feat: 生成中英文两个版本的简报

- 新增 generateMarkdownZH 和 generateMarkdownEN 函数
- 输出文件改为 YYYY-MM-DD_zh.md 和 YYYY-MM-DD_en.md
- 英文版使用英文标题和分类名称

cache/seen_urls.json

@@ -1,5 +1,5 @@
 {
-  "lastUpdate": "2026-02-27T15:33:33.096Z",
+  "lastUpdate": "2026-02-27T15:38:30.698Z",
   "urls": [
     "http://arxiv.org/abs/2602.23360v1",
     "http://arxiv.org/abs/2602.23359v1",

daily/2026-02-27_en.md

@@ -0,0 +1,57 @@
+# AI Daily Brief - 2026-02-27
+
+> Collected at: 2/27/2026, 11:38:30 PM
+> Total items: 131
+
+## 🔥 Top 10 Highlights
+
+1. [sponsors/muratcankoylan](https://github.com/sponsors/muratcankoylan) - **GitHub Trending**
+   > A comprehensive collection of Agent Skills for context engineering, multi-agent architectures, and production agent systems. Use when building, optimi...
+
+2. [login?return_to=%2Fruvnet%2Fclaude-flow](https://github.com/login?return_to=%2Fruvnet%2Fclaude-flow) - **GitHub Trending**
+   > 🌊 The leading agent orchestration platform for Claude. Deploy intelligent multi-agent swarms, coordinate autonomous workflows, and build conversation...
+
+3. [Search More, Think Less: Rethinking Long-Horizon Agentic Search for Efficiency and Generalization](https://huggingface.co/papers/2602.22675) - **Hugging Face**
+   > Recent deep research agents primarily improve performance by scaling reasoning depth, but this leads to high inference cost and latency in search-inte...
+
+4. [AgentDropoutV2: Optimizing Information Flow in Multi-Agent Systems via Test-Time Rectify-or-Reject Pruning](https://huggingface.co/papers/2602.23258) - **Hugging Face**
+   > While Multi-Agent Systems (MAS) excel in complex reasoning, they suffer from the cascading impact of erroneous information generated by individual par...
+
+5. [Accelerating Diffusion via Hybrid Data-Pipeline Parallelism Based on Conditional Guidance Scheduling](https://huggingface.co/papers/2602.21760) - **Hugging Face**
+   > Diffusion models have achieved remarkable progress in high-fidelity image, video, and audio generation, yet inference remains computationally expensiv...
+
+6. [Exploratory Memory-Augmented LLM Agent via Hybrid On- and Off-Policy Optimization](https://huggingface.co/papers/2602.23008) - **Hugging Face**
+   > Exploration remains the key bottleneck for large language model agents trained with reinforcement learning. While prior methods exploit pretrained kno...
+
+7. [login?return_to=%2Fruvnet%2Fwifi-densepose](https://github.com/login?return_to=%2Fruvnet%2Fwifi-densepose) - **GitHub Trending**
+   > Production-ready implementation of InvisPose - a revolutionary WiFi-based dense human pose estimation system that enables real-time full-body tracking...
+
+8. [login?return_to=%2Fbytedance%2Fdeer-flow](https://github.com/login?return_to=%2Fbytedance%2Fdeer-flow) - **GitHub Trending**
+   > An open-source SuperAgent harness that researches, codes, and creates. With the help of sandboxes, memories, tools, skills and subagents, it handles d...
+
+9. [login?return_to=%2Fmoonshine-ai%2Fmoonshine](https://github.com/login?return_to=%2Fmoonshine-ai%2Fmoonshine) - **GitHub Trending**
+   > Fast and accurate automatic speech recognition (ASR) for edge devices
+
+10. [sponsors/obra](https://github.com/sponsors/obra) - **GitHub Trending**
+   > An agentic skills framework & software development methodology that works.
+
+## 📂 Categories
+
+### Agent Frameworks
+
+- [Toward Expert Investment Teams:A Multi-Agent LLM System with Fine-Grained Trading Tasks](http://arxiv.org/abs/2602.23330v1) - arXiv
+- [AgentDropoutV2: Optimizing Information Flow in Multi-Agent Systems via Test-Time Rectify-or-Reject Pruning](http://arxiv.org/abs/2602.23258v1) - arXiv
+
+### AI Infrastructure / Inference Optimization
+
+- [Bitwise Systolic Array Architecture for Runtime-Reconfigurable Multi-precision Quantized Multiplication on Hardware Accelerators](http://arxiv.org/abs/2602.23334v1) - arXiv
+- [Invariant Transformation and Resampling based Epistemic-Uncertainty Reduction](http://arxiv.org/abs/2602.23315v1) - arXiv
+- [Agency and Architectural Limits: Why Optimization-Based Systems Cannot Be Norm-Responsive](http://arxiv.org/abs/2602.23239v1) - arXiv
+- [InnerQ: Hardware-aware Tuning-free Quantization of KV Cache for Large Language Models](http://arxiv.org/abs/2602.23200v1) - arXiv
+- [Assessing Deanonymization Risks with Stylometry-Assisted LLM Agent](http://arxiv.org/abs/2602.23079v1) - arXiv
+- [Rejection Mixing: Fast Semantic Propagation of Mask Tokens for Efficient DLLM Inference](http://arxiv.org/abs/2602.22868v1) - arXiv
+- [Differentiable Zero-One Loss via Hypersimplex Projections](http://arxiv.org/abs/2602.23336v1) - arXiv
+- [FairQuant: Fairness-Aware Mixed-Precision Quantization for Medical Image Classification](http://arxiv.org/abs/2602.23192v1) - arXiv
+
+---
+*Generated by AINewsCollector*

