antigravity-ai-kit

v3.10.1

Published

13 days ago

🚀 Trust-Grade AI development framework with a 29-module runtime engine — 20 Agents, 34 Skills, 37 Commands, 21 Workflows, 9 Rules, 349 Tests. Workflow enforcement, task governance, agent reputation, self-healing, and skill marketplace.

0High
0Medium
0Low

emredursun

ai agents gemini claude antigravity development workflow skills commands devtools productivity cli

🚀 Antigravity AI Kit

📚 Table of Contents

🤔 What is Antigravity AI Kit?

Antigravity AI Kit transforms your IDE into a virtual engineering team with:

✨ Key Features

🔒 Trust-Grade Governance: /explore → /plan → /work → /review — Each iteration builds context
🤖 Multi-Agent System: 20 specialized agents that collaborate (Mobile Developer, DevOps, Database Architect, Sprint Orchestrator...)
⚙️ Runtime Engine: 29 modules enforcing workflow transitions, task governance, agent reputation, self-healing, and marketplace
📦 Context as Artifact: Persistent markdown files for plans, specs, and decisions
🔄 Continuous Learning: PAAL cycle extracts patterns from every session
🛡️ Security First: Built-in secret detection, vulnerability scanning, and compliance checks

Core Philosophy

"Trust > Optimization. Safety > Growth. Explainability > Performance."

⚡ Quick Start

Option 1: Create New Project (Recommended)

npx create-antigravity-app my-project
npx create-antigravity-app my-api --template node-api
npx create-antigravity-app my-app --template nextjs

Creates a new project with .agent/ pre-configured. Templates: minimal, node-api, nextjs.

Option 2: Add to Existing Project

npx antigravity-ai-kit init

🔄 Updating

ag-kit update             # Non-destructive — preserves your customizations
ag-kit update --dry-run   # Preview changes without applying

Prefer ag-kit update over ag-kit init --force. The update command preserves your session data, ADRs, learning contexts, and customizations. Use init --force only for clean reinstalls.

✅ Verify Installation

ag-kit verify     # Manifest integrity check
ag-kit scan       # Security scan

🛡️ Safety Guarantees

Antigravity AI Kit is designed to never touch your project files. All operations are scoped to the .agent/ directory.

| Your Project Files | Safe? | Details | |:---|:---|:---| | Source code (src/, lib/, app/) | ✅ Never touched | Init/update only operates on .agent/ | | Config files (.env, package.json) | ✅ Never touched | No project config is read or written | | Documentation (docs/, README.md) | ✅ Never touched | Only .agent/ docs are managed | | Tests (tests/, __tests__/) | ✅ Never touched | Kit tests are internal to the package | | Platform files (android/, ios/) | ✅ Never touched | No platform-specific operations |

init --force safety features (v3.3.1+):

🔒 Auto-backup: Creates timestamped backup of existing .agent/ before overwriting
⚛️ Atomic copy: Uses temp directory + rename to prevent corruption on failure
🔗 Symlink guard: Skips symbolic links to prevent path traversal attacks
⚠️ Session warning: Alerts if active work-in-progress would be destroyed
🔍 Dry-run preview: --dry-run --force shows exactly which user files would be overwritten

update preserved files:

session-context.md — Your active session notes
session-state.json — Your session metadata
decisions/ — Your Architecture Decision Records
contexts/ — Your learning data and plan quality logs

