Antigravity CLI (agy) is Google's terminal agent and the successor to Gemini CLI for personal accounts. It reads AGENTS.md and GEMINI.md at the workspace root, discovers skills in .agents/skills/<name>/SKILL.md and turns each one into a slash command.
MCP servers live in their own file, .agents/mcp_config.json, and remote servers must use serverUrl rather than url or httpUrl. Agent Loadout generates that file from each loadout's mcp.json, so a loadout with MCP servers installs for Antigravity CLI without hand-editing.
AGENTS.md (read natively; GEMINI.md is also read)
.agents/skills/<name>/SKILL.md
.agents/mcp_config.json
39 native
Fail-closed PreToolUse, post-edit and Stop hooks that block force-pushes, .env reads and wild deletes, check each edit and run tests before done.
discord.js v14 rules and an add-a-slash-command skill for building a Discord bot that registers commands properly and never leaks its token.
GDScript conventions, jam-scope discipline and a ship-to-itch.io skill so your agent helps you finish a Godot 4 game in 48 hours.
Rules and an add-a-tool skill for building a Model Context Protocol server in TypeScript that agents can actually use well — clear tools, safe inputs, no stdout noise.
Rules, an RLS audit skill and a read-only Supabase MCP config for vibe-coding a Next.js App Router SaaS without leaking your database.
A launch-check skill and always-on security rules that catch leaked keys, open databases and missing auth before your vibe-coded app goes live.
A stack-agnostic working style for coding agents — plan first, small diffs, never delete tests to go green — plus handoff notes and up-to-date docs via MCP.
Review-only rule files for Claude Code Review, Cursor Bugbot and Copilot, built from one list: severity, a nit cap, what to skip and what to always check.
Astro conventions for fast, zero-JS-by-default sites with content collections, plus a skill for deploying to Cloudflare Pages with headers and redirects.
Make the agent load every UI change in a real browser via Playwright MCP, exercise it, fail on console or network errors and attach screenshots at two widths.
Manifest V3 rules for service workers, messaging and minimal permissions, plus a skill that gets your extension through Chrome Web Store review.
Workers rules for agents: bindings from env, wrangler types, prepared D1 queries, waitUntil, secrets via wrangler secret, no deploys or --remote writes unasked.
How to decide between AGENTS.md, Cursor .mdc rules, skills and hooks, write front matter that loads, and check that a rule is really attached.
Django rules and a safe-migration skill: migrations checked in CI, two-step destructive changes, N+1 caught with query-count tests, settings from the environment.
Build LLM and agent evals the practitioner way: read real traces first, tally failure categories, then turn each into code-checked regression cases.
Expo rules for agents plus Expo's official MCP server: npx expo install, Expo Router, no secrets in EXPO_PUBLIC_, no production EAS releases, check on a simulator.
Rules and an endpoint skill for FastAPI with uv and Pydantic v2: Annotated dependencies, typed responses, async only when awaited, lifespan, tested routes.
Go rules for agents: wrapped errors with errors.Is, context first, owned goroutines, 1.22+ routing, timeouts, and gofmt, vet and go test -race as the gate.
Drive Godot 4 and Blender from one agent: Godot MCP to run the game and read errors, MCP for Blender in safe mode to model assets, and a glTF pipeline between them.
Run an agent in a Ralph-style loop with guard rails: a fresh process per pass, state in files, iteration and no-progress limits, and a check it cannot fake.
Rules and a GitHub Actions template for unattended claude -p or codex exec: bare mode, read-only, a spend cap, pinned versions, PR text as untrusted data.
Let the agent build, test, run and screenshot iOS and macOS apps through MobileBuildMCP, pinned with telemetry off, with Swift concurrency and pbxproj rules.
Notebook rules for agents: edit Jupytext-paired .py files instead of JSON, run top to bottom in a fresh kernel, keep outputs out of git with nbstripout, check data first.
Keep one AGENTS.md as the source of truth, with thin imports and a single MCP list, so Claude Code, Codex, Cursor, Copilot and Antigravity CLI all load the same rules.
Rules and two skills for running several coding agents at once, one task per git worktree: isolation, ports, dependencies, rebasing and clean-up.
