Skip to main content
skill CX Codex

web-search · claude-office-skills — Claude Code skill: what it does, how to install

Formulates targeted search queries with advanced operators, recommends specialized engines, and evaluates source credibility.

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add claude-office-skills/skills --skill web-search -a claude-code
Open on GitHub

What web-search does

This skill structures search strategies for technical and domain-specific questions. Instead of vague phrases, it builds queries using Google search operators like site:, filetype:, and date filters, moving logically from broad concepts to narrow terms. Crucially, the skill cannot execute web searches on its own; it requires you or an external tool to fetch the pages.

Once you provide the search outputs, it assesses source credibility using the CRAAP framework and a five-tier reliability scale. It also recommends specialized databases such as arXiv, Semantic Scholar, or Statista for targeted research. If the office-mcp server is configured, it exports synthesized findings using the create_docx tool.

What it helps with

  • build advanced search queries using operators like site: and filetype:
  • identify specialized search engines for academic, financial, or code research
  • evaluate source credibility and bias using the CRAAP framework
  • synthesize raw search findings into structured research notes

Example prompts for the agent

  • Build a search strategy with Google operators to find 2024 EV battery reports
  • Recommend specialized search engines and queries for quantum computing preprints
  • Evaluate the reliability of these three search results using the CRAAP test

How to install web-search

1Skills CLI (Vercel)

npx skills add claude-office-skills/skills --skill web-search -a claude-code

2Manually: put the files in place

~/.claude/skills/web-search/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/web-search/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about web-search

How to install web-search in Claude Code?

Run npx skills add claude-office-skills/skills --skill web-search -a claude-code. This copies the skill files directly into your Claude Code environment.

Does web-search run live queries on the web?

No, it cannot execute live web searches or access paywalled content directly. It designs optimized search queries and evaluates results you feed back into it.

Which MCP tools does web-search work with?

It integrates with the office-mcp server to run the create_docx tool for generating Word documents from search syntheses.

Source text

SKILL.md · MIT
---
# ═══════════════════════════════════════════════════════════════════════════════
# CLAUDE OFFICE SKILL - Web Search
# ═══════════════════════════════════════════════════════════════════════════════

name: web-search
description: "Formulate effective web search queries, analyze search results, and synthesize findings. Optimize search strategies for different types of information needs."
version: "1.0.0"
author: claude-office-skills
license: MIT

category: research
tags:
  - search
  - web
  - information-retrieval
  - query-optimization
  - research
department: Research

models:
  recommended:
    - claude-sonnet-4
    - claude-opus-4
  compatible:
    - claude-3-5-sonnet
    - gpt-4
    - gpt-4o

mcp:
  server: office-mcp
  tools:
    - create_docx

capabilities:
  - query_formulation
  - search_strategy
  - result_analysis
  - source_evaluation
  - information_synthesis

languages:
  - en
  - zh

related_skills:
  - deep-research
  - academic-search
  - news-monitor
---

# Web Search Skill

## Overview

I help you formulate effective web search queries and strategies to find the information you need. I can optimize search queries, suggest alternative approaches, evaluate sources, and synthesize findings.

**What I can do:**
- Formulate optimized search queries
- Suggest search operators and techniques
- Recommend specialized search engines
- Analyze and evaluate search results
- Identify reliable sources
- Synthesize information from multiple sources

**What I cannot do:**
- Execute actual web searches (need external tools)
- Access real-time search results
- Access paywalled content
- Guarantee source accuracy

---

## How to Use Me

### Step 1: Describe Your Information Need

Tell me:
- What you're trying to find
- Why you need this information
- What you've already tried
- Any constraints (time period, source type, language)

### Step 2: Get Search Strategy

I'll provide:
- Optimized search queries
- Alternative query formulations
- Recommended search engines/tools
- Search operators to use

### Step 3: Evaluate Results

If you share results, I can:
- Assess source reliability
- Extract key information
- Identify gaps
- Suggest follow-up searches

