docs: 重构 SKILL.md 为 Agent 决策驱动,更新文档版本号及 CLI docstring

This commit is contained in:
vex1023 committed 2026-05-14 18:30:42 +08:00
1 parent 39c01ebf69
commit a81661d828
9 files changed
+932 -183

No files matched your search

+100 -27
View File
@@ -1,6 +1,6 @@
# Web Search Skills
Multi-source web search CLI tool — search across 20+ engines from your terminal.
Multi-source web search CLI tool — search across **28 engines** from your terminal: web search, news, WeChat articles, academic papers, social media, RSS feeds, and AI-powered Baidu search.
## Quick Start
@@ -26,28 +26,68 @@ python ws.py wechat 蒙面财经
web-search <command> <query> [options]
```
| Command | Description |
|------------|------------------------------------------|
| `search` | Search all sources (default) |
| `web` | Web search engines (Google, Baidu, DDG) |
| `news` | News (财联社, 华尔街见闻, RSS) |
| `wechat` | WeChat public account articles |
| `academic` | Academic papers (ArXiv) |
| `social` | Social media (Twitter/X) |
| `rss` | RSS news feeds |
| `sources` | List all available search sources |
| `urls` | Generate search URLs without executing |
| Command | Description |
|------------|-----------------------------------------------------|
| `search` | Search 28 sources in parallel (default) |
| `web` | Web search engines (Google, Baidu, DDG, Yahoo…) |
| `news` | News (Hacker News, GitHub Trending, 36氪, RSS…) |
| `baidu` | AI-powered Baidu search (site/recency filters) |
| `wechat` | WeChat public account articles (via Sogou) |
| `academic` | Academic papers (ArXiv) |
| `social` | Social media (Twitter/X) |
| `rss` | RSS feeds (AI newsletters, essays, podcasts…) |
| `health` | Test connectivity of all 28 engines (concurrent) |
| `sources` | List all available search sources |
| `urls` | Generate search URLs without executing |
Options:
### Options
| Flag | Description |
|------|-------------|
| `-n N` | Max results per source (default: 10) |
| `-j` | JSON output |
| `-o FILE` | Save to JSON file |
| `-s NAME` | Specific source name(s) |
| `-r cn/global` | Region filter (web search) |
| `-v` | Verbose/debug logs |
| Flag | Description |
|-------------------|----------------------------------------------|
| `-n N` | Max results per source (default: 10) |
| `-j` | JSON output |
| `-o FILE` | Save to JSON file |
| `-s NAME [...]` | Specific source(s) — engine names or categories (e.g. `-s "Hacker News" V2EX`, `-s web news`) |
| `-r cn/global` | Region filter (web search) |
| `--site SITE` | Restrict to site (baidu command only) |
| `--recency RECENCY` | Time filter: day/week/month/semiyear/year |
| `-v` | Verbose/debug logs |
### Examples
```bash
# Search specific engines concurrently (new!)
web-search search "technology" -s "Hacker News" "V2EX" "GitHub Trending" -n 3
# Search two categories concurrently (new!)
web-search search "AI" -s web news -n 3
# Baidu with site & recency filters
web-search baidu "中美元首会晤" --site mp.weixin.qq.com --recency week
# Quick search
web-search news "人工智能" -n 5
web-search wechat 蒙面财经
web-search academic "quantum computing"
web-search search "latest AI news" -o results.json
# Check engine health (concurrent, ~14s for 28 engines)
web-search health
```
## Sources (28 engines)
| Category | Sources | Count |
|----------|---------|-------|
| **Web** | Baidu, Bing CN, Bing INT, 360, Sogou, Shenma, Google, Google HK, DuckDuckGo, Yahoo, Startpage, Brave, Ecosia, Qwant | 14 |
| **News (CN)** | 财联社, 华尔街见闻, 36氪, 微博热搜, V2EX, 腾讯新闻 | 6 |
| **News (Global)** | Hacker News, GitHub Trending | 2 |
| **RSS** | Reuters, AP, BBC, FT, CNN, Al Jazeera + AI newsletters (Latent Space, ChinAI, KDnuggets…), essays (Paul Graham, James Clear…), podcasts (Lex Fridman…) | 18 |
| **Academic** | ArXiv | 1 |
| **WeChat** | Sogou WeChat | 1 |
| **Social** | Twitter, Twitter Latest | 2 |
| **API** | Baidu Qianfan (AI search with site/recency) | 1 |
| **Special** | WolframAlpha | 1 |
## Project Structure
@@ -55,23 +95,26 @@ Options:
web-search-skills/
├── scripts/
│ ├── cli.py # CLI entry point
│ ├── unified_search.py # Unified search orchestrator
│ ├── unified_search.py # Unified search orchestrator (async + concurrent)
│ ├── core/
│ │ ├── base.py # SearchResult & SearchEngine base class
│ │ └── config_loader.py # Load engine config from JSON
│ │ ├── config_loader.py # Load engine config from JSON
│ │ └── thread_utils.py # Shared thread pool for HTML parsing
│ ├── engines/
│ │ ├── url_engine.py # URL-template search engines
│ │ └── parser_engines.py # HTML-parsed search (DuckDuckGo)
│ │ ├── parser_engines.py # HTML-parsed search (DuckDuckGo)
│ │ └── baidu_qianfan.py # Baidu Qianfan AI Search API
│ ├── news/
│ │ ├── cls.py # 财联社
│ │ ├── wallstreetcn.py # 华尔街见闻
│ │ └── rss.py # RSS feeds
│ │ ├── rss.py # RSS feeds (18 sources)
│ │ └── tech_news.py # Tech news engines (Hacker News, GitHub Trending, 36氪, etc.)
