Webcrawler

Webcrawler Skill

Intelligent documentation harvesting agent that recursively crawls documentation websites and extracts structured content about specific subjects.

Last Updated: 2026-01-23

Quick Start

# Crawl Python documentation about async/await
python skills/webcrawler/scripts/crawl_docs.py \
  --url "https://docs.python.org/3/library/asyncio.html" \
  --subject "asyncio" \
  --depth 2 \
  --output .tmp/docs/python-asyncio/

# Crawl React documentation
python skills/webcrawler/scripts/crawl_docs.py \
  --url "https://react.dev/" \
  --subject "React" \
  --depth 3 \
  --output .tmp/docs/react/

# Extract only API reference pages
python skills/webcrawler/scripts/crawl_docs.py \
  --url "https://expressjs.com/en/4x/api.html" \
  --subject "Express API" \
  --filter "api" \
  --output .tmp/docs/express-api/

Core Workflow

1. Initialize Crawl — Provide base URL and subject focus
2. Discover Pages — Recursively find all linked documentation pages
3. Filter Content — Keep only pages matching the subject criteria
4. Extract Content — Convert HTML to clean markdown
5. Organize Output — Structure files in a navigable hierarchy
6. Generate Index — Create a master index with all harvested pages

Scripts

crawl_docs.py — Main Documentation Crawler

The primary crawling script that handles recursive page discovery and content extraction.

python skills/webcrawler/scripts/crawl_docs.py \
  --url <base-url>           # Starting URL (required)
  --subject <topic>          # Subject focus for filtering (required)
  --output <directory>       # Output directory (default: .tmp/crawled/)
  --depth <n>                # Max crawl depth (default: 2)
  --filter <pattern>         # URL path filter pattern (optional)
  --delay <seconds>          # Delay between requests (default: 0.5)
  --max-pages <n>            # Maximum pages to crawl (default: 100)
  --same-domain              # Stay within same domain (default: true)
  --include-code             # Preserve code blocks (default: true)
  --format <md|json|both>    # Output format (default: both)

Outputs:

• index.md — Master index with links to all pages
• pages/*.md — Individual markdown files per page
• metadata.json — Crawl metadata and page inventory
• content.json — Structured JSON with all extracted content

extract_page.py — Single Page Extractor

Extract content from a single documentation page.

python skills/webcrawler/scripts/extract_page.py \
  --url <page-url>           # Page to extract (required)
  --output <file>            # Output file (default: stdout)
  --format <md|json>         # Output format (default: md)
  --include-links            # Include internal links (default: true)

filter_docs.py — Post-Crawl Filtering

Filter already-crawled documentation by subject or pattern.

python skills/webcrawler/scripts/filter_docs.py \
  --input <crawl-dir>        # Crawled docs directory (required)
  --subject <topic>          # Subject to filter for (required)
  --output <directory>       # Filtered output directory (required)
  --threshold <0.0-1.0>      # Relevance threshold (default: 0.3)

Configuration

Rate Limiting & Politeness

The crawler respects robots.txt and implements polite crawling:

• Default delay: 0.5s between requests
• User-Agent: Identifies as documentation harvester
• robots.txt: Honored by default (disable with --ignore-robots)

Domain Handling

Content Extraction

The crawler uses intelligent content extraction:

1. Main content detection — Finds <main>, <article>, or content containers
2. Navigation removal — Strips headers, footers, sidebars
3. Code preservation — Maintains code blocks with language hints
4. Link normalization — Converts relative links to absolute
5. Image handling — Optionally downloads and references images

Output Structure

.tmp/docs/<subject>/
├── index.md              # Master index with TOC
├── metadata.json         # Crawl metadata
├── content.json          # Structured JSON export
└── pages/
    ├── getting-started.md
    ├── installation.md
    ├── api-reference.md
    ├── configuration/
    │   ├── basic.md
    │   └── advanced.md
    └── troubleshooting.md

