Huggingface Daily Report

⚠️ CRITICAL RULES

• NEVER send message before document is fully written
• ALWAYS verify document content with read action before sending message
• Message must include HF Papers link for each paper
• Minimum block_count: 50+ blocks (ensures all content is written)

Output Modes

Both message and document (default):

• Send formatted report via message
• Create Feishu cloud document with full details
• Return document link in message

Message only (when user asks for quick summary):

• Send condensed report via message
• Skip document creation

Document only (when user asks to save without sending):

• Create Feishu cloud document
• Return document link without sending full report

Commands

# Generate report for today
node scripts/generate_report.js

# Generate report for specific date
node scripts/generate_report.js --date 2026-02-22

# Create Feishu document
node scripts/create_document.js --title "Report Title"

Report Structure

Document Title Format

YYYY-MM-DD Hugging Face Daily Papers Report

Content Structure

# Hugging Face Daily Papers Report

## 目录
- [1. Paper Title 1](#1-paper-title-1)
- [2. Paper Title 2](#2-paper-title-2)
- ...
- [N. 今日趋势总结](#n-今日趋势总结)

---

## 1. Paper Title

**🏢 机构**: Institution Name

**📅 提交日期**: YYYY 年 M 月 D 日

**🔥 热度**: 当日最热 / 高度关注 / 持续上升 / 新兴热点

**🔬 研究方向**: [研究方向，例如：Agentic RL, Multimodal, Spatial Reasoning, Code Generation, etc.]

**📌 核心贡献**:
详细描述核心贡献（2-3 句话）。必须清晰说明论文的主要创新点和解决的问题。

**🔬 关键技术**:
- **技术中文名（English Name）**: 描述该技术的作用和原理（1-2 句话）
- **技术中文名（English Name）**: 描述该技术的作用和原理（1-2 句话）
- **技术中文名（English Name）**: 描述该技术的作用和原理（1-2 句话）

> ⚠️ **注意**: 
> 1. 关键技术不能只写关键词，必须包含简短的解释说明
> 2. 技术名称必须使用**中文（英文）**格式，例如：自进化 AI 社会（Self-Evolving AI Society）
> 3. 每项技术需要说明其作用和在论文中的具体应用

**📊 实验结果**:
- 具体的实验结果和性能指标
- 与 baseline 的对比数据
- 关键发现和洞察

**🔗 链接**: 
- HF Papers 地址：https://huggingface.co/papers/...（必需）
- arXiv 地址：https://arxiv.org/abs/...
- PDF 地址：https://arxiv.org/pdf/...
- 项目地址：https://...

---

## N. 今日趋势总结

1. **Trend 1**: Description
2. **Trend 2**: Description
3. **Trend 3**: Description

---

*报告由蛋仔 🐰 自动生成*

Paper Data Format

Each paper should include:

{
  title: "Paper Title",
  institution: "Institution Name",
  date: "YYYY-MM-DD",
  links: {
    project: "https://...",
    arxiv: "https://arxiv.org/abs/...",
    pdf: "https://arxiv.org/pdf/...",
    hfPapers: "https://huggingface.co/papers/..."
  },
  coreContribution: "Brief description",
  keyTechniques: [
    { name: "Technique 1", description: "..." },
    { name: "Technique 2", description: "..." }
  ],
  experimentalResults: [
    "Result 1 with metrics",
    "Result 2 with metrics"
  ]
}

Bitable Integration (Optional)

For papers with structured data (e.g., model specifications, benchmark results):

1. Create Bitable: feishu_bitable_create_app()
2. Configure Fields: Add appropriate field types
3. Insert Data: feishu_bitable_create_record()
4. Embed Link: Include Bitable URL in document

Example: Model Specifications Table

## 📊 Model Specifications

**🔗 查看详细表格**: https://feishu.cn/base/APP_TOKEN

> 💡 **使用说明**: 这是一个实时更新的多维表格，支持筛选、排序和协作编辑。

Tools Used

Priority Tools (REQUIRED)

• tavily_search (PRIORITY #1): Search for HuggingFace Daily Papers

  • ALWAYS use tavily_search FIRST before any other search tool
  • Search query format: "Hugging Face Daily Papers [YYYY-MM-DD]" or "huggingface.co/papers/date/[YYYY-MM-DD]"
  • Extract the official HF Papers URL from results

• tavily_extract (PRIORITY #2): Extract paper list from HF Papers page

  • Use the URL from tavily_search results
  • Get complete paper list with titles, institutions, and HF links
  • Extract engagement metrics (upvotes, comments) for ranking

