AgentSkillsCN

parallel-web-search

使用 Parallel AI 执行代理式网络搜索。当用户需要获取最新的网络信息、进行研究、事实核查、查阅新闻,或获取超出训练截止日期的实时数据时,可选用此方法。

SKILL.md
--- frontmatter
name: parallel-web-search
description: Performs agentic web search using Parallel AI. Use when user needs current web information, research, fact-checking, news, or real-time data beyond training cutoff.
allowed-tools: Bash, Read

Parallel Web Search

Agentic web search optimized for LLM workflows using Parallel AI API.

When to Use

  • User needs current/real-time information
  • User asks about recent events or news
  • User needs to fact-check or verify claims
  • User wants to research a topic with citations
  • User needs information beyond training data cutoff

Requirements

Environment variable must be set:

  • PARALLEL_API_KEY - Parallel AI API key

Command

bash
./scripts/parallel_search.py search -o "objective" [-q "query"] [-n limit]

Options

OptionDescription
-o, --objectiveNatural language search goal (required)
-q, --queryAdditional keyword queries (can repeat)
-n, --limitMax results 1-20 (default: 10)
-c, --max-charsMax chars per excerpt (default: 500)
-d, --domainRestrict to domains (can repeat)
-f, --formatOutput: json (default) or table

Output Format

Default JSON for LLM parsing:

json
{
  "objective": "Find recent AI regulation news",
  "queries": ["AI regulation 2024", "EU AI Act"],
  "results": [
    {
      "title": "EU AI Act Implementation Timeline",
      "url": "https://example.com/article",
      "excerpt": "The European Union's AI Act...",
      "publish_date": "2024-12-15"
    }
  ]
}

Examples

Basic search:

bash
./scripts/parallel_search.py search -o "What are the latest developments in fusion energy?"

With keyword queries (improves results):

bash
./scripts/parallel_search.py search \
  -o "Recent breakthroughs in quantum computing" \
  -q "quantum computing 2024" \
  -q "quantum supremacy"

Restrict to specific domains:

bash
./scripts/parallel_search.py search \
  -o "Climate change research findings" \
  -d "nature.com" \
  -d "science.org" \
  -n 5

Human-readable table output:

bash
./scripts/parallel_search.py search -o "AI safety news" -f table

Best Practices

  1. Use both objective AND queries - Objective provides context, queries ensure keyword coverage
  2. Be specific - Include timeframes, sources preferences, or content types
  3. Limit results for efficiency - Use -n 5 for quick lookups
  4. Domain filtering - Use -d for authoritative sources on specific topics