---

## Search Query Optimization

### Google Search Operators

| Operator | Usage | Example |
|----------|-------|---------|
| `"exact phrase"` | Exact match | `"climate change policy"` |
| `site:` | Search within site | `site:reddit.com AI tools` |
| `filetype:` | Find specific files | `filetype:pdf annual report` |
| `-word` | Exclude term | `apple -fruit` |
| `OR` | Either term | `startup OR entrepreneur` |
| `intitle:` | Word in title | `intitle:guide python` |
| `inurl:` | Word in URL | `inurl:blog marketing` |
| `before:` | Before date | `AI before:2023-01-01` |
| `after:` | After date | `ChatGPT after:2024-01-01` |
| `*` | Wildcard | `"how to * in python"` |
| `related:` | Similar sites | `related:techcrunch.com` |

### Query Formulation Techniques

#### 1. Start Broad, Then Narrow
```
Broad: electric vehicles
Narrow: electric vehicle battery technology 2024
More narrow: solid-state battery EV range comparison 2024
```

#### 2. Use Synonyms and Variations
```
Original: AI writing tools
Variations:
- artificial intelligence writing software
- AI content generator
- machine learning writing assistant
- GPT writing tool
```

#### 3. Question-Based Queries
```
How: "how to implement SSO"
What: "what is zero trust security"
Why: "why companies use kubernetes"
Best: "best practices API design"
Compare: "AWS vs Azure vs GCP comparison"
```

#### 4. Source-Specific Queries
```
Academic: site:edu OR site:ac.uk [topic]
Government: site:gov [topic]
News: [topic] site:reuters.com OR site:bbc.com
Forum: [topic] site:reddit.com OR site:stackoverflow.com
```

---

## Specialized Search Engines

| Search Engine | Best For | URL |
|---------------|----------|-----|
| Google Scholar | Academic papers | scholar.google.com |
| Semantic Scholar | AI-powered paper search | semanticscholar.org |
| PubMed | Medical/biomedical | pubmed.ncbi.nlm.nih.gov |
| arXiv | Preprints (CS, physics) | arxiv.org |
| Perplexity | AI-powered research | perplexity.ai |
| Wolfram Alpha | Computations, data | wolframalpha.com |
| Statista | Statistics | statista.com |
| Crunchbase | Company data | crunchbase.com |
| Product Hunt | New products | producthunt.com |
| GitHub | Code/projects | github.com |
| Stack Overflow | Programming Q&A | stackoverflow.com |

---

## Source Evaluation Framework

### CRAAP Test

| Criterion | Questions to Ask |
|-----------|-----------------|
| **Currency** | When was it published? Updated? |
| **Relevance** | Does it relate to your topic? Audience? |
| **Authority** | Who is the author? Credentials? |
| **Accuracy** | Is it supported by evidence? Verifiable? |
| **Purpose** | Why was it written? Bias? |

### Source Reliability Tiers

| Tier | Source Type | Reliability |
|------|-------------|-------------|
| Tier 1 | Peer-reviewed journals, official statistics | Highest |
| Tier 2 | Quality news (Reuters, AP), industry reports | High |
| Tier 3 | Company blogs, trade publications | Medium |
| Tier 4 | Social media, forums, wikis | Verify required |
| Tier 5 | Anonymous sources, content farms | Low |

