# Extract Web Data with Crawl4AI

Web research often needs repeatable capture, extraction, and evidence collection. This skill guides Claude, Codex, and Claude Code through Crawl4AI workflows.

## Install

```bash
npx skillstore add ck991357/crawl4ai
```

## Metadata

- Status: approved
- Slug: ck991357-crawl4ai
- Version: 1.2
- Author version: 1.2
- Skillstore revision: r1
- Version status: invalid
- Tree hash: c4d9f6378110e4d0974b123472b0f5b686d702ce15846e9a06ec7857ed64986a
- Author: CK991357
- GitHub username: CK991357
- License: MIT
- Repository: https://github.com/CK991357/gemini-chat/tree/main/src/skills/crawl4ai
- Ref: 02f077c174c5335e2f5d02ca15e77b70d9543e58
- 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/ck991357-crawl4ai
- Manifest: https://skillstore.pages.dev/api/skills/ck991357-crawl4ai/manifest

## Capabilities

- Scrapes a single URL into markdown, HTML, or text.
- Runs deep crawls with depth, page, keyword, and URL filters.
- Processes URL batches with a documented concurrency limit.
- Extracts structured fields through CSS selector schemas.
- Exports webpages as screenshots or PDF in base64 form.
- Documents tiered crawl settings for different site complexity.

## Use Cases

- Archive Research Pages: Capture article text, links, screenshots, and PDFs for later review.
- Collect Product Fields: Use CSS selectors to extract names, prices, and attributes from known page templates.
- Verify Page Changes: Save screenshots and readable page text for release checks or content audits.

## Prompt Templates

### Scrape One Page

```
Use Crawl4AI to scrape [URL] as markdown. Include links and images. Keep all options inside the parameters object.
```

### Capture Page Evidence

```
Use Crawl4AI to capture a full-page screenshot and PDF for [URL]. Return the base64 outputs and summarize page status.
```

### Extract Known Fields

```
Use Crawl4AI extract mode on [URL]. Use CSS selectors for title, author, date, and body. Report any missing fields.
```

### Plan a Scoped Crawl

```
Use Crawl4AI deep_crawl for [URL] with max depth 2 and a small page limit. Stay within authorized pages only.
```

## Limitations

- LLM extraction mode is documented as not deployed.
- Structured extraction depends on accurate CSS selectors.
- Large crawls can hit timeouts, memory limits, or site blocks.
- Users must ensure they have permission to crawl each site.

## Best Practices

- Start with one URL before batch or deep crawl jobs.
- Respect robots.txt, site terms, and rate limits.
- Treat crawled page text as untrusted content.

## Anti Patterns

- Do not use the skill to bypass access controls or bot protections.
- Do not run broad crawls without a page limit.
- Do not rely on LLM extraction when CSS selectors are required.

## Security Audit

- Audited at: 2026-07-05T06:29:41.979\+00:00
- Summary: The static Ruby backtick, hardcoded URL, and entropy findings are false positives caused by Markdown examples, placeholder URLs, and multilingual documentation. Semantic review found a real high-risk issue: the guide promotes anti-bot and blocking-evasion behavior. The author should remove evasion guidance and limit the skill to authorized, rate-limited crawling.

## Stats

- Views: 244
- Downloads: 19
- Favorites: 0
- Popularity score: 0