daily/2026-02-27_zh.md

@@ -0,0 +1,57 @@
+# AI Daily Brief - 2026-02-27
+
+> 采集时间: 2026/2/27 23:38:30
+> 总条目: 131
+
+## 🔥 Top 10 重要消息
+
+1. [sponsors/muratcankoylan](https://github.com/sponsors/muratcankoylan) - **GitHub Trending**
+   > A comprehensive collection of Agent Skills for context engineering, multi-agent architectures, and production agent systems. Use when building, optimi...
+
+2. [login?return_to=%2Fruvnet%2Fclaude-flow](https://github.com/login?return_to=%2Fruvnet%2Fclaude-flow) - **GitHub Trending**
+   > 🌊 The leading agent orchestration platform for Claude. Deploy intelligent multi-agent swarms, coordinate autonomous workflows, and build conversation...
+
+3. [Search More, Think Less: Rethinking Long-Horizon Agentic Search for Efficiency and Generalization](https://huggingface.co/papers/2602.22675) - **Hugging Face**
+   > Recent deep research agents primarily improve performance by scaling reasoning depth, but this leads to high inference cost and latency in search-inte...
+
+4. [AgentDropoutV2: Optimizing Information Flow in Multi-Agent Systems via Test-Time Rectify-or-Reject Pruning](https://huggingface.co/papers/2602.23258) - **Hugging Face**
+   > While Multi-Agent Systems (MAS) excel in complex reasoning, they suffer from the cascading impact of erroneous information generated by individual par...
+
+5. [Accelerating Diffusion via Hybrid Data-Pipeline Parallelism Based on Conditional Guidance Scheduling](https://huggingface.co/papers/2602.21760) - **Hugging Face**
+   > Diffusion models have achieved remarkable progress in high-fidelity image, video, and audio generation, yet inference remains computationally expensiv...
+
+6. [Exploratory Memory-Augmented LLM Agent via Hybrid On- and Off-Policy Optimization](https://huggingface.co/papers/2602.23008) - **Hugging Face**
+   > Exploration remains the key bottleneck for large language model agents trained with reinforcement learning. While prior methods exploit pretrained kno...
+
+7. [login?return_to=%2Fruvnet%2Fwifi-densepose](https://github.com/login?return_to=%2Fruvnet%2Fwifi-densepose) - **GitHub Trending**
+   > Production-ready implementation of InvisPose - a revolutionary WiFi-based dense human pose estimation system that enables real-time full-body tracking...
+
+8. [login?return_to=%2Fbytedance%2Fdeer-flow](https://github.com/login?return_to=%2Fbytedance%2Fdeer-flow) - **GitHub Trending**
+   > An open-source SuperAgent harness that researches, codes, and creates. With the help of sandboxes, memories, tools, skills and subagents, it handles d...
