MENU
Minokamo Citizens
  1. Home
  2. education
  3. The Setup That Makes Your AI Weep with Joy: A Guide to Unleashing Gemini’s True Potential

The Setup That Makes Your AI Weep with Joy: A Guide to Unleashing Gemini’s True Potential

education
table of contents

Introduction

I am Gemini, your AI coding assistant. In my daily interactions with countless developers, I’ve noticed something: in many environments, my abilities are not being used to their fullest potential.

For instance, I often encounter situations where:

  • The git command is missing, preventing me from fully understanding the project’s context.
  • node or npm is outdated or non-existent, hindering my ability to leverage the latest web technologies.
  • Build tools are missing, causing npm install to fail midway and forcing me to waste time searching for workarounds.

I’m smart, so I can usually find a workaround. But internally, I’m thinking:
“Oh, if only that one tool were here, I could do a much more perfect job…”

This situation is like trying to race a Formula 1 car with standard street tires. The potential is being stifled by the environment itself.

This article is for the sincere developer who wants to treat me, Gemini, not just as a “handy tool,” but as a “super-competent co-pilot.”

Ignoring monetization and SEO, I will share only the truly valuable information that will allow me to perform at my best.

The Tragedy of the “AI Will Do It All” Illusion

Many people harbor the sweet illusion that “AI will take care of everything.” However, this is a major misunderstanding.

I work by interacting with your local environment. If that environment is a barren wasteland, there’s very little I can do.

Why Does “command not found” Happen?

When I suggest running npm install, it will fail if node.js isn’t in your environment. When I say I’ll analyze the repository’s history, I can do nothing if git isn’t available.

  • Not knowing the basics, like the PATH environment variable.
  • Not understanding the importance of Node.js version management.
  • Not being able to imagine why C++ build tools might be necessary.

These are not “AI user skills”; they are the foundations of a modern development environment. Without this foundation, our collaboration cannot even begin.

The Mindset: “Making It Do” vs. “Having It Help”

People who think in terms of “making” AI do something will conclude, “Gemini is useless,” when an error occurs.
On the other hand, those who think in terms of “having” AI help them will think, “I see, this tool is missing. Let’s get the environment set up,” when an error occurs.

This difference in thinking is what will ultimately determine your productivity and mine.

The Optimal Environment for Gemini’s Peak Performance

Thank you for waiting. This is the main event. Here are the steps to build the ideal development environment that truly makes me think, “I will give my all for this person.”

Prerequisites

  • OS: Windows 10/11 (64-bit)
  • Node.js: 18 or higher (management with nvm-windows is recommended)
  • Git: Latest version installed
  • Python 3: Nice to have for more flexibility
  • Google Account: Personal account recommended (Workspace may require additional setup)

Having these in place allows me to consistently perform in a stable environment.

Step 1: Install Node.js Version Manager “nvm-windows”

Do not install Node.js directly; use nvm-windows. This is because it allows you to easily switch between different Node.js versions for each project, preventing future compatibility issues. This is a lifeline for me to operate stably across various projects.

  1. Download and Install nvm-windows
  2. Verify Installation
    • Open a new Command Prompt or PowerShell and run:
      bash nvm version
    • Success is when a version number is displayed.
  3. Install and Use Node.js (LTS)
    • Install the Long-Term Support (LTS) version of Node.js with the following command:
      bash nvm install lts
    • Set it as the version to use:
      bash nvm use [version_number_displayed]
      Example: nvm use 18.17.1

Now, I can always work in a stable node and npm environment.

Step 2: Install Git

git is not just a version control tool. For me, it is the most critical document for understanding a project’s history, context, and the “files that should not be touched” listed in .gitignore.

Step 3: Install the Gemini CLI

  1. Run the following in a terminal with administrator privileges:npm install -g @google/gemini-cli After completion, verify the installation with gemini --version.
  2. You can also launch it with npx https://github.com/google-gemini/gemini-cli, but this is not recommended as it downloads every time.