🏗️ Architecture Overview

┌─────────────────────────────────────────────────────────────────────┐
│                    USER INTERFACE LAYER                              │
│  ┌─────────────────────────┐  ┌─────────────────────────┐          │
│  │  Slash Commands (37)    │  │  Workflows (21)         │          │
│  └────────────┬────────────┘  └────────────┬────────────┘          │
├───────────────┼────────────────────────────┼────────────────────────┤
│               ▼         INTELLIGENCE LAYER ▼                        │
│  ┌──────────────────┐  ┌──────────────────┐  ┌──────────────────┐  │
│  │ Intelligent      │  │ Workflow State   │  │ Context Budget   │  │
│  │ Router           │  │ Machine          │  │ Engine           │  │
│  └────────┬─────────┘  └──────────────────┘  └──────────────────┘  │
├───────────┼────────────────────────────────────────────────────────-┤
│           ▼              AGENT LAYER (20)                            │
│  ┌──────────────────┐  ┌──────────────────┐  ┌──────────────────┐  │
│  │ Core Agents      │  │ Domain Agents    │  │ Support Agents   │  │
│  │ Planner          │  │ Mobile Dev       │  │ Security         │  │
│  │ Architect        │  │ Frontend         │  │ Performance      │  │
│  │ Code Reviewer    │  │ Backend          │  │ Docs, Explorer   │  │
│  │ TDD Specialist   │  │ DB, DevOps       │  │ Knowledge        │  │
│  └────────┬─────────┘  └────────┬─────────┘  └────────┬─────────┘  │
├───────────┼─────────────────────┼─────────────────────┼────────────┤
│           ▼              SKILL LAYER (34)              ▼            │
│  ┌──────────────────┐  ┌──────────────────┐  ┌──────────────────┐  │
│  │ Orchestration    │  │ Operational      │  │ Domain Skills    │  │
│  │ Routing, Modes   │  │ Verification     │  │ API, Testing     │  │
│  │ Parallel Agents  │  │ Learning, Budget │  │ Security, MCP    │  │
│  └────────┬─────────┘  └────────┬─────────┘  └────────┬─────────┘  │
├───────────┼─────────────────────┼─────────────────────┼────────────┤
│           ▼           GOVERNANCE LAYER                 ▼            │
│  ┌──────────────────┐  ┌──────────────────┐  ┌──────────────────┐  │
│  │ rules/ (8)       │  │ hooks.json       │  │ manifest.json    │  │
│  │ Governance       │  │ 7 Event Hooks    │  │ Integrity Check  │  │
│  └──────────────────┘  └──────────────────┘  └──────────────────┘  │
├─────────────────────────────────────────────────────────────────────┤
│                    RUNTIME ENGINE (29 modules)                      │
│  workflow-engine · session-manager · task-governance                │
│  agent-reputation · self-healing · marketplace · + 15 more         │
└─────────────────────────────────────────────────────────────────────┘

How It Works: The Autonomy Engine

Antigravity AI Kit uses a 7-phase workflow state machine that guides development:

EXPLORE → PLAN → IMPLEMENT → VERIFY → CHECKPOINT → REVIEW → DEPLOY

| Phase | What Happens | Transition Guard | |:------|:-------------|:-----------------| | EXPLORE | Codebase discovery, research | Exploration artifact exists | | PLAN | Implementation plan with user approval | Plan approved by user | | IMPLEMENT | Code generation with agent routing | Auto on commit | | VERIFY | Quality gates, tests, lint | All gates pass | | CHECKPOINT | Developer decision gate | Explicit developer choice | | REVIEW | Code review (human or Copilot) | Review approved | | DEPLOY | Production deployment | Deployment checklist complete |

Intelligent Routing: The kit analyzes your request keywords (including implicit security triggers like "login", "payment", "upload") and automatically loads the right agents and skills (max 4 agents + 8 skills per session to stay within context budgets). Planning workflows use protected budget enforcement — mandatory skills survive trimming even when over budget.

🤖 Agents (20)

Core Development

Domain Specialists

Support & Intelligence

Autonomy Agents

⌨️ Commands (37)

Core Workflow

Development

Documentation & Git

Exploration & Research

Quality & Security

Integration & Deployment

Context Management

🛠️ Skills (34)

Operational Skills (6)

Orchestration Skills (4)

Domain Skills (13)

Development Skills (11)

⚙️ Runtime Engine (29 Modules)

Antigravity AI Kit v3.2.0 includes a full runtime engine built across 4 phases — all using Node.js built-ins with zero external dependencies.

