Meta-programming framework for Claude Code skills
Project description
๐จ SkillForge
Meta-Programming Framework for Claude Code
Transform Claude Code from a generic AI assistant into an expert team member who knows YOUR stack, YOUR conventions, and YOUR patterns.
Features โข Why SkillForge โข Installation โข Quick Start โข Documentation โข Contributing
๐ Table of Contents
- What is SkillForge?
- Why SkillForge?
- Features
- How It Works
- Installation
- Quick Start
- Usage
- Architecture
- Roadmap
- Contributing
- FAQ
- License
๐ฏ What is SkillForge?
SkillForge is an intelligent meta-programming framework that extends Claude Code's native skill system. It automatically generates, orchestrates, and continuously improves personalized skills tailored to YOUR development workflow.
In one sentence: SkillForge turns Claude Code into an expert developer who codes exactly the way YOU do, using YOUR stack, YOUR conventions, and YOUR patterns.
The Problem
Claude Code's native skills are powerful but generic:
- โ Don't know YOUR tech stack (Next.js? FastAPI? Supabase?)
- โ Don't know YOUR conventions (camelCase? Absolute imports? Feature-based structure?)
- โ Don't learn from YOUR patterns (Always use Zod? Prefer server actions?)
- โ No intelligent orchestration (Claude must guess which skills to load)
- โ Static and never improve over time
The Solution
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ SkillForge โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโค
โ โ
โ Generator โ Creates personalized skills โ
โ Orchestrator โ Loads the right skills smartly โ
โ Learner โ Learns from your usage patterns โ
โ Optimizer โ Improves skills continuously โ
โ โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ
~/.claude/skills/
โโโ nextjs-fullstack/ โ YOUR Next.js patterns
โโโ supabase-integration/ โ YOUR auth setup
โโโ git-workflow/ โ YOUR commit style
๐ก Why SkillForge?
Without SkillForge
You: "Create a login component with Supabase auth"
Claude Code:
โโ Uses generic knowledge (possibly outdated)
โโ Doesn't know if you use App Router or Pages Router
โโ Doesn't know your UI library (Tailwind? Material UI?)
โโ Doesn't know your state management (Zustand? Redux?)
โโ Generates generic code
Result: OK code, but you need 15-30 minutes to:
โโ Fix naming conventions
โโ Adjust import paths
โโ Modify styling approach
โโ Adapt to your patterns
โโ ๐ Time wasted: 15-30 minutes
With SkillForge
You: "Create a login component with Supabase auth"
SkillForge (invisible, in milliseconds):
โโ Analyzes intent: "login component, Supabase, auth"
โโ Identifies skills: nextjs-fullstack + supabase-integration
โโ Loads YOUR personalized skills
โโ Claude receives perfect context
Claude Code generates:
โโ Next.js 15 App Router (you use this)
โโ Shadcn/ui components (your UI library)
โโ Zustand store (your state management)
โโ @/ import aliases (your convention)
โโ camelCase naming (your convention)
โโ Zod validation (learned pattern after 20+ uses)
โโ Error boundary wrapper (learned pattern)
โโ Feature-based structure (your preference)
Result: Production-ready code, 0 modifications needed โ
๐ Time saved: 100%
โจ Features
๐จ Smart Skill Generation
- Interactive Wizard: Answer questions about YOUR preferences (naming, imports, patterns)
- Template System: Pre-built templates for popular stacks (Next.js, Vue, React, FastAPI, etc.)
