Exploring the Gemini CLI

July 2, 2025 :: Wietse Venema

I have just started working with the Gemini CLI, and here are a few things I discovered that helped me be more productive. If you haven't tried out the tool, consider doing that! The free tier is very generous.

For reference, you can find Gemini CLI on GitHub. Here's how to install it: 

npm install -g @google/gemini-cli
gemini

Here are six things I discovered this week:

1. Provide context first

Always start your session by giving Gemini context about your project. If your very first prompt is something like the following, you’ll be off to a great start.

Example prompt: "Read the source code and identify all files relevant to ..." (add the task you're about to do)

2. Start with a plan

Instead of jumping straight into coding, begin by outlining your goal and collaborating on a plan. This helps you be more specific in your instructions, which leads to better results.

Example prompt: "I want to implement feature X. Let's discuss it together and create an implementation plan before you start coding."

3. Create strong feedback loops

Context is everything. The more relevant information you provide, the better it can assist you. Don't copy and paste errors, make sure Gemini sees them before you do!

You can build these loops in a few ways. First, insist on including unit tests to verify the code's functionality, then run them within the CLI or direct Gemini to execute them after making changes. Second, request thorough logging to trace execution flows and pinpoint issues during runs. Finally, execute scripts directly in the CLI using the ! prefix (for example, !python script.py), allowing you to feed any runtime errors straight back into your next prompt with a quick instruction to fix them.

4. Use the escape key

Don't be afraid to reject a suggestion if it doesn't seem right. You are in control of the process.

If Gemini proposes an action (like writing a file or generating code) that is heading down the wrong path, you can interrupt it immediately by hitting the ESC key. Once canceled, just continue the conversation to guide it back on track. For instance, you could say, “That doesn’t seem right, can you do a web search first?” to direct its next steps.

5. Your on-demand scripting expert

Use Gemini to generate shell commands you can't quite remember. This saves you from having to search for the right syntax and flags. If it doesn't seem to know the command, just tell it to figure it out through using the --help flag.

Example Prompt: "Write a bash script to convert all .HEIC files in the current directory to .jpg and resize them to a maximum width of 1200 pixels."

6. Finish the task with Git: commit and push

When you've finished implementing a feature, you can hand off the entire Git workflow. Gemini will:

  1. Summarize the work completed.
  2. Check the git status.
  3. Stage all the new and modified files.
  4. Write a detailed, conventional commit message based on the context of the tasks it just performed.
  5. Confirm with you before committing and pushing.

Example Prompt: "Okay, it's done. Please commit and push the changes."

Signature