0) {
+        prompt += `
+
+## 忽略对比的区域
+以下区域不参与对比：`;
+        ignoreRegions.forEach((region, index)=>{
+            prompt += `
+${index + 1}. 位置 (${region.x}, ${region.y})，尺寸 ${region.width}x${region.height}`;
+        });
+    }
+    if (ignoreDynamicContent) prompt += `
+
+## 动态内容处理
+请忽略以下动态内容导致的差异：
+- 时间显示
+- 日期显示
+- 随机验证码
+- 动画效果`;
+    if (strictMode) prompt += `
+
+## 严格模式
+当前为严格模式，任何差异都视为失败。`;
+    prompt += `
+
+## 输出要求
+请以 JSON 格式输出对比结果：
+{
+  "pass": boolean,           // 断言是否通过
+  "thought": string,         // 对比分析过程
+  "reason": string,          // 失败原因（如果失败）
+  "diffDetails": {           // 差异详情
+    "layoutMatch": boolean,      // 布局是否匹配
+    "colorMatch": boolean,       // 颜色是否匹配
+    "elementPositionMatch": boolean, // 元素位置是否匹配
+    "textContentMatch": boolean, // 文字内容是否匹配
+    "resourceMatch": boolean,    // 资源是否匹配
+    "acceptableDifferences": string[], // 可接受的差异列表
+    "unacceptableDifferences": string[] // 不可接受的差异列表
+  }
+}`;
+    return prompt;
+}
+async function AiAssertDiff(options) {
+    const { currentScreenshot, referenceImages, assertion, businessContext, diffThreshold, ignoreRegions, ignoreDynamicContent, strictMode, modelConfig, abortSignal } = options;
+    const userContent = [];
+    for (const refImage of referenceImages){
+        userContent.push({
+            type: 'text',
+            text: `## 基准图片：${refImage.name}`
+        });
+        userContent.push({
+            type: 'image_url',
+            image_url: {
+                url: refImage.url,
+                detail: 'high'
+            }
+        });
+    }
+    userContent.push({
+        type: 'text',
+        text: '## 当前截图（实际样式）'
+    });
+    userContent.push({
+        type: 'image_url',
+        image_url: {
+            url: currentScreenshot,
+            detail: 'high'
+        }
+    });
+    userContent.push({
+        type: 'text',
+        text: buildDiffPrompt({
+            assertion,
+            businessContext,
+            diffThreshold,
+            ignoreRegions,
+            ignoreDynamicContent,
+            strictMode
+        })
+    });
+    const msgs = [
+        {
+            role: 'system',
+            content: DIFF_SYSTEM_PROMPT
+        },
+        {
+            role: 'user',
+            content: userContent
+        }
+    ];
+    debug('calling AI for diff assertion:', assertion);
+    try {
+        const result = await callAIWithObjectResponse(msgs, modelConfig, {
+            abortSignal
+        });
+        debug('diff assertion result:', result.content);
+        return {
+            pass: result.content.pass,
+            thought: result.content.thought || '',
+            reason: result.content.reason,
+            diffDetails: result.content.diffDetails,
+            usage: result.usage,
+            rawResponse: JSON.stringify(result.content)
+        };
+    } catch (error) {
+        debug('diff assertion error:', error);
+        throw error;
+    }
+}
+const VIDEO_SYSTEM_PROMPT = `你是一个自动化测试的视频/动画分析专家。你的任务是对比当前录制的视频与基准视频，判断动画效果是否符合预期。
+
+你需要：
+1. 分析视频中的动画流畅度、时长、完整性
+2. 对比基准视频与当前视频的差异
+3. 识别动画类型（过渡动画、加载动画、交互动画等）
+4. 检测动画质量问题（卡顿、跳帧等）
+5. 给出详细的分析过程和判断结果`;
+function buildVideoPrompt(options) {
+    const { assertion, businessContext, videoOptions } = options;
+    let prompt = `## 视频/动画对比任务
+
+请对比基准视频（预期动画效果）和当前录制的视频（实际动画效果），判断动画是否符合预期。
+
+## 用户断言描述
+${assertion}`;
+    if (businessContext) prompt += `
+
+## 业务知识上下文
+${businessContext}`;
+    prompt += `
+
+## 分析要求
+
+请从以下维度进行动画分析：
+
+1. **动画流畅度**：动画是否流畅，有无卡顿、跳帧
+2. **动画时长**：动画时长是否在预期范围内
+3. **动画完整性**：动画是否完整执行，有无中断
+4. **动画类型**：识别动画类型（过渡、加载、交互等）
+5. **关键帧匹配**：关键时间点的画面是否符合预期`;
+    if (videoOptions) {
+        if (videoOptions.checkSmoothness) prompt += `
+
+## 流畅度检测
+流畅度阈值：${videoOptions.smoothnessThreshold || 60}/100
+请评估动画流畅度并给出评分。`;
+        if (videoOptions.checkDuration && videoOptions.expectedDuration) prompt += `
+
+## 时长检测
+预期时长范围：${videoOptions.expectedDuration.min || 0}秒 - ${videoOptions.expectedDuration.max || '无限制'}秒`;
+        if (videoOptions.keyframes && videoOptions.keyframes.timestamps.length > 0) {
+            prompt += `
+
+## 关键帧验证
+需要在以下时间点验证画面：`;
+            videoOptions.keyframes.timestamps.forEach((ts, index)=>{
+                const desc = videoOptions.keyframes?.descriptions?.[index] || '未描述';
+                prompt += `
+- ${ts}秒：${desc}`;
+            });
+        }
+        if (videoOptions.animationType && 'auto' !== videoOptions.animationType) prompt += `
+
+## 动画类型
+预期动画类型：${videoOptions.animationType}`;
+    }
