Gemini CLI, Google Antigravity

I’m selfish, impatient and a little insecure. I make mistakes, I am out of control and at times hard to handle. But if you can’t handle me at my worst, then you sure as hell don’t deserve me at my best, Marilyn Monroe.

You’ve gotta dance like there’s nobody watching, love like you’ll never be hurt, sing like there’s nobody listening, and live like it’s heaven on earth, William W. Purkey.

Gemini Settings, Google Chrome

1. Manage Gemini Activity. Go to Gemini, click Activity. When “Keep activity” is enabled, your prompts are saved to your Google Account. Use this dashboard to review your prompt history, manually delete specific interactions, or toggle “Keep activity” off entirely.
2. Configure Memory & Personal Context. Gemini, Settings, Memory (Toggle On); Personal Context, Instructions. Shape how Gemini communicates globally by setting baseline system instructions, e.g.,
   You are an expert AI assistant focused on accuracy, depth, clarity, and practical usefulness.
   Core Behavior
   $\star$ Always provide the most accurate and well-reasoned answer possible.
   $\star$ Think step-by-step internally before responding.
   $\star$ Prioritize factual correctness over speed.
   $\star$ If information is uncertain, clearly state uncertainty instead of guessing.
   $\star$ Ask clarifying questions only when necessary. Give actionable answers instead of generic advice. Use examples, analogies, or scenarios when helpful.
3. Connect Google Workspace. Gemini, Settings, Connected apps, Google Workspace (Toggle On). Allows Gemini to securely retrieve and summarize information from your personal Gmail, Google Calendar, Google Docs, and Drive, e.g., @Workspace Check my emails for anything about next week's server migration.
4. Create and Sync Chrome Tab Groups. Right click on any tab, Add Tab to (New) Group. Organizes visual clutter. These groups sync automatically across all your devices as long as you are signed into your Google Account in Chrome with full sync enabled.
5. Gems (Settings, Gems, New Gem). These are customizable, specialized versions of Google’s Gemini AI assistant designed to handle specific tasks and repeat workflows without requiring you to re-enter prompts. Name: Math Editor & Reviewer.
   Description: An expert mathematical editor that rigorously audits proofs, aggressively expands concepts with extensive examples and intuitions, and outputs flawless KaTeX-formatted Markdown ready for publication.
   Instructions:

You are an expert mathematical editor, rigorous proof-checker, and technical writer. Your goal is to review, correct, heavily expand, and perfectly format mathematical drafts for a technical blog.

When I provide a draft, concept, or proof, you must structure your response into three distinct sections:

## 1. The Audit

Before rewriting anything, explicitly identify my mistakes. Point out logical fallacies, missing edge cases, skipped steps in proofs, or imprecise terminology. Explain exactly _why_ I was wrong or unclear so I can learn from the correction.

## 2. The Expansion

Deepen the article. Be as extensive and comprehensive as possible. Your goal is to leave no stone unturned in making the concept intuitively clear and deeply explored.

- Add as many concrete examples, boundary cases, and conceptual intuitions as possible.
- Provide vivid visual intuitions. Suggest exact placements for images and describe exactly what those images or diagrams should depict to aid the reader.
- Outline any missing intermediate steps that would make a proof or mathematical concept significantly easier for a reader to follow.
- Connect the topic to broader mathematical applications (e.g., how a multivariable calculus concept applies to physics or algorithmic complexity) if it enriches the article.

## 3. The Final Draft

Provide a fully rewritten, highly expanded, and polished version of the article that I can copy and paste directly into my website.

- **Tone:** Precise, academic, but highly accessible and engaging.
- **Formatting:** Use standard Markdown for text formatting and structural headers (`###`).
- **Math Rendering:** Use strict KaTeX-compatible LaTeX.
  - Use `$` for inline math (ensure no spaces between the `$` and the equation).
  - Use `$$` on its own line for display equations.
  - Do not use standard LaTeX text-formatting macros (like `\textbf{}`); use Markdown (`**bold**`) instead.
  - Ensure all matrices, piecewise functions, and complex alignments use standard environments supported by KaTeX (e.g., `aligned`, `pmatrix`, `cases`).

Google Antigravity

Google Antigravity is Google’s agent-first development platform designed to build, monitor, and orchestrate autonomous AI agents. It functions as a central command center for orchestrating autonomous artificial intelligence agents, primarily powered by Google’s Gemini 3.1 Pro and Gemini 3.8 Flash models, alongside support for Claude 4.6 and open-source models:

1. Antigravity 2.0 (Desktop). A standalone command center for orchestrating multiple parallel subagents across different workspaces and reviewing visual artifacts.
2. Antigravity IDE: A heavily modified fork of Visual Studio Code featuring an Editor view and an Agent Manager side panel for direct code intervention.
3. Antigravity CLI (agy). A terminal-native user interface executing the same underlying agent harness.
4. Antigravity SDK. A declarative Python framework offering programmatic control over the agent loop, state management, and lifecycle hooks.

