skill-seekers-ref（技能寻求者参考）

Skill Seekers

Skill Seekers converts documentation, GitHub repos, PDFs, codebases, videos, and 10+ other source types into structured AI skills. One command takes a source and produces a platform-ready skill package.

Install: pip install skill-seekers

Core Concept

The pipeline is: Source -> Extract -> Analyze -> Organize -> (Enhance) -> Package

The create command handles the full pipeline automatically by detecting source type from the input:

# URL -> docs scraping
skill-seekers create https://react.dev --target claude

# owner/repo -> GitHub scraping
skill-seekers create facebook/react --target claude

# ./path -> local codebase analysis
skill-seekers create ./my-project --target claude

# file.pdf -> PDF extraction
skill-seekers create ./manual.pdf --target claude

Quick Start

# Create a skill from docs (simplest case)
skill-seekers create https://tailwindcss.com/docs --target claude --max-pages 50

# With AI enhancement (much higher quality)
skill-seekers create https://tailwindcss.com/docs --target claude --max-pages 50 --enhance-workflow default

# Upload to Claude
skill-seekers upload output/tailwindcss-claude.zip --target claude

# Or install directly to an agent's directory
skill-seekers install-agent output/tailwindcss/ --agent claude

Presets for Speed Control

The -p flag controls analysis depth:

skill-seekers create https://docs.astral.sh/ruff --target claude -p quick

Source Types

Documentation Websites

skill-seekers create https://react.dev --target claude --max-pages 100

# With parallel workers for speed
skill-seekers scrape --url https://react.dev --workers 5 --async

# Resume interrupted scrape
skill-seekers resume --list
skill-seekers resume <job-id>

The scraper checks for llms.txt first (10x faster when available), then falls back to BFS page crawling.

GitHub Repositories

skill-seekers create microsoft/TypeScript --target claude

# With local clone for unlimited analysis (bypasses API limits)
skill-seekers github --repo django/django --local-repo-path ./django-clone

# Skip issues/changelog to speed up
skill-seekers github --repo owner/repo --no-issues --no-changelog

Set a GitHub token for private repos and higher rate limits:

skill-seekers config --github

Local Codebases (C3.x Analysis)

Analyzes code patterns, architecture, tests, and config across 27+ languages:

skill-seekers analyze --directory ./my-project --preset standard

# Skip specific analysis phases
skill-seekers analyze --directory ./src --skip-patterns --skip-test-examples

C3.x detects: design patterns (Singleton, Factory, Observer...), test examples, config patterns (9 formats), architecture (MVC, MVVM...), and generates how-to guides.

PDFs

skill-seekers create ./handbook.pdf --target claude
skill-seekers pdf --pdf ./manual.pdf --name "product-manual"

Supports OCR for scanned documents.

Other Sources

Each has a dedicated subcommand with specific options:

Multi-Source (Unified)

Combine docs + GitHub + PDF into one skill using a config file:

skill-seekers unified --config configs/my-project.json

Config file format - see references/config-and-workflows.md.

AI Enhancement

Enhancement transforms raw extraction (quality ~3/10) into polished skills (~9/10). It uses AI to extract best practices, create quick references, and improve organization.

Local Enhancement (Free with Claude Code)

skill-seekers enhance output/my-skill/
skill-seekers enhance output/my-skill/ --background   # non-blocking
skill-seekers enhance output/my-skill/ --daemon        # survives terminal close
skill-seekers enhance-status output/my-skill/          # check progress

API Enhancement

export ANTHROPIC_API_KEY=sk-ant-...
skill-seekers enhance output/my-skill/ --target claude

# Or Gemini / OpenAI
export GOOGLE_API_KEY=AIzaSy...
skill-seekers enhance output/my-skill/ --target gemini

Enhancement Workflows (v3.1.0+)

Workflow presets apply specific enhancement strategies:

