/skill-from-docs

Purpose

Generate a Claude Code skill from a documentation website. Implements smart scraping, categorization, pattern extraction, and AI enhancement.

Syntax

/skill-from-docs <url> [--name <skill-name>] [--max-pages <n>] [--output <path>]

Parameters

Parameter	Type	Required	Default	Description
url	string	Yes	-	Documentation website URL
--name	string	No	derived	Skill name
--max-pages	integer	No	500	Maximum pages to scrape
--output	string	No	~/.coditect/skills/{name}/	Output directory
--async	flag	No	true	Enable async scraping
--workers	integer	No	8	Async worker count
--estimate-only	flag	No	false	Only estimate page count
--skip-enhance	flag	No	false	Skip AI enhancement
--target	string	No	claude	Target platform (claude/gemini/openai/markdown)
--rate-limit	float	No	0.5	Requests per second

Examples

Basic Usage

# Generate skill from React documentation
/skill-from-docs https://react.dev/

# With custom name
/skill-from-docs https://fastapi.tiangolo.com/ --name fastapi

Advanced Usage

# Estimate pages first
/skill-from-docs https://docs.python.org/ --estimate-only

# Large documentation with limits
/skill-from-docs https://godotengine.org/docs/ --max-pages 2000 --workers 8

# Custom output location
/skill-from-docs https://flask.palletsprojects.com/ --output ~/my-skills/flask/

# For Google Gemini
/skill-from-docs https://react.dev/ --target gemini

Workflow Examples

# 1. Estimate → Scrape → Enhance → Package
/skill-from-docs https://react.dev/ --estimate-only
# Output: ~180 pages estimated

/skill-from-docs https://react.dev/
# Output: SKILL.md + references/ created

# 2. Quick iteration (skip enhancement)
/skill-from-docs https://flask.palletsprojects.com/ --skip-enhance

# 3. Multi-platform generation
/skill-from-docs https://react.dev/ --target claude
/skill-from-docs https://react.dev/ --target gemini
/skill-from-docs https://react.dev/ --target openai

Output

~/.coditect/skills/{name}/
├── SKILL.md                    # AI-enhanced (300+ lines)
├── references/
│   ├── index.md                # Category index
│   ├── getting_started/        # Categorized content
│   ├── api_reference/
│   ├── tutorials/
│   └── patterns/
├── examples/
│   ├── basic/
│   └── advanced/
└── metadata.json
    {
      "source_url": "...",
      "pages_scraped": 180,
      "scrape_date": "2026-01-23",
      "quality_score": 8.7
    }

What Happens

1. URL Validation: Verify URL is accessible
2. Strategy Detection: Check for llms.txt, sitemap, or BFS
3. Page Scraping: Extract content with rate limiting
4. Categorization: Smart category assignment
5. Pattern Extraction: Find code patterns and examples
6. AI Enhancement: Generate comprehensive SKILL.md
7. Quality Check: Validate against CODITECT standards

Success Output

✅ COMMAND COMPLETE: /skill-from-docs

Scraping Summary:
- [x] URL: https://react.dev/
- [x] Strategy: llms.txt (10x faster)
- [x] Pages: 180/180 (100%)
- [x] Categories: 6
- [x] Patterns: 23

Output: ~/.coditect/skills/react/
- SKILL.md (342 lines, AI-enhanced)
- references/ (6 categories)
- examples/ (12 examples)

Quality Score: 8.7/10

When NOT to Use

Do NOT use this command when:

• Documentation requires authentication (OAuth, API keys, login) - use manual download instead
• Target site has aggressive anti-bot protection (Cloudflare, captchas) - will fail
• Documentation is behind a paywall - not supported
• Source is a video tutorial - not supported
• Documentation is a single PDF file - use /pdf skill instead
• Need to combine docs with code analysis - use skill-generator-orchestrator instead
• Documentation is not in English - limited support
• Quick one-off lookup - just ask Claude directly

Completion Checklist

Before marking this command as complete, verify:

• URL validated and accessible
• Scraping strategy determined (llms.txt, sitemap, or BFS)
• All pages scraped within configured limits
• Content categorized into correct directories
• Code blocks extracted with language detection
• Patterns identified and documented
• SKILL.md generated (300+ lines if AI-enhanced)
• Output directory structure matches specification
• metadata.json created with scraping metrics

Failure Indicators

• ❌ URL not accessible (404, 403, timeout)
• ❌ robots.txt blocks scraping
• ❌ Zero pages scraped
• ❌ Rate limited by target site
• ❌ SKILL.md under 100 lines
• ❌ Categorization confidence below 50%
• ❌ Output directory not created

Anti-Patterns (Avoid)

Anti-Pattern	Problem	Solution
No rate limiting	IP blocked, incomplete scrape	Use --rate-limit 0.5 minimum
Ignoring robots.txt	Legal issues, site blocks	Command respects robots.txt by default
Scraping auth pages	Empty content	Use manual download + PDF analysis
Too many workers	Server overload, blocks	Limit --workers 8 maximum
No max-pages limit	Hours of scraping	Set reasonable --max-pages
Skipping estimation	Unexpected large sites	Run --estimate-only first
Skipping AI enhancement	Generic output	Only use --skip-enhance for iteration

Verification

After execution, verify success:

# 1. Check output directory exists
ls -la ~/.coditect/skills/{name}/

# 2. Verify SKILL.md length (should be 200+ lines)
wc -l ~/.coditect/skills/{name}/SKILL.md

# 3. Check categorization
ls -la ~/.coditect/skills/{name}/references/

# 4. Validate metadata
cat ~/.coditect/skills/{name}/metadata.json | python3 -m json.tool

# 5. Count extracted code blocks
grep -r '```' ~/.coditect/skills/{name}/references/ | wc -l

Related

• Agent: doc-to-skill-converter
• Orchestrator: skill-generator-orchestrator
• Companion: /skill-from-repo
• Skill: multi-source-skill-generation