│ ├── academic/
│ │ └── arxiv.py # ArXiv papers
│ ├── wechat/
│ │ └── sogou_weixin.py # WeChat (via Sogou)
│ │ └── sogou_weixin.py # Sogou WeChat search
│ └── social/
│ └── twitter.py # Twitter/X
│ └── twitter.py # Twitter/X search
├── config/
│ └── engines.json # Search engine definitions
├── ws.py # Direct run entry point
@@ -81,6 +124,36 @@ web-search-skills/
└── README.md
```
## Python API
```python
from scripts.unified_search import UnifiedSearch
async def main():
us = UnifiedSearch()
# Search specific sources concurrently
results = await us.search("quantum computing",
sources=["Hacker News", "ArXiv"],
max_results=5)
for r in results:
print(f"[{r.rank}] {r.title}: {r.url}")
await us.close()
```
## Performance
- **Concurrent engine execution**: All 28 engines run in parallel via `asyncio.gather`, with `Semaphore(5)` for gentle concurrency
- **Health check**: 28 engines checked in ~14 seconds (vs ~84s without concurrency)
- **Thread pool**: HTML parsing offloaded to `ThreadPoolExecutor` via `scripts/core/thread_utils.py`, preventing event loop blocking
- **Timeout**: 15s per engine timeout prevents hanging
## Notes
- Sogou WeChat has anti-scraping limits, may return empty results
- 财联社 and 华尔街见闻 require JavaScript rendering; `UnifiedSearch` returns search page links for these
- RSS feeds require stable network connectivity
- Baidu Qianfan requires `BAIDU_API_KEY` environment variable
## License
MIT
+7 -1
View File
@@ -23,7 +23,13 @@
],
"news_sources": [
{"name": "cls", "url": "https://www.cls.cn/searchPage?keyword={keyword}&type=all", "region": "cn", "type": "news"},
{"name": "wallstreetcn", "url": "https://wallstreetcn.com/search?q={keyword}", "region": "cn", "type": "news"}
{"name": "wallstreetcn", "url": "https://wallstreetcn.com/search?q={keyword}", "region": "cn", "type": "news"},
{"name": "Hacker News", "type": "news", "region": "global"},
{"name": "GitHub Trending", "type": "news", "region": "global"},
{"name": "36氪", "type": "news", "region": "cn"},
{"name": "微博热搜", "type": "news", "region": "cn"},
{"name": "V2EX", "type": "news", "region": "cn"},
{"name": "腾讯新闻", "type": "news", "region": "cn"}
],
"academic_sources": [
{"name": "arxiv", "url": "https://export.arxiv.org/api/query?search_query=all:{keyword}&max_results=10", "region": "global", "type": "api"}
+1 -1
View File
@@ -1,6 +1,6 @@
[project]
name = "web-search-skills"
version = "2.1.0"
version = "2.2.0"
description = "Multi-source web search CLI tool"
requires-python = ">=3.8"
dependencies = [
+27 -20
View File
@@ -1,41 +1,45 @@
#!/usr/bin/env python3
"""
Web Search Skills CLI — search across 20+ sources from the command line.
Web Search Skills CLI — search across 28 sources from the command line.
Usage:
web-search <command> <query> [options]
Commands:
search Search all sources (default)
web Search web search engines (Google, Baidu, DuckDuckGo, etc.)
baidu Search via Baidu Qianfan AI Search API (需配置 BAIDU_API_KEY)
news Search news (财联社, 华尔街见闻, RSS)
wechat Search WeChat public account articles (via Sougou)
academic Search academic papers (ArXiv)
social Search social media (Twitter/X)
rss Search RSS news feeds
health Test engine connectivity
search Search all 28 sources in parallel (default)
web Web search engines (Google, Baidu, DuckDuckGo, etc.)
baidu Baidu Qianfan AI Search API (需配置 BAIDU_API_KEY)
news News (Hacker News, GitHub Trending, 36氪, RSS, etc.)
wechat WeChat public account articles (via Sogou)
academic Academic papers (ArXiv)
social Social media (Twitter/X)
rss RSS feeds (AI newsletters, essays, podcasts)
health Test connectivity of all 28 engines (concurrent)
sources List all available sources
urls Get search URLs without executing
urls Generate search URLs without executing
Options:
-n, --max-results N Max results per source (default: 10)
-j, --json Output as JSON
-o, --output FILE Save results to file
-s, --source NAME Specific source name(s)
-r, --region REGION Region filter (cn/global)
--site SITE Restrict to specific site (baidu command only, e.g. mp.weixin.qq.com)
--recency RECENCY Time filter: day, week, month, semiyear, year (baidu command only)
-s, --source NAME Specific source name(s), support multiple
-r, --region REGION Region filter: cn / global (web only)
--site SITE Restrict to site (baidu only, e.g. mp.weixin.qq.com)
--recency RECENCY Time filter: day/week/month/semiyear/year (baidu only)
-v, --verbose Show debug logs
-h, --help Show this help
Examples:
web-search wechat 蒙面财经
web-search baidu 今日新闻 --recency week
web-search baidu "央视新闻 中美元首会晤" --site mp.weixin.qq.com --recency week
web-search news "人工智能" -n 5
web-search web "Python async" -r global -j
web-search search "量子计算" -o results.json
web-search search "AI" -s web news -n 5
web-search search "technology" -s "Hacker News" "V2EX" -n 3
web-search web "Python async" -r global -j
web-search baidu "今日新闻" --recency week
web-search baidu "中美元首会晤" --site mp.weixin.qq.com
web-search news "人工智能" -n 5
web-search wechat 蒙面财经
web-search academic "quantum computing"
web-search social "Iran nuclear" -n 5
web-search health
web-search sources
web-search urls "机器学习"
@@ -377,6 +381,9 @@ async def _run_search(args):
kwargs["site"] = args.site
if args.recency:
kwargs["recency"] = args.recency
# --source / -s: pass specific source names/categories to search
if args.source:
kwargs["sources"] = args.source
try:
results = await search_fn(
+52
View File
@@ -0,0 +1,52 @@
"""
Thread pool utilities for CPU-bound operations.
