@digimetalab/agentkit

The Enterprise-Grade Skill Manager for AI Code Agents

211 Skills · 5 Meta-Bundles · 18 Platforms · Zero Lock-in

📖 Table of Contents

• Introduction
• Why AgentKit?
• Architecture
• Quick Start
• Meta-Bundles
• Supported Platforms
• CLI Reference
• Programmatic API
• Project Structure
• Documentation
• Contributing
• License

📖 Introduction

AgentKit is a standardized infrastructure layer that transforms generic AI code assistants into specialized Senior Engineering Partners.

Instead of relying on an AI's generalized training data—which often leads to hallucinated libraries, legacy patterns, or security flaws—AgentKit injects 211 battle-tested engineering skills directly into your agent's context.

These skills act as Standard Operating Procedures (SOPs), ensuring your AI agent follows strict, production-ready guidelines for everything from React architecture to OWASP security audits, from RAG pipelines to growth marketing strategies.

💡 Why AgentKit?

| Feature | Without AgentKit | With AgentKit | |---|---|---| | Context | Generic, often outdated training data | Project-specific context & latest best practices | | Consistency | Varies by prompt phrasing | Deterministically follows engineering standards | | Complex Tasks | Often gets lost in multi-step tasks | Follows strict step-by-step SOPs | | Security | Frequently suggests vulnerable code | Includes automated security auditing skills | | Onboarding | Requires manual prompt engineering | Instant specialized persona via Meta-Bundles | | Platform | Locked to one agent | Works with 18 AI agents simultaneously |

🏗️ Architecture

AgentKit uses a Universal Platform Architecture (UPA) — a single skill library that syncs across all your AI tools.

graph TD
    A[AgentKit CLI] -->|Installs| B[.agent/skills]
    A -->|Detects stack| F[agentkit sync]
    A -->|Generates proxies| G[agentkit workflow proxy]
    
    B -->|Consumed by| C[Antigravity]
    B -->|Consumed by| D[Cursor / Windsurf / Claude Code]
    B -->|Proxied to| E[Gemini / Copilot / Cody / 15+ more]
    
    subgraph "Your Project"
    B
    H[.agent/agentkit.json] -.->|Config| B
    end

🚀 Quick Start

1. Installation

# Recommended — run without install
npx @digimetalab/agentkit

# Or install globally
npm install -g @digimetalab/agentkit

2. Interactive Setup

Run the command in your project root:

agentkit

The wizard will:

1. Detect your project stack (React, Node, Python, etc.)
2. Select your AI agent — Antigravity, Cursor, Windsurf, Claude Code, and 14 more.
3. Choose a Meta-Bundle — Foundation, Builder, Guardian, Brain, or Strategist.
4. Install skills to your agent's native directory.

3. Auto-Sync (Smart Install)

Let AgentKit detect your project stack and install the right skills automatically:

agentkit sync

This scans your package.json, requirements.txt, etc. and installs only the relevant skills for your tech stack.

4. Usage

Open your AI agent chat and work as usual. Your agent will automatically detect the installed skills and follow the engineering SOPs.

Example: "Build a user dashboard with authentication."

Your agent now follows the React Patterns SOP, API Security Best Practices, and Authentication Flow guidelines — automatically.

📦 Meta-Bundles

AgentKit organizes 211 skills into 5 Meta-Bundles — curated collections based on engineering role.

🟢 The Foundation (foundation)

Architecture, Standards, & Documentation

The bedrock skills every project needs. Clean code principles, Git workflows, architectural decision records, and documentation templates.

agentkit install --pack foundation

🔵 The Builder (builder)

Frontend, Backend, & Infrastructure — 52 Skills

Full-stack development SOPs. React/Vue/Svelte patterns, Node.js architecture, API design (REST/GraphQL), Docker containers, database design, Redis caching, serverless deployment, and more.

agentkit install --pack builder

🔴 The Guardian (guardian)

Security, Quality, & Reliability — 34 Skills

Defensive engineering. OWASP Top 10, penetration testing methodology, API security, performance profiling, code quality auditing, TDD workflows, and incident response procedures.

agentkit install --pack guardian

🟣 The Brain (brain)

AI, Data, & Orchestration — 58 Skills

AI engineering stack. Prompt engineering, RAG pipelines, LLM fine-tuning, multi-agent architecture, MCP server building, embeddings, vector search, computer vision, NLP, and ML operations.

agentkit install --pack brain

🟠 The Strategist (strategist)

Product, Growth, & Content — 67 Skills

Business and growth skills. SEO optimization, A/B testing, conversion rate optimization, content strategy, copywriting, paid ads, analytics tracking, ASO, and product management frameworks.