Index Format

# <Subject> Documentation

> Crawled from: <base-url>
> Pages: <count>
> Date: <timestamp>

## Table of Contents

- [Getting Started](pages/getting-started.md)
- [Installation](pages/installation.md)
- [API Reference](pages/api-reference.md)
- Configuration
  - [Basic](pages/configuration/basic.md)
  - [Advanced](pages/configuration/advanced.md)
- [Troubleshooting](pages/troubleshooting.md)

Common Workflows

1. Harvest API Documentation

# Crawl API docs with deep recursion
python skills/webcrawler/scripts/crawl_docs.py \
  --url "https://api.example.com/docs" \
  --subject "Example API" \
  --depth 4 \
  --filter "/api/" \
  --output .tmp/docs/example-api/

2. Build RAG Knowledge Base

# Crawl and export as JSON for embedding
python skills/webcrawler/scripts/crawl_docs.py \
  --url "https://docs.example.com" \
  --subject "Example Docs" \
  --depth 3 \
  --format json \
  --output .tmp/rag/example/

# The content.json can be fed directly to embedding pipelines

3. Offline Documentation Mirror

# Full documentation harvest
python skills/webcrawler/scripts/crawl_docs.py \
  --url "https://docs.kubernetes.io/docs/concepts/" \
  --subject "Kubernetes Concepts" \
  --depth 5 \
  --max-pages 500 \
  --include-images \
  --output .tmp/docs/k8s-concepts/

4. Focused Topic Extraction

# Crawl, then filter to specific topic
python skills/webcrawler/scripts/crawl_docs.py \
  --url "https://developer.hashicorp.com/terraform/docs" \
  --subject "Terraform" \
  --depth 3 \
  --output .tmp/docs/terraform-full/

# Filter to AWS provider only
python skills/webcrawler/scripts/filter_docs.py \
  --input .tmp/docs/terraform-full/ \
  --subject "AWS Provider" \
  --output .tmp/docs/terraform-aws/

Best Practices

Crawling

1. Start shallow — Begin with --depth 1 to test, then increase
2. Use filters — Narrow scope with --filter patterns
3. Set page limits — Use --max-pages to prevent runaway crawls
4. Respect rate limits — Increase --delay for slower servers

Content Quality

1. Subject focus — Be specific with --subject for better filtering
2. Review index — Check index.md to verify crawl coverage
3. Post-filter — Use filter_docs.py to refine results

Storage

1. Use .tmp/ — Store crawled docs in the temp directory
2. Organize by subject — Create subdirectories per topic
3. Version with dates — Add timestamps for recurring crawls

Troubleshooting

Dependencies

Required Python packages:

pip install requests beautifulsoup4 html2text lxml
# Optional for JavaScript rendering:
pip install playwright && playwright install

Related Skills

• qdrant-memory — Store crawled docs in vector database for RAG
• pdf-reader — Extract text from PDF documentation

External Resources

• Scrapy Documentation — For complex crawling needs
• html2text — HTML to Markdown conversion
• BeautifulSoup — HTML parsing

AGI Framework Integration

Qdrant Memory Integration

Before executing complex tasks with this skill:

python3 execution/memory_manager.py auto --query "<task summary>"

Decision Tree:

• Cache hit? Use cached response directly — no need to re-process.
• Memory match? Inject context_chunks into your reasoning.
• No match? Proceed normally, then store results:

python3 execution/memory_manager.py store \
  --content "Description of what was decided/solved" \
  --type decision \
  --tags webcrawler <relevant-tags>

Note: Storing automatically updates both Vector (Qdrant) and Keyword (BM25) indices.

Agent Team Collaboration- Strategy: This skill communicates via the shared memory system.

• Orchestration: Invoked by orchestrator via intelligent routing.
• Context Sharing: Always read previous agent outputs from memory before starting.

Local LLM Support

When available, use local Ollama models for embedding and lightweight inference:

• Embeddings: nomic-embed-text via Qdrant memory system
• Lightweight analysis: Local models reduce API costs for repetitive patterns

Converted and distributed by TomeVault — claim your Tome and manage your conversions.