- Documentation Fetching: Auto-fetches latest docs via Context7 MCP
- Convention Injection: Every generated skill includes YOUR personal conventions
๐ง Intelligent Orchestration
- Intent Analysis: Understands what you're trying to do
- Smart Loading: Loads only relevant skills with token optimization
- Progressive Disclosure: Loads metadata โ core โ full content as needed
- Dependency Management: Automatically loads related skills
๐ Pattern Learning
- Usage Tracking: Tracks every skill usage and outcome
- Pattern Detection: Identifies YOUR recurring patterns (confidence-based)
- Auto-Application: Updates skills when patterns reach 80%+ confidence
- Cross-Project Learning: Learns from ALL your projects simultaneously
โก Continuous Optimization
- Token Reduction: Compresses verbose content without losing meaning
- Redundancy Detection: Identifies overlapping skills and suggests merges
- Documentation Updates: Auto-updates with latest library versions
- Unused Skill Detection: Suggests removal of rarely-used skills
๐๏ธ How It Works
Architecture Overview
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ User Profile โ
โ (YOUR preferences, global) โ
โ ~/.claude/skills/skillforge/data/user_profile.json โ
โ โ
โ - Naming conventions (camelCase, PascalCase) โ
โ - Import style (absolute with @/) โ
โ - Preferred libraries (Zustand, Zod, React Query) โ
โ - Code patterns (feature-based structure) โ
โ - Workflow (conventional commits) โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ
Injected into every skill
โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ Generated Skills โ
โ (Tech-specific + YOUR preferences) โ
โ ~/.claude/skills/ โ
โ โ
โ nextjs-fullstack/ โ
โ โโ SKILL.md โ Next.js patterns + YOUR prefs โ
โ โโ templates/ โ Code templates โ
โ โโ docs/ โ Reference docs โ
โ โ
โ supabase-integration/ โ
โ โโ SKILL.md โ Supabase setup + YOUR prefs โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ
Used at runtime per project
โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ Runtime (Per Project) โ
โ โ
โ Project A: Next.js 15 + Supabase โ
โ โโ Loads: nextjs-fullstack + supabase-integration โ
โ โโ Applies: YOUR conventions to THIS project stack โ
โ โ
โ Project B: Vue.js + Firebase โ
โ โโ Loads: vue-spa + firebase-integration โ
โ โโ Applies: YOUR conventions to THIS project stack โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
The Four Core Components
1. Generator
Creates personalized skills from templates + latest documentation + YOUR preferences.
User answers wizard questions
โ
Fetch latest docs (Context7)
โ
Process template with YOUR preferences
โ
Generate SKILL.md (2000+ lines)
2. Orchestrator
Intelligently selects and loads the right skills for each request.
Analyze user request
โ
Identify relevant skills
โ
Calculate token budget
โ
Progressive loading (metadata โ core โ full)
โ
Claude executes with perfect context
3. Learner
Detects patterns in your usage and updates skills automatically.
Track every usage
โ
After 10+ uses, detect patterns
โ
If confidence โฅ 80%, apply pattern
โ
Update skills automatically
โ
Notify user of learned pattern
4. Optimizer
Continuously improves skills for performance and quality.
Analyze all skills weekly
โ
Find optimization opportunities:
โโ Token reduction
โโ Redundancy removal
โโ Documentation updates
โโ Unused skills
โ
Apply optimizations (with user approval)
๐ฆ Installation
Prerequisites
- Python 3.11 or higher
- Claude Code installed
- pipx (recommended) or pip
Install SkillForge
# Install globally using pipx (recommended)
pipx install skillforge
# Or using pip
pip install skillforge
Initialize
# Initialize SkillForge in Claude Code
skillforge install
This creates the SkillForge directory structure in ~/.claude/skills/skillforge/.
๐ Quick Start
1. Run the Setup Wizard
Open Claude Code in your most representative project:
cd ~/projects/my-main-project
claude
In Claude Code, run the wizard:
> /sf:wizard
The wizard will:
- โ Ask about YOUR personal preferences (naming, imports, patterns)
- โ Detect your current project's tech stack (if applicable)
- โ Fetch latest documentation for your technologies
- โ
Generate personalized skills in
~/.claude/skills/
Example wizard session:
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ ๐ง SkillForge Setup Wizard โ
โ Building YOUR personal development profile โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
๐ Auto-detecting from current directory...
/Users/you/projects/my-nextjs-app
โ
Detected:
โโ Next.js 15.0.0
โโ TypeScript 5.3
โโ Tailwind CSS
โโ Supabase
Use these as defaults? [Y/n] Y
Great! Let me ask a few more questions about YOUR preferences...