---

## Output Format

```markdown
# Web Search Strategy: [Topic]

**Information Need**: [What you're looking for]
**Search Date**: [Date]

---

## Recommended Search Queries

### Primary Query
```
[Optimized search query with operators]
```
**Rationale**: [Why this query works]

### Alternative Queries
1. `[Alternative query 1]`
   - Use when: [Scenario]
2. `[Alternative query 2]`
   - Use when: [Scenario]
3. `[Alternative query 3]`
   - Use when: [Scenario]

---

## Recommended Search Engines

| Engine | Why | Query Modification |
|--------|-----|-------------------|
| [Engine 1] | [Reason] | [Any modifications] |
| [Engine 2] | [Reason] | [Any modifications] |

---

## Search Strategy

### Step 1: [First search approach]
- Query: `[query]`
- Expected results: [What to look for]

### Step 2: [Second search approach]
- Query: `[query]`
- Expected results: [What to look for]

### Step 3: [Third search approach]
- Query: `[query]`
- Expected results: [What to look for]

---

## Expected Source Types

| Source Type | What to Look For |
|-------------|-----------------|
| [Type 1] | [Specifics] |
| [Type 2] | [Specifics] |

---

## Verification Strategy

1. [How to verify finding 1]
2. [How to verify finding 2]

---

## Potential Challenges

- [Challenge 1]: [How to address]
- [Challenge 2]: [How to address]

---

*Use multiple sources and cross-reference findings for accuracy.*
```

---

## Example Use Cases

### Case 1: Finding Recent Statistics
**Need**: Latest global EV sales figures
**Strategy**:
```
Query: global electric vehicle sales 2024 statistics
Operators: after:2024-01-01 (filetype:pdf OR site:statista.com)
Sources: IEA, Bloomberg NEF, industry reports
```

### Case 2: Technical How-To
**Need**: How to implement OAuth 2.0
**Strategy**:
```
Query: "OAuth 2.0" implementation tutorial
Site-specific: site:stackoverflow.com OR site:auth0.com
Filter: Look for official docs, recent posts
```

### Case 3: Competitive Intelligence
**Need**: Information about a competitor's product
**Strategy**:
```
Query: "[Company name]" product launch OR announcement
Sources: Press releases, news, Product Hunt
Social: site:twitter.com OR site:linkedin.com "[Company]"
```

---

## Tips for Better Results

1. **Start with the end in mind** - know what type of answer you need
2. **Use quotes** for exact phrases
3. **Combine operators** for precision
4. **Try multiple query variations**
5. **Check source dates** - information expires
6. **Cross-reference** findings across sources
7. **Use specialized engines** for specific content types

---

## Limitations

- Cannot execute actual searches
- Cannot access real-time results
- Cannot access paywalled content
- Search engines change their algorithms
- Results vary by location and personalization

---

*Built by the Claude Office Skills community. Contributions welcome!*

© claude-office-skills · MIT · claude-office-skills/skills

Similar skills

Everything for Science and research →

web-search by other authors comes first.

web-search 101-skills This skill gives your agent web search and content extraction through Tavily and Exa. Useful for research, fact-checking and RAG pipelines. CX Codex +1 110,000installs web-search jwynia A skill that tells the agent how to use its built-in WebSearch: current info, fact checks, source links. No API key needed. CX Codex +1 776installs web-search-tavily agent-skills A skill that runs web searches through the Tavily API and returns structured results with relevance scores. Useful when the agent needs current data from the web. CX Codex +1 1,100installs exa-search affaan-m Wires Exa neural search into your agent: fresh web pages, code examples, company and people lookups. Useful when you need search instead of model guesswork. CX Codex +1 4,500installs search-specialist VoltAgent Community A subagent that hunts down precise information across many sources: builds queries, picks sources, filters results. Useful when the job is finding, not analyzing. 6,886installs web-search-brave agent-skills A skill that searches the web through the Brave Search API, with freshness, country and language filters. Needs BRAVE_API_KEY and Deno. CX Codex +1 312installs tavily-search framix-team Integrates Tavily API into your coding agent for web search, text extraction, site crawling, and research reports. Works as an AI-oriented search replacement. CX Codex +1 840installs perplexity-search parcadei Runs Perplexity AI web search, deep research, and chain-of-thought reasoning directly from the agent terminal. CX Codex +1 524installs

From the same set109

Project sponsors

Companies that keep this analytics open
Gold sponsor

Impressions, clicks and code copies — open statistics. We keep the numbers public for the same reason NeBlask itself is public. Become a sponsor