Phase 1 — Foundation Hardening

| Module | Purpose | |:---|:---| | workflow-engine | Runtime state machine enforcement | | session-manager | Active session state management | | verify | Manifest integrity verification | | updater | Diff-based CLI update | | error-budget | Error budget tracking with metrics |

Phase 2 — Runtime Engine

| Module | Purpose | |:---|:---| | workflow-persistence | Persistent state + checkpoints | | agent-registry | Agent contract validation | | loading-engine | Keyword matching + implicit triggers + context budget | | hook-system | Event-driven lifecycle hooks | | task-model | Task CRUD with status tracking |

Phase 3 — Collaboration & Security

| Module | Purpose | |:---|:---| | identity | Developer identity system | | task-governance | Locking, assignment, audit trail | | skill-sandbox | Runtime skill permission enforcement | | conflict-detector | Agent conflict detection | | security-scanner | Runtime injection & secret scanning | | plugin-system | Full plugin lifecycle management |

Phase 4 — Platform Leadership

| Module | Purpose | |:---|:---| | agent-reputation | Score tracking, trends, rankings | | engineering-manager | Sprint planning, auto-assignment | | self-healing | CI failure detection & JSON patch generation | | marketplace | Community skill search & install | | cli-commands | Extracted CLI handlers for marketplace & heal |

🔄 Workflows (21)

⚖️ Operating Constraints

Immutable Rules

Trust > Optimization — Never compromise trust for speed
Safety > Growth — Prevent harm before enabling capability
No Memory of Previous Sessions — Treat each session as fresh
Explainability > Performance — Be transparent about decisions
Human Override Always Available — User can always interrupt

Governance Protocol

/explore → /plan → /work → /review → /deploy

Each phase requires explicit approval before proceeding.

🔄 Session Management

The secret to 10x productivity: Never lose context between sessions.

Antigravity AI Kit includes a robust Session Management Architecture that ensures continuity across work sessions. This is what separates casual AI usage from Trust-Grade AI development.

How It Works

┌─────────────────────────────────────────────────────────────────────┐
│                     SESSION LIFECYCLE                                │
├─────────────────────────────────────────────────────────────────────┤
│                                                                      │
│  ┌──────────────┐       ┌──────────────┐       ┌──────────────┐     │
│  │ Session      │  ───► │   WORK       │  ───► │ Session      │     │
│  │ Start Hook   │       │   SESSION    │       │ End Hook     │     │
│  └──────────────┘       └──────────────┘       └──────────────┘     │
│         │                      │                      │              │
│         ▼                      ▼                      ▼              │
│  ┌──────────────┐       ┌──────────────┐       ┌──────────────┐     │
│  │ Load Context │       │ Pre-Commit   │       │ Save State   │     │
│  │ Verify Env   │       │ Quality Gate │       │ Handoff Docs │     │
│  └──────────────┘       └──────────────┘       └──────────────┘     │
│                                                                      │
└─────────────────────────────────────────────────────────────────────┘

Key Components

Usage

Starting a Session:

Follow the session-start checklist

The AI will:

✅ Load previous session context
✅ Verify git status and branch
✅ Check dependencies and build
✅ Resume from last open task

During Work:

/verify  # Run quality checks before commits

Ending a Session:

Follow the session-end checklist

The AI will:

✅ Update session-context.md with progress
✅ Document open items and next steps
✅ Commit all changes
✅ Create handoff notes

Productivity Benefits

Example Session Context

# AI Session Context

## Last Session Summary

**Date**: February 5, 2026
**Focus**: User authentication feature

### What Was Done

- ✅ Implemented JWT refresh tokens
- ✅ Added login/logout endpoints
- [ ] Email verification (in progress)

### Open Items

1. [ ] Complete email verification
2. [ ] Add password reset flow

## Quick Resume

cd my-project && npm run dev

🔧 How to Extend

Adding Custom Agents

## <!-- .agent/agents/my-agent.md -->

name: my-agent
description: Custom agent description
triggers: [keyword1, keyword2]

---

# My Agent

Instructions for the agent...