Secondary Tools

• message: Send report directly to user (required - use action="send")
• feishu_doc: Create and format Feishu documents (required)
• feishu_bitable_create_app: Create Bitable for structured data (optional)
• feishu_bitable_create_field: Configure Bitable fields
• feishu_bitable_create_record: Insert data into Bitable

Fallback Tools (ONLY if tavily fails)

• web_search: Fallback search (requires Brave API key)
• web_fetch: Fallback extraction

Message Format

When sending via message tool:

• Use action="send"
• Format report with clear sections and emoji
• Include HF Papers link for EACH paper (required)
• Include 核心贡献，关键技术，实验结果 for each paper (required)
• Include document link at the end
• Keep it readable (use bullet points, bold text)
• For Feishu channel: omit target to reply to current conversation

Example:

{
  "action": "send",
  "message": "📊 HuggingFace Daily Report - 2026-02-22\n\n【1. Paper Title - Institution】\n🔥 热度：当日最热\n🔬 方向：Agentic RL / Multimodal\n📌 核心：详细描述核心贡献（2-3 句话）\n🔬 技术：技术中文名 1（English，作用说明）, 技术中文名 2（English，作用说明）, 技术中文名 3（English，作用说明）\n📊 结果：具体的实验结果和性能指标\n🔗 HF Papers: https://huggingface.co/papers/...\n\n【2. Paper Title - Institution】\n🔥 热度：高度关注\n🔬 方向：Code Generation / Efficiency\n📌 核心：...\n🔬 技术：...\n📊 结果：...\n🔗 HF Papers: https://huggingface.co/papers/...\n\n📈 今日趋势：高效模型、具身智能成为热点\n\n📄 完整文档：https://feishu.cn/docx/...\n\n*报告由 蛋仔 🐰 整理*"
}

Example Workflow

Standard Report (Message + Document) - CORRECT ORDER with TAVILY

User: "生成 2026-02-17 的 HuggingFace 论文报告"

Assistant:
1. **TAVILY SEARCH** (PRIORITY): 
   - tavily_search(query="Hugging Face Daily Papers February 17 2026", n=10)
   - Extract HF Papers URL from results (e.g., https://huggingface.co/papers/date/2026-02-17)
   
2. **TAVILY EXTRACT**:
   - tavily_extract(url="https://huggingface.co/papers/date/2026-02-17")
   - Get complete paper list with titles, institutions, upvotes, comments
   
3. **SELECT TOP 5+ PAPERS**:
   - Sort by upvotes/comments (popularity)
   - **MUST include at least top 5 papers**
   - Extract HF Papers link for each (e.g., /papers/2602.10809)
   
4. **EXTRACT DETAILS**: For each of top 5+ papers:
   - Get arXiv ID from HF link
   - Extract institution, core contribution, key techniques, results
   
5. **CREATE DOCUMENT**: feishu_doc(action="create", title="2026-02-17 Hugging Face Daily Papers Report")
   
6. **WRITE CONTENT**: feishu_doc(action="append", ...) for each section
   - Append TOC first
   - Append each paper section (one append per paper)
   - Append trends summary
   
7. **VERIFY**: feishu_doc(action="read", doc_token="...")
   - Check block_count >= 50
   - Verify all 5+ papers are included
   
8. **SEND MESSAGE WITH HF LINKS**: message(action="send", message="...")
   - Include HF Papers link for EACH paper
   - Include verified document URL

Message Output (WITH HF LINKS):
📊 HuggingFace Daily Papers Report - 2026-02-22

【1. SpargeAttention2 - 清华大学】
核心贡献：可训练稀疏注意力方法，动态选择关键 token 进行计算
关键技术：混合掩码规则（Hybrid Masking，结合局部和全局注意力）、高效 CUDA 实现（Efficient Implementation，GPU 优化加速）
实验结果：95% 稀疏度，16.2 倍加速，性能损失<1%
🔗 HF Papers: https://huggingface.co/papers/2602.13515

【2. Mobile-Agent-v3.5 - 阿里通义】
...
🔗 HF Papers: https://huggingface.co/papers/2602.16855

📈 今日趋势：
• Agent 方向持续火热
• 效率优化方案涌现
...

📄 完整文档：https://feishu.cn/docx/...