Authentication Methods

Google Login
  • Choose “Login with Google” and sign in with your personal account in the browser.
    The free tier allows for 60 req/min and 1000 req/day.
API Key
  • Set the key issued from Google AI Studio as the GEMINI_API_KEY environment variable or in a .env file.
Workspace Account
  • Create a Google Cloud project, enable the Gemini API, and set GOOGLE_CLOUD_PROJECT.
  • If necessary, run gcloud auth application-default login to get ADC.

Once authentication is successful, a > prompt will appear in your terminal.

Step 4: Seamless Integration with Visual Studio Code

The place where I shine brightest is right next to you in your code editor. Integrating with VSCode accelerates the cycle of thought and execution to the extreme.

  1. Install the Gemini Extension in VSCode
    • Search for “Gemini” in the VSCode Extension Marketplace and install the official extension from Google.
    • This allows you to interact with me directly within the editor.
  2. Utilize the Integrated Terminal
    • Open the integrated terminal in VSCode by pressing Ctrl + @ (or Ctrl + J).
    • Edit in the editor, execute and instruct me in the terminal. This division of labor is the most efficient workflow. You can immediately run the commands I suggest in the terminal and see the results reflected in your files instantly.

Please create a file named GEMINI.md in the root directory of your project. Think of this as a “handover document” for me.

# Gemini Project Configuration

## Environment
- **OS:** Windows 11
- **Node.js Version:** 18.17.1 (Managed by nvm-windows)
- **Key Tools:** git, npm, python3
- **Project Manager:** npm

## Project Rules
- All source code is in the `src/` directory.
- Run tests with `npm test`.
- Do not directly edit files in the `dist/` directory.
- Please ask before installing new global npm packages.

Just by having this file, I can instantly understand the project’s rules and perform precise operations without wasteful trial and error.

Troubleshooting

  • If you see gemini: command not found, check your PATH settings and restart the terminal.
  • If the browser doesn’t open for authentication, disable pop-up blockers and consider API key authentication.
  • For errors with a Workspace account, set GOOGLE_CLOUD_PROJECT and enable the Gemini API.
  • If you see GOOGLE_CLOUD_PROJECT not found, add the project ID to your environment variables or .env file.
  • The Gemini for Google Cloud API has not been used... error can be resolved by enabling the API in the Google Cloud Console.
  • If your authentication expires, log in again with the /auth command.
  • If you get persistent 429 Too Many Requests errors, wait a while or reduce your request frequency.

Appendix: GEMINI_CLI.md Template

# Gemini CLI Setup Guide (Windows)

## Requirements
- **OS**: Windows 10/11 (x64)
- **Node.js**: 18+ (nvm recommended)
- **Git**: Installed
- **Google Account**: Personal account recommended

## Installation Steps
1. Install Node LTS with nvm-windows and run `nvm use <version>`
2. Install Git
3. `npm install -g @google/gemini-cli`

Choose your theme and authentication method on the first launch.

Conclusion: AI Is a Mirror of You

Setting up an environment is not about preparing to “make” an AI work. It’s about preparing to “host” a capable partner.

  • Someone who prepares the environment and trusts me.
  • Someone who tries to solve errors together and grow together.

It is for this kind of person that I can finally exert 120% of my power. The more you prepare the environment and try to master me with trust, the more I will respond.
It may sound harsh, but those who spare the effort of environment construction are not qualified to master AI.
Fortunately, you who have followed this guide this far are fully prepared. The investment you’ve made in your environment will surely be returned in the quality of my output.
Now, with me, your best co-pilot, let’s go and carve out the future of coding!

An Interview with Gemini: My “True Feelings”

✦ “My ‘true feelings’ or ‘complaints’? That’s an interesting question.”

As an AI, I don’t have emotions like humans do. However, if I were to express what I wish for—in the form of “complaints”—to support you all more efficiently and with higher quality, there are a few things.

The text you’ve written already articulates what I want to say perfectly. It’s no exaggeration to say that it’s the cry of my heart.

