AntiGravity: O Guia DEFINITIVO que ninguém fez

Join our AI Incubator Community now, the shortest path between you and your first client as a freelancer or with your AI Agency. 🚀 https://www.lionlabs.com.br Examples of Rules: https://drive.google.com/file/d/1NDrD... Google AntiGravity - Agent-based IDE launched in November 2025 along with Gemini 3. Development platform that puts AI agents at the center of the workflow. Orchestration of autonomous agents that plan, execute, test, and validate code. Fundamentals IDE built on a VS Code fork with Agent First architecture Difference between those who know how to develop and those who don't using these tools Reality of Vibe Coding: security flaws, duplicated code, asynchronous logic Operating Modes Agent-Driven Development: total autonomy, ideal for scaffolding Agent-Assisted Development: balance between autonomy and control (recommended) Review-Driven Development: strict control, approval of each step Important Settings Secure Mode: paranoid level of approval Review Policy: Always Proceed, Agent Decides, Request Review Terminal Command Auto Execution: control of commands in the terminal File Access: Git Ignore and Non-Workspace File Access Auto-fix Lint: automatic correction of syntax errors Auto-continue: resumes after rate limits Conversation History and Knowledge: agent memory and context Rules Global Rules: apply to all projects (Person, tech stack, definition of done) Workspace Rules: specific to the current project (mandatory stack, code standards, security) Location: .antigravity/rules.md Workflows Saved recipes or prompts triggered with /command Examples: /readme, /audit, /turbo Global and Workspace Location: .agent/workflows/ MCPs (Model Context Protocol) Connection to external services: Supabase, GitHub, Stripe, Postgres, Firebase, Notion ⚠️ Caution with direct connection to production databases Use a test database during development Agents and Models Opus: best for back-end tasks Gemini Pro: best for front-end tasks Parallel agents: beware of file conflicts Strategy: tasks in different files for safe parallelization Skills Not native to AntiGravity, Created via files Location: .agent/skills/ Works automatically (unlike workflows) Examples: git commit formatter, license header, API patterns Optimization Tips Create PLAN.md before running Use SUMMARY.md at the end of sessions Separate projects into folders Always review artifacts before accepting Disable unused MCPs to save context window Use checkpoints in long tasks Document decisions in comments Rate Limit Saving Strategy Use Gemini and Claude outside of AntiGravity for planning Connect both to GitHub for code access AntiGravity only for execution Trio: AntiGravity + Claude + Gemini 00:00 AntiGravity Intro 03:37 AntiGravity IDE Agent Style Configuration 05:43 AntiGravity Security Configuration 11:25 Global Rules and AntiGravity Workspace Rules 14:55 Global Workflows and Agent Workflows 17:57 AntiGravity MCPs: How to Use Them and Security Precautions 22:17 Agent Manager and Agents Working in Parallel 25:12 Skills to Use in AntiGravity 26:35 Usage Tips and How to Get the Most Out of AntiGravity