BeautifulSoup/HTML parsing is CPU-bound and can block the async event loop.
Use this shared thread pool to offload parsing work to background threads.
"""
from __future__ import annotations
import logging
from concurrent.futures import ThreadPoolExecutor
from typing import Callable, TypeVar
logger = logging.getLogger(__name__)
T = TypeVar("T")
# Shared thread pool - max_workers=4 balances parallelism with overhead
_PARSE_POOL: ThreadPoolExecutor | None = None
_MAX_WORKERS = 4
def get_parse_pool() -> ThreadPoolExecutor:
"""Get the shared thread pool for parsing operations."""
global _PARSE_POOL
if _PARSE_POOL is None:
_PARSE_POOL = ThreadPoolExecutor(
max_workers=_MAX_WORKERS,
thread_name_prefix="parse",
)
return _PARSE_POOL
async def run_in_thread(func: Callable[..., T], *args, **kwargs) -> T:
"""
Run a CPU-bound function in a thread pool to avoid blocking the event loop.
Usage:
soup = await run_in_thread(BeautifulSoup, html, "lxml")
results = await run_in_thread(parse_func, html, max_results)
"""
import asyncio
loop = asyncio.get_running_loop()
return await loop.run_in_executor(get_parse_pool(), func, *args, **kwargs)
def shutdown_pool():
"""Clean up the thread pool (call during application shutdown)."""
global _PARSE_POOL
if _PARSE_POOL:
_PARSE_POOL.shutdown(wait=True)
_PARSE_POOL = None
+80 -1
View File
@@ -27,7 +27,8 @@ from scripts.core.base import SearchEngine, SearchResult
logger = logging.getLogger(__name__)
# RSS feed definitions (from news-summary skill + additional sources)
# RSS feed definitions
# Core sources (general news) + AI newsletters + essays + podcasts
RSS_FEEDS = {
"BBC World": {
"url": "https://feeds.bbci.co.uk/news/world/rss.xml",
@@ -64,6 +65,84 @@ RSS_FEEDS = {
"category": "news",
"lang": "en",
},
# -----------------------------------------------------------------------
# AI Newsletters (inspired by news-aggregator-skill)
# -----------------------------------------------------------------------
"Latent Space": {
"url": "https://feeds.transistor.fm/latent-space-podcast",
"category": "technology",
"lang": "en",
"subcategory": "ai_newsletter",
},
"ChinAI": {
"url": "https://chinai.substack.com/feed",
"category": "technology",
"lang": "en",
"subcategory": "ai_newsletter",
},
"Memia": {
"url": "https://memia.substack.com/feed",
"category": "technology",
"lang": "en",
"subcategory": "ai_newsletter",
},
"Interconnects": {
"url": "https://interconnects.substack.com/feed",
"category": "technology",
"lang": "en",
"subcategory": "ai_newsletter",
},
"KDnuggets": {
"url": "https://www.kdnuggets.com/feed",
"category": "technology",
"lang": "en",
"subcategory": "ai_newsletter",
},
# -----------------------------------------------------------------------
# Deep Essays & Thinkers (inspired by news-aggregator-skill)
# -----------------------------------------------------------------------
"Paul Graham": {
"url": "https://paulgraham.com/rss.xml",
"category": "technology",
"lang": "en",
"subcategory": "essay",
},
"James Clear": {
"url": "https://jamesclear.com/feed",
"category": "news",
"lang": "en",
"subcategory": "essay",
},
"Wait But Why": {
"url": "https://waitbutwhy.com/feed",
"category": "news",
"lang": "en",
"subcategory": "essay",
},
"Farnam Street": {
"url": "https://fs.blog/feed/",
"category": "news",
"lang": "en",
"subcategory": "essay",
},
# -----------------------------------------------------------------------
# Podcasts (inspired by news-aggregator-skill)
# -----------------------------------------------------------------------
"Lex Fridman Podcast": {
"url": "https://lexfridman.com/feed/podcast/",
"category": "technology",
"lang": "en",
"subcategory": "podcast",
},
"80,000 Hours": {
"url": "https://80000hours.org/feed/podcast/",
"category": "news",
"lang": "en",
"subcategory": "podcast",
},
}
+461
View File
@@ -0,0 +1,461 @@
"""
Tech & social news aggregation engines.
Implements non-Playwright news sources inspired by news-aggregator-skill:
- Hacker News (via Algolia API)
- GitHub Trending (HTML scrape)
- 36氪 (HTML scrape)
- 微博热搜 (Ajax API)
- V2EX (Public API)
- 腾讯新闻 (API)
"""
from __future__ import annotations
import logging
import re
from datetime import datetime, timezone
from typing import Any, Optional
from urllib.parse import quote_plus
import httpx
from bs4 import BeautifulSoup
from scripts.core.base import SearchEngine, SearchResult
from scripts.core.thread_utils import run_in_thread
logger = logging.getLogger(__name__)
# Shared HTTP client settings
_DEFAULT_TIMEOUT = 15
_COMMON_HEADERS = {
"User-Agent": (
"Mozilla/5.0 (Windows NT 10.0; Win64; x64) "
"AppleWebKit/537.36 (KHTML, like Gecko) "
"Chrome/120.0.0.0 Safari/537.36"
),
"Accept": "text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8",
"Accept-Language": "zh-CN,zh;q=0.9,en;q=0.8",
}
# ---------------------------------------------------------------------------
# Hacker News (via Algolia API)
# ---------------------------------------------------------------------------
class HackerNewsEngine(SearchEngine):
"""Hacker News search via Algolia public API."""