Adding Custom Skills

## <!-- .agent/skills/my-skill/SKILL.md -->

name: my-skill
description: What this skill does
triggers: [context, keywords]

---

# My Skill

## Overview

...

## Workflow

...

Adding Custom Commands

## <!-- .agent/commands/my-command.md -->

## description: What this command does

# /my-command

Usage and instructions...

📁 Repository Structure

antigravity-ai-kit/
├── .agent/                    # Core AI Kit
│   ├── agents/               # 20 specialized agents
│   ├── commands/             # 37 slash commands
│   ├── skills/               # 34 capability modules
│   ├── workflows/            # 21 process templates
│   ├── engine/               # Autonomy Engine (state machine, loading rules, configs)
│   ├── hooks/                # 8 event hooks (runtime + git-hook)
│   ├── rules/                # 9 modular governance rules
│   ├── checklists/           # Verification checklists (4)
│   ├── templates/            # ADR, feature-request, bug-report templates
│   ├── decisions/            # Architecture Decision Records
│   └── manifest.json         # Machine-readable capability registry
├── lib/                       # Runtime Engine (29 modules)
│   ├── workflow-engine.js    # State machine enforcement
│   ├── task-governance.js    # Locking, audit trail, decision timeline
│   ├── agent-reputation.js   # Score tracking & rankings
│   ├── self-healing.js       # CI failure detection & patch generation
│   ├── marketplace.js        # Community skill marketplace
│   └── + 16 more modules     # Identity, plugins, hooks, registry...
├── bin/                       # CLI (ag-kit)
├── create-antigravity-app/    # Project scaffolder (separate npm package)
├── tests/                     # Test suites (349 tests, 34 files)
│   ├── unit/                 # Module tests (loading-engine, self-healing, plugins...)
│   ├── structural/           # Inventory + schema validation
│   └── security/             # Injection scan + leakage detection
├── docs/                      # MkDocs documentation site
├── .github/workflows/         # CI pipeline
├── .githooks/                 # Secret detection pre-commit hook
├── README.md                  # This file
├── LICENSE                    # MIT License
└── CHANGELOG.md               # Version history

🤝 Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

Development Workflow

Fork the repository
Create a feature branch
Make your changes
Run verification: /verify
Submit a pull request

📄 License

MIT License — See LICENSE for details.

👤 Author

Emre Dursun — Full-Stack Automation Engineer | AI Development Specialist

Creator of BeSync and the Trust-Grade AI Governance framework

🔗 Links

Repository: github.com/besync-labs/antigravity-ai-kit
Documentation: besync-labs.github.io/antigravity-ai-kit
Origin: Derived from BeSync Trust-Grade AI Governance

📖 Contributor Guide

Want to use Antigravity AI Kit in your project? The Contributor Guide covers the full end-to-end lifecycle:

| Phase | What You’ll Learn | |:---|:---| | Onboard | Install with ag-kit init, verify with ag-kit status | | Identity | Set up developer identity with role-based access | | Sprint Planning | Create sprints, define tasks, auto-assign agents | | Task Management | Lifecycle: pending → in_progress → review → completed | | Development | Trust-Grade workflow: /explore → /plan → /work → /review | | Quality Gates | 5-gate pipeline: lint, types, tests, security, build | | Completion | Sprint retrospective, tracking updates, next sprint |

👉 Read the full guide →

🙏 Acknowledgments

Initial patterns from vudovn/antigravity-kit
Command structure inspired by nth5693/gemini-kit
Context-driven development from Google Conductor
Hook concepts from everything-claude-code

Antigravity AI Kit v3.5.0 extends these foundations with a 29-module runtime engine, Trust-Grade governance, session management, and 100+ capabilities.