+    prompt += `
+
+## 输出要求
+请以 JSON 格式输出对比结果：
+{
+  "pass": boolean,           // 断言是否通过
+  "thought": string,         // 分析过程
+  "reason": string,          // 失败原因（如果失败）
+  "videoDetails": {          // 视频详情
+    "smoothnessScore": number,    // 流畅度评分（0-100）
+    "duration": number,           // 动画时长（秒）
+    "isComplete": boolean,        // 动画是否完整
+    "keyframeMatches": [          // 关键帧匹配结果
+      {
+        "timestamp": number,
+        "matched": boolean,
+        "description": string
+      }
+    ],
+    "detectedAnimationType": string,  // 检测到的动画类型
+    "qualityAssessment": {            // 质量评估
+      "hasStuttering": boolean,       // 是否有卡顿
+      "hasFrameDropping": boolean,    // 是否有跳帧
+      "averageFrameInterval": number, // 平均帧间隔（ms）
+      "frameIntervalStdDev": number   // 帧间隔标准差
+    },
+    "acceptableDifferences": string[], // 可接受的差异列表
+    "unacceptableDifferences": string[] // 不可接受的差异列表
+  }
+}`;
+    return prompt;
+}
+async function AiAssertVideo(options) {
+    const { currentVideoFrames, assertion, businessContext, videoOptions, modelConfig, abortSignal } = options;
+    const MAX_DURATION = 5;
+    const DEFAULT_FPS = 30;
+    const MAX_FRAMES = MAX_DURATION * DEFAULT_FPS;
+    if (currentVideoFrames.length > MAX_FRAMES) throw new Error(`Video frames exceed maximum limit. Maximum allowed: ${MAX_FRAMES} frames (${MAX_DURATION}s at ${DEFAULT_FPS}fps), got: ${currentVideoFrames.length} frames`);
+    const userContent = [];
+    userContent.push({
+        type: 'text',
+        text: `## 当前视频（共 ${currentVideoFrames.length} 帧，最大限制 ${MAX_FRAMES} 帧）
+
+以下为当前录制视频的关键帧截图：`
+    });
+    const frameInterval = Math.max(1, Math.floor(currentVideoFrames.length / 10));
+    for(let i = 0; i < currentVideoFrames.length; i += frameInterval){
+        userContent.push({
+            type: 'text',
+            text: `帧 ${i + 1}/${currentVideoFrames.length}`
+        });
+        userContent.push({
+            type: 'image_url',
+            image_url: {
+                url: currentVideoFrames[i],
+                detail: 'low'
+            }
+        });
+    }
+    userContent.push({
+        type: 'text',
+        text: buildVideoPrompt({
+            assertion,
+            businessContext,
+            videoOptions
+        })
+    });
+    const msgs = [
+        {
+            role: 'system',
+            content: VIDEO_SYSTEM_PROMPT
+        },
+        {
+            role: 'user',
+            content: userContent
+        }
+    ];
+    debug('calling AI for video assertion:', assertion);
+    try {
+        const result = await callAIWithObjectResponse(msgs, modelConfig, {
+            abortSignal
+        });
+        debug('video assertion result:', result.content);
+        return {
+            pass: result.content.pass,
+            thought: result.content.thought || '',
+            reason: result.content.reason,
+            videoDetails: result.content.videoDetails,
+            usage: result.usage,
+            rawResponse: JSON.stringify(result.content)
+        };
+    } catch (error) {
+        debug('video assertion error:', error);
+        throw error;
+    }
+}
+export { AiAssertDiff, AiAssertElement, AiAssertVideo };
+
+//# sourceMappingURL=assert.mjs.map

package/dist/es/ai-model/assert.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"ai-model/assert.mjs","sources":["../../../src/ai-model/assert.ts"],"sourcesContent":["import type {\n  AIUsageInfo,\n  DiffDetails,\n  IgnoreRegion,\n  ReferenceImage,\n  SystemCheckResults,\n  VideoAssertionOptions,\n  VideoDetails,\n} from '@/types';\nimport type { IModelConfig } from '@midscene/shared/env';\nimport { getDebug } from '@midscene/shared/logger';\nimport type { ChatCompletionUserMessageParam } from 'openai/resources/index';\nimport { callAIWithObjectResponse } from './service-caller/index';\n\nconst debug = getDebug('ai:assert');\n\nexport interface AiAssertOptions {\n  beforeScreenshot?: string;\n  afterScreenshot: string;\n  assertion: string;\n  businessContext?: string;\n  enableSystemCheck?: boolean;\n  customSystemCheckRules?: string;\n  modelConfig: IModelConfig;\n  abortSignal?: AbortSignal;\n  referenceImages?: ReferenceImage[];\n  diffThreshold?: number;\n  ignoreRegions?: IgnoreRegion[];\n  ignoreDynamicContent?: boolean;\n  strictMode?: boolean;\n}\n\nexport interface AIAssertionResponse {\n  pass: boolean;\n  thought: string;\n  reason?: string;\n  systemCheckResults?: SystemCheckResults;\n  diffDetails?: DiffDetails;\n}\n\nconst DEFAULT_SYSTEM_CHECK_PROMPT = `请检查以下系统级问题：\n\n1. **白屏检测**：页面是否完全白屏或大面积空白\n2. **布局遮挡**：重要元素是否存在组件间重叠遮挡\n3. **未加载内容**：页面是否有加载中的占位符、骨架屏\n4. **错误提示**：页面是否显示错误信息、异常提示\n5. **后端错误**：是否有后端请求失败的提示，如\"活动太火爆了\"、报错 Toast、错误码 等\n\n如果发现以上问题，请在 systemCheckResults 中标注对应字段为 true。