A visual, example-driven curriculum from everyday prompts to multi-agent /boost & /teamwork-preview orchestration, Generative UI artifacts, OS-sandboxed hooks, native plugins, and the Python SDK.
English | Tiếng Việt
Quick Start • Learning Roadmap • Quick Reference • Template Catalog • Concepts Guide • Topic Index
Most developers use only a fraction of Google Antigravity and Gemini—typing free-form chat prompts while missing the customization primitives and autonomous orchestration engines that turn Antigravity into a programmable engineering platform:
- Progressive-Disclosure Rules & Memory (
GEMINI.md,AGENTS.md,.agents/rules/*.mdwithalways_on,model_decision,glob, andmanualtriggers inside a 20,000-token budget) - Dual-Mode Agent Skills (
.agents/skills/<name>/SKILL.mdthat work simultaneously as/<name>slash commands and semantically auto-discovered capabilities) - Worktree-Isolated Subagents (
research,browser,self, and custom.agents/agents/*.mdrunning in parallel acrossinherit,branch, orshareworkspaces) - Deterministic Lifecycle Hooks (
.agents/hooks.jsoninterceptingSessionStart,PreToolUse,PostToolUse,PreInvocation,PostInvocation, andStop) - Rich Artifacts, Generative UI & Sidecars (Implementation Plans with inline review comments, theme-aware
<agent-embed>HTML widgets, andsidecar.jsoncron automations) - Frontier Multi-Agent Modes (
/boost3-phase deep reasoning,/teamwork-previewautonomous engineering teams with adversarialCritic/Challenger/Auditorgates, and/goal)
Every module in this repository includes Mermaid architecture diagrams, production-ready copy-paste templates, Do's and Don'ts, and interactive quizzes.
When you open this repository in the Antigravity App or Antigravity CLI (agy), two interactive skills in .agents/skills/ are automatically available:
/self-assessment # Take a 10-question diagnostic to get a personalized learning path
/lesson-quiz 01 # Test your mastery of any module (01 through 10) with a 5-question quiz
Tip
Both skills use Antigravity's native interactive ask_question UI modal—giving you clickable multiple-choice options and instant architectural explanations.
flowchart TB
subgraph Surfaces["Multi-Surface Antigravity Ecosystem"]
App["Antigravity App v2.24<br/>(Desktop & Web)"]
CLI["Antigravity CLI v1.4<br/>(agy TUI & -p Headless)"]
IDE["IDE Extensions & Antigravity IDE<br/>(VS Code, JetBrains, Zed, Xcode)"]
SDK["Python SDK v0.1.22<br/>(google-antigravity)"]
end
subgraph Core["Agent Runtime & Frontier Models"]
Models["Gemini 4 Argon • Gemini 3.8 Flash • Gemini 3.1 Pro • Nano Banana 2 • Gemma 4 (LiteRT)"]
Modes["Execution Modes: Interactive • /plan • /goal • /boost (3-Phase) • /teamwork-preview (2-Phase)"]
end
subgraph Customizations["Workspace (.agents/) & Global (~/.gemini/config/) Customizations"]
Rules["02. Rules & Memory<br/>GEMINI.md, AGENTS.md,<br/>.agents/rules/*.md"]
Skills["01 & 03. Skills & /Commands<br/>.agents/skills/<name>/SKILL.md"]
Agents["04. Subagents<br/>.agents/agents/*.md<br/>(inherit | branch | share)"]
MCP["05. MCP Servers<br/>mcp_config.json<br/>(Stdio & Streamable HTTP)"]
Hooks["06. Lifecycle Hooks<br/>.agents/hooks.json<br/>(6 Lifecycle Events)"]
Plugins["07. Native Plugins<br/>plugin.json Bundles"]
end
subgraph StateAndSafety["State, Visual Output & Kernel Security"]
Artifacts["08. Checkpoints, Implementation Plans,<br/>Generative UI (<agent-embed>) & Sidecars"]
Sandbox["09. Permissions (Deny > Ask > Allow)<br/>& OS Sandbox (Seatbelt / nsjail / AppContainer)"]
end
Surfaces --> Core
Core --> Customizations
Core --> StateAndSafety
| Module | Topic | Difficulty | Time | What You Will Build & Master |
|---|---|---|---|---|
