SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsai-daily-report

ai-daily-report

Agents & models

View repository

Updated 10-15 19:58

About this skill

AI Daily Report

This skill is used to collect and generate the daily popular report in the AI field.

Trigger scenarios (generalized)

All of the following expressions can trigger this skill:

  • "Collect today's AI news" / "What AI news is there"
  • "Get AI information" / "Help me check AI updates"
  • "What happened in AI today" / "What's new in AI recently"
  • "Generate an AI report" / "Make an AI summary"
  • "AI today's updates" / "AI weekly report"
  • "See what's been happening in the AI world recently"

Supported parameters

Parameter Type Required Description
领域 string No Specific AI field, such as: LLM, large models, CV, computer vision, NLP, natural language processing, robotics, autonomous driving, multimodal, AI Agent, embodied intelligence, generative AI, AIGC, Stable Diffusion, Midjourney, etc. If unspecified, collect across all fields
条数 number No Number of items to collect; defaults to 10 if unspecified
时间段 string No Time range to collect, such as "today", "this week", "this month", "the past week", "the past month", etc. Defaults to "today" if unspecified

Time period parameter description

User expression Parsed result Description
"What AI news is there today" Today Default option
"This week's AI updates" This week The last 7 days
"This month's AI report" This month The last 30 days
"AI papers from the past week" The past week The last 7 days
"Help me organize this month's AI technology" The past month The last 30 days
"Last week's AI news" Last week 7-14 days ago

Anti-freeze mechanism (important)

To prevent the task from getting stuck or timing out, the following rules must be followed during execution:

Timeout control

Operation Timeout limit Description
web_fetch Single 10 seconds Timeout for fetching a single URL
web_search Single 8 seconds Timeout for a single search
Single information source 15 seconds Timeout limit for the entire source
Entire collection phase 120 seconds Total time limit for step two

Error handling

  1. Skip on failure - A single source failure does not affect the whole; continue to the next one
  2. Fail fast - If 3 sources fail consecutively, lower the collection target
  3. Fallback content - If a certain type of content is insufficient, supplement with existing content
  4. Never block - Every operation must complete within the timeout or be abandoned

Parallel strategy

  • Use sessions_spawn to launch subtasks that fetch multiple sources in parallel
  • Each subtask sets its own timeout independently
  • The main task aggregates results; sources that do not return before the timeout are ignored

Progress feedback

Each time an information source is completed, output a brief prompt:

[1/5] Hacker News ✓
[2/5] arXiv 论文列表 ✓
...

Workflow

Step 1: Parse user instructions (fast)

  • Identify trigger keywords
  • Extract domain parameters (optional, defaults to all domains)
  • Extract item count parameters (optional, defaults to 10 items)
  • Extract time period parameters(optional, defaults to "Today")
  • Estimate the time required and inform the user

Time period parsing rules:

Keyword Time range
Today/today/today's news Same day (0 days)
Yesterday/yesterday Previous 1 day
This week/last 7 days/most recent week 0-7 days
Last week 7-14 days ago
This month/last 30 days/most recent month 0-30 days
Last month 30-60 days ago
Specified date such as "March 15" Specific date

Time period impact:

  • File names include a time period identifier
  • Use time filter parameters when searching
  • Consider time relevance when filtering content

Step 2: Multi-dimensional information collection

Timeout limit: 120 seconds

Fetch information sources by priority, with each source having its own timeout:

Priority 1 - Must succeed (choose 2-3):

  • Hacker News or Reddit
  • arXiv paper list

Priority 2 - Try to succeed (choose 2-3):

  • The Verge AI
  • VentureBeat
  • 36kr AI

Priority 3 - Optional (choose 1-2):

  • Papers with Code
  • GitHub Trending

Execution rules:

  1. Each web_fetch call sets maxChars: 8000 to limit content
  2. Each web_fetch implicitly has a 10-second timeout
  3. If a single source does not respond within 15 seconds, skip it
  4. After 120 seconds, force termination and enter the organization phase

Example:

// 并行抓取多个源，设置超时
await Promise.race([
  web_fetch({ url: "...", maxChars: 8000 }),
  new Promise(r => setTimeout(() => r(null), 15000)) // 15秒超时
])

