Web Search

Purpose

Guide the agent to break down complex search queries into focused sub-queries, execute systematic web searches for each, and synthesize comprehensive answers.

When to Use

Load this skill when:

• User explicitly requests online search or information lookup
• Agent encounters unknown information during task execution
• Agent needs to verify potentially incorrect understanding
• Current knowledge may be outdated or incomplete
• Technical details, APIs, or documentation need verification

Core Approach

1. Analyze the Query

Examine the search query to determine if it contains multiple aspects based on 5W1H framework:

• What: What is X?
• Why: Why does X happen?
• When: When did/will X occur?
• Where: Where is X located/used?
• Who: Who created/uses X?
• How: How does X work?

If the query contains only one 5W1H aspect, proceed directly with a single search.

If the query contains multiple 5W1H aspects, break it down into focused sub-queries.

2. Break Down Complex Queries

For queries with multiple aspects, decompose them using 5W1H:

Example:

• Original: "What is FastAPI and why is it popular?"
• Sub-query 1 (What): "What is FastAPI?"
• Sub-query 2 (Why): "Why is FastAPI popular?"

Example:

• Original: "History of LLM agents and future trends"
• Sub-query 1 (When - past): "History of LLM agents"
• Sub-query 2 (When - future): "Future trends of LLM agents"

Principles:

• Each sub-query should focus on one 5W1H aspect
• Keep sub-queries clear and specific
• Maintain the original language and context

3. Execute Searches

For each sub-query (or the single query if no breakdown needed):

Use subagent tool:
- subrecipe: "web-search"
- parameters: {"search_query": "[sub-query]"}
- summary: true

4. Synthesize Results

After collecting all search results, synthesize a comprehensive answer that:

• Addresses the original query completely
• Integrates information from all sub-queries
• Maintains coherence and flow
• Preserves important details and nuances
• Cites sources when relevant

The final output should be a direct answer to the original query.

Key Principles

• Smart decomposition: Only break down queries that genuinely have multiple aspects
• Focused searches: Each sub-query targets one specific aspect
• Comprehensive synthesis: Combine results into a unified answer
• Maintain context: Preserve the original question's intent during breakdown

Notes

• This is a guidance skill, not a rigid workflow
• Use judgment to determine if breakdown is beneficial
• Simple, focused queries don't need decomposition
• The goal is to answer the user's original question effectively