Welcome to the Agent Playbook

This is my official log of exploring Agentic Workflows.

In this blog, I will be document my journey of building alongside AI agents, exploring the boundaries of what’s possible when the editor becomes an active collaborator.

What to expect:

• Deep dives into Antigravity features.
• Real-world automated workflows.
• Technical translated guides.

Stay tuned as we build the future!

Google Antigravity: A Beginner’s Guide to AI-Powered Coding

Coding used to be about memorizing syntax and typing thousands of lines of text. Google Antigravity changes that. It’s a new tool that turns coding into a conversation, making it accessible even if you aren’t a technical wizard.

What is Antigravity?

Think of Antigravity as a “Smart Editor”.

• Old Way: You open a blank text file and type code manually. If you make a typo, everything breaks.
• Antigravity Way: You have a chat window where you talk to an expert AI. You say, “Create a website for my coffee shop with a menu and contact form,” and the AI writes the code, creates the files, and even runs it for you.

It’s like having a senior software engineer sitting right next to you, ready to do the heavy lifting while you direct the project.

Why Should You Use It?

1. Focus on Ideas, Not Typos

You don’t need to stress about missing semicolons or specific commands. Focus on what you want to build (e.g., “A personal blog”), and let Antigravity handle how to build it.

2. Learn as You Go

Antigravity doesn’t just do the work; it explains it. You can ask, “Why did you use that file?” or “How do I change the color?” and it will teach you.

3. Build Faster

Projects that used to take weeks can now be set up in minutes. You can go from a blank screen to a working prototype in a single coffee break.

How to Install It

Getting started is as easy as installing a web browser.

Step 1: Download

Visit the official website at antigravity.google and click the Download button for your computer (Mac, Windows, or Linux).

Step 2: Install

Open the downloaded file and follow the standard installation prompts (dragging to Applications on Mac, or clicking “Next” on Windows).

Step 3: Sign In

Launch Antigravity and sign in with your Google account.

Step 4: Your First Project

Once open, you’ll see a chat box. Type something simple like:

“Create a simple ‘Hello World’ website with a blue background.”

Watch the magic happen! ✨

---

⚡ Summary / Quick Recap

1. Download: Go to antigravity.google and get the installer.
2. Install: Run the installer like any other app.
3. Login: Open the app and sign in with Google.
4. Create: Type your idea in the chat and hit Enter! 🚀

A brief guide to understanding the unique features of Google Antigravity, the next-generation AI-powered development environment.

What is Antigravity?

Antigravity is an agentic IDE that transforms how developers work. Instead of writing code line-by-line, you describe tasks at a higher level and let AI agents handle the implementation. The agent can:

• Work across multiple files simultaneously
• Run terminal commands
• Browse the web for research
• Create implementation plans
• Verify its own work

---

📁 GEMINI.md - The Agent’s Instructions

GEMINI.md files are markdown files that tell the agent how to behave for your project.

Types of GEMINI.md

Location	Scope	Example Use Case
~/.gemini/GEMINI.md	Global - All projects	”Always use TypeScript”, “Prefer pnpm over npm”
.gemini/GEMINI.md	Workspace - Current project	”Use Next.js App Router”, “Follow our API patterns”

Example GEMINI.md

# Project Guidelines

## Tech Stack

- Framework: Next.js 14 with App Router
- Styling: Tailwind CSS
- Database: Supabase

## Coding Standards

- Use TypeScript strict mode
- Write tests for all new features
- Follow conventional commits

## Don't

- Never use `any` type
- Don't commit .env files

---

🧠 Skills - Extending Agent Capabilities

Skills are reusable instruction sets that teach the agent how to perform specialized tasks. They follow the agentskills.io standard.

Where Skills Live

~/.gemini/antigravity/skills/    # Global skills (all projects)
.agent/skills/                    # Project-specific skills

Skill Structure

skills/
└── deploy-to-vercel/
    ├── SKILL.md           # Required - main instructions
    ├── scripts/           # Optional - helper scripts
    └── examples/          # Optional - reference code

Example SKILL.md

---
name: Deploy to Vercel
description: Deploy projects to Vercel with optimal settings
tags: [deployment, vercel, nextjs]
---

## Overview

This skill enables deployment to Vercel.

## Instructions

1. Check for vercel.json configuration
2. Run `vercel --prod` for production deploys
3. Verify deployment URL is accessible

## Examples

- `vercel deploy --prod`
- `vercel env pull .env.local`

---

📚 Knowledge - Agent’s Memory

The agent can remember important information across sessions.

Types of Knowledge

Type	Description
Facts	Things you’ve told the agent (“I prefer tabs over spaces”)
Preferences	Your working style and preferences
Project Context	Information about your codebase learned over time

How to Use

Just tell the agent something important and it will remember:

You: "Remember that we always use pnpm in this project"
Agent: "Got it! I'll use pnpm for package management."

---

🎛️ Agent Manager - Control Center

The Agent Manager is the main interface for interacting with the AI agent. It’s where conversations happen.

