# Search and Extract Web Research

Finding current web evidence can slow down research and agent workflows. This skill runs Tavily and Exa searches through belt for sourced answers and extracted content.

## Install

```bash
npx skillstore add inference-skills/web-search
```

## Metadata

- Status: approved
- Slug: inference-skills-web-search
- Skillstore revision: r1
- Version status: missing
- Tree hash: 6d46abaac42f3b0b029123656ae450ef4d441d582b6de07c9b95c252f32507d7
- Author: inference-skills
- GitHub username: inference-skills
- License: MIT
- Repository: https://github.com/inference-skills/skills/tree/main/tools/llm/web-search/
- Ref: a25199bc7d6b82598536822d1738eb5d5f54025b
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: allowed
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands, network
- Quality score: 70
- Public page: https://skillstore.pages.dev/skills/inference-skills-web-search
- Manifest: https://skillstore.pages.dev/api/skills/inference-skills-web-search/manifest

## Capabilities

- Runs Tavily Search Assistant through the belt CLI for web questions.
- Extracts readable text and images from URLs with Tavily Extract.
- Runs Exa Search, Exa Answer, and Exa Extract workflows.
- Supports research pipelines that save search or extraction output for later analysis.
- Shows how to combine search results with Claude models through OpenRouter apps.

## Use Cases

- Current Topic Research: Search recent web sources and collect links, answers, and page context for a research question.
- RAG Source Collection: Find and extract pages that can support a retrieval-augmented generation workflow.
- Claim Verification: Compare search results and source pages before accepting a factual claim.

## Prompt Templates

### Basic Web Search

```
Search the web for recent information about [topic] using Tavily. Return the answer, source links, and useful images.
```

### Extract Page Content

```
Extract readable content from these URLs: [urls]. Summarize the key points and list the source pages.
```

### Compare Providers

```
Run both Tavily and Exa searches for [query]. Compare the top results, source quality, and conflicting claims.
```

### Build Research Pipeline

```
Search for [topic], extract the most relevant pages, then summarize trends with Claude. Keep source citations separate from analysis.
```

## Limitations

- Requires the inference.sh belt CLI to be installed and authenticated.
- Sends queries, URLs, and extracted content to external service providers.
- Depends on provider availability, account permissions, and network access.
- Does not independently verify the trustworthiness of returned sources.

## Best Practices

- Use precise queries and ask for source links with every answer.
- Review extracted pages before using their content in reports or agents.
- Remove secrets and private identifiers before sending prompts or URLs to providers.

## Anti Patterns

- Do not treat search summaries as verified facts without source review.
- Do not send confidential documents or private URLs through external extraction tools.
- Do not use broad research prompts when a narrow query would reduce irrelevant results.

## Security Audit

- Audited at: 2026-07-05T17:29:37.065\+00:00
- Summary: The static backtick and hardcoded URL findings are false positives caused by Markdown formatting, documentation links, and examples. The remaining material risk is the intended use of a broad belt CLI allowlist with external search, extraction, and LLM providers. No prompt injection attempt or malicious instruction was found in SKILL.md.

## Stats

- Views: 98
- Downloads: 12
- Favorites: 0
- Popularity score: 0