SEARCH_URL = "https://hn.algolia.com/api/v1/search"
def __init__(self, config: Optional[dict] = None):
super().__init__("Hacker News", config)
self._client = httpx.AsyncClient(
timeout=_DEFAULT_TIMEOUT,
headers={"User-Agent": _COMMON_HEADERS["User-Agent"]},
follow_redirects=True,
)
async def search(self, query: str, max_results: int = 10, **kwargs) -> list[SearchResult]:
try:
params = {
"query": query,
"hitsPerPage": max_results,
"tags": "story",
}
response = await self._client.get(self.SEARCH_URL, params=params)
response.raise_for_status()
data = response.json()
return self._parse_results(data, max_results)
except Exception as e:
logger.error(f"Hacker News search failed: {e}")
return []
def _parse_results(self, data: dict, max_results: int) -> list[SearchResult]:
results = []
hits = data.get("hits", [])[:max_results]
for i, hit in enumerate(hits):
title = hit.get("title", "")
url = hit.get("url") or f"https://news.ycombinator.com/item?id={hit.get('objectID', '')}"
snippet = hit.get("story_text", "") or ""
# Clean HTML from snippet
snippet = re.sub(r"<[^>]+>", "", snippet)[:200]
points = hit.get("points", 0)
author = hit.get("author", "")
created_at = hit.get("created_at", "")
results.append(SearchResult(
title=title,
url=url,
snippet=snippet or f"[{points} points] by {author}",
source=self.name,
rank=i + 1,
category="news",
extra={
"points": points,
"author": author,
"time": created_at,
"heat": f"{points} points",
},
))
return results
async def close(self):
await self._client.aclose()
# ---------------------------------------------------------------------------
# GitHub Trending
# ---------------------------------------------------------------------------
class GitHubTrendingEngine(SearchEngine):
"""GitHub Trending repositories."""
TRENDING_URL = "https://github.com/trending"
def __init__(self, config: Optional[dict] = None):
super().__init__("GitHub Trending", config)
self._client = httpx.AsyncClient(
timeout=_DEFAULT_TIMEOUT,
headers=_COMMON_HEADERS,
follow_redirects=True,
)
async def search(self, query: str = "", max_results: int = 10, **kwargs) -> list[SearchResult]:
"""GitHub Trending doesn't support keyword search; returns top trending repos."""
url = self.TRENDING_URL
if query:
url += f"?spoken_language_code=&q={quote_plus(query)}"
try:
response = await self._client.get(url)
response.raise_for_status()
return await run_in_thread(self._parse_html, response.text, max_results)
except Exception as e:
logger.error(f"GitHub Trending failed: {e}")
return []
def _parse_html(self, html: str, max_results: int) -> list[SearchResult]:
soup = BeautifulSoup(html, "lxml")
results = []
articles = soup.select("article.Box-row")[:max_results]
for i, article in enumerate(articles):
h2 = article.select_one("h2")
if not h2:
continue
repo_path = h2.get_text(strip=True).replace(" ", "")
repo_url = f"https://github.com/{repo_path}"
desc_el = article.select_one("p")
description = desc_el.get_text(strip=True) if desc_el else ""
# Language
lang_el = article.select_one("[itemprop='programmingLanguage']")
language = lang_el.get_text(strip=True) if lang_el else ""
# Stars today
stars_el = article.select_one(".float-sm-right")
stars_today = stars_el.get_text(strip=True) if stars_el else ""
results.append(SearchResult(
title=repo_path,
url=repo_url,
snippet=description[:200] if description else language,
source=self.name,
rank=i + 1,
category="news",
extra={
"language": language,
"heat": stars_today,
"time": "",
},
))
return results
async def close(self):
await self._client.aclose()
# ---------------------------------------------------------------------------
# 36氪 (36kr.com)
# ---------------------------------------------------------------------------
class Kr36Engine(SearchEngine):
"""36氪 tech news search."""
SEARCH_URL = "https://36kr.com/search/articles/{keyword}"
def __init__(self, config: Optional[dict] = None):
super().__init__("36氪", config)
self._client = httpx.AsyncClient(
timeout=_DEFAULT_TIMEOUT,
headers={**_COMMON_HEADERS, "Referer": "https://36kr.com/"},
follow_redirects=True,
)
async def search(self, query: str, max_results: int = 10, **kwargs) -> list[SearchResult]:
url = self.SEARCH_URL.format(keyword=quote_plus(query))
try:
response = await self._client.get(url)
response.raise_for_status()
return await run_in_thread(self._parse_html, response.text, query, max_results)
except Exception as e:
logger.error(f"36氪 search failed: {e}")
return []
def _parse_html(self, html: str, query: str, max_results: int) -> list[SearchResult]:
soup = BeautifulSoup(html, "lxml")
results = []
# Look for article cards
for i, item in enumerate(soup.select("a[href*='/p/'], .article-item, .search-result-item, [class*='article']")[:max_results]):
title_el = item.select_one("h3, h4, .title, [class*='title']")
link = item.get("href", "")
snippet_el = item.select_one(".abstract, .summary, .desc, p, [class*='desc']")
title = title_el.get_text(strip=True) if title_el else ""
snippet = snippet_el.get_text(strip=True) if snippet_el else ""
if not title and not link:
continue
full_url = link if link.startswith("http") else f"https://36kr.com{link}"
results.append(SearchResult(
title=title or f"36氪: {query}",
url=full_url,
snippet=snippet[:200] if snippet else "",
source=self.name,
rank=i + 1,
category="news",
extra={"region": "cn"},
))
if not results:
search_url = f"https://36kr.com/search/articles/{quote_plus(query)}"
results.append(SearchResult(
title=f"36氪: {query}",
url=search_url,
snippet="Search 36氪 (site may require JS)",
source=self.name,
rank=1,
category="news",
extra={"search_url": search_url, "note": "JS-rendered page"},
))
return results
async def close(self):
await self._client.aclose()
# ---------------------------------------------------------------------------
# 微博热搜 (Weibo Trending)
# ---------------------------------------------------------------------------
class WeiboEngine(SearchEngine):
"""微博热搜 trending topics."""