| 01. Slash Commands | Built-in & Custom /commands |
Beginner | 30m | Master 35+ built-in commands (/boost, /teamwork-preview, /goal, /plan, /grill-me, /learn, /btw) and build 8 custom skill-backed slash commands (/optimize, /pr, /commit, etc.). |
| 02. Rules & Memory | GEMINI.md, AGENTS.md, .agents/rules/ |
Beginner | 40m | Configure hierarchical GEMINI.md/AGENTS.md, modular rules with 4 triggers (always_on, model_decision, glob, manual), @[label](path) includes, the 20k token budget, .agents/rules.json, and /learn. |
| 03. Agent Skills | .agents/skills/<name>/SKILL.md |
Intermediate | 45m | Build multi-file Open Agent Skills (agentskills.io) with 3-stage progressive disclosure, bundled scripts/, references/, and 6 production skill packages. |
| 04. Subagents | .agents/agents/*.md & Worktrees |
Intermediate | 45m | Orchestrate research, browser, self, and 9 custom specialist subagents across isolated inherit, branch, and share workspaces with reactive wakeup messaging. |
| 05. MCP Protocol | mcp_config.json (Stdio & HTTP) |
Intermediate | 45m | Connect GitHub, PostgreSQL, Sentry, and filesystem MCP servers via Stdio and Streamable HTTP (${VAR} header expansion), scope mcp(server/tool) permissions, and use Fix with Agent. |
| 06. Lifecycle Hooks | .agents/hooks.json (6 Events) |
Intermediate–Adv | 60m | Intercept Se
6870
ssionStart, PreToolUse, PostToolUse, PreInvocation, PostInvocation, and Stop using the camelCase JSON contract to block secrets, auto-format code, and enforce test gates. |
| 07. Native Plugins | plugin.json & agy plugin CLI |
Advanced | 50m | Package skills, subagents, rules, hooks.json, mcp_config.json, and sidecars into distributable plugins (pr-review, devops-automation, documentation) and validate with agy plugin validate. |
| 08. Checkpoints, Artifacts & Sidecars | /rewind, /fork, Generative UI, sidecar.json |
Intermediate | 45m | Rewind code/chat state, review Implementation Plans with inline comments, build theme-aware <agent-embed> HTML widgets, and run sidecar.json daemons & CRON_TZ scheduled automations. |
| 09. Advanced Features | /boost, /teamwork-preview, Permissions, Sandbox |
Advanced | 60m | Deploy 3-phase /boost and 2-phase /teamwork-preview agent teams, tune Gemini 4 Argon effort, author Deny > Ask > Allow permissions with regex:, enforce OS kernel sandboxing, and use agy remote-control. |
| 10. CLI & Python SDK | agy v1.4 & google-antigravity v0.1.22 |
All Levels | 60m | Master agy TUI shortcuts (alt+m, Ctrl+J/Ctrl+K), -p headless CI/CD pipelines, and programmatic agents in Python (Agent, @beta.workflows.define, LiteRTAgentConfig for local Gemma 4). |
If you previously used Claude Code (claude-howto), here is how every concept maps—and upgrades—in Google Antigravity & Gemini:
| Capability | Claude Code | Google Antigravity & Gemini | Key Antigravity Upgrade |
|---|---|---|---|
| CLI Binary | claude |
agy (~/.local/bin/agy) |
Alt-Screen/Inline TUI, alt+m live queue toggle, Ctrl+J subagent jump, OSC 7501 tab status |
| Workspace Memory | CLAUDE.md, .claude/rules/ |
GEMINI.md, AGENTS.md, .agents/rules/*.md |
4 explicit triggers (always_on, model_decision, glob, manual), @[label](path) inlining, .agents/rules.json inheritance, /learn |
| Custom Slash Commands | .claude/commands/*.md |
.agents/skills/<name>/SKILL.md |
Unified model: every Skill is automatically BOTH a /<name> slash command AND semantically auto-discovered |
| Subagent Isolation | .claude/agents/*.md |
.agents/agents/*.md + invoke_subagent |
Per-invocation Workspace: "inherit" | "branch" | "share", dynamic define_subagent, reactive wakeup (no polling) |