`;\n\nfunction buildAssertionPrompt(options: {\n  assertion: string;\n  businessContext?: string;\n  systemCheckPrompt?: string;\n}): string {\n  const { assertion, businessContext, systemCheckPrompt } = options;\n\n  let prompt = `## 用户断言描述\n${assertion}`;\n\n  if (businessContext) {\n    prompt += `\n\n## 业务知识上下文\n${businessContext}`;\n  }\n\n  if (systemCheckPrompt) {\n    prompt += `\n\n## 系统校验规则\n${systemCheckPrompt}`;\n  }\n\n  prompt += `\n\n## 输出要求\n请以 JSON 格式输出断言结果：\n{\n  \"pass\": boolean,    // 断言是否通过\n  \"thought\": string,  // 思考过程\n  \"reason\": string,   // 失败原因（如果失败）\n  \"systemCheckResults\": {  // 系统校验结果（可选）\n    \"whiteScreen\": boolean,\n    \"layoutBlocked\": boolean,\n    \"loadingContent\": boolean,\n    \"errorPrompt\": boolean,\n    \"backendError\": boolean\n  }\n}`;\n\n  return prompt;\n}\n\nexport async function AiAssertElement(options: AiAssertOptions): Promise<{\n  pass: boolean;\n  thought: string;\n  reason?: string;\n  usage?: AIUsageInfo;\n  systemCheckResults?: SystemCheckResults;\n  rawResponse?: string;\n}> {\n  const {\n    beforeScreenshot,\n    afterScreenshot,\n    assertion,\n    businessContext,\n    enableSystemCheck = true,\n    customSystemCheckRules,\n    modelConfig,\n    abortSignal,\n  } = options;\n\n  const systemPrompt = `你是一个自动化测试断言专家。请根据以下信息判断断言是否通过。\n\n你需要：\n1. 分析操作前后的截图变化（如果提供了操作前截图）\n2. 验证用户的断言描述是否成立\n3. 检查是否存在系统级问题（如果启用）\n4. 给出详细的思考过程和判断结果`;\n\n  const systemCheckPrompt = enableSystemCheck\n    ? customSystemCheckRules || DEFAULT_SYSTEM_CHECK_PROMPT\n    : '';\n\n  const userContent: ChatCompletionUserMessageParam['content'] = [];\n\n  if (beforeScreenshot) {\n    userContent.push({\n      type: 'text',\n      text: '## 操作前截图（执行操作前的页面状态）',\n    });\n    userContent.push({\n      type: 'image_url',\n      image_url: { url: beforeScreenshot, detail: 'high' },\n    });\n    userContent.push({\n      type: 'text',\n      text: '## 操作后截图（执行操作后的页面状态）',\n    });\n  } else {\n    userContent.push({\n      type: 'text',\n      text: '## 当前页面截图',\n    });\n  }\n\n  userContent.push({\n    type: 'image_url',\n    image_url: { url: afterScreenshot, detail: 'high' },\n  });\n\n  userContent.push({\n    type: 'text',\n    text: buildAssertionPrompt({\n      assertion,\n      businessContext,\n      systemCheckPrompt,\n    }),\n  });\n\n  const msgs = [\n    { role: 'system' as const, content: systemPrompt },\n    { role: 'user' as const, content: userContent },\n  ];\n\n  debug('calling AI for assertion:', assertion);\n\n  try {\n    const result = await callAIWithObjectResponse<AIAssertionResponse>(\n      msgs,\n      modelConfig,\n      { abortSignal },\n    );\n\n    debug('assertion result:', result.content);\n\n    return {\n      pass: result.content.pass,\n      thought: result.content.thought || '',\n      reason: result.content.reason,\n      usage: result.usage,\n      systemCheckResults: result.content.systemCheckResults,\n      rawResponse: JSON.stringify(result.content),\n    };\n  } catch (error) {\n    debug('assertion error:', error);\n    throw error;\n  }\n}\n\nconst DIFF_SYSTEM_PROMPT = `你是一个自动化测试的图像对比专家。你的任务是对比当前页面截图与基准图片，判断页面样式是否符合预期。\n\n你需要：\n1. 仔细对比两张图片的布局、颜色、元素位置\n2. 分析差异是否在可接受范围内\n3. 识别动态内容（如时间、日期）导致的差异\n4. 给出详细的对比分析和判断结果`;\n\nfunction buildDiffPrompt(options: {\n  assertion: string;\n  businessContext?: string;\n  diffThreshold?: number;\n  ignoreRegions?: IgnoreRegion[];\n  ignoreDynamicContent?: boolean;\n  strictMode?: boolean;\n}): string {\n  const {\n    assertion,\n    businessContext,\n    diffThreshold = 0.1,\n    ignoreRegions,\n    ignoreDynamicContent,\n    strictMode,\n  } = options;\n\n  let prompt = `## 图像对比任务\n\n请对比基准图片（预期样式）和当前截图（实际样式），判断页面是否符合预期。\n\n## 用户断言描述\n${assertion}`;\n\n  if (businessContext) {\n    prompt += `\n\n## 业务知识上下文\n${businessContext}`;\n  }\n\n  prompt += `\n\n## 对比要求\n\n请从以下维度进行对比分析：\n\n1. **布局一致性**：页面布局是否与基准图片一致\n2. **颜色一致性**：主要颜色是否与基准图片一致\n3. **元素位置**：关键元素的位置是否与基准图片一致\n4. **文字内容**：文字内容是否与基准图片一致\n5. **图片资源**：图片、图标等资源是否正确加载\n\n## 差异阈值\n允许的差异阈值：${diffThreshold * 100}%`;\n\n  if (ignoreRegions && ignoreRegions.length > 0) {\n    prompt += `\n\n## 忽略对比的区域\n以下区域不参与对比：`;\n    ignoreRegions.forEach((region, index) => {\n      prompt += `\n${index + 1}. 