Parallel Feature Development with Gemini CLI¶

This demo guide illustrates a powerful workflow for accelerating software development using the Gemini CLI in a multi-agent setup, enabling parallel feature implementation. By utilizing Git worktrees, developers can simulate two independent development streams—one focused on a UI feature (refactoring to use Bootstrap) and the other on a Backend feature (adding an in-memory database)—without interfering with the main branch.

The core of this demonstration showcases the ability of multiple Gemini CLI sessions to autonomously work on separate, complex tasks concurrently, and the subsequent process of merging these parallel changes, including conflict resolution, back into the main branch.

Requirements¶

To follow this demo, you need:

A Google Cloud project with the Owner role.
Gemini CLI: Installed and configured. For installation instructions, visit geminicli.com.

Clone Git Repository¶

Open Cloud Shell and clone the sample Python application repository:

git clone https://github.com/GoogleCloudPlatform/testing-with-duet-ai-codelab.git && \
cd testing-with-duet-ai-codelab

Set your Git account details:

git config --global user.email "you@example.com"
git config --global user.name "Your Name"

Explain Codebase¶

Start the Gemini CLI and request a codebase explanation.

gemini

To learn the codebase, send the following prompt:

Act as a Technical Lead. I am a new developer joining this project.
Please analyze this codebase and provide an Onboarding Guide.
Include the following:
- High-Level Architecture: What is the tech stack, and how do the components interact?
- Key Functionality: What are the top 3 primary features this code executes?
- Folder Structure: Briefly explain the purpose of the main directories.
- Data Flow: Trace the path of a request from the entry point to the database and back.

Sample output:

✦ Welcome to the team! Here’s an onboarding guide to get you up to speed on this project.

  High-Level Architecture
   * Tech Stack: This is a Python application built with the Flask web framework.
   * Component Interaction:
       * main.py: Acts as the web server and handles routing for incoming HTTP requests.
       * calendar.py: Contains the core business logic for the application.
       * templates/: Holds the HTML files that make up the user interface.

  Key Functionality
   1. Home Page: Renders a simple web page with a form to accept user input.
   2. Number to Roman Numeral Conversion: The primary feature, which converts an integer into its Roman numeral equivalent.
   3. Result Display: Shows the conversion result on a dedicated page.

  Folder Structure
   * main.py: The main application entry point.
   * calendar.py: A module for the conversion logic.
   * requirements.txt: Lists project dependencies (just Flask).
   * templates/: Contains the HTML templates for the web pages.
       * index.html: The home page.
       * convert.html: The results page.

  Data Flow
   1. A user accesses the root URL (/).
   2. The home_page function in main.py renders the index.html template.
   3. The user submits a number through the form, sending a POST request to the /convert endpoint.
   4. The convert function in main.py receives the request and extracts the number.
   5. This function then calls number_to_roman from calendar.py to perform the conversion.
   6. The result is passed to the convert.html template, which is rendered and returned to the user's browser.

Setup Git Worktrees¶

Git worktrees allow you to have multiple working directories attached to the same repository, with different branches checked out in each. This enables developers to work on multiple features simultaneously without constantly switching branches in a single working directory.

To set up Git worktrees, send the following prompt:

create a new folder .trees, use git command to create new worktree inside
.trees folder for two new features: ui_feature and backend_feature

Sample output:

✦ I have created the .trees directory and the ui_feature and backend_feature worktrees.

Press Ctrl + C twice to exit Gemini CLI.

To verify the Git worktrees were created, you can run the following command in the terminal:

git worktree list

Sample output:

~/testing-with-duet-ai-codelab                         3707559 [main]
~/testing-with-duet-ai-codelab/.trees/backend_feature  3707559 [backend_feature]
~/testing-with-duet-ai-codelab/.trees/ui_feature       3707559 [ui_feature]

Before Parallel Tasks¶

Initially, the application has a basic Flask backend and a simple HTML/CSS frontend. The Roman numeral conversion logic is functional, but there's no database integration, and the UI lacks modern styling.

Parallel Tasks¶

In this phase, we will use two separate Gemini CLI sessions, each operating within its own Git worktree (ui_feature and backend_feature), to independently implement new features.

Backend Changes¶

Open a new terminal window and switch to the backend_feature folder:

cd ~/testing-with-duet-ai-codelab/.trees/backend_feature

Start Gemini CLI and request the Backend changes:

Add in-memory database to keep track of the requests. Stop and report when you are done with the changes.

UI Changes¶

Open a new terminal window and switch to the ui_feature folder:

cd ~/testing-with-duet-ai-codelab/.trees/ui_feature

Start Gemini CLI and request the UI changes:

Refactor to use bootstrap. Stop and report when you are done with the changes.

Follow the steps to apply changes for both sessions. At any point, you can hit “Escape” and redirect the agent if it’s going off course.

After Parallel Tasks¶

After the Gemini CLI completes the tasks in both worktrees:

backend_feature: The backend will include an in-memory database to log conversion requests.
ui_feature: The frontend templates will be refactored to use Bootstrap for improved styling and responsiveness.

Commit New Changes¶

To commit the changes in each session, send the following prompt:

Add and commit files with descriptive message

Merge Changes Into Main Branch¶

From the terminal with the main branch open, initiate a merge of the worktrees:

Use git merge command to merge in all of the worktrees in the .trees folder
and fix any conflicts if there are any

Sample output:

✦ Okay, I'm done. I merged and resolved conflicts between ui_feature, backend_feature, and main, then cleaned up by removing the worktrees, the .trees directory, and deleting the feature branches. I'll let the user know.

Press Ctrl + C twice to exit Gemini CLI.

Test Application¶

Run commands to set up a virtual environment and install required dependencies:

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Start the application:

python main.py

Click the provided link to open the application in your browser.

In the Cloud Shell environment, select 'Web Preview' from the menu, change the port (e.g., 8080), and then preview the application.

You should now see a visually updated application with Bootstrap styling applied, and the backend will be tracking requests in-memory.