Key Features

Feature	Description
Conversation History	All your chats organized by session
Model Selection	Switch between Gemini, Claude, GPT models
Mode Toggle	Switch between Planning and Fast mode
Task Progress	See what the agent is working on

Modes

Mode	Best For	Behavior
Planning Mode	Complex tasks	Creates implementation plans, organizes into task groups
Fast Mode	Quick fixes	Direct execution, minimal planning

---

📋 Artifacts - Agent’s Work Products

Artifacts are structured documents the agent creates to maintain transparency and control.

Types of Artifacts

Artifact	Purpose
Task List	Live progress tracker (Research → Implementation → Verification)
Implementation Plan	Detailed proposal of changes (requires your approval!)
Walkthrough	Guided tour of changes after completion

Implementation Plan Example

# Implementation Plan: Add Dark Mode

## Overview

Add dark mode toggle to the application.

## Changes Required

### 1. Create Theme Context

- File: `src/context/ThemeContext.tsx`
- Add React context for theme state

### 2. Update CSS Variables

- File: `src/styles/globals.css`
- Define dark mode color palette

### 3. Add Toggle Component

- File: `src/components/ThemeToggle.tsx`
- Create toggle button with icon

## Verification

- [ ] Toggle switches theme
- [ ] Preference persists in localStorage
- [ ] No flash of wrong theme on load

---

🔄 Workflows - Reusable Procedures

Workflows are step-by-step procedures the agent follows for common tasks.

Where Workflows Live

.agent/workflows/           # Project workflows
~/.gemini/workflows/        # Global workflows

Workflow Format

---
description: How to deploy to production
---

## Steps

1. Run tests
   ```bash
   npm run test
   ```

2. Build the application

   npm run build

3. Deploy to production // turbo

   vercel --prod

> 💡 The `// turbo` annotation allows auto-execution of that step

---

## 📏 Rules - Behavioral Constraints

**Rules** are constraints that guide agent behavior. More specific than GEMINI.md, they're for targeted guidance.

### Location

.gemini/rules/ # Project rules

### Example Rule

```markdown
---
description: API response formatting rules
---

## API Responses

All API endpoints must:
1. Return JSON with consistent structure
2. Include `success` boolean field
3. Include `data` or `error` field
4. Use HTTP status codes correctly

## Response Format

```json
{
  "success": true,
  "data": { ... }
}

{
  "success": false,
  "error": {
    "code": "NOT_FOUND",
    "message": "Resource not found"
  }
}

---

## 🔌 MCP Servers - External Connections

**MCP (Model Context Protocol)** allows the agent to connect to external tools and services.

### Supported Connections

| Type | Examples |
|------|----------|
| **Databases** | SQLite, PostgreSQL, MySQL |
| **Services** | Sentry, Linear, GitHub, GitLab |
| **Local Tools** | Custom scripts, CLIs |

### Configuration

MCP servers are configured in your settings. Once connected, the agent can:

- Query your database directly
- Create issues in Linear
- Read error logs from Sentry
- Interact with your GitHub repos

### Example Use Cases

You: “Show me the latest errors from Sentry” Agent: [queries Sentry via MCP] “Found 3 new errors in the last hour…”

You: “Create a Linear issue for this bug” Agent: [creates issue via MCP] “Created issue DEV-123: Fix login timeout”

---

## 🤖 Available Models

Antigravity supports multiple AI models that you can switch between:

| Model | Provider | Best For |
|-------|----------|----------|
| **Gemini 3 Pro** | Google | Complex reasoning, large context |
| **Gemini 3 Flash** | Google | Fast responses, everyday tasks |
| **Claude 4.5 Sonnet** | Anthropic | Coding, nuanced understanding |
| **Claude 4.5 Opus** | Anthropic | Most capable, complex tasks |
| **GPT-OSS** | OpenAI | General purpose |

### Switching Models

Use the model selector in the Agent Manager, or specify in your message:

/model gemini-3-pro

---

## 🌐 Browser Subagent - Web Automation

The **Browser Subagent** is a specialized agent that can control a web browser.

### Capabilities

| Action | Example |
|--------|---------|
| **Navigate** | Open URLs, follow links |
| **Read** | Extract content from web pages |
| **Interact** | Click buttons, fill forms |
| **Screenshot** | Capture visual state |
| **Test** | Run UI tests, verify deployments |

### Use Cases

You: “Check if our staging site is working” Agent: [opens browser, navigates to site, checks for errors] “Staging is up. Homepage loads in 1.2s. No console errors.”

You: “Read the latest Next.js docs on caching” Agent: [browses docs, extracts information] “Here’s a summary of the caching strategies…”

You: “Create an issue on GitHub for this bug” Agent: [navigates to GitHub, creates issue] “Created issue #42: Fix authentication timeout”

---

## 🏁 Quick Reference

| Feature | Location | Purpose |
|---------|----------|---------|
| GEMINI.md | `.gemini/` or `~/.gemini/` | Project/global instructions |
| Skills | `.agent/skills/` | Extend agent capabilities |
| Knowledge | Agent memory | Persistent facts & preferences |
| Workflows | `.agent/workflows/` | Step-by-step procedures |
| Rules | `.gemini/rules/` | Behavioral constraints |
| MCP | Settings | External service connections |

---

## 🔗 Resources

- **Documentation**: [antigravity.google/docs](https://antigravity.google/docs)
- **Skills Standard**: [agentskills.io](https://agentskills.io)

---

*Build the new way! 🚀*