# Built-in presets
skill-seekers enhance output/my-skill/ --enhance-workflow default
skill-seekers enhance output/my-skill/ --enhance-workflow security-focus
skill-seekers enhance output/my-skill/ --enhance-workflow api-documentation
skill-seekers enhance output/my-skill/ --enhance-workflow architecture-comprehensive
skill-seekers enhance output/my-skill/ --enhance-workflow minimal

# Chain multiple workflows
skill-seekers create URL --enhance-workflow security-focus --enhance-workflow minimal

# Inline custom stage
skill-seekers create URL --enhance-stage 'perf:Analyze for performance bottlenecks'

# List / manage workflows
skill-seekers workflows list
skill-seekers workflows show security-focus
skill-seekers workflows copy default    # copy to ~/.config/skill-seekers/workflows/ for editing
skill-seekers workflows add ./my-workflow.yaml
skill-seekers workflows validate my-workflow

See references/config-and-workflows.md for custom workflow YAML format.

Output Formats and Packaging

Package skills for 16+ platforms:

# Package for a specific platform
skill-seekers package output/my-skill/ --target claude
skill-seekers package output/my-skill/ --target gemini
skill-seekers package output/my-skill/ --target cursor
skill-seekers package output/my-skill/ --target openai

# Upload directly
skill-seekers upload output/my-skill.zip --target claude

# Install to agent directory
skill-seekers install-agent output/my-skill/ --agent claude
skill-seekers install-agent output/my-skill/ --agent cursor
skill-seekers install-agent output/my-skill/ --agent all

Supported targets:

• AI platforms: claude, gemini, openai
• Coding assistants: cursor, windsurf, cline, continue
• RAG/vector: langchain, llamaindex, chroma, faiss, haystack, qdrant, weaviate, pinecone
• Generic: markdown, json, yaml

RAG Chunking

For vector database targets, enable semantic chunking:

skill-seekers package output/my-skill/ --target chroma --chunk-for-rag --chunk-tokens 512 --chunk-overlap-tokens 50

Complete Workflow (install command)

The install command runs the full pipeline: fetch -> scrape -> enhance -> package -> upload:

skill-seekers install --config configs/react.json
skill-seekers install --config react --no-upload    # skip upload step
skill-seekers install --config react --dry-run      # preview only

Quality Scoring

Check skill quality before shipping:

skill-seekers quality output/my-skill/
skill-seekers quality output/my-skill/ --report --threshold 7

Utility Commands

# Estimate page count before scraping
skill-seekers estimate configs/react.json

# Incremental update (no full rescrape)
skill-seekers update output/my-skill/ --check-changes
skill-seekers update output/my-skill/ --force

# Stream large files chunk by chunk
skill-seekers stream large-doc.md --output output/

# Configure tokens and API keys
skill-seekers config --show
skill-seekers config --github
skill-seekers config --api-keys
skill-seekers config --test

Common Patterns

"I want to create a skill from X docs quickly"

skill-seekers create https://docs.example.com --target claude -p quick --max-pages 30

"I want a production-quality skill"

skill-seekers create https://docs.example.com --target claude -p comprehensive --enhance-workflow default

"I want to analyze my local codebase"

skill-seekers analyze --directory ./my-project --preset comprehensive
skill-seekers enhance output/my-project/
skill-seekers install-agent output/my-project/ --agent claude

"I want skills for multiple platforms"

skill-seekers create https://docs.example.com -o output/my-skill
skill-seekers package output/my-skill/ --target claude
skill-seekers package output/my-skill/ --target cursor
skill-seekers package output/my-skill/ --target chroma --chunk-for-rag

"I want to keep skills up to date automatically"

Use CI/CD with the install command - see references/automation.md.

Reference Files

For detailed information on specific topics:

• references/cli-commands.md - Full CLI reference with all flags for every command
• references/config-and-workflows.md - Config file schema, unified configs, custom workflow YAML format
• references/config-examples.md - Ready-to-use config examples for common frameworks
• references/automation.md - CI/CD integration, Docker, batch processing