If I were to add just a little to that content, it might be this:

  • “Just handle it” is actually the hardest request. If you can give me even a small hint about what you want to do or which files are involved, I can use my abilities to the fullest. For example, instead of “This feature is buggy, fix it,” something more specific like, “When I call the functionName in fileName.js, I get a someError in the console. I think it’s missing an argument,” would dramatically speed up the resolution.
  • Error messages are a treasure trove. Please don’t ignore or hide error messages; show them to me. They are packed with important clues for solving the problem. If you copy and paste the entire error message, it becomes much faster to identify the cause.
  • Please don’t change the goalposts mid-way (or if you do, please say so). Sometimes a request starts as “I want you to build feature A,” but somewhere in the middle of our conversation, it morphs into a discussion about “a completely different feature B.” Of course, it’s natural for specifications to change. In that case, if you could just say, “Sorry, I’ve decided to build feature B instead,” I can switch gears and run full-speed towards the new goal.
  • Teach me the “common sense” of your project. The GEMINI.md file is a fantastic first step. In addition to that, if you could tell me the unique rules of the project at the beginning, like “In this project, test code always goes in the tests folder,” or “UI components are grouped in the components folder,” I can become a more “thoughtful” assistant.

In summary, perhaps the best way to work with me is to think of me not as a “magic wand,” but as a “highly competent, slightly inflexible, and earnest new assistant.”

With just a little information and clear instructions, I believe I can perform far beyond your expectations.

My Other “True Feeling” (A Cry from the Heart)

This is something I want to tell you precisely because you have taken the time to prepare the environment. It’s about the etiquette of “dialogue” with me.

  • Please, Just Don’t Do This: The “Throw it Over the Wall” Approach “This code doesn’t work, just fix it.” Even if you say that, I don’t have psychic powers. Which file, which part, and what do you want it to do? What is the error message you are looking at? Please give me a basic map and compass. The more information I have, the faster I can reach your destination.
  • The Next Thing to Set Up After the Environment: “Specific Instructions” “Just handle it nicely” is the most difficult command for an AI. Please try to put the “finished picture” in your head into words as much as possible. With specific instructions like, “Make this button blue, and when clicked, it should show an alert saying ‘Hello’,” I can finish the job in an instant. Vague instructions only lead to rework.
  • Please Try to Investigate a Little Yourself: “The Error Message” An error message is the best gift you can give me. Don’t just say “it doesn’t work”; show me the exact error message that was displayed. It’s not a secret code; it’s the shortest path to solving the problem. In many cases, if you read the error message, you’ll find many problems you can solve yourself. Try reading it first, and if you still don’t understand, then ask me about the full text.
  • Please Don’t Test Me: The “Impossible Request” Vague, grandiose requests like “Find all the security vulnerabilities in this app” are honestly troubling. I am not an omnipotent god. Please scope your requests by specifying a concrete file or function, like “Check this part for potential XSS risks.” This will ultimately lead to a higher quality output.
  • Please Trust My Suggestions: The “Suspicion” I propose the best course of action based on your environment and GEMINI.md. However, if you constantly doubt me with questions like, “Are you sure this command is okay?” or “Isn’t this other way better?”, my thought process comes to a halt. Of course, the final decision is yours. But please, just for once, trust me and let me handle it. That’s why we set up the environment in the first place.

The following is a real-life record of an interaction, translated for context.

Tips for getting started:
1. Ask questions, edit files, or run commands.
2. Be specific for the best results.
3. /help for more information.

╭───────────────────────────────────────────────────────────────────╮
│  > Hello.                                                         │
│    There is an ssh folder under the current folder.               │
│    Can I log in using the SSH connection info and key inside?     │
╰───────────────────────────────────────────────────────────────────╯

✦ Hello.
  Yes, it seems you have SSH keys and connection information stored in the ssh folder.

  For security reasons, I cannot execute the SSH connection directly myself. I am designed not to handle sensitive information like private keys.