Step 3: Filtering and organization

  • Filter content according to the specified domain and item count
  • When the collected content is insufficient, lower the item count requirement
  • Ensure that news, papers, and tools are all represented

Step 4: AI domain knowledge summary

Core principle: Summarize based on AI domain knowledge; do not directly translate the original text

Summary writing requirements:

  • Write in Chinese, but do not machine translate
  • Interpret by combining AI domain background knowledge
  • Extract technical points, innovations, and significance
  • Explain the position and impact of the content in AI development
  • If it involves a paper, explain the innovative method, experimental results, and applicable scenarios
  • If it involves a tool, explain its features, applicable scenarios, and comparison with competitors

Each piece of content includes:

  • Title
  • Summary (100 words or more, incorporating analysis of the AI field)
  • Source link
  • Publication date

Step 5: Overall Summary

Includes:

  • Core trends (3-5 items) - interpreted from the perspective of the AI field
  • Issues that need attention
  • One-sentence summary

Step 6: Generate a multi-format report

Timeout limit: 60 seconds

Save to:~/.openclaw/workspace/ai-reports/

  1. Markdown version (.md) - save directly
  2. HTML version (.html) - rendered and beautified HTML
  3. PDF version (.pdf) - converted via HTML (optional; failure does not affect the main process)

Call the script to convert formats:

python3 ~/.openclaw/workspace/skills/ai-daily-report/scripts/md_converter.py \
  ~/.openclaw/workspace/ai-reports/AI-报告-{领域}-{时间段}-{日期}.md \
  ~/.openclaw/workspace/ai-reports/

Filename format:

  • Today's report:AI-报告-LLM-今日-2026-03-18.md
  • This week's report:AI-报告-LLM-本周-2026-03-18.md
  • This month's report:AI-报告-LLM-本月-2026-03-18.md

PDF conversion failure does not affect report generation, HTML is sufficient for reading.

Output format (Markdown version)

# 🔥 AI领域热门速递（{领域}）— {时间段} YYYY年M月D日

> 收集领域：{领域} | 收集时间段：{时间段} | 收集条数：{N}条

## 📰 {N}条热门新闻

### 1. [标题]
> [基于AI领域知识的摘要，100字以上...]
> 📅 发布日期 | 🔗 [来源](URL)

### 2. [标题]
...

## 📄 {N}条热门论文

### 1. [论文标题]
> [AI领域视角的论文解读，包括创新方法、实验结果等，100字以上]...
> 📅 arXiv ID | 🔗 [来源](URL)

## 🛠️ {N}条热门技术/工具

### 1. [技术/工具名称]
> [功能特点、适用场景、竞品对比，100字以上]...
> 🔗 [项目地址](URL)

## 📊 今日AI领域总体总结

### 【核心趋势】
1. [从AI领域视角解读]
2. ...

### 【需要关注的问题】
- ...

### 【一句话总结】
...

HTML template

Script md_converter.py has a built-in beautiful HTML template, including:

  • Gradient purple theme header
  • Card-style content layout
  • Responsive design (adapts to mobile/computer)
  • Categorized display of news/papers/tools
  • Print-friendly styles

PDF generation

The script will automatically detect and install PDF conversion dependencies:

  1. Automatic installation - The script will automatically try to install the following Python libraries (in priority order):

    • weasyprint ⭐ Recommended, best results
    • pdfkit
    • xhtml2pdf
    • fpdf2 - Pure Python, simplified styling

  2. No manual installation required - On first run, missing libraries will be downloaded automatically

  3. Fallback option - If all installations fail, it will prompt the user to print to PDF using a browser

Notes

  • Summaries must be based on knowledge of the AI field; direct translation is prohibited
  • Each summary must be at least 100 words
  • The source URL must be attached
  • The overall summary must include in-depth analysis
  • Write in Chinese
  • Field names need to be standardized (e.g., "large model" → "LLM")
  • After saving the report, show the file path to the user
  • A timeout must be set during execution to prevent freezing
  • PDF generation failure does not affect the HTML report
  • The time period defaults to "Today" and can be adjusted according to user needs