Phaser 4 rules that send the agent to the skills Phaser ships in node_modules, not v3 tutorials: pooled objects, delta movement, play-tested changes.
Commit and PR rules for coding agents: small PRs, a four-part body, screenshots with gh --attach and a leak check for session links and secrets before pushing.
Rules for reproducible, single-file Python data scripts with uv and inline dependencies, plus a skill that sanity-checks a dataset before anyone trusts the numbers.
Luau conventions for Rojo projects with a server-authoritative mindset, plus a skill that reviews every RemoteEvent for exploits before you publish.
Rust rules for agents: typed errors instead of unwrap, clippy -D warnings, fmt and tests as the done-gate, no unsafe without a SAFETY note, no surprise MSRV bumps.
Before calling work done, the agent gets a read-only review of its diff from a different model, checks each finding and stops after two rounds.
Lightweight spec-driven flow: a SPEC.md with testable acceptance criteria, small tickets, test-first builds in parallel waves and a review against the spec.
Rules for when to delegate, which model and effort each subagent gets, and a brief template, with ready subagents for Claude Code and Codex.
Supabase MCP scoped to one dev project, read-only, database and docs tools only; every change as a reviewed migration with RLS, policy and expand-contract rules.
Tauri 2 rules and a permissions skill: capabilities for named windows, scoped fs and http, no shell execute, a CSP kept, plus a checker that flags wide grants.
Rules and a command skill for Telegram bots on python-telegram-bot v22: token from the environment, async-safe handlers, testable logic, escaped output, webhook secrets.
Let agents write Terraform and produce plans but never apply: a guard hook for apply, destroy and state changes, a plan summary that flags deletes, pinned providers.
Project rules for using Unity's official agent plugin safely in Claude Code or Codex: commit first, Editor-driven scene edits, .meta discipline, compile and test checks.
Unity C# conventions that respect the editor, .meta files and serialisation, plus a skill for scripted web builds ready for itch.io.
Connect the read-only Agent Loadout MCP server and ask Antigravity CLI for "a loadout for my Godot jam". It can search, read files and get install paths for Antigravity CLI without you copying anything. Set-up for every agent.
{"mcpServers":{"agent-loadout":{"serverUrl":"https://loadout.jnw-it.com/mcp"}}} Antigravity CLI reads AGENTS.md (read natively; GEMINI.md is also read). Agent Loadout keeps the shared rules in AGENTS.md so one set of rules works across agents.
Each skill is a folder with a SKILL.md file. For Antigravity CLI they go in .agents/skills/<skill-name>/SKILL.md, relative to the project root.
Project MCP servers go in .agents/mcp_config.json. Antigravity CLI reads workspace MCP servers from .agents/mcp_config.json. Agent Loadout generates that file from mcp.json: remote servers use serverUrl (url and httpUrl are not supported). Check the servers with /mcp. Never commit real tokens: version 1.4.0 (16 October 2026) fixed ${VAR} expansion in headers for plugin MCP servers, but the MCP docs do not cover workspace files, so keep tokens in the global ~/.gemini/config/mcp_config.json.
Add the read-only server at https://loadout.jnw-it.com/mcp: {"mcpServers":{"agent-loadout":{"serverUrl":"https://loadout.jnw-it.com/mcp"}}}. It needs no sign-in, and lets Antigravity CLI search loadouts and fetch exact install steps itself.
No. Antigravity CLI reads AGENTS.md at the workspace root directly, so loadouts install AGENTS.md and skills only. If you also keep a GEMINI.md for Gemini CLI, Antigravity reads both; keep GEMINI.md to an import of AGENTS.md so rules are not duplicated.
Move workspace skills from .gemini/skills/ to .agents/skills/, and MCP servers from .gemini/settings.json into .agents/mcp_config.json with serverUrl for remote servers. agy plugin import gemini converts Gemini CLI extensions into Antigravity plugins.
Other agents: Claude Code · Codex CLI · Cursor · Gemini CLI · GitHub Copilot. By stack: Bash · Cloudflare · C# · GDScript · Git · GitHub · Godot · iOS · JavaScript · jq · Node.js · Postgres · pytest · Python · Rust · Supabase · TypeScript · Unity · uv. Antigravity CLI documentation.