OpenSkillIndex← back to search
claude-code

hot-monitor

install
source · Clone the upstream repo
git clone https://github.com/liyupi/yupi-hot-monitor
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/liyupi/yupi-hot-monitor "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/hot-monitor" ~/.claude/skills/liyupi-yupi-hot-monitor-hot-monitor && rm -rf "$T"
manifest: skills/hot-monitor/SKILL.md
source content

Hot Monitor — AI 热点监控技能

Search and analyze trending topics across 8+ sources without any server or database. Scripts handle data collection; use your own AI capabilities for analysis.

Quick Start

All scripts are in 

scripts/
. Install dependencies first:

pip install requests beautifulsoup4

Set optional env vars for Twitter (other sources need no keys):

export TWITTER_API_KEY=your_key   # optional, for Twitter search

Core Workflow

1. Understand User Intent

Determine what the user needs:

  • Broad discovery: "最近AI有什么热点" → search multiple sources with broad keywords
  • Specific tracking: "帮我关注GPT-5动态" → targeted search with specific keyword
  • Report generation: "生成今日热点报告" → full pipeline + formatted output

2. Execute Search

Run search scripts based on scope. Always combine international + Chinese sources for comprehensive coverage.

International sources (no API keys needed):

python scripts/search_web.py "AI programming" --sources bing,hackernews,duckduckgo

Chinese sources (no API keys needed):

python scripts/search_china.py "AI编程" --sources sogou,bilibili,weibo

Twitter (requires 

TWITTER_API_KEY
):

python scripts/search_twitter.py "AI programming"

All scripts output JSON to stdout. Combine results for multi-source analysis.

3. Analyze Results

After collecting search results, apply the analysis framework yourself (no external AI API needed). For each result, evaluate:

  1. Authenticity (
    isReal
    ): Is this genuine news or clickbait/rumor?
  2. Relevance (0-100): How related is this to the user's interest area?
  3. Importance (low/medium/high/urgent): How significant is this development?
  4. Summary: One-sentence Chinese summary of the core information

See references/analysis-guide.md for detailed evaluation criteria.

4. Present Results

Format output as structured report, sorted by importance. Use this template:

## 🔥 热点监控报告 — {keyword}
> 扫描时间: {timestamp} | 数据源: {sources_used}

### 🚨 紧急 (Urgent)
- **{title}** — {summary}
  来源: {source} | 相关性: {relevance}% | [原文链接]({url})

### 🔴 重要 (High)
...

### 🟡 一般 (Medium)
...

### 🟢 低优先级 (Low)
...

---
共发现 {total} 条热点，其中紧急 {urgent} 条，重要 {high} 条

Script Reference

ScriptSourcesAPI KeyOutput
search_web.py
Bing, Google, DuckDuckGo, HackerNewsNoneJSON array of 
{title, content, url, source, publishedAt?}
search_china.py
Sogou, Bilibili, WeiboNoneJSON array (same schema + engagement metrics)
search_twitter.py
Twitter/X
TWITTER_API_KEY
JSON array (same schema + author info)
generate_report.py
—NoneReads JSON from stdin, outputs Markdown report

Common Options

All search scripts support:

  • --sources
    : Comma-separated list of sources to use (default: all)
  • --limit
    : Max results per source (default: 20)
  • --json
    : Output raw JSON (default, always JSON to stdout)

Error Handling

Scripts output 

[]
(empty JSON array) on failure and print errors to stderr. Always check for empty results and inform the user which sources failed.

Advanced Patterns

Keyword Expansion

For better coverage, expand the user's keyword into variants before searching:

  • English/Chinese translations (e.g., "人工智能" ↔ "AI" ↔ "Artificial Intelligence")
  • Abbreviations (e.g., "GPT-5" → also search "GPT5", "OpenAI GPT")
  • Related terms (e.g., "Claude" → also search "Anthropic", "Claude 4")

Bilibili Account Detection

search_china.py
supports 
--detect-account
flag. If the keyword matches a Bilibili creator's name, it fetches their latest videos instead of doing a generic search:

python scripts/search_china.py "程序员鱼皮" --detect-account

Multi-keyword Batch

For monitoring multiple keywords, run searches sequentially with a 3-second delay between keywords to respect rate limits.

Reference Files