# Antigravity.md

> The definitive guide to Google Antigravity: mission control for autonomous AI coding agents.

## Overview

Google Antigravity is an AI-powered IDE that fundamentally reimagines software development from line-by-line coding to agent orchestration. Instead of being a text editor with AI assistance, it's a mission control center where you manage autonomous agents that execute tasks asynchronously.

## Key Concepts

### Agent-First Development
Traditional IDEs (Cursor, VS Code Copilot) are "editor-centric" - they wait for keystrokes and offer suggestions. Antigravity is "agent-centric" - you give high-level objectives to autonomous agents that work independently while you focus on architecture.

### Swarm Development
Multiple specialized agents work in parallel: one refactors authentication, another writes E2E tests, a third debugs memory leaks - all simultaneously. You orchestrate and approve their work.

### Configuration: GEMINI.md vs AGENTS.md
- **GEMINI.md**: Antigravity-native configuration for the Gemini model
- **AGENTS.md**: Universal standard that works across tools (Cursor, Windsurf, etc.)
- Priority: AGENTS.md → GEMINI.md → Built-in defaults
- Most developers use AGENTS.md for portability

### Model Context Protocol (MCP)
MCP connects agents to external systems: databases (Postgres, MySQL, MongoDB), Git, Slack, Linear, and APIs. Extends agents beyond local file access to interact with your entire development ecosystem.

### Skills Architecture
Reusable procedural knowledge packages that teach agents specific tasks. Instead of explaining error handling to every agent, create a SKILL.md that encapsulates the pattern.

Skills are discovered in:
- `.claude/skills/` (project-specific)
- `~/.config/claude/skills/` (global)
- Public repositories

### Production Workflows

**NotebookLM + Antigravity Pipeline:**
1. Research phase (NotebookLM): Feed requirements and docs, generate implementation plan
2. Review: Export plan as structured document
3. Orchestrate (Antigravity): Import plan, assign tasks to specialized agents
4. Iterate: Agents execute, you review, approve, loop continues

## When to Use Antigravity

- Multi-module projects requiring parallel development
- Greenfield projects and rapid prototyping
- Teams comfortable with agent orchestration workflows
- Projects with clear boundaries between components

## Safety Considerations

Implement proper guardrails (see guardrails.md):
- Clear instructions in AGENTS.md
- Approval workflows for destructive operations
- GUARDRAILS.md to capture learned constraints
- Isolated agent domains to prevent conflicts
- Human checkpoints for critical paths

## Common Patterns

**Test-Driven Agent Development:**
- Agent A: Write comprehensive E2E tests from requirements
- Agent B: Implement features until tests pass
- Agent C: Refactor for performance while maintaining green tests

**Parallel Feature Development:**
- Agent 1 (feature/auth): OAuth2 implementation
- Agent 2 (feature/payments): Stripe integration
- Agent 3 (feature/admin): Admin dashboard
- You orchestrate merges after reviewing artifacts

## Troubleshooting

- **Agent stuck in loop**: Check GUARDRAILS.md to prevent repeated failures
- **Merge conflicts**: Better domain isolation via `@workspace_scope` directives
- **Slow performance**: Reduce MCP server count, use lazy-loading gateways
- **Context pollution**: Force context reset, use GUARDRAILS.md for failure patterns

## Related Resources

- Official site: https://antigravity.google/
- Documentation: https://antigravity.google/docs
- Safety protocols: https://guardrails.md
- AGENTS.md spec: https://agents.md
- MCP documentation: https://modelcontextprotocol.io/
- Why AGENTS.md matters more than Skills (Vercel): https://jpcaparas.medium.com/vercel-says-agents-md-matters-more-than-skills-should-we-listen-d83d7dc2d978

---

Curated by NMA.vc | https://nma.vc