Background
With generative AI entering daily development, the AI assistant in the terminal has also ushered in an epic update! If you are a loyal supporter of the original Gemini CLI, you may already know that this tool will be officially retired on June 18, 2026.
Taking over the torch of this era is Google's stunning launch at I/O 2026, the next-generation lightweight, Go language-driven multi-agent terminal UI assistant —— Antigravity CLI (called agy in the terminal)!
However, the launch of new tools is always accompanied by various pitfalls and surprises. This article will focus on Antigravity CLI (agy), revealing how to deal with the "invisible color scheme hell", how to enable the addictive YOLO no-confirmation frenzy mode, and those terminal black technologies and setting secrets hidden deep within settings.json!
🛠️ Step 1: Antigravity CLI (agy) Color Scheme Savior for Invisible Text!
When installing and launching agy for the first time, the first blow that many developers accustomed to macOS / Linux dark background terminals usually face is: "The font is all black, and the text is completely invisible!"
This is because the agy default configuration file may be configured with a light (Light) theme. We don't need to compromise and change our favorite terminal background, just modify settings.json and it can be saved with one click!
🛠️ Steps to Fill the Pit
Find the global configuration file for Antigravity CLI, the path is usually:
~/.gemini/antigravity-cli/settings.jsonModify the
"colorScheme"setting value from"light"to"dark":After saving the file and restarting the terminal, all outputs will automatically convert to a high-contrast dark mode color scheme, and your eyes will be saved instantly!
🔥 The Main Event: YOLO Mode —— Unlock the "No Confirmation" Ultimate Move for Unlimited Automated Execution
When using AI to write code, the most annoying thing is that every time you make a file modification and execute a git command, the CLI will pop up a question: "Are you sure you want to perform this operation? (y/N)". This is simply a double torment of fingers and spirit when performing large-scale refactoring or batch tasks.
To this end, agy provides two levels of YOLO (You Only Live Once) no-confirmation automatic execution mode, allowing AI to smoothly and continuously execute autonomously until the task is completed:
1. ⚡ Extreme YOLO: --dangerously-skip-permissions Parameter
If you are in a completely isolated and secure sandbox environment, or have 100% confidence in the instructions generated by AI, you can add this ultimate move when starting:
agy --dangerously-skip-permissions
Once this Flag is added, agy will completely skip all tool authorization and command execution confirmation prompts, and enter an "all the way to the top" automatic execution state. Suitable for letting it run complex automated tests or file migrations on its own!
2. 🛡️ Moderate Control: /permissions Fine-grained Settings
If you don't want to risk the AI executing rm -rf, you can directly enter /permissions in the CLI or directly modify settings.json. Through a whitelist mechanism, only specific commands or paths are automatically approved:
{
"permissions": {
"allow": [
"read_file(/Users/al03034132/Documents)",
"command(git)",
"command(npm test)"
],
"deny": [
"command(rm -rf)"
]
}
}
This way, you can allow Git operations and unit tests to enter the YOLO no-confirmation state, while also ensuring the security of the core file system!
🤫 Those Unknown agy Black Technologies and Setting Secrets
As Google's latest official Code-first agent weapon, agy also has several hidden functions deep within the configuration file and commands, which are rarely seen in newspapers:
🧩 1. Asynchronous Subagents
This is definitely agy's most revolutionary multi-agent architecture! You can directly call multiple subagents in the terminal to run complex tasks in the background:
- For example: one subagent goes online to check the latest API documentation, one runs unit tests in the background, and one performs code refactoring.
- And your main terminal will not be blocked at all! You can enter
/agentsto monitor the health and execution progress of all subagents in the background at any time.
🧠 2. Change Brains at Any Time: /model Secret
agy not only supports the Gemini series models on Vertex AI, but if you need to, you can also use the built-in /model slash command to directly switch seamlessly between Gemini, Claude, and even other open-source models with one click, helping you verify the same bug with different thinking models, which is super convenient!
🛡️ 3. Multi-Operating System Level Security Sandbox (Terminal Sandbox)
In order to prevent AI from running out of control malicious code in YOLO mode, agy silently implements operating system level sandbox protection at the bottom!
-
nsjailisolation will be automatically enabled on Linux. - macOS will automatically call the system's native
sandbox-exec. - Even if AI writes a script that pollutes the file system, it will be perfectly confined in the sandbox and unable to move!
📦 4. Upgrade Old Things: Seamless Migration Mechanism from Gemini CLI
Although Gemini CLI has gone down in history, agy has thoughtfully designed a "one-click import tool". When you start agy for the first time, it will automatically scan the old configuration path and perfectly align and migrate your original plugins, custom skills, and settings.json accumulated in Gemini CLI!
Summary and Suggestions
The upgrade from Gemini CLI to Antigravity CLI (agy) is not just a change in the command line name, but a leap forward from single-model question and answer to Multi-Agent Workflows.
By properly setting permissions in settings.json, combined with the no-confirmation function of YOLO mode, developers can allow AI to automatically and smoothly complete various medium and large tasks while ensuring the security of the host.
Quickly open your terminal and enter agy --dangerously-skip-permissions to experience this futuristic development artifact! See you next time for the actual combat!
Top comments (0)