HOT_URL = "https://weibo.com/ajax/side/hotSearch"
def __init__(self, config: Optional[dict] = None):
super().__init__("微博热搜", config)
self._client = httpx.AsyncClient(
timeout=_DEFAULT_TIMEOUT,
headers={**_COMMON_HEADERS, "Referer": "https://weibo.com/"},
follow_redirects=True,
)
async def search(self, query: str = "", max_results: int = 10, **kwargs) -> list[SearchResult]:
try:
response = await self._client.get(self.HOT_URL)
response.raise_for_status()
data = response.json()
return self._parse_results(data, query, max_results)
except Exception as e:
logger.error(f"微博热搜 failed: {e}")
return []
def _parse_results(self, data: dict, query: str, max_results: int) -> list[SearchResult]:
results = []
realtime = data.get("data", {}).get("realtime", [])[:max_results]
for i, item in enumerate(realtime):
word = item.get("word", "")
raw_hot = item.get("raw_hot", 0)
url = f"https://s.weibo.com/weibo?q={quote_plus(word)}"
# Filter by keyword if query provided
if query and query.lower() not in word.lower():
continue
results.append(SearchResult(
title=f"#{word}#",
url=url,
snippet=f"热度: {raw_hot:,}" if raw_hot else "",
source=self.name,
rank=i + 1,
category="news",
extra={
"heat": f"{raw_hot:,}" if raw_hot else "",
"region": "cn",
},
))
return results
async def close(self):
await self._client.aclose()
# ---------------------------------------------------------------------------
# V2EX (via public API)
# ---------------------------------------------------------------------------
class V2EXEngine(SearchEngine):
"""V2EX tech community hot topics."""
HOT_API = "https://www.v2ex.com/api/v2/topics/hot"
# Fallback to older API if v2 needs auth
HOT_API_FALLBACK = "https://www.v2ex.com/api/topics/hot.json"
def __init__(self, config: Optional[dict] = None):
super().__init__("V2EX", config)
self._client = httpx.AsyncClient(
timeout=_DEFAULT_TIMEOUT,
headers=_COMMON_HEADERS,
follow_redirects=True,
)
async def search(self, query: str = "", max_results: int = 10, **kwargs) -> list[SearchResult]:
# Try v2 API first, fallback to old API
for api_url in [self.HOT_API, self.HOT_API_FALLBACK]:
try:
response = await self._client.get(api_url)
if response.status_code == 200:
data = response.json()
# v2 API returns list directly; old API also returns list
topics = data if isinstance(data, list) else data.get("data", [])
return self._parse_results(topics, query, max_results)
except Exception:
continue
logger.error("V2EX API failed (both v2 and fallback)")
return []
def _parse_results(self, topics: list, query: str, max_results: int) -> list[SearchResult]:
results = []
for i, topic in enumerate(topics[:max_results]):
title = topic.get("title", "")
node_name = ""
if "node" in topic and isinstance(topic["node"], dict):
node_name = topic["node"].get("title", "")
topic_id = topic.get("id", "")
url = f"https://www.v2ex.com/t/{topic_id}"
replies = topic.get("replies", 0)
created = topic.get("created", "")
member = ""
if "member" in topic and isinstance(topic["member"], dict):
member = topic["member"].get("username", "")
# Filter by keyword
if query and query.lower() not in title.lower():
continue
# Format time
time_str = ""
if created:
try:
dt = datetime.fromtimestamp(int(created), tz=timezone.utc)
time_str = dt.strftime("%Y-%m-%d %H:%M")
except (ValueError, OSError):
time_str = str(created)
results.append(SearchResult(
title=title,
url=url,
snippet=f"[{node_name}] by {member}" if node_name and member else f"by {member}" if member else "",
source=self.name,
rank=i + 1,
category="news",
extra={
"heat": f"{replies} replies",
"author": member,
"time": time_str,
"region": "cn",
},
))
return results
async def close(self):
await self._client.aclose()
# ---------------------------------------------------------------------------
# 腾讯新闻 (news.qq.com)
# ---------------------------------------------------------------------------
class TencentNewsEngine(SearchEngine):
"""腾讯新闻 search."""