| Hooks Config | settings.json (hooks key) |
.agents/hooks.json & ~/.gemini/config/hooks.json |
Dedicated hooks.json with 6 events (SessionStart, PreToolUse, PostToolUse, PreInvocation, PostInvocation, Stop), tool input overwrite, and injectSteps |
| Plugins | .claude-plugin/plugin.json |
<plugin-root>/plugin.json |
Bundles skills/, agents/, rules/, root hooks.json, root mcp_config.json, and sidecars/ + agy plugin validate --strict |
| Visual Output & Plans | Terminal Markdown | Artifacts & Generative UI | Implementation Plans with inline line-by-line review comments, PDF export, and live <agent-embed> HTML/Tailwind widgets |
| Background & Cron | /loop or OS crontab |
Sidecars (sidecar.json) & /schedule |
Managed background daemons (restart_policy), sidecar:// UI panels, and timezone-aware CRON_TZ automations with scoped access_grants |
| Deep Multi-Agent Modes | Manual subagents | /boost, /teamwork-preview, /goal |
First-class 3-phase hypothesis/worktree exploration (/boost) and 2-phase autonomous teams with adversarial Critic/Challenger/Auditor gates (/teamwork-preview) |
| Programmatic SDK | Agent SDK | google-antigravity Python SDK |
Supports Gemini 4 Argon, declarative @beta.workflows.define (parallel, pipeline), and offline on-device Gemma 4 via LiteRTAgentConfig |
# macOS / Linux — Install Antigravity CLI (v1.4.0+)
curl -fsSL https://antigravity.google/cli/install.sh | bash
# Windows (PowerShell)
irm https://antigravity.google/cli/install.ps1 | iex
# Optional: Install the Google Antigravity Python SDK (v0.1.22+)
pip install google-antigravitygit clone https://github.com/minhxuvi/gemini-howto.git
cd gemini-howto
# Launch Antigravity CLI in this workspace
agyInside the Antigravity prompt (App or CLI), run:
/self-assessment
# 1. Copy the workspace GEMINI.md & modular rules into your project
cp 02-memory/project-GEMINI.md /path/to/your-project/GEMINI.md
mkdir -p /path/to/your-project/.agents/rules
cp 02-memory/rules/*.md /path/to/your-project/.agents/rules/
# 2. Add a custom /optimize slash-command skill
mkdir -p /path/to/your-project/.agents/skills/optimize
cp 01-slash-commands/optimize.md /path/to/your-project/.agents/skills/optimize/SKILL.md
# 3. Add a read-only Security Reviewer subagent
mkdir -p /path/to/your-project/.agents/agents
cp 04-subagents/secure-reviewer.md /path/to/your-project/.agents/agents/
# 4. Enable secret-scanning & auto-formatting lifecycle hooks
cp 06-hooks/hooks.json /path/to/your-project/.agents/hooks.json
mkdir -p /path/to/your-project/.agents/hooks
cp 06-hooks/*.sh 06-hooks/*.py /path/to/your-project/.agents/hooks/
chmod +x /path/to/your-project/.agents/hooks/*.sh /path/to/your-project/.agents/hooks/*.pyNot sure where to begin? Pick the track that matches your role, or see the full Learning Roadmap:
- Track A — Everyday Productive Developer (~2 hours):
- Track B — Workflow & Team Automator (~3.5 hours):
- Track C — Staff / Platform Architect (~4 hours):
LEARNING-ROADMAP.md— Visual dependency graphs, time estimates, and milestone checklists.QUICK_REFERENCE.md— Single-page cheat sheet for every slash command, frontmatter field, hook JSON schema, permission rule, and CLI flag.CATALOG.md— Browse and copy all 50+ ready-to-use templates across Modules 01–10.gemini_concepts_guide.md— Deep architectural mental models: how Antigravity assembles the context window, when to choose a Rule vs Skill vs Subagent vs Hook vs Plugin, and how/boostand/teamwork-previewcoordinate multi-agent verification.INDEX.md— Alphabetical and problem-oriented index ("How do I blockgit push --force?", "How do I run Gemma 4 offline?").
Distributed under the MIT License.