+
+9. [login?return_to=%2Fmoonshine-ai%2Fmoonshine](https://github.com/login?return_to=%2Fmoonshine-ai%2Fmoonshine) - **GitHub Trending**
+   > Fast and accurate automatic speech recognition (ASR) for edge devices
+
+10. [sponsors/obra](https://github.com/sponsors/obra) - **GitHub Trending**
+   > An agentic skills framework & software development methodology that works.
+
+## 📂 分类汇总
+
+### Agent 框架
+
+- [Toward Expert Investment Teams:A Multi-Agent LLM System with Fine-Grained Trading Tasks](http://arxiv.org/abs/2602.23330v1) - arXiv
+- [AgentDropoutV2: Optimizing Information Flow in Multi-Agent Systems via Test-Time Rectify-or-Reject Pruning](http://arxiv.org/abs/2602.23258v1) - arXiv
+
+### AI 基础设施 / 推理优化
+
+- [Bitwise Systolic Array Architecture for Runtime-Reconfigurable Multi-precision Quantized Multiplication on Hardware Accelerators](http://arxiv.org/abs/2602.23334v1) - arXiv
+- [Invariant Transformation and Resampling based Epistemic-Uncertainty Reduction](http://arxiv.org/abs/2602.23315v1) - arXiv
+- [Agency and Architectural Limits: Why Optimization-Based Systems Cannot Be Norm-Responsive](http://arxiv.org/abs/2602.23239v1) - arXiv
+- [InnerQ: Hardware-aware Tuning-free Quantization of KV Cache for Large Language Models](http://arxiv.org/abs/2602.23200v1) - arXiv
+- [Assessing Deanonymization Risks with Stylometry-Assisted LLM Agent](http://arxiv.org/abs/2602.23079v1) - arXiv
+- [Rejection Mixing: Fast Semantic Propagation of Mask Tokens for Efficient DLLM Inference](http://arxiv.org/abs/2602.22868v1) - arXiv
+- [Differentiable Zero-One Loss via Hypersimplex Projections](http://arxiv.org/abs/2602.23336v1) - arXiv
+- [FairQuant: Fairness-Aware Mixed-Precision Quantization for Medical Image Classification](http://arxiv.org/abs/2602.23192v1) - arXiv
+
+---
+*Generated by AINewsCollector*

skill/ai-news-collector/collect.js

@@ -168,11 +168,7 @@ function sortItems(items, topics) {
 
 // ============ 输出模块 ============
 
-function generateMarkdown(items, topCount, topics) {
+function generateMarkdownZH(items, topCount, topics, date) {
-  const date = new Date().toLocaleDateString('zh-CN', {
-    year: 'numeric', month: '2-digit', day: '2-digit', timeZone: 'Asia/Shanghai'
-  }).replace(/\//g, '-');
-  
   const top10 = items.slice(0, topCount);
   
   let md = `# AI Daily Brief - ${date}\n\n`;
@@ -206,7 +202,63 @@ function generateMarkdown(items, topCount, topics) {
   
   md += `---\n*Generated by AINewsCollector*\n`;
   