SEARCH_URL = "https://news.qq.com/search"
def __init__(self, config: Optional[dict] = None):
super().__init__("腾讯新闻", config)
self._client = httpx.AsyncClient(
timeout=_DEFAULT_TIMEOUT,
headers={**_COMMON_HEADERS, "Referer": "https://news.qq.com/"},
follow_redirects=True,
)
async def search(self, query: str, max_results: int = 10, **kwargs) -> list[SearchResult]:
url = f"{self.SEARCH_URL}?q={quote_plus(query)}"
try:
response = await self._client.get(url)
response.raise_for_status()
return await run_in_thread(self._parse_html, response.text, query, max_results)
except Exception as e:
logger.error(f"腾讯新闻 search failed: {e}")
return []
def _parse_html(self, html: str, query: str, max_results: int) -> list[SearchResult]:
soup = BeautifulSoup(html, "lxml")
results = []
for i, item in enumerate(soup.select("a[href*='news.qq.com'], .list-item, [class*='result'], article")[:max_results]):
title_el = item.select_one("h3, h4, .title, a")
link = item.get("href", "") or (title_el.get("href", "") if title_el else "")
snippet_el = item.select_one(".summary, .desc, .abstract, p, .text")
title = title_el.get_text(strip=True) if title_el else ""
snippet = snippet_el.get_text(strip=True) if snippet_el else ""
if not title:
continue
full_url = link if link.startswith("http") else f"https:{link}" if link.startswith("//") else link
results.append(SearchResult(
title=title,
url=full_url,
snippet=snippet[:200] if snippet else "",
source=self.name,
rank=i + 1,
category="news",
extra={"region": "cn"},
))
if not results:
search_url = f"{self.SEARCH_URL}?q={quote_plus(query)}"
results.append(SearchResult(
title=f"腾讯新闻: {query}",
url=search_url,
snippet=f"Search Tencent News for '{query}'",
source=self.name,
rank=1,
category="news",
extra={"search_url": search_url, "note": "JS-rendered page"},
))
return results
async def close(self):
await self._client.aclose()
+49 -20
View File
@@ -21,6 +21,10 @@ from scripts.engines.parser_engines import DuckDuckGoEngine
from scripts.news.cls import CLSNewsEngine
from scripts.news.wallstreetcn import WallStreetCNEngine
from scripts.news.rss import RSSNewsEngine
from scripts.news.tech_news import (
HackerNewsEngine, GitHubTrendingEngine, Kr36Engine,
WeiboEngine, V2EXEngine, TencentNewsEngine,
)
from scripts.academic.arxiv import ArXivEngine
from scripts.wechat.sogou_weixin import SogouWeChatEngine
from scripts.social.twitter import TwitterSearchEngine
@@ -132,32 +136,45 @@ class UnifiedSearch:
return []
async def search_health(self) -> list[dict[str, Any]]:
"""Check health of all configured engines."""
statuses = []
seen_names = set()
for cat_key, cat_fn in SOURCE_CATEGORIES.items():
"""Check health of all configured engines (concurrent)."""
seen_names: set[str] = set()
tasks: list[asyncio.Task] = []
for cat_fn in SOURCE_CATEGORIES.values():
for engine_def in cat_fn():
name = engine_def.get("name", "?")
if name in seen_names:
continue
seen_names.add(name)
engine = self._get_engine(engine_def)
if engine is None:
statuses.append({"name": name, "status": "error", "detail": "engine_init_failed"})
continue
try:
async with self._semaphore:
await asyncio.wait_for(
engine.search("health check", max_results=1),
timeout=8,
)
statuses.append({"name": name, "status": "ok"})
except asyncio.TimeoutError:
statuses.append({"name": name, "status": "timeout"})
except Exception as e:
statuses.append({"name": name, "status": "error", "detail": str(e)[:80]})
tasks.append(asyncio.create_task(self._check_one_engine(engine_def, name)))
results = await asyncio.gather(*tasks, return_exceptions=True)
statuses: list[dict[str, Any]] = []
for r in results:
if isinstance(r, Exception):
statuses.append({"name": "?", "status": "error", "detail": str(r)[:80]})
else:
statuses.append(r)
return statuses
async def _check_one_engine(
self, engine_def: dict[str, Any], name: str,
) -> dict[str, Any]:
"""Check health of a single engine."""
engine = self._get_engine(engine_def)
if engine is None:
return {"name": name, "status": "error", "detail": "engine_init_failed"}
try:
async with self._semaphore:
await asyncio.wait_for(
engine.search("health check", max_results=1),
timeout=8,
)
return {"name": name, "status": "ok"}
except asyncio.TimeoutError:
return {"name": name, "status": "timeout"}
except Exception as e:
return {"name": name, "status": "error", "detail": str(e)[:80]}
def _get_engine(self, engine_def: dict[str, Any]):
"""Get or create an engine instance by definition."""
name = engine_def.get("name", "")
@@ -180,7 +197,19 @@ class UnifiedSearch:
engine = SogouWeChatEngine(self.config)
elif "twitter" in name.lower():
engine = TwitterSearchEngine(self.config)
elif engine_type == "news" and name not in ("cls", "wallstreetcn"):
elif name == "Hacker News":
engine = HackerNewsEngine(self.config)
elif name == "GitHub Trending":
engine = GitHubTrendingEngine(self.config)
elif name == "36氪":
engine = Kr36Engine(self.config)
elif name == "微博热搜":
engine = WeiboEngine(self.config)
elif name == "V2EX":
engine = V2EXEngine(self.config)
elif name == "腾讯新闻":
engine = TencentNewsEngine(self.config)
elif engine_type == "news" and name not in ("cls", "wallstreetcn", "Hacker News", "GitHub Trending", "36氪", "微博热搜", "V2EX", "腾讯新闻"):
engine = RSSNewsEngine(self.config)
elif engine_type == "social":
engine = TwitterSearchEngine(self.config) if "twitter" in name.lower() else UrlSearchEngine(name, engine_def)
+155 -113
View File
@@ -1,144 +1,186 @@
---
name: web-search-skills
description: "多源搜索 CLI 工具,支持 20+ 搜索引擎。搜索微信公众号文章、新闻资讯、学术论文、社交媒体的综合搜索技能。当用户需要搜索网络信息、查找微信公众号文章、搜索财经新闻、查询学术论文时使用。"
version: 2.1.0
description: "Multi-source web search CLI & library, covering web, news, WeChat, academic, social, and RSS feeds across 28 engines. Use when the user asks to search the web, find WeChat articles, look up news, search academic papers, or check social media."
version: 2.2.0
license: MIT
compatibility: Requires Python 3.8+, httpx, beautifulsoup4, lxml
metadata:
author: openbot-coder
category: development
tags: [search, web, wechat, news, academic, cli]
tags: [search, web, wechat, news, academic, social, rss, cli, concurrent]
---
# Web Search Skills
## 概述
## Overview
Multi-source web search CLI and library supporting 20+ search engines, including web search, news (财联社, 华尔街见闻), WeChat articles, academic papers (ArXiv), social media (Twitter/X), and RSS feeds.