agentkit install --pack strategist

🔌 Supported Platforms

AgentKit supports 18 AI agent platforms with native integration and automatic proxy generation.

| Agent | Type | Skills Path | Proxy Format | |---|---|---|---| | Antigravity | Native | .agent/skills/ | .md | | Cursor | IDE | .cursor/rules/ | .mdc | | Windsurf | IDE | .agent/rules/ | .md | | Claude Code | CLI | .claude/skills/ | .md | | Gemini CLI | CLI | .gemini/skills/ | .md | | GitHub Copilot | IDE | .github/ | copilot-instructions.md | | Sourcegraph Cody | IDE | .github/ | cody-instructions.md | | Tabnine | IDE | .github/ | tabnine-instructions.md | | Codex CLI | CLI | .codex/skills/ | .md | | Trae | IDE | .trae/rules/ | .md | | Continue.dev | IDE | .continue/ | config.json | | Cline | IDE | .cline/rules/ | .md | | Aider | CLI | . | .aider.conf.yml | | PearAI | IDE | .pearai/rules/ | .md | | Goose | CLI | .goose/skills/ | .md | | Void | IDE | .void/rules/ | .md | | OpenCode | IDE | .opencode/rules/ | .md | | Project IDX | IDE | .idx/skills/ | .md |

Generate proxy files for all your platforms at once:

agentkit workflow proxy

🛠️ CLI Reference

| Command | Arguments | Description | |---|---|---| | agentkit | (none) | Launch the interactive Setup Wizard. | | agentkit list | --skills | List all Meta-Bundles. Use --skills to see all 211 skills. | | agentkit install | <skill> | Install a single skill (e.g., agentkit install react-best-practices). | | | --pack <bundle> | Install a Meta-Bundle (foundation, builder, guardian, brain, strategist). | | | --all | Install the entire library (211 skills). | | | --out <path> | Specify custom output directory (default: ./.agent/skills). | | agentkit search | <query> | Search for skills by keyword (e.g., react, security, rag). | | agentkit sync | --out <path> | Auto-detect project stack and install matching skills. | | agentkit doctor | --fix | Diagnose installation issues. Use --fix to auto-repair. | | agentkit status | (none) | Show current installation status and detected agents. | | agentkit platform list | (none) | List all 18 supported platforms and detection status. | | agentkit platform install | <name> | Initialize configuration for a specific platform. | | agentkit platform sync | (none) | Sync skills to all detected platforms. | | agentkit workflow proxy | (none) | Generate platform-specific proxy files for installed skills. | | agentkit about | (none) | Show AgentKit info and credits. |

💻 Programmatic API

Use AgentKit programmatically in your own Node.js tools.

const agentkit = require('@digimetalab/agentkit');

// Install specific skills programmatically
await agentkit.commands.install('react-best-practices', { out: './my-skills' });

// List available bundles
const bundles = await agentkit.utils.getBundles();
console.log(bundles['builder']);

📁 Project Structure

When installed, AgentKit creates a non-intrusive structure in your project:

my-project/
├── .agent/                  # AgentKit Core Directory
│   ├── skills/              # Installed Skills (Markdown SOPs)
│   │   ├── react-best-practices/
│   │   ├── api-security-best-practices/
│   │   ├── prompt-engineering-master/
│   │   └── ... (211 skills available)
│   ├── workflows/           # Workflow Proxies
│   └── agentkit.json        # Configuration & Metadata
├── .cursor/rules/           # (If Cursor detected) Auto-generated proxies
├── .github/                 # (If Copilot/Cody) Instruction files
└── ...

📚 Documentation

Detailed documentation is available in the docs/ directory:

| Document | Description | |---|---| | Getting Started | Installation and first steps | | Skills Catalog | Full list of all 211 skills | | Bundles Guide | Meta-Bundle details and selection guide | | Visual Guide | Visual walkthrough of features | | Skill Anatomy | How to write your own skills | | Examples | Practical usage scenarios | | FAQ | Frequently Asked Questions | | Quality Bar | Skill validation standards | | Security Policy | Reporting vulnerabilities | | Security Guardrails | Offensive skill usage policy | | Contributing | Contribution guidelines | | Changelog | Version history |

🤝 Contributing

We welcome contributions! Please see our Contributing Guide for details.

1. Fork the repository.
2. Create your feature branch (git checkout -b feature/amazing-skill).
3. Follow the Skill Anatomy guide for new skills.
4. Ensure your skill passes the Quality Bar.
5. Open a Pull Request.

📄 License

This project is licensed under the MIT License. See the LICENSE file for details.

Built with ❤️ for the AI Developer Community
By digimetalab