1๏ธโฃ YOUR naming convention for variables?
[1] camelCase
[2] snake_case
> 1
2๏ธโฃ YOUR import style?
[1] Absolute imports with alias (@/...)
[2] Relative imports (../../...)
> 1
3๏ธโฃ YOUR preferred state management?
[1] Zustand
[2] Redux Toolkit
[3] Context API
> 1
... (8-10 more questions) ...
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ โ๏ธ Generating Skills... โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
[1/3] ๐ฆ Fetching Next.js documentation...
โโ โ
Documentation fetched (Context7)
[2/3] ๐จ Generating nextjs-fullstack skill...
โโ Processing template
โโ Injecting YOUR preferences
โโ Creating SKILL.md (2341 lines)
โโ โ
Skill generated
[3/3] ๐จ Generating supabase-integration skill...
โโ โ
Skill generated
โ
Setup Complete!
Skills installed:
~/.claude/skills/
โโ nextjs-fullstack/
โโ supabase-integration/
โโ git-workflow/
Your skills will improve as you use them!
2. Use Claude Code Normally
SkillForge works invisibly in the background:
You: "Create a login component with Supabase auth"
[SkillForge automatically loads your personalized skills]
Claude Code: [generates perfect code using YOUR conventions]
3. Generate Additional Skills
When you work on projects with different stacks:
> /sf:generate vue-spa
> /sf:generate firebase-integration
> /sf:generate python-fastapi
๐ Usage
Available Commands
Core Commands
# Run the setup wizard (first time or update)
> /sf:wizard
# Generate a specific skill
> /sf:generate <skill-name>
# Generate skill from current project
> /sf:generate --from-current
# List all installed skills
> /sf:list
# View SkillForge status
> /sf:status
Analysis & Optimization
# Analyze skills and usage patterns
> /sf:analyze
# Optimize existing skills
> /sf:optimize
# Update skills with latest documentation
> /sf:update <skill-name>
> /sf:update --all
# Check for available updates
> /sf:update --check
Debug & Introspection
# Debug orchestration decisions
> /sf:introspect --orchestration
# Debug pattern learning
> /sf:introspect --learning
Skill Templates
SkillForge includes templates for popular stacks:
Frontend:
nextjs-fullstack- Next.js App Router full-stacknextjs-pages- Next.js Pages Routerreact-vite- React + Vite SPAvue-spa- Vue.js 3 + Vitesvelte-kit- SvelteKit
Backend:
python-fastapi- FastAPI backendpython-django- Djangonode-express- Express.js
Integration:
supabase-integration- Supabase (auth, db, storage)firebase-integration- Firebasepostgresql-integration- PostgreSQLmongodb-integration- MongoDB
Workflow:
git-workflow- Git conventions & commit styletesting-vitest- Vitest testingtesting-jest- Jest testing
๐ง Configuration
User Profile Structure
Your preferences are stored in ~/.claude/skills/skillforge/data/user_profile.json:
{
"version": "1.0",
"preferences": {
"naming": {
"variables": "camelCase",
"components": "PascalCase",
"constants": "UPPER_SNAKE_CASE"
},
"imports": {
"style": "absolute",
"alias": "@/"
},
"technologies": {
"state_management": "Zustand",
"validation": "Zod",
"testing": "Vitest"
},
"patterns": {
"folder_structure": "feature-based",
"component_style": "functional",
"async_handling": "async/await"
},
"workflow": {
"git_commits": "conventional"
}
}
}
Manual Editing
You can manually edit your profile:
# Edit profile
code ~/.claude/skills/skillforge/data/user_profile.json
# Regenerate skills with new preferences
> /sf:wizard --update
๐๏ธ Architecture
Directory Structure
~/.claude/skills/skillforge/
โ
โโโ SKILL.md # Entry point (meta-skill)
โ
โโโ core/ # Behavioral files
โ โโโ SKILLFORGE.md # Core configuration