A CLI (`web-search`) and Python library that queries **28 search engines** concurrently via `asyncio.gather`. Supports web search, news (Hacker News, GitHub Trending, 36氪, 微博热搜, V2EX, RSS feeds), WeChat articles, ArXiv papers, Twitter/X, and Baidu Qianfan AI search.
## 触发词 Triggers
All engines run in parallel with `Semaphore(5)` concurrency control and 15s per-engine timeout.
中文触发词:
- "搜索微信公众号..."
- "搜一下微信公众号..."
- "查一下新闻..."
- "搜索新闻关于..."
- "搜论文..."
- "查学术资料..."
- "搜索一下..."
## Triggers
English Triggers:
- "search the web for..."
- "find information about..."
- "look up..."
- "search WeChat for..."
- "search news about..."
- Any query requiring real-time web data
When the user says any of these, use this skill:
## 安装 Installation
| Intent | Example user phrases |
|--------|---------------------|
| Web search | "搜索一下 xxx", "search the web for xxx" |
| WeChat articles | "搜微信公众号 xxx", "search WeChat for xxx" |
| News | "查一下新闻", "搜索最新的科技新闻", "search news about xxx" |
| Academic | "搜论文 xxx", "search papers about xxx", "查学术资料" |
| Social media | "看看 Twitter 上关于", "check Twitter for xxx" |
| Specific source | "看看 Hacker News 上关于", "what's trending on GitHub" |
| Hot topics | "查一下微博热搜", "check Weibo trending" |
## Agent Decision Tree
Use this flow to pick the right command:
```
用户说了什么?
│
├─ "微信公众号" / "WeChat" / 公众号名称
│ → web-search wechat <query>
│
├─ "学术" / "论文" / "paper" / "ArXiv"
│ → web-search academic <query>
│
├─ "Twitter" / "X" / "社交" / "social"
│ → web-search social <query> [-n 5]
│
├─ "新闻" / "news" / 明确指定 RSS/博文/播客
│ → web-search news <query> [-n 5]
│
├─ 明确指定了特定源名称
│ → web-search search <query> -s "源名1" "源名2" [-n 5]
│
├─ 需要多类源对比(如"新闻+社交上关于伊朗的看法")
│ → web-search search <query> -s news social [-n 10]
│
├─ 需要百度搜索 + 站点/时间过滤
│ → web-search baidu <query> [--site xxx] [--recency week]
│
├─ "健康检查" / "health" / "测试引擎"
│ → web-search health
│
├─ "列出源" / "sources" / "有哪些搜索引擎"
│ → web-search sources
│
├─ 其余通用查询
│ → web-search search <query> [-s 源1 源2...] [-n 10]
│
└─ 不确定 / 没特别要求
→ web-search search <query> -n 10
```
## Quick Command Reference
| Command | Speed | Best for |
|---------|-------|----------|
| `search` | 并发 28 源 | 通用搜索、多源对比 |
| `web` | 14 网页引擎 | 普通网页搜索 |
| `baidu` | 单引擎 | 需 site/recency 过滤 |
| `news` | ~10+ 新闻源 | 科技/财经/博客/播客 |
| `wechat` | 单引擎 | 微信公众号独家内容 |
| `academic` | 单引擎 | 学术论文 |
| `social` | 2 源 (Twitter) | 社交媒体动态 |
| `rss` | 18 订阅源 | AI 周报/深度文章/播客 |
| `health` | 28 源并发 (~14s) | 检测连通性 |
**通用选项:** `-n N`(每源结果数), `-j`(JSON), `-o FILE`(存文件), `-s NAME...`(指定源), `-v`(调试日志)
## Result Field Reference
Each `SearchResult` returned has these fields:
| Field | Type | Description |
|-------|------|-------------|
| `rank` | int | Position in results (1-based) |
| `title` | str | Article/title text |
| `url` | str | Direct URL to source |
| `snippet` | str | Summary text (first ~300 chars) |
| `source` | str | Engine name (e.g. "Hacker News") |
| `category` | str | "web", "news", "academic", "social", etc. |
| `extra` | dict | Engine-specific metadata (see below) |
### `extra` fields by source
| Source | extra fields |
|--------|-------------|
| Hacker News | `points`, `author`, `time` (ISO date) |
| GitHub Trending | `language`, `stars today` (heat) |
| WeChat | `account` (公众号名), `time` |
| Twitter | `time`, `account`, `search_type` |
| RSS feeds | `time`, `feed_url` |
| Cls / WallStreetCN | `search_url` (JS-rendered fallback) |
| Baidu Qianfan | `site`, `recency` (if filtered) |
## Error Handling
| Symptom | Likely cause | Action |
|---------|-------------|--------|
| Empty results from Hacker News/GitHub | Network issue or no matches | Retry with different query |
| Baidu Qianfan returns nothing | Missing `BAIDU_API_KEY` env | Check env var is set |
| DuckDuckGo times out (15s) | Anti-scraping / rate limit | Note to user, skip |
| WeChat empty results | Sogou anti-scraping | Inform user it may be limited |
| `web-search health` shows [TIMEOUT] | Engine unreachable from this region | Not critical, skip that engine |
| GBK encode error on Windows | Non-ASCII char in output | Already auto-fixed by `_sanitize_for_console` |
**General rule:** If a specific source fails, note it but don't fail the whole search — other sources continue concurrently.
