| name | human-search |
| description | Comprehensive human-like web search and scraping meta-skill. Mimics human behavior to bypass blocks/rate-limits. Auto-installs dependencies, auto-repairs failures, cascades through 5 tiers: (1) Browser CLI (agent-browser/playwright), (2) Scraping APIs (crawl4ai/jina.ai), (3) Python scraper (unlimited), (4) Quick extract (searxng/tavily), (5) Discovery search. Includes Yandex search. For Telegram login: telegram-scraper. Use for any "search", "scrape", "extract", "find articles" request.
|
| compatibility | Auto-installs Python (crawl4ai, requests, bs4, lxml, markdownify, tiktoken), npm (agent-browser, playwright-cli), fixes Docker. Yandex via direct curl. |
| allowed-tools | Bash(pip *), Bash(npm *), Bash(docker *), Bash(curl *), Bash(python *), Read, Glob, Grep, Write, Edit |
Human Search
Robust meta-skill for human-like web search/scraping. Auto-fixes issues, bypasses blocks, cascades intelligently. No API keys required (free, unlimited).
Philosophy
- Human-like — Avoids bot detection
- Self-healing — Auto-installs, restarts, fixes configs
- Token-efficient — Max content per call
- Cascade — Tier 1 → Tier 5 fallback
- Yandex — RU-specific search included
Tier Cascade
Tier 0: jina.ai/reader → Free, instant markdown (no install)
Tier 1: Browser CLI → agent-browser/playwright (JS, human-like)
Tier 2: Scraping APIs → crawl4ai (Playwright + readability)
Tier 3: Python Scraper → BS4 + markdownify (unlimited)
Tier 4: Quick Extract → searxng/tavily (2500-4000 chars)
Tier 5: Discovery → searxng-search/Yandex/websearch
Auto-Setup (Run First)
python references/test_deps.py
Tier 0: jina.ai/reader (Instant, Free)
curl https://r.jina.ai/https://example.com | head -500
When: Simple articles, quickest test.
Tier 1: Browser CLI (Human-like, JS)
Auto-install:
npm install -g agent-browser playwright-cli
Test:
agent-browser --version
npx playwright --version
Usage:
agent-browser open https://example.com && agent-browser snapshot -i && agent-browser get text body
npx playwright-cli open https://example.com && npx playwright-cli snapshot
Fallback: If blocked → Tier 2
Tier 2: crawl4ai (Anti-bot, JS)
Auto-install:
pip install crawl4ai
Test:
python -c "from crawl4ai import AsyncWebCrawler; print('OK')"
Usage:
from crawl4ai import AsyncWebCrawler
crawler = AsyncWebCrawler()
result = crawler.arun('https://example.com')
print(result.markdown)
Fallback: Heavy sites → Tier 3
Tier 3: Python Scraper (Unlimited)
Auto-install:
pip install requests beautifulsoup4 lxml markdownify tiktoken
Quick scrape:
python references/quick_scrape.py https://example.com
Batch (sitemap):
python references/cmwlab_crawl4ai_ingest.py
Fallback: Rate-limited → Tier 4
Tier 4: Quick Extract
Docker check/fix:
curl -s http://localhost:8000/health || (cd D:/Repo/searxng-docker-tavily-adapter && docker compose up -d)
docker restart searxng
Usage:
curl -s -X POST http://localhost:8000/search -H "Content-Type: application/json" -d '{"query": "site:example.com", "max_results": 1, "include_raw_content": true}'
Fallback: Empty → Tier 5
Tier 5: Discovery Search (Yandex + Fallbacks)
Yandex (RU-specific):
curl "https://yandex.com/search/?text=QUERY&lr=213" | grep -o 'https://[^"]*'
SearXNG:
curl -s -X POST http://localhost:8000/search -d '{"query": "QUERY", "max_results": 5}'
Websearch fallback: Native tool.
Auto-Repair Commands
Docker Fix:
docker restart searxng tavily-adapter
docker logs searxng --tail 20 | grep ERROR
Config Fix (rate-limits):
docker restart searxng
Python Fix:
pip install requests beautifulsoup4 lxml markdownify tiktoken crawl4ai --upgrade
Node Fix:
npm install -g agent-browser playwright-cli
Full Test:
python references/test_deps.py
Telegram Note
Public t.me: Use Tier 1-3. Private login: telegram-scraper
See Also