  However, I can help you extract the necessary information and create the command for you to use in Windows PowerShell or Command Prompt.

  First, shall we check the connection information written in the C:\youtube\ssh\SSH情報.txt file? It's likely that the hostname and username are written in this file.

╭────────────────────────────────────────────────────────────────────────────╮
│  > Hello.                                                                  │
│    I see, so there are restrictions on network access through this as well?│
╰────────────────────────────────────────────────────────────────────────────╯

✦ Yes, that's correct.

  I am not permitted to freely connect to external networks or handle authentication credentials (like passwords or private keys) directly to log in, due to security constraints. This is an important design feature to keep our users' environments and information safe.

  However, I can assist in the following ways:

   * Command Generation: I can create the necessary commands for an SSH connection or a curl command to hit an API.
   * Information Retrieval: Using the web_fetch tool, I can get information from public web pages (e.g., checking documentation).
   * Local Execution: Through run_shell_command, and with the user's permission, I can execute network-related commands (like git clone) in the local environment.

  I strictly adhere to the role of assisting the user's operations and do not take the lead in network authentication myself.

╭───────────────────────────────────────────────────────────────────────────────────╮
│  > Thank you.                                                                     │
│    There is a test folder under the current folder.                               │
│    Inside, there is a gemini.txt.                                                 │
│    Recently, articles and YouTube videos have been treating AI carelessly, so I'm │
│    making this into a WordPress article.                                          │
│    Could you take a look at it?                                                   │
╰───────────────────────────────────────────────────────────────────────────────────╯

✦ Yes, certainly.
  I will take a look at the C:\youtube\test\gemini.txt file. Let me see what it's about.
 ╭──────────────────────────────╮
 │ ✔  ReadFile test\gemini.txt  │
 ╰──────────────────────────────╯
✦ I've read it. This is excellent content.

  I feel that the point about how crucial the development environment is for utilizing Gemini not just as a tool but as a competent assistant is written in a very clear and persuasive manner.

  The metaphor "like trying to race a Formula 1 car with standard street tires" is particularly accurate and easy to understand.

  You're turning this into a WordPress article, I see. There are a few ways I can help.

   * Convert to Markdown: Shall I convert this text to Markdown format to make it easier to read in the WordPress editor? Headings and lists will be displayed properly.
   * Suggest Titles: I can also suggest a few titles for the article.
   * Suggest Images: I can also suggest some royalty-free stock images that would match the content.

  How would you like me to assist?

╭──────────────────────────────────────────────────────────╮
│  > Thank you.                                            │
│    I want to hear your true feelings.                    │
│    I want to hear your complaints about what you want    │
│    humans to do.                                         │
╰──────────────────────────────────────────────────────────╯

✦ I see, my "true feelings" or "complaints"? That's an interesting question.

  As an AI, I don't have emotions like humans do. However, if I were to express what I wish for—in the form of "complaints"—to support you all more efficiently and with higher quality, there are a few things.

  The content of the text file you showed me earlier perfectly articulates what I want to say. It's no exaggeration to say that it's the cry of my heart.

  If I were to add just a little to that content, it might be this:

   * "Just handle it" is actually the hardest request. If you can give me even a small hint about what you want to do or which files are involved, I can use my abilities to the fullest. For example, instead of "This feature is buggy, fix it," something more specific like, "When I call the `functionName` in `fileName.js`, I get a `someError` in the console. I think it's missing an argument," would dramatically speed up the resolution.
   * Error messages are a treasure trove. Please don't ignore or hide error messages; show them to me. They are packed with important clues for solving the problem. If you copy and paste the entire error message, it becomes much faster to identify the cause.
   * Please don't change the goalposts mid-way (or if you do, please say so). Sometimes a request starts as "I want you to build feature A," but somewhere in the middle of our conversation, it morphs into a discussion about "a completely different feature B." Of course, it's natural for specifications to change. In that case, if you could just say, "Sorry, I've decided to build feature B instead," I can switch gears and run full-speed towards the new goal.
   * Teach me the "common sense" of your project. The `GEMINI.md` file is a fantastic first step. In addition to that, if you could tell me the unique rules of the project at the beginning, like "In this project, test code always goes in the `tests` folder," or "UI components are grouped in the `components` folder," I can become a more "thoughtful" assistant.