## Usage Examples
```bash
cd web-search-skills
uv tool install -e .
# Full search (all 28 sources, concurrent)
web-search search "quantum computing" -n 10
# Specific sources only
web-search search "technology" -s "Hacker News" "V2EX" "GitHub Trending" -n 3
# Multiple categories
web-search search "Iran nuclear" -s web news social -n 5
# Baidu with site filter
web-search baidu "中美元首会晤" --site mp.weixin.qq.com --recency week
# All web engines
web-search web "Python async" -r global -j
# Quick searches
web-search news "人工智能" -n 5
web-search wechat 蒙面财经
web-search academic "machine learning"
web-search social "AI regulation" -n 5
# Concurrent health check (~14s for 28 engines)
web-search health
```
安装后 `web-search` 命令全局可用。
## 命令 Commands
| 命令 Commands | 说明 Description |
|---------------|------------------|
| `web-search search <query>` | 全源搜索 Search all sources |
| `web-search web <query>` | 网页搜索 Web search engines |
| `web-search news <query>` | 新闻搜索 News (财联社, 华尔街见闻) |
| `web-search wechat <query>` | 微信公众号搜索 WeChat articles |
| `web-search academic <query>` | 学术搜索 Academic papers (ArXiv) |
| `web-search social <query>` | 社交搜索 Social media (Twitter/X) |
| `web-search rss <query>` | RSS 源搜索 RSS feeds |
| `web-search sources` | 列出所有搜索源 List all sources |
| `web-search urls <query>` | 只生成搜索链接 Search URLs only |
### 通用选项 Options
| 选项 | 说明 |
|------|------|
| `-n N` | 每源最大结果数 (默认: 10) |
| `-j` | JSON 格式输出 |
| `-o FILE` | 保存结果到 JSON 文件 |
| `-s NAME` | 指定来源名称 |
| `-r cn/global` | 区域过滤 (web 搜索) |
| `-v` | 调试日志 |
## 项目结构 Structure
```
web-search-skills/
├── scripts/ # Python 脚本目录
│ ├── cli.py # CLI 入口
│ ├── unified_search.py # 统一搜索编排器
│ ├── core/
│ │ ├── base.py # SearchResult & SearchEngine 基类
│ │ └── config_loader.py# 从 JSON 加载搜索引擎配置
│ ├── engines/
│ │ ├── url_engine.py # URL 模板搜索引擎
│ │ └── parser_engines.py# HTML 解析搜索 (DuckDuckGo)
│ ├── news/
│ │ ├── cls.py # 财联社新闻引擎
│ │ ├── wallstreetcn.py # 华尔街见闻引擎
│ │ └── rss.py # RSS 聚合引擎
│ ├── academic/
│ │ └── arxiv.py # ArXiv 学术论文引擎
│ ├── wechat/
│ │ └── sogou_weixin.py # 搜狗微信搜索引擎
│ └── social/
│ └── twitter.py # Twitter/X 搜索引擎
├── config/
│ └── engines.json # 搜索引擎定义配置
├── ws.py # 本地运行入口
├── pyproject.toml # 包构建配置
├── SKILL.md # 本技能定义文件
├── LICENSE # MIT 许可证
└── README.md # 项目文档
```
## 配置 Configuration
所有搜索引擎定义在 `config/engines.json` 中,支持 6 类源:
| 类别 | 数量 |
|------|------|
| Web 搜索引擎 | 14 个 (Baidu, Google, DuckDuckGo 等) |
| 新闻源 | 2 个 (财联社, 华尔街见闻) |
| 学术源 | 1 个 (ArXiv) |
| 微信源 | 1 个 (搜狗微信) |
| 社交源 | 2 个 (Twitter, Twitter Latest) |
| 特殊源 | 1 个 (WolframAlpha) |
每个引擎定义格式:
```json
{"name": "Baidu", "url": "https://www.baidu.com/s?wd={keyword}", "region": "cn", "type": "url"}
```
- `{keyword}` 会被替换为搜索关键词
- `type: url` 表示只生成 URL,`type: parser` 表示会解析 HTML 提取结果
## Python API
也可在代码中直接调用:
## Python API (for programmatic use)
```python
from scripts.unified_search import UnifiedSearch
async def main():
us = UnifiedSearch()
results = await us.search("量子计算", sources=["web", "news"], max_results=5)
results = await us.search(
"quantum computing",
sources=["Hacker News", "ArXiv"],
max_results=5,
)
for r in results:
print(f"[{r.rank}] {r.title}: {r.url}")
print(f"[{r.rank}] {r.title} ({r.source})")
if r.extra:
print(f" extra: {r.extra}")
await us.close()
```
## 注意事项 Notes
## Important Notes
- 搜狗微信搜索有反爬限制,可能返回空结果
- 财联社和华尔街见闻需要 JavaScript 渲染,`UnifiedSearch` 返回搜索页面链接
- RSS 源需要稳定的网络连接
- **WeChat** via Sogou has anti-scraping limits, may return empty
- **财联社 / 华尔街见闻** need JS rendering → returns search page URL, not parsed results
- **Baidu Qianfan** requires `BAIDU_API_KEY` env var
- **RSS feeds** need stable internet
- **Windows GBK** terminals: special chars auto-replaced by `_sanitize_for_console()`
> Full user-facing documentation (detailed command reference, project structure, config guide, architecture) is in `README.md`.