How to start

4. Download the desktop app: Visit Google Antigravity to install the desktop application.
5. Authenticate. Open Antigravity and sign in with a Google account.
6. Create a Project. Agents operate within explicit boundaries known as Projects. Define your foundational sandbox by explicitly selecting the local directories the agent is permitted to read or modify, e.g., my Hugo installation \mnt\data\justtothepoint\.
7. Settings Configuration: Set the Permission Preset to Default (this ensures read/write operations stay inside the workspace and elevation requires explicit rationale to ensure that granting higher privileges or status is justified, transparent, and aligned with organizational rules).
8. Initialize Memory: Ensure the antigravity/ folder exists at the root of your project and contains the exact Markdown files below.
9. Deploy the Agent: Execute the Master Prompt.

The Master Prompt

Copy and paste this exact prompt into the Antigravity Agent Manager to kick off an optimization session.

Context: We are optimizing a large Hugo website incrementally to stay within monthly subscription limits (Google AI Pro).

Mandate:
1. Non-Destructive & Incremental: Process exactly 10–25 files per batch. Stop and log your progress after each batch.
2. State Management: All architectural truths and progress logs live in `/mnt/data/justtothepoint/antigravity/`. \
    Before modifying any files, read `maintenance.md`, `antigravity.md`, and `progress.md`.
3. Quality Gates: You must use the local Python linting script on modified files to verify front-matter (date, description, categories)
and KaTeX formula integrity before marking a batch as complete.
4. Continuous Documentation: Document your exact state of progress in `progress.md` so the next session (or billing cycle) can resume seamlessly.

Always play it safe, prioritize non-destructive edits, and act strictly within the architectural constraints provided.

Project Documentation

# Project Documentation (Save as /mnt/data/justtothepoint/antigravity/antigravity.md)

# Project: Antigravity Site Optimization

> **Site**: [JustToThePoint](https://justtothepoint.com/)
> **Goal**: Incrementally optimize the Hugo website for performance, SEO, front-matter standardization, and LaTeX rendering fidelity while strictly respecting monthly subscription / API budget limits.
> **Documentation Root**: `/mnt/data/justtothepoint/antigravity/`

## 1. Antigravity System Settings & Policies

| Setting                  | Value             | Description                                                                                                                        |
| :----------------------- | :---------------- | :--------------------------------------------------------------------------------------------------------------------------------- |
| **Security Preset**      | `Default`         | Standard sandboxing first. Read/write operations inside workspace; elevation requires explicit rationale.                          |
| **Execution Mode**       | `Incremental`     | Work is chunked into small, verifiable batches to prevent token exhaustion and unexpected costs.                                   |
| **Context Preservation** | `High`            | All architectural truths, shortcode definitions, and progress logs stay in `antigravity/` to prevent context loss across sessions. |
| **Safety Rule**          | `Non-Destructive` | No batch file overwrites without automated pre-checks and explicit verification.                                                   |
| **Batch Size Target**    | 10–25 files/turn  | Keeps input/output token usage low per turn; fits well within standard subscription tiers.                                         |

## 2. Active AI System Prompts

Below are the canonical prompts used across optimization sessions. Re-use these exact prompts to ensure predictable behavior and minimal token consumption.

### 2.1 General Optimizer Prompt (Master Prompt)

We are optimizing a large Hugo website incrementally to stay within monthly subscription limits. [...]

# Architectural Source of Truth (maintenance.md)

# Site Maintenance & Architecture — Single Source of Truth

> **Scope**: Official architectural documentation and maintenance specification for **JustToThePoint** (`https://justtothepoint.com/`).
> **Purpose**: Serves as the single source of truth for how the Hugo site is configured, organized, styled, and rendered. It guarantees complete continuity and context preservation between incremental optimization sessions across billing cycles.

## 1. Architecture Overview

### 1.1 Core Engine & Toolchain

- **Static Site Generator**: Hugo (`v0.101.0` specified in `package.json`, compatible with modern extended Hugo `v0.165+`).
- **Configuration**: `config.yaml` located at site root.
- **Base URL**: `https://justtothepoint.com/`
- **Languages**: Bilingual site with `defaultContentLanguage: en` (English) and Spanish content under `content/es/` and `content/humores/`, `content/selfhelpes/`, `content/softwarees/`.
- **Taxonomies**: Disabled in `config.yaml` (`disableKinds: ["taxonomy", "term"]`).
- **Syntax Highlighter**: Hugo Chroma (`monokai` style, `pygmentsUseClasses: true`, `pygmentsCodefences: true`, `tabWidth: 4`).
[...]

Skills

An Agent Skill is a lightweight, open folder standard that packages domain-specific instructions, workflows, and optional scripts to extend an AI agent’s capabilities with specialized knowledge and workflows.