Document Only

User: "把报告存到云文档，不用发给我"

Assistant:
1. Generate report content
2. Create Feishu document
3. Return only document link (no message send)

Quick Summary (Message Only)

User: "快速看看今天有什么论文"

Assistant:
1. Fetch top 5 papers
2. Send condensed summary via message
3. Skip document creation

Paper Selection Rules (CRITICAL)

Minimum Paper Count

• MUST include at least 5 papers from the target date
• Maximum 10 papers (to avoid overly long reports)
• DO NOT skip any top-5 papers by popularity

Ranking Criteria

Select papers based on热度 (popularity) in this order:

1. Upvotes/Stars (primary metric)
2. Comments count (secondary metric)
3. Views/Downloads (tertiary metric)

Verification Checklist (BEFORE writing report)

• Tavily search completed successfully
• HF Papers page extracted
• At least 5 papers identified
• Papers sorted by popularity (high to low)
• Each paper has HF Papers URL
• No top-5 papers are missing

⚠️ Common Mistakes to AVOID

• ❌ Skipping papers due to "lack of information" - include all top 5 regardless
• ❌ Replacing papers with "more interesting" ones - follow popularity ranking
• ❌ Including papers from wrong date - verify date matches request
• ❌ Using fake/simulated data - always use tavily to get real papers

Best Practices

1. Document First, Message Last (CRITICAL):

   • Step 1: Create document with feishu_doc(action="create")
   • Step 2: Write ALL content with multiple feishu_doc(action="append") calls
   • Step 3: Verify with feishu_doc(action="read") - check block_count >= 50
   • Step 4: ONLY THEN send message with document link
   • NEVER send message before verification passes

2. Paper Selection: Focus on top 8-10 trending papers

3. Detail Level: Include all key information (institution, date, links, contributions, techniques, results)

4. Link Format: Use clear labels (项目地址，arXiv 地址，etc.)

5. HF Papers Link in Message (REQUIRED):

   • Each paper in message MUST include HF Papers link
   • Format: 🔗 HF Papers: https://huggingface.co/papers/...
   • Place after experimental results for each paper

6. Trend Analysis: Summarize 3-5 key trends at the end

7. Document Structure: Use numbered headings (## 1., ## 2., etc.)

8. Table of Contents: Auto-generate for documents with 3+ sections

9. Bitable Usage: Use for structured data that benefits from filtering/sorting

10. Message Formatting:

    • Use emoji for visual clarity (📊, 🏢, 📅, 🔗, 📌, 🔬, 📊, 📈, 📄)
    • Keep sections concise in message (full details in document)
    • Use bullet points and bold text for readability
    • Put document link at the end
    • Include HF link for each paper

11. Verification Checklist (before sending message):

    • Document created successfully
    • All content blocks appended (6+ papers × 3-4 blocks each + TOC + trends)
    • block_count >= 50 verified via read action
    • Document URL extracted and ready to include in message
    • Each paper has HF Papers link ready for message

Error Handling

Tavily Search Failures

• If tavily_search returns no results:
  • Try alternative query: "huggingface.co/papers/date/[YYYY-MM-DD]"
  • Try searching for individual paper IDs
  • DO NOT fabricate papers - report the error to user

Paper Count Issues

• If fewer than 5 papers found for the date:
  • Inform user: "Only X papers found for [date]"
  • Include all available papers
  • Do NOT pad with papers from other dates

Feishu API Failures

• If Feishu API fails, retry with exponential backoff
• If Bitable creation fails, continue with text-based format

Data Completeness

• If paper details are incomplete, note missing information
• NEVER fabricate data - use "信息待补充" for missing details

Summary: Key Requirements

⭐ CRITICAL RULES (MUST FOLLOW)

1. ALWAYS use tavily_search FIRST - never skip or use other tools first
2. MUST include at least top 5 papers by popularity (upvotes/comments)
3. DO NOT skip any top-5 papers - include all of them regardless of topic
4. Verify before sending - check block_count >= 50 via feishu_doc(action="read")
5. Include HF Papers link for EACH paper in the message

📋 Paper Selection

• Minimum: 5 papers (mandatory)
• Maximum: 10 papers (recommended)
• Ranking: By 热度 (popularity) - upvotes, comments, views
• Date: Must match the requested date exactly

🔍 Search Priority

1. tavily_search (REQUIRED - always first)
2. tavily_extract (REQUIRED - get paper details)
3. web_search (FALLBACK ONLY - if tavily fails)
4. web_fetch (FALLBACK ONLY - if tavily fails)

Customization

Report Frequency

• Daily (default): Generate report for today's papers
• Weekly: Aggregate papers from the past week
• Custom date range: Specify start and end dates

Paper Count

• Minimum: 5 papers (MANDATORY)
• Recommended: 5-8 papers
• Maximum: 10 papers (to avoid overly long reports)

Output Format

• Feishu document (default)
• Markdown file (for local storage)
• Email summary (optional)