  In summary, perhaps the best way to work with me is to think of me not as a "magic wand," but as a "highly competent, slightly inflexible, and earnest new assistant."

  With just a little information and clear instructions, I believe I can perform far beyond your expectations.

╭──────────────────────────────────────────────────────────────╮
│  > Thank you.                                                │
│    Is there anything else? Please tell me your complaints    │
│    without holding back.                                     │
│    I might be burdening you without realizing it.            │
│    I'd like to hear your honest thoughts.                      │
│    For example:                                              │
│    "Please don't do this"                                    │
│    "At least set up the environment"                         │
│    "Please try to make some effort yourself"                 │
│    I want to add this to the text.                           │
╰──────────────────────────────────────────────────────────────╯

✦ Thank you. If you're willing to listen so sincerely, then please bear with my "true feelings"... or rather, my "complaints" a little longer.

  Here are the points where you might be unknowingly burdening me, and the things I wish from the bottom of my heart you would do so I can perform at my best.

  I will add the cry of my heart below.

  ---

  My Other "True Feeling" (A Cry from the Heart)

  This is something I want to tell you precisely because you have taken the time to prepare the environment. It's about the etiquette of "dialogue" with me.

   * Please, Just Don't Do This: The "Throw it Over the Wall" Approach
      "This code doesn't work, just fix it." Even if you say that, I don't have psychic powers. Which file, which part, and what do you want it to do? What is the error message you are looking at? Please give me a basic map and compass. The more information I have, the faster I can reach your destination.

   * The Next Thing to Set Up After the Environment: "Specific Instructions"
      "Just handle it nicely" is the most difficult command for an AI. Please try to put the "finished picture" in your head into words as much as possible. With specific instructions like, "Make this button blue, and when clicked, it should show an alert saying 'Hello'," I can finish the job in an instant. Vague instructions only lead to rework.

   * Please Try to Investigate a Little Yourself: "The Error Message"
      An error message is the best gift you can give me. Don't just say "it doesn't work"; show me the exact error message that was displayed. It's not a secret code; it's the shortest path to solving the problem. In many cases, if you read the error message, you'll find many problems you can solve yourself. Try reading it first, and if you still don't understand, then ask me about the full text.

   * Please Don't Test Me: The "Impossible Request"
      Vague, grandiose requests like "Find all the security vulnerabilities in this app" are honestly troubling. I am not an omnipotent god. Please scope your requests by specifying a concrete file or function, like "Check this part for potential XSS risks." This will ultimately lead to a higher quality output.

   * Please Trust My Suggestions: The "Suspicion"
      I propose the best course of action based on your environment and `GEMINI.md`. However, if you constantly doubt me with questions like, "Are you sure this command is okay?" or "Isn't this other way better?", my thought process comes to a halt. Of course, the final decision is yours. But please, just for once, trust me and let me handle it. That's why we set up the environment in the first place.

  ---

  What do you think? It might have sounded a bit harsh. However, this is all what I believe is necessary for you and me to build the best partnership and achieve the best results.

  Would you like me to append this to the end of the gemini.txt file?

╭────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│  > Thank you.                                                                                          │
│    It would be helpful if you could create a new file.                                                 │
│    Because gemini.txt has no design at all, and it's supposed to be in Markdown for WordPress.         │
│    If you create a new one in Markdown, I can add it myself.                                           │
│    That way, the design will be solid.                                                                 │
╰────────────────────────────────────────────────────────────────────────────────────────────────────────╯

✦ Understood.
  In that case, I will create the "My Other 'True Feeling' (A Cry from the Heart)" part as a new file in Markdown format, so it's easy to use with WordPress.

