# Search and Extract Web Content

Current web research often requires separate tools for search, extraction, and summarization. This skill gives Claude, Codex, and Claude Code reusable Tavily and Exa commands through the inference.sh CLI.

## Install

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

## Metadata

- Status: approved
- Slug: inference-sh-skills-web-search
- Skillstore revision: r1
- Version status: missing
- Tree hash: 63bfc5912981c4558644704f559773e22738dafc5457f0e2833b114600a5f6b3
- Author: inference-sh-skills
- GitHub username: inference-sh-skills
- License: MIT
- Repository: https://github.com/inference-sh-skills/skills/tree/main/tools/llm/web-search/
- Ref: a25199bc7d6b82598536822d1738eb5d5f54025b
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: confirmation\_required
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands, network
- Quality score: 38
- Quality tier: warning
- Public page: https://skillstore.pages.dev/skills/inference-sh-skills-web-search
- Manifest: https://skillstore.pages.dev/api/skills/inference-sh-skills-web-search/manifest

## Capabilities

- Runs Tavily search-assistant queries through the belt CLI.
- Extracts clean text and images from public URLs with tavily/extract.
- Runs Exa search, answer, and extract apps through inference.sh.
- Shows workflows that combine search results with Claude model calls.
- Provides documentation links for agent search and research workflows.

## Use Cases

- Gather Source-Backed Research: Search public web sources and collect links, snippets, and answer summaries for a research question.
- Prepare RAG Inputs: Extract readable text from public URLs before passing the content into retrieval or summarization workflows.
- Check Claims Against Current Sources: Use search and answer tools to compare a claim with recent public sources and supporting links.

## Prompt Templates

### Find Recent Sources

```
Search the web for [topic]. Return a concise answer, the strongest source links, and any date-sensitive caveats.
```

### Extract Public Pages

```
Extract readable content from these public URLs: [URLs]. Summarize each page and preserve source titles.
```

### Compare Search Evidence

```
Use Tavily or Exa to compare [option A] and [option B]. Include source quality notes and unresolved questions.
```

### Build a Research Brief

```
Search for [topic], extract the best sources, and prepare a source-grounded brief for [audience]. Separate facts, uncertainty, and next steps.
```

## Limitations

- Requires the inference.sh belt CLI and an authenticated account.
- Sends queries, URLs, and extracted content to remote services.
- Results depend on Tavily, Exa, source availability, and provider limits.
- Does not provide interactive browser automation or authenticated site scraping.

## Best Practices

- Review source URLs before relying on extracted or summarized content.
- Use specific search questions and date constraints when freshness matters.
- Avoid sending secrets, private URLs, or confidential text to remote search tools.

## Anti Patterns

- Do not treat generated answers as verified facts without checking sources.
- Do not extract private or access-controlled pages without permission.
- Do not paste tokens, credentials, or confidential documents into search inputs.

## Security Audit

- Audited at: 2026-07-05T18:07:01.267\+00:00
- Summary: The static findings are mostly Markdown code spans, fenced command examples, and documentation links rather than Ruby backtick execution. I found no prompt injection text in SKILL.md, but the skill grants broad Bash\(belt \*\) access and sends user inputs to remote services. Tighten allowed tool scope and add clear privacy and command-quoting guidance.

## Stats

- Views: 89
- Downloads: 9
- Favorites: 0
- Popularity score: 0