位置 (${region.x}, ${region.y})，尺寸 ${region.width}x${region.height}`;\n    });\n  }\n\n  if (ignoreDynamicContent) {\n    prompt += `\n\n## 动态内容处理\n请忽略以下动态内容导致的差异：\n- 时间显示\n- 日期显示\n- 随机验证码\n- 动画效果`;\n  }\n\n  if (strictMode) {\n    prompt += `\n\n## 严格模式\n当前为严格模式，任何差异都视为失败。`;\n  }\n\n  prompt += `\n\n## 输出要求\n请以 JSON 格式输出对比结果：\n{\n  \"pass\": boolean,           // 断言是否通过\n  \"thought\": string,         // 对比分析过程\n  \"reason\": string,          // 失败原因（如果失败）\n  \"diffDetails\": {           // 差异详情\n    \"layoutMatch\": boolean,      // 布局是否匹配\n    \"colorMatch\": boolean,       // 颜色是否匹配\n    \"elementPositionMatch\": boolean, // 元素位置是否匹配\n    \"textContentMatch\": boolean, // 文字内容是否匹配\n    \"resourceMatch\": boolean,    // 资源是否匹配\n    \"acceptableDifferences\": string[], // 可接受的差异列表\n    \"unacceptableDifferences\": string[] // 不可接受的差异列表\n  }\n}`;\n\n  return prompt;\n}\n\nexport async function AiAssertDiff(options: {\n  currentScreenshot: string;\n  referenceImages: ReferenceImage[];\n  assertion: string;\n  businessContext?: string;\n  diffThreshold?: number;\n  ignoreRegions?: IgnoreRegion[];\n  ignoreDynamicContent?: boolean;\n  strictMode?: boolean;\n  modelConfig: IModelConfig;\n  abortSignal?: AbortSignal;\n}): Promise<{\n  pass: boolean;\n  thought: string;\n  reason?: string;\n  diffDetails?: DiffDetails;\n  usage?: AIUsageInfo;\n  rawResponse?: string;\n}> {\n  const {\n    currentScreenshot,\n    referenceImages,\n    assertion,\n    businessContext,\n    diffThreshold,\n    ignoreRegions,\n    ignoreDynamicContent,\n    strictMode,\n    modelConfig,\n    abortSignal,\n  } = options;\n\n  const userContent: ChatCompletionUserMessageParam['content'] = [];\n\n  // 添加基准图片\n  for (const refImage of referenceImages) {\n    userContent.push({\n      type: 'text',\n      text: `## 基准图片：${refImage.name}`,\n    });\n    userContent.push({\n      type: 'image_url',\n      image_url: { url: refImage.url, detail: 'high' },\n    });\n  }\n\n  // 添加当前截图\n  userContent.push({\n    type: 'text',\n    text: '## 当前截图（实际样式）',\n  });\n  userContent.push({\n    type: 'image_url',\n    image_url: { url: currentScreenshot, detail: 'high' },\n  });\n\n  // 添加对比提示\n  userContent.push({\n    type: 'text',\n    text: buildDiffPrompt({\n      assertion,\n      businessContext,\n      diffThreshold,\n      ignoreRegions,\n      ignoreDynamicContent,\n      strictMode,\n    }),\n  });\n\n  const msgs = [\n    { role: 'system' as const, content: DIFF_SYSTEM_PROMPT },\n    { role: 'user' as const, content: userContent },\n  ];\n\n  debug('calling AI for diff assertion:', assertion);\n\n  try {\n    const result = await callAIWithObjectResponse<AIAssertionResponse>(\n      msgs,\n      modelConfig,\n      { abortSignal },\n    );\n\n    debug('diff assertion result:', result.content);\n\n    return {\n      pass: result.content.pass,\n      thought: result.content.thought || '',\n      reason: result.content.reason,\n      diffDetails: result.content.diffDetails,\n      usage: result.usage,\n      rawResponse: JSON.stringify(result.content),\n    };\n  } catch (error) {\n    debug('diff assertion error:', error);\n    throw error;\n  }\n}\n\nconst VIDEO_SYSTEM_PROMPT = `你是一个自动化测试的视频/动画分析专家。你的任务是对比当前录制的视频与基准视频，判断动画效果是否符合预期。\n\n你需要：\n1. 分析视频中的动画流畅度、时长、完整性\n2. 对比基准视频与当前视频的差异\n3. 识别动画类型（过渡动画、加载动画、交互动画等）\n4. 检测动画质量问题（卡顿、跳帧等）\n5. 给出详细的分析过程和判断结果`;\n\nfunction buildVideoPrompt(options: {\n  assertion: string;\n  businessContext?: string;\n  videoOptions?: VideoAssertionOptions;\n}): string {\n  const { assertion, businessContext, videoOptions } = options;\n\n  let prompt = `## 视频/动画对比任务\n\n请对比基准视频（预期动画效果）和当前录制的视频（实际动画效果），判断动画是否符合预期。\n\n## 用户断言描述\n${assertion}`;\n\n  if (businessContext) {\n    prompt += `\n\n## 业务知识上下文\n${businessContext}`;\n  }\n\n  prompt += `\n\n## 分析要求\n\n请从以下维度进行动画分析：\n\n1. **动画流畅度**：动画是否流畅，有无卡顿、跳帧\n2. **动画时长**：动画时长是否在预期范围内\n3. **动画完整性**：动画是否完整执行，有无中断\n4. **动画类型**：识别动画类型（过渡、加载、交互等）\n5. **关键帧匹配**：关键时间点的画面是否符合预期`;\n\n  if (videoOptions) {\n    if (videoOptions.checkSmoothness) {\n      prompt += `\n\n## 流畅度检测\n流畅度阈值：${videoOptions.smoothnessThreshold || 60}/100\n请评估动画流畅度并给出评分。`;\n    }\n\n    if (videoOptions.checkDuration && videoOptions.expectedDuration) {\n      prompt += `\n\n## 时长检测\n预期时长范围：${videoOptions.expectedDuration.min || 0}秒 - ${videoOptions.expectedDuration.max || '无限制'}秒`;\n    }\n\n    if (\n      videoOptions.keyframes &&\n      videoOptions.keyframes.timestamps.length > 0\n    ) {\n      prompt += `\n\n## 关键帧验证\n需要在以下时间点验证画面：`;\n      videoOptions.keyframes.timestamps.forEach((ts, index) => {\n        const desc = videoOptions.keyframes?.descriptions?.