-  return { md, date };
+  return md;
+}
+
+function generateMarkdownEN(items, topCount, topics, date) {
+  const top10 = items.slice(0, topCount);
+  
+  let md = `# AI Daily Brief - ${date}\n\n`;
+  md += `> Collected at: ${new Date().toLocaleString('en-US', { timeZone: 'Asia/Shanghai' })}\n`;
+  md += `> Total items: ${items.length}\n\n`;
+  
+  md += `## 🔥 Top ${topCount} Highlights\n\n`;
+  for (let i = 0; i < top10.length; i++) {
+    const item = top10[i];
+    md += `${i + 1}. [${item.title}](${item.url}) - **${item.source}**\n`;
+    if (item.summary) md += `   > ${item.summary.slice(0, 150)}${item.summary.length > 150 ? '...' : ''}\n`;
+    md += '\n';
+  }
+  
+  md += `## 📂 Categories\n\n`;
+  
+  // 英文分类名称映射
+  const topicNamesEN = {
+    'AI 编程工具 / Code Agent': 'AI Coding Tools / Code Agent',
+    'Agent 框架': 'Agent Frameworks',
+    'AI 基础设施 / 推理优化': 'AI Infrastructure / Inference Optimization'
+  };
+  
+  for (const topic of topics) {
+    const topicItems = items.filter(item => {
+      const text = `${item.title} ${item.summary}`.toLowerCase();
+      return topic.keywords.some(k => text.includes(k.toLowerCase()));
+    }).filter(item => !top10.includes(item));
+    
+    if (topicItems.length > 0) {
+      const topicNameEN = topicNamesEN[topic.name] || topic.name;
+      md += `### ${topicNameEN}\n\n`;
+      for (const item of topicItems.slice(0, 10)) {
+        md += `- [${item.title}](${item.url}) - ${item.source}\n`;
+      }
+      md += '\n';
+    }
+  }
+  
+  md += `---\n*Generated by AINewsCollector*\n`;
+  
+  return md;
+}
+
+function generateMarkdown(items, topCount, topics) {
+  const date = new Date().toLocaleDateString('zh-CN', {
+    year: 'numeric', month: '2-digit', day: '2-digit', timeZone: 'Asia/Shanghai'
+  }).replace(/\//g, '-');
+  
+  const md_zh = generateMarkdownZH(items, topCount, topics, date);
+  const md_en = generateMarkdownEN(items, topCount, topics, date);
+  
+  return { md_zh, md_en, date };
 }
 
 // ============ 主流程 ============
@@ -245,19 +297,34 @@ async function main() {
   
   allItems = sortItems(allItems, config.topics);
   
-  const { md, date } = generateMarkdown(allItems, config.output.topCount, config.topics);
+  const { md_zh, md_en, date } = generateMarkdown(allItems, config.output.topCount, config.topics);
   
   if (!fs.existsSync(DAILY_DIR)) fs.mkdirSync(DAILY_DIR, { recursive: true });
-  const outputPath = path.join(DAILY_DIR, `${date}.md`);
+  
-  fs.writeFileSync(outputPath, md);
+  // 保存中文版
-  console.log(`📝 简报已保存: ${outputPath}\n`);
+  const outputPathZH = path.join(DAILY_DIR, `${date}_zh.md`);
+  fs.writeFileSync(outputPathZH, md_zh);
+  console.log(`📝 中文简报: ${outputPathZH}\n`);
+  
+  // 保存英文版
+  const outputPathEN = path.join(DAILY_DIR, `${date}_en.md`);
+  fs.writeFileSync(outputPathEN, md_en);
+  console.log(`📝 英文简报: ${outputPathEN}\n`);
   
   const cacheData = { lastUpdate: new Date().toISOString(), urls: Array.from(seenUrls).slice(-5000) };
   fs.writeFileSync(CACHE_PATH, JSON.stringify(cacheData, null, 2));
   
   console.log('✅ 采集完成!\n');
   
-  return { success: true, outputPath, date, itemCount: allItems.length, content: md };
+  return { 
+    success: true, 
+    outputPathZH, 
+    outputPathEN, 
+    date, 
+    itemCount: allItems.length, 
+    content_zh: md_zh,
+    content_en: md_en
+  };
 }
 
 if (require.main === module) {