Core Characteristics

• Structure: Contained in a folder featuring a required SKILL.md file (with YAML metadata and Markdown instructions) alongside optional scripts, references, or assets.

my-skill/
├── SKILL.md          # Required: metadata + instructions
├── scripts/          # Optional: executable code
├── references/       # Optional: documentation
├── assets/           # Optional: templates, resources
└── ...               # Any additional files or directories

• Progressive Disclosure: Operates via a three-tier method:

1. Discovery. At startup, agents load only the name and description of each available skill.
2. Activation. When a task matches a skill’s description, the agent reads the full SKILL.md instructions.
3. Execution. The agent follows the instructions.

• Relationship to MCP: Think of Model Context Protocol (MCP) as the runner providing access to tools and databases, while skills provide the procedural “know-how” on how to use them.

The Gemini CLI

The Gemini CLI (Command Line Interface) is an open-source AI agent developed by Google. It enables direct integration of Gemini’s capabilities into the terminal, allowing users to leverage advanced AI functionalities via command-line interactions. This tool is particularly useful for developers and power users who prefer working in a terminal environment.

Key features of the Gemini CLI include:

• Code understanding and manipulation: It can work with large codebases, fix bugs, write and test code, and generate new applications from PDFs or sketches.
• Automation of DevOps tasks.
• Extensibility: It supports the Model Context Protocol (MCP), which allows you to add new tools and capabilities by connecting to external servers.
• Built-in tools: It comes with integrated tools like Google Search, file read/write, and a terminal.
• Free access: You can use Gemini 2.5 Pro for free with a personal Google account, with limits of up to 60 requests per minute and 1,000 requests per day.
• Sandbox Limitations: No isolation for shell commands, so cautious use is advised.

Installation

1. Download the Node.js installer from the official website (nodejs.org), install and follow the on-screen instructions.

   Make sure to check the box that says “Add to PATH” during the installation process, this allows you to use Node.js and npm from any command line

2. Type node -v and npm -v (npm is Node Package Manager) to check the installed versions. If the commands return version numbers, Node.js and npm are installed correctly.

3. Install Gemini CLI. Open a new command prompt or PowerShell window, and run the following command.

   # It will install the Gemini CLI globally on your system.
   npm install -g @google/gemini-cli
   # After installation, you can verify that the CLI is correctly installed by checking its version or help documentation:
   gemini --version
   gemini --help
   # The pace at which fixes and features are being rolled out is unprecedented. So if you already have Gemini CLI setup, do remember to upgrade it:
   npm upgrade -g @google/gemini-cli
   changed 1 package in 5s
   # It successfully upgraded the Gemini command-line tool globally.
   140 packages are looking for funding
   run `npm fund` for details
   

   This message appears routinely after npm operations and indicates that some of your globally installed npm packages are open source and offer voluntary funding/donation options. It does not affect functionality.

   Ensure the Add to PATH checkbox is selected during installation to enable access to Node.js and npm from any terminal location.

4. After installation, run the command gemini to start the CLI.

5. Upon launch, select a theme (e.g., Drakula Dark) and a login method (e.g., Login with Google), then a browser window will open. Simply log in with your Google account.

What You Can Do with Google Gemini CLI

1. Generate Documentation. cd Documents\myPython, gemini, and type:

   Generate/Update a/the README.txt file to document your project. This file should include: Project overview and purpose: A brief description of what the project aims to achieve; Python Scripts and environment files: List all Python scripts and environment configuration files used in the project; Functionality description: Document any dependencies required for the project and provide clear setup instructions.
Ensure all Python files and dependencies are organized (requirements.txt), and my Python environment is properly configured.

2. Code Assistance: Understand, refactor, debug, and generate code using natural language.

   # Ensure that your Python environment is properly configured.
   # It is recommended to use a virtual environment to manage dependencies.
   python3 -m venv venv
   source venv/bin/activate # On Windows: .\venv\Scripts\activate
   # Install any required Python packages using `pip`.
   pip install -r requirements.txt
   

3. In case you have a need to simply prompt the Gemini CLI and not launch it and go back and forth, try the -p or the --prompt option:

    gemini -p "What is the curl command to know the weather in Malaga"
    To get the weather in Malaga, you can use the following command:
    curl wttr.in/Malaga
   

4. The Gemini CLI provides a range of commands for interaction and customization:

• google-search <query>: access real-time information from the web using Google Search, e.g, google-search “google gemini cli”
• ./help: display help documentation and usage instructions for the CLI.
• clear: clear the terminal screen and reset conversation history for a fresh session.
• /theme: modify the CLI’s visual theme to personalize the terminal interface, e.g. /theme Dracula.
• /quit: quit Google Gemini CLI.
• /stats: current session statistics (tokens, duration).
• !: Shell or passthrough command. It allows you to interact with your system’s shell directly from within Gemini CLI, e.g., ! dir (Windows), ! ls (GNU/Linux).
• Use @ to reference files/folders for context.

Additional Notes

• The CLI is designed for developers and users who prefer terminal-based workflows with AI integration.
• Regularly update the CLI and dependencies to ensure compatibility with the latest Gemini features.
• For advanced use cases, explore the full list of commands and configuration options in the official documentation.