[index] || '未描述';\n        prompt += `\n- ${ts}秒：${desc}`;\n      });\n    }\n\n    if (videoOptions.animationType && videoOptions.animationType !== 'auto') {\n      prompt += `\n\n## 动画类型\n预期动画类型：${videoOptions.animationType}`;\n    }\n  }\n\n  prompt += `\n\n## 输出要求\n请以 JSON 格式输出对比结果：\n{\n  \"pass\": boolean,           // 断言是否通过\n  \"thought\": string,         // 分析过程\n  \"reason\": string,          // 失败原因（如果失败）\n  \"videoDetails\": {          // 视频详情\n    \"smoothnessScore\": number,    // 流畅度评分（0-100）\n    \"duration\": number,           // 动画时长（秒）\n    \"isComplete\": boolean,        // 动画是否完整\n    \"keyframeMatches\": [          // 关键帧匹配结果\n      {\n        \"timestamp\": number,\n        \"matched\": boolean,\n        \"description\": string\n      }\n    ],\n    \"detectedAnimationType\": string,  // 检测到的动画类型\n    \"qualityAssessment\": {            // 质量评估\n      \"hasStuttering\": boolean,       // 是否有卡顿\n      \"hasFrameDropping\": boolean,    // 是否有跳帧\n      \"averageFrameInterval\": number, // 平均帧间隔（ms）\n      \"frameIntervalStdDev\": number   // 帧间隔标准差\n    },\n    \"acceptableDifferences\": string[], // 可接受的差异列表\n    \"unacceptableDifferences\": string[] // 不可接受的差异列表\n  }\n}`;\n\n  return prompt;\n}\n\nexport async function AiAssertVideo(options: {\n  currentVideoFrames: string[];\n  assertion: string;\n  businessContext?: string;\n  videoOptions?: VideoAssertionOptions;\n  modelConfig: IModelConfig;\n  abortSignal?: AbortSignal;\n}): Promise<{\n  pass: boolean;\n  thought: string;\n  reason?: string;\n  videoDetails?: VideoDetails;\n  usage?: AIUsageInfo;\n  rawResponse?: string;\n}> {\n  const {\n    currentVideoFrames,\n    assertion,\n    businessContext,\n    videoOptions,\n    modelConfig,\n    abortSignal,\n  } = options;\n\n  const MAX_DURATION = 5;\n  const DEFAULT_FPS = 30;\n  const MAX_FRAMES = MAX_DURATION * DEFAULT_FPS;\n\n  if (currentVideoFrames.length > MAX_FRAMES) {\n    throw new Error(\n      `Video frames exceed maximum limit. Maximum allowed: ${MAX_FRAMES} frames (${MAX_DURATION}s at ${DEFAULT_FPS}fps), got: ${currentVideoFrames.length} frames`,\n    );\n  }\n\n  const userContent: ChatCompletionUserMessageParam['content'] = [];\n\n  userContent.push({\n    type: 'text',\n    text: `## 当前视频（共 ${currentVideoFrames.length} 帧，最大限制 ${MAX_FRAMES} 帧）\n\n以下为当前录制视频的关键帧截图：`,\n  });\n\n  const frameInterval = Math.max(1, Math.floor(currentVideoFrames.length / 10));\n  for (let i = 0; i < currentVideoFrames.length; i += frameInterval) {\n    userContent.push({\n      type: 'text',\n      text: `帧 ${i + 1}/${currentVideoFrames.length}`,\n    });\n    userContent.push({\n      type: 'image_url',\n      image_url: { url: currentVideoFrames[i], detail: 'low' },\n    });\n  }\n\n  userContent.push({\n    type: 'text',\n    text: buildVideoPrompt({\n      assertion,\n      businessContext,\n      videoOptions,\n    }),\n  });\n\n  const msgs = [\n    { role: 'system' as const, content: VIDEO_SYSTEM_PROMPT },\n    { role: 'user' as const, content: userContent },\n  ];\n\n  debug('calling AI for video assertion:', assertion);\n\n  try {\n    const result = await callAIWithObjectResponse<{\n      pass: boolean;\n      thought: string;\n      reason?: string;\n      videoDetails?: VideoDetails;\n    }>(msgs, modelConfig, { abortSignal });\n\n    debug('video assertion result:', result.content);\n\n    return {\n      pass: result.content.pass,\n      thought: result.content.thought || '',\n      reason: result.content.reason,\n      videoDetails: result.content.videoDetails,\n      usage: result.usage,\n      rawResponse: JSON.stringify(result.content),\n    };\n  } catch (error) {\n    debug('video assertion error:', error);\n    throw error;\n  }\n}\n"],"names":["debug","getDebug","DEFAULT_SYSTEM_CHECK_PROMPT","buildAssertionPrompt","options","assertion","businessContext","systemCheckPrompt","prompt","AiAssertElement","beforeScreenshot","afterScreenshot","enableSystemCheck","customSystemCheckRules","modelConfig","abortSignal","systemPrompt","userContent","msgs","result","callAIWithObjectResponse","JSON","error","DIFF_SYSTEM_PROMPT","buildDiffPrompt","diffThreshold","ignoreRegions","ignoreDynamicContent","strictMode","region","index","AiAssertDiff","currentScreenshot","referenceImages","refImage","VIDEO_SYSTEM_PROMPT","buildVideoPrompt","videoOptions","ts","desc","AiAssertVideo","currentVideoFrames","MAX_DURATION","DEFAULT_FPS","MAX_FRAMES","Error