โ โโโ ORCHESTRATION.md # Orchestration logic
โ โโโ GENERATOR.md # Generation system
โ โโโ PATTERNS.md # Pattern learning
โ โโโ MCP_INTEGRATION.md # MCP integrations
โ
โโโ generators/ # Python generators
โ โโโ wizard_engine.py # Interactive wizard
โ โโโ skill_generator.py # Skill generator
โ โโโ doc_fetcher.py # Context7 integration
โ
โโโ analyzers/ # Intelligence
โ โโโ intent_analyzer.py # Analyze requests
โ โโโ skill_discovery.py # Find relevant skills
โ โโโ pattern_detector.py # Detect patterns
โ โโโ skill_optimizer.py # Optimize skills
โ
โโโ templates/ # Skill templates
โ โโโ tech-stack/
โ โ โโโ nextjs-fullstack.template
โ โ โโโ react-spa.template
โ โ โโโ python-api.template
โ โโโ workflow/
โ โโโ git-workflow.template
โ
โโโ data/ # Persistent data
โโโ user_profile.json # Your preferences
โโโ usage_analytics.json # Usage stats
โโโ learned_patterns.json # Learned patterns
Integration with Claude Code
SkillForge extends Claude Code's native skill system:
Claude Code loads skills from ~/.claude/skills/
When you make a request:
1. SkillForge analyzes intent
2. Identifies relevant skills
3. Loads them with token optimization
4. Claude receives perfect context
5. SkillForge tracks usage for learning
MCP Integration (Context7)
SkillForge uses Context7 MCP server to fetch up-to-date documentation:
# Fetch latest Next.js docs
docs = context7_get_library_docs(
library_id="/vercel/next.js",
topic="App Router",
tokens=2000
)
# Extract and use in skill generation
best_practices = extract_best_practices(docs)
code_examples = extract_code_examples(docs)
๐บ๏ธ Roadmap
Current Status: v1.0.0 Production Release ๐
โ Phase 1-10: Complete (100%)
Foundation & Core Systems
- Core architecture design
- Complete briefing documentation
- Generator implementation (WizardEngine, SkillGenerator, TemplateProcessor, DocFetcher)
- Orchestrator implementation (IntentAnalyzer, SkillDiscovery)
- Learner implementation (UsageTracker, PatternDetector, LearningPipeline)
- Optimizer implementation (SkillOptimizer)
User Interface & Commands
- Slash commands system (10 commands: /sf:wizard, /sf:generate, /sf:analyze, etc.)
- Visual output system (VERBOSE, COMPACT, SILENT modes)
- Beautiful installation experience
Quality & Testing
- Comprehensive test suite (unit, integration, e2e)
- Test coverage implementation
- Performance optimization
Documentation & Release
- Complete user documentation
- API documentation
- README and guides
- Package published to PyPI
- Installation scripts
- Production-ready v1.0.0
๐ฎ Future Enhancements (Post v1.0)
- Additional framework templates (Vue, Angular, FastAPI, Django)
- Web UI for skill management
- Skill marketplace for sharing and collaboration
- Advanced analytics dashboard
- Multi-project skill sharing
- Team collaboration features
- Import/export profiles
- CI/CD integration hooks
- IDE extensions (VSCode, JetBrains)
๐ค Contributing
SkillForge is in production (v1.0.0) and we welcome contributions!
Ways to Contribute
- ๐ Report bugs: Open an issue
- ๐ก Suggest features: Start a discussion
- ๐ Improve docs: Submit documentation PRs
- ๐จ Add templates: Contribute skill templates for new stacks
- ๐งช Test: Try SkillForge and provide feedback
Development Setup
# Clone repository
git clone https://github.com/yourusername/skillforge.git
cd skillforge
# Create virtual environment
python -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate
# Install dependencies
pip install -e ".[dev]"
# Run tests
pytest
# Run linter
ruff check .