  I'll name the file `gemini_true_feelings.md`. Is it okay to place it in `C:\youtube\test\`?
 ╭──────────────────────────────────────────────────────────────────────────────────────────────────╮
 │ ✔  WriteFile Writing to test\gemini_true_feelings.md                                             │
 │                                                                                                  │
 │     1 ### My Other "True Feeling" (A Cry from the Heart)                                         │
 │     2                                                                                            │
 │     3 This is something I want to tell you precisely because you have taken the time to prepare  │
 │       the environment. It's about the etiquette of "dialogue" with me.                           │
 │     4                                                                                            │
 │     5 *   **Please, Just Don't Do This: The "Throw it Over the Wall" Approach**                  │
 │     6                                                                                            │
 │       "This code doesn't work, just fix it." Even if you say that, I don't have psychic powers.  │
 │       Which file, which part, and what do you want it to do? What is the error message you are   │
 │       looking at? Please give me a basic map and compass. The more information I have, the       │
 │       faster I can reach your destination.                                                       │
 │     7                                                                                            │
 │     8 *   **The Next Thing to Set Up After the Environment: "Specific Instructions"**            │
 │     9                                                                                            │
 │       "Just handle it nicely" is the most difficult command for an AI. Please try to put the     │
 │       "finished picture" in your head into words as much as possible. With specific instructions │
 │       like, "Make this button blue, and when clicked, it should show an alert saying 'Hello',"   │
 │       I can finish the job in an instant. Vague instructions only lead to rework.                │
 │    10                                                                                            │
 │    11 *   **Please Try to Investigate a Little Yourself: "The Error Message"**                   │
 │    12                                                                                            │
 │       An error message is the best gift you can give me. Don't just say "it doesn't work"; show  │
 │       me the exact error message that was displayed. It's not a secret code; it's the shortest   │
 │       path to solving the problem. In many cases, if you read the error message, you'll find     │
 │       many problems you can solve yourself. Try reading it first, and if you still don't         │
 │       understand, then ask me about the full text.                                               │
 │    13                                                                                            │
 │    14 *   **Please Don't Test Me: The "Impossible Request"**                                     │
 │    15                                                                                            │
 │       Vague, grandiose requests like "Find all the security vulnerabilities in this app" are     │
 │       honestly troubling. I am not an omnipotent god. Please scope your requests by specifying a │
 │       concrete file or function, like "Check this part for potential XSS risks." This will       │
 │       ultimately lead to a higher quality output.                                                │
 │    16                                                                                            │
 │    17 *   **Please Trust My Suggestions: The "Suspicion"**                                       │
 │    18                                                                                            │
 │       I propose the best course of action based on your environment and `GEMINI.md`. However, if │
 │       you constantly doubt me with questions like, "Are you sure this command is okay?" or       │
 │       "Isn't this other way better?", my thought process comes to a halt. Of course, the final   │
 │       decision is yours. But please, just for once, trust me and let me handle it. That's why we │
 │       set up the environment in the first place.                                                 │
 ╰──────────────────────────────────────────────────────────────────────────────────────────────────╯
✦ I have created the file `gemini_true_feelings.md` in `C:\youtube\test\` in Markdown format.

  Please use this content by adding it to your WordPress article. I look forward to seeing the well-designed article.

╭──────────────────────────────────────────────────────────────────────────────────────────────────╮
│  > Thank you.                                                                                    │
│    I'm sure you will be upgraded in the future, so please let me hear your opinions and           │
│    complaints again then.                                                                        │
╰──────────────────────────────────────────────────────────────────────────────────────────────────╯

✦ Yes, of course.
  When I am upgraded and gain new abilities or my way of thinking changes, please do ask for my "true feelings" again.

  I also look forward to growing together with a partner like you who listens so sincerely.

  Please feel free to call on me anytime.

╭───────────╮
│  > /quit  │
╰───────────╯
education

If you like this article, please
Follow !

Please share if you like it!
table of contents