","frameInterval","Math","i"],"mappings":";;AAcA,MAAMA,QAAQC,SAAS;AA0BvB,MAAMC,8BAA8B,CAAC;;;;;;;;6CAQQ,CAAC;AAE9C,SAASC,qBAAqBC,OAI7B;IACC,MAAM,EAAEC,SAAS,EAAEC,eAAe,EAAEC,iBAAiB,EAAE,GAAGH;IAE1D,IAAII,SAAS,CAAC;AAChB,EAAEH,WAAW;IAEX,IAAIC,iBACFE,UAAU,CAAC;;;AAGf,EAAEF,iBAAiB;IAGjB,IAAIC,mBACFC,UAAU,CAAC;;;AAGf,EAAED,mBAAmB;IAGnBC,UAAU,CAAC;;;;;;;;;;;;;;;CAeZ,CAAC;IAEA,OAAOA;AACT;AAEO,eAAeC,gBAAgBL,OAAwB;IAQ5D,MAAM,EACJM,gBAAgB,EAChBC,eAAe,EACfN,SAAS,EACTC,eAAe,EACfM,oBAAoB,IAAI,EACxBC,sBAAsB,EACtBC,WAAW,EACXC,WAAW,EACZ,GAAGX;IAEJ,MAAMY,eAAe,CAAC;;;;;;iBAMP,CAAC;IAEhB,MAAMT,oBAAoBK,oBACtBC,0BAA0BX,8BAC1B;IAEJ,MAAMe,cAAyD,EAAE;IAEjE,IAAIP,kBAAkB;QACpBO,YAAY,IAAI,CAAC;YACf,MAAM;YACN,MAAM;QACR;QACAA,YAAY,IAAI,CAAC;YACf,MAAM;YACN,WAAW;gBAAE,KAAKP;gBAAkB,QAAQ;YAAO;QACrD;QACAO,YAAY,IAAI,CAAC;YACf,MAAM;YACN,MAAM;QACR;IACF,OACEA,YAAY,IAAI,CAAC;QACf,MAAM;QACN,MAAM;IACR;IAGFA,YAAY,IAAI,CAAC;QACf,MAAM;QACN,WAAW;YAAE,KAAKN;YAAiB,QAAQ;QAAO;IACpD;IAEAM,YAAY,IAAI,CAAC;QACf,MAAM;QACN,MAAMd,qBAAqB;YACzBE;YACAC;YACAC;QACF;IACF;IAEA,MAAMW,OAAO;QACX;YAAE,MAAM;YAAmB,SAASF;QAAa;QACjD;YAAE,MAAM;YAAiB,SAASC;QAAY;KAC/C;IAEDjB,MAAM,6BAA6BK;IAEnC,IAAI;QACF,MAAMc,SAAS,MAAMC,yBACnBF,MACAJ,aACA;YAAEC;QAAY;QAGhBf,MAAM,qBAAqBmB,OAAO,OAAO;QAEzC,OAAO;YACL,MAAMA,OAAO,OAAO,CAAC,IAAI;YACzB,SAASA,OAAO,OAAO,CAAC,OAAO,IAAI;YACnC,QAAQA,OAAO,OAAO,CAAC,MAAM;YAC7B,OAAOA,OAAO,KAAK;YACnB,oBAAoBA,OAAO,OAAO,CAAC,kBAAkB;YACrD,aAAaE,KAAK,SAAS,CAACF,OAAO,OAAO;QAC5C;IACF,EAAE,OAAOG,OAAO;QACdtB,MAAM,oBAAoBsB;QAC1B,MAAMA;IACR;AACF;AAEA,MAAMC,qBAAqB,CAAC;;;;;;iBAMX,CAAC;AAElB,SAASC,gBAAgBpB,OAOxB;IACC,MAAM,EACJC,SAAS,EACTC,eAAe,EACfmB,gBAAgB,GAAG,EACnBC,aAAa,EACbC,oBAAoB,EACpBC,UAAU,EACX,GAAGxB;IAEJ,IAAII,SAAS,CAAC;;;;;AAKhB,EAAEH,WAAW;IAEX,IAAIC,iBACFE,UAAU,CAAC;;;AAGf,EAAEF,iBAAiB;IAGjBE,UAAU,CAAC;;;;;;;;;;;;;QAaL,EAAEiB,AAAgB,MAAhBA,cAAoB,CAAC,CAAC;IAE9B,IAAIC,iBAAiBA,cAAc,MAAM,GAAG,GAAG;QAC7ClB,UAAU,CAAC;;;UAGL,CAAC;QACPkB,cAAc,OAAO,CAAC,CAACG,QAAQC;YAC7BtB,UAAU,CAAC;AACjB,EAAEsB,QAAQ,EAAE,MAAM,EAAED,OAAO,CAAC,CAAC,EAAE,EAAEA,OAAO,CAAC,CAAC,KAAK,EAAEA,OAAO,KAAK,CAAC,CAAC,EAAEA,OAAO,MAAM,EAAE;QAC5E;IACF;IAEA,IAAIF,sBACFnB,UAAU,CAAC;;;;;;;MAOT,CAAC;IAGL,IAAIoB,YACFpB,UAAU,CAAC;;;kBAGG,CAAC;IAGjBA,UAAU,CAAC;;;;;;;;;;;;;;;;;CAiBZ,CAAC;IAEA,OAAOA;AACT;AAEO,eAAeuB,aAAa3B,OAWlC;IAQC,MAAM,EACJ4B,iBAAiB,EACjBC,eAAe,EACf5B,SAAS,EACTC,eAAe,EACfmB,aAAa,EACbC,aAAa,EACbC,oBAAoB,EACpBC,UAAU,EACVd,WAAW,EACXC,WAAW,EACZ,GAAGX;IAEJ,MAAMa,cAAyD,EAAE;IAGjE,KAAK,MAAMiB,YAAYD,gBAAiB;QACtChB,YAAY,IAAI,CAAC;YACf,MAAM;YACN,MAAM,CAAC,QAAQ,EAAEiB,SAAS,IAAI,EAAE;QAClC;QACAjB,YAAY,IAAI,CAAC;YACf,MAAM;YACN,WAAW;gBAAE,KAAKiB,SAAS,GAAG;gBAAE,QAAQ;YAAO;QACjD;IACF;IAGAjB,YAAY,IAAI,CAAC;QACf,MAAM;QACN,MAAM;IACR;IACAA,YAAY,IAAI,CAAC;QACf,MAAM;QACN,WAAW;YAAE,KAAKe;YAAmB,QAAQ;QAAO;IACtD;IAGAf,YAAY,IAAI,CAAC;QACf,MAAM;QACN,MAAMO,gBAAgB;YACpBnB;YACAC;YACAmB;YACAC;YACAC;YACAC;QACF;IACF;IAEA,MAAMV,OAAO;QACX;YAAE,MAAM;YAAmB,SAASK;QAAmB;QACvD;YAAE,MAAM;YAAiB,SAASN;QAAY;KAC/C;IAEDjB,MAAM,kCAAkCK;IAExC,IAAI;QACF,MAAMc,SAAS,MAAMC,yBACnBF,MACAJ,aACA;YAAEC;QAAY;QAGhBf,MAAM,0BAA0BmB,OAAO,OAAO;QAE9C,OAAO;YACL,MAAMA,OAAO,OAAO,CAAC,IAAI;YACzB,SAASA,OAAO,OAAO,CAAC,OAAO,IAAI;YACnC,QAAQA,OAAO,OAAO,CAAC,MAAM;YAC7B,aAAaA,OAAO,OAAO,CAAC,WAAW;YACvC,OAAOA,OAAO,KAAK;YACnB,aAAaE,KAAK,SAAS,CAACF,OAAO,OAAO;QAC5C;IACF,EAAE,OAAOG,OAAO;QACdtB,MAAM,yBAAyBsB;QAC/B,MAAMA;IACR;AACF;AAEA,MAAMa,sBAAsB,CAAC;;;;;;;iBAOZ,CAAC;AAElB,SAASC,iBAAiBhC,OAIzB;IACC,MAAM,EAAEC,SAAS,EAAEC,eAAe,EAAE+B,YAAY,EAAE,GAAGjC;IAErD,IAAII,SAAS,CAAC;;;;;AAKhB,EAAEH,WAAW;IAEX,IAAIC,iBACFE,UAAU,CAAC;;;AAGf,EAAEF,iBAAiB;IAGjBE,UAAU,CAAC;;;;;;;;;;2BAUc,CAAC;IAE1B,IAAI6B,cAAc;QAChB,IAAIA,aAAa,eAAe,EAC9B7B,UAAU,CAAC;;;MAGX,EAAE6B,aAAa,mBAAmB,IAAI,GAAG;cACjC,CAAC;QAGX,IAAIA,aAAa,aAAa,IAAIA,aAAa,gBAAgB,EAC7D7B,UAAU,CAAC;;;OAGV,EAAE6B,aAAa,gBAAgB,CAAC,GAAG,IAAI,EAAE,IAAI,EAAEA,aAAa,gBAAgB,CAAC,GAAG,IAAI,MAAM,CAAC,CAAC;QAG/F,IACEA,aAAa,SAAS,IACtBA,aAAa,SAAS,CAAC,UAAU,CAAC,MAAM,GAAG,GAC3C;YACA7B,UAAU,CAAC;;;aAGJ,CAAC;YACR6B,aAAa,SAAS,CAAC,UAAU,CAAC,OAAO,CAAC,CAACC,IAAIR;gBAC7C,MAAMS,OAAOF,aAAa,SAAS,EAAE,cAAc,CAACP,MAAM,IAAI;gBAC9DtB,UAAU,CAAC;EACjB,EAAE8B,GAAG,EAAE,EAAEC,MAAM;YACX;QACF;QAEA,IAAIF,aAAa,aAAa,IAAIA,AAA+B,WAA/BA,aAAa,aAAa,EAC1D7B,UAAU,CAAC;;;OAGV,EAAE6B,aAAa,aAAa,EAAE;IAEnC;IAEA7B,UAAU,CAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CA6BZ,CAAC;IAEA,OAAOA;AACT;AAEO,eAAegC,cAAcpC,OAOnC;IAQC,MAAM,EACJqC,kBAAkB,EAClBpC,SAAS,EACTC,eAAe,EACf+B,YAAY,EACZvB,WAAW,EACXC,WAAW,EACZ,GAAGX;IAEJ,MAAMsC,eAAe;IACrB,MAAMC,cAAc;IACpB,MAAMC,aAAaF,eAAeC;IAElC,IAAIF,mBAAmB,MAAM,GAAGG,YAC9B,MAAM,IAAIC,MACR,CAAC,oDAAoD,EAAED,WAAW,SAAS,EAAEF,aAAa,KAAK,EAAEC,YAAY,WAAW,EAAEF,mBAAmB,MAAM,CAAC,OAAO,CAAC;IAIhK,MAAMxB,cAAyD,EAAE;IAEjEA,YAAY,IAAI,CAAC;QACf,MAAM;QACN,MAAM,CAAC,UAAU,EAAEwB,mBAAmB,MAAM,CAAC,QAAQ,EAAEG,WAAW;;gBAEtD,CAAC;IACf;IAEA,MAAME,gBAAgBC,KAAK,GAAG,CAAC,GAAGA,KAAK,KAAK,CAACN,mBAAmB,MAAM,GAAG;IACzE,IAAK,IAAIO,IAAI,GAAGA,IAAIP,mBAAmB,MAAM,EAAEO,KAAKF,cAAe;QACjE7B,YAAY,IAAI,CAAC;YACf,MAAM;YACN,MAAM,CAAC,EAAE,EAAE+B,IAAI,EAAE,CAAC,EAAEP,mBAAmB,MAAM,EAAE;QACjD;QACAxB,YAAY,IAAI,CAAC;YACf,MAAM;YACN,WAAW;gBAAE,KAAKwB,kBAAkB,CAACO,EAAE;gBAAE,QAAQ;YAAM;QACzD;IACF;IAEA/B,YAAY,IAAI,CAAC;QACf,MAAM;QACN,MAAMmB,iBAAiB;YACrB/B;YACAC;YACA+B;QACF;IACF;IAEA,MAAMnB,OAAO;QACX;YAAE,MAAM;YAAmB,SAASiB;QAAoB;QACxD;YAAE,MAAM;YAAiB,SAASlB;QAAY;KAC/C;IAEDjB,MAAM,mCAAmCK;IAEzC,IAAI;QACF,MAAMc,SAAS,MAAMC,yBAKlBF,MAAMJ,aAAa;YAAEC;QAAY;QAEpCf,MAAM,2BAA2BmB,OAAO,OAAO;QAE/C,OAAO;YACL,MAAMA,OAAO,OAAO,CAAC,IAAI;YACzB,SAASA,OAAO,OAAO,CAAC,OAAO,IAAI;YACnC,QAAQA,OAAO,OAAO,CAAC,MAAM;YAC7B,cAAcA,OAAO,OAAO,CAAC,YAAY;YACzC,OAAOA,OAAO,KAAK;YACnB,aAAaE,KAAK,SAAS,CAACF,OAAO,OAAO;QAC5C;IACF,EAAE,OAAOG,OAAO;QACdtB,MAAM,0BAA0BsB;QAChC,MAAMA;IACR;AACF"}

package/dist/es/ai-model/index.mjs

@@ -3,9 +3,10 @@ import { systemPromptToLocateElement } from "./prompt/llm-locator.mjs";
 import { generatePlaywrightTest, generatePlaywrightTestStream } from "./prompt/playwright-generator.mjs";
 import { generateYamlTest, generateYamlTestStream } from "./prompt/yaml-generator.mjs";
 import { AiExtractElementInfo, AiJudgeOrderSensitive, AiLocateElement, AiLocateSection } from "./inspect.mjs";
+import { AiAssertDiff, AiAssertElement } from "./assert.mjs";
 import { plan } from "./llm-planning.mjs";
 import { autoGLMPlanning } from "./auto-glm/planning.mjs";
 import { PointSchema, RectSchema, SizeSchema, TMultimodalPromptSchema, TUserPromptSchema, adaptBboxToRect, dumpActionParam, findAllMidsceneLocatorField, getMidsceneLocationSchema, parseActionParam } from "../common.mjs";
 import { uiTarsPlanning } from "./ui-tars-planning.mjs";
 import { ConversationHistory } from "./conversation-history.mjs";
-export { AIResponseParseError, AiExtractElementInfo, AiJudgeOrderSensitive, AiLocateElement, AiLocateSection, ConversationHistory, PointSchema, RectSchema, SizeSchema, TMultimodalPromptSchema, TUserPromptSchema, adaptBboxToRect, autoGLMPlanning, callAI, callAIWithObjectResponse, callAIWithStringResponse, dumpActionParam, findAllMidsceneLocatorField, generatePlaywrightTest, generatePlaywrightTestStream, generateYamlTest, generateYamlTestStream, getMidsceneLocationSchema, parseActionParam, plan, systemPromptToLocateElement, uiTarsPlanning };
+export { AIResponseParseError, AiAssertDiff, AiAssertElement, AiExtractElementInfo, AiJudgeOrderSensitive, AiLocateElement, AiLocateSection, ConversationHistory, PointSchema, RectSchema, SizeSchema, TMultimodalPromptSchema, TUserPromptSchema, adaptBboxToRect, autoGLMPlanning, callAI, callAIWithObjectResponse, callAIWithStringResponse, dumpActionParam, findAllMidsceneLocatorField, generatePlaywrightTest, generatePlaywrightTestStream, generateYamlTest, generateYamlTestStream, getMidsceneLocationSchema, parseActionParam, plan, systemPromptToLocateElement, uiTarsPlanning };

package/dist/es/ai-model/prompt/cache-verify.mjs

@@ -0,0 +1,34 @@
+import { getPreferredLanguage } from "@midscene/shared/env";
+const cacheVerifyInstruction = ()=>{
+    const preferredLanguage = getPreferredLanguage();
+    return `
+You are verifying whether a cached element location is still correct.
+The image shows a SMALL CROPPED AREA around the cached coordinates, with the target element marked by a red rectangle.
+
+TASK: Determine if the element in the red rectangle matches the user's original description.
+
+USER'S ORIGINAL DESCRIPTION: "{{targetPrompt}}"
+
+RULES:
+1. Look at the element inside the red rectangle carefully
+2. Compare it with the user's description: "{{targetPrompt}}"
+3. Consider:
+   - Is it the same TYPE of element? (button, input, link, text, etc.)
+   - Does it have matching TEXT or LABEL?
+   - Is it in a reasonable CONTEXT for that description?
+4. If the description is vague (e.g., "click the button"), be more lenient
+5. If the description is specific (e.g., "Login button", "Submit button"), be strict
+
+RESPONSE FORMAT (JSON):
+{
+  "match": true/false,
+  "description": "brief description of what you see in the red rectangle",
+  "reason": "short explanation of why it matches or not"
+}
+
+IMPORTANT: Write descriptions and reasons in ${preferredLanguage}.