Contribution Guidelines
- Fork the repository
- Create a feature branch (
git checkout -b feature/amazing-feature) - Commit changes (
git commit -m 'feat: add amazing feature') - Push to branch (
git push origin feature/amazing-feature) - Open a Pull Request
Please follow Conventional Commits for commit messages.
โ FAQ
General Questions
Q: Is SkillForge free?
A: Yes, SkillForge is open-source and free to use under the MIT license.
Q: Does SkillForge require Claude Code?
A: Yes, SkillForge is specifically designed as an extension for Claude Code's native skill system.
Q: Does SkillForge work with other AI coding assistants?
A: Not currently. SkillForge is built specifically for Claude Code's skill architecture.
Q: Is my data private?
A: Yes. All your preferences, patterns, and skills are stored locally in ~/.claude/skills/. Nothing is sent to external servers except when fetching public documentation via Context7.
Usage Questions
Q: Do I need to run the wizard for each project?
A: No! Run the wizard once to set up YOUR preferences. Skills are global and reusable across all projects.
Q: What if my project uses a different stack?
A: Simply generate additional skills: /sf:generate vue-spa or /sf:generate python-fastapi. Your personal preferences are applied to every skill.
Q: Can I share skills with my team?
A: Yes! You can export your profile and skills, and team members can import them. Team collaboration features are planned for v1.1.
Q: How much disk space does SkillForge use?
A: Approximately 50-100MB depending on how many skills you generate.
Q: Does SkillForge slow down Claude Code?
A: No. The orchestration happens in milliseconds, and intelligent token optimization ensures optimal performance.
Technical Questions
Q: How does pattern learning work?
A: SkillForge tracks your usage across all projects. When a pattern appears consistently (80%+ confidence), it's automatically applied to your skills.
Q: Can I disable pattern learning?
A: Yes, you can configure this in your profile or use /sf:introspect --learning to review before application.
Q: How often are skills updated?
A: Documentation updates can be run manually (/sf:update --all) or scheduled weekly via the optimizer.
Q: Can I customize skill templates?
A: Yes! Templates are stored in ~/.claude/skills/skillforge/templates/ and can be edited or extended.
๐ License
MIT License - see LICENSE file for details.
๐ Acknowledgments
- Anthropic for Claude Code and the native skills system
- Context Labs for Context7 MCP server
- Community contributors who help improve SkillForge
๐ Links
- Documentation: See docs/ folder
- GitHub Repository: https://github.com/omarpiosedev/SkillForge
- Issues: https://github.com/omarpiosedev/SkillForge/issues
Built with โค๏ธ by developers, for developers
If SkillForge helps you, please โญ star the repo!
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file skillforge-1.2.0.tar.gz.
File metadata
- Download URL: skillforge-1.2.0.tar.gz
- Upload date:
- Size: 430.1 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.13.7
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
76243f1d41c1f9c332dbdc48a87aa1113b37a45852c260c71d7b1533e04e936d
|
|
| MD5 |
cfb6f3e7ceb1fa8be17f815779d9b4d0
|
|
| BLAKE2b-256 |
23729226b8e6e53ca6dd9f4458224a352b31da05d6793196c83c729ac74e9fb0
|
File details
Details for the file skillforge-1.2.0-py3-none-any.whl.
File metadata
- Download URL: skillforge-1.2.0-py3-none-any.whl
- Upload date:
- Size: 262.6 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.13.7
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
cecd37896bfa363423c519d5a04b288571db7be14aab96916c12801c8c7fae2e
|
|
| MD5 |
1eee39bdc1de753a8d649c8e600f3673
|
|
| BLAKE2b-256 |
0c5d9a5a30d9234ebbadf389496cf7858acba9336b1d0f1e2ff4c9af0d5ed77c
|