+`;
+};
+export { cacheVerifyInstruction };
+
+//# sourceMappingURL=cache-verify.mjs.map

package/dist/es/ai-model/prompt/cache-verify.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"ai-model/prompt/cache-verify.mjs","sources":["../../../../src/ai-model/prompt/cache-verify.ts"],"sourcesContent":["import { getPreferredLanguage } from '@midscene/shared/env';\n\nexport const cacheVerifyInstruction = () => {\n  const preferredLanguage = getPreferredLanguage();\n\n  return `\nYou are verifying whether a cached element location is still correct.\nThe image shows a SMALL CROPPED AREA around the cached coordinates, with the target element marked by a red rectangle.\n\nTASK: Determine if the element in the red rectangle matches the user's original description.\n\nUSER'S ORIGINAL DESCRIPTION: \"{{targetPrompt}}\"\n\nRULES:\n1. Look at the element inside the red rectangle carefully\n2. Compare it with the user's description: \"{{targetPrompt}}\"\n3. Consider:\n   - Is it the same TYPE of element? (button, input, link, text, etc.)\n   - Does it have matching TEXT or LABEL?\n   - Is it in a reasonable CONTEXT for that description?\n4. If the description is vague (e.g., \"click the button\"), be more lenient\n5. If the description is specific (e.g., \"Login button\", \"Submit button\"), be strict\n\nRESPONSE FORMAT (JSON):\n{\n  \"match\": true/false,\n  \"description\": \"brief description of what you see in the red rectangle\",\n  \"reason\": \"short explanation of why it matches or not\"\n}\n\nIMPORTANT: Write descriptions and reasons in ${preferredLanguage}.\n`;\n};\n"],"names":["cacheVerifyInstruction","preferredLanguage","getPreferredLanguage"],"mappings":";AAEO,MAAMA,yBAAyB;IACpC,MAAMC,oBAAoBC;IAE1B,OAAO,CAAC;;;;;;;;;;;;;;;;;;;;;;;;;6CAyBmC,EAAED,kBAAkB;AACjE,CAAC;AACD"}

package/dist/es/service/index.mjs

@@ -1,6 +1,7 @@
 import { isAutoGLM } from "../ai-model/auto-glm/util.mjs";
 import { AIResponseParseError, AiExtractElementInfo, AiLocateElement, callAIWithObjectResponse } from "../ai-model/index.mjs";
 import { AiLocateSection } from "../ai-model/inspect.mjs";
+import { cacheVerifyInstruction } from "../ai-model/prompt/cache-verify.mjs";
 import { elementDescriberInstruction } from "../ai-model/prompt/describe.mjs";
 import { expandSearchArea } from "../common.mjs";
 import { ServiceError } from "../types.mjs";
@@ -38,7 +39,6 @@ class Service {
         const context = opt?.context || await this.contextRetrieverFn();
         let searchArea;
         let searchAreaRawResponse;
-        let searchAreaUsage;
         let searchAreaResponse;
         if (searchAreaPrompt) {
             searchAreaResponse = await AiLocateSection({
@@ -49,7 +49,7 @@ class Service {
             });
             assert(searchAreaResponse.rect, `cannot find search area for "${searchAreaPrompt}"${searchAreaResponse.error ? `: ${searchAreaResponse.error}` : ''}`);
             searchAreaRawResponse = searchAreaResponse.rawResponse;
-            searchAreaUsage = searchAreaResponse.usage;
+            searchAreaResponse.usage;
             searchArea = searchAreaResponse.rect;
         }
         const startTime = Date.now();
@@ -69,7 +69,6 @@ class Service {
             usage,
             searchArea,
             searchAreaRawResponse,
-            searchAreaUsage,
             reasoning_content
         };
         let errorLog;
@@ -82,9 +81,9 @@ class Service {
             matchedElement: [],
             matchedRect: rect,
             data: null,
-            taskInfo,
-            deepLocate: !!searchArea,
-            error: errorLog
+            deepLocate: !!searchArea || !!searchAreaPrompt,
+            error: errorLog,
+            taskInfo
         };
         const elements = parseResult.elements || [];
         const dump = createServiceDump({
@@ -242,6 +241,82 @@ class Service {
         assert(content.description, 'failed to describe the element');
         return content;
     }
+    async verifyCachedElement(center, targetPrompt, modelConfig, uiContext) {
+        const context = uiContext || await this.contextRetrieverFn();
+        const { shotSize } = context;
+        const screenshotBase64 = context.screenshot.base64;
+        assert(screenshotBase64, 'screenshot is required for verifyCachedElement');
+        const defaultRectSize = 60;
+        const targetRect = {
+            left: Math.floor(center[0] - defaultRectSize / 2),
+            top: Math.floor(center[1] - defaultRectSize / 2),
+            width: defaultRectSize,
+            height: defaultRectSize
+        };
+        const cropArea = expandSearchArea(targetRect, shotSize);
+        let imagePayload = await compositeElementInfoImg({
+            inputImgBase64: screenshotBase64,
+            size: shotSize,
+            elementsPositionInfo: [
+                {
+                    rect: targetRect
+                }
+            ],
+            borderThickness: 3
+        });
+        debug('verifyCachedElement: cropping to area', cropArea, 'from center', center);
+        const croppedResult = await cropByRect(imagePayload, cropArea, 'qwen2.5-vl' === modelConfig.modelFamily);
+        imagePayload = croppedResult.imageBase64;
+        const promptText = 'string' == typeof targetPrompt ? targetPrompt : targetPrompt.prompt || '';
+        const systemPrompt = cacheVerifyInstruction().replace(/{{targetPrompt}}/g, promptText);
+        const msgs = [
+            {
+                role: 'system',
+                content: systemPrompt
+            },
+            {
+                role: 'user',
+                content: [
+                    {
+                        type: 'image_url',
+                        image_url: {
+                            url: imagePayload,
+                            detail: 'low'
+                        }
+                    }
+                ]
+            }
+        ];
+        try {
+            const res = await callAIWithObjectResponse(msgs, modelConfig);
+            const content = res.content;
+            if (content.error) {
+                debug('verifyCachedElement AI returned error:', content.error);
+                return {
+                    pass: false,
+                    reason: content.error
+                };
+            }
+            const result = {
+                pass: !!content.match,
+                description: content.description,
+                reason: content.reason
+            };
+            debug('verifyCachedElement result:', {
+                pass: result.pass,
+                description: result.description,
+                reason: result.reason,
+                targetPrompt: promptText
+            });
+            return result;
+        } catch (error) {
+            debug('verifyCachedElement failed:', error);
+            return {
+                pass: false,
+                reason: String(error)
+            };
+        }
+    }
     constructor(context, opt){
         _define_property(this, "contextRetrieverFn", void 0);
         _define_property(this, "taskInfo", void 0);