i’ve always found writing git commit messages to be a bit of a chore, especially when i’m deep in coding and don’t want to interrupt my flow. to scratch my itch, i wrote a script called gpt_commit that automates the generation of commit messages using openai.

how it works

the script retrieves the diff of your staged changes and sends it to the openai with a prompt to generate a commit message. it then provides you with options to accept the generated message, edit it, regenerate a new one, or abort the commit.

here’s a simplified overview of the process:

1. stage your changes:

   git add .
   

2. run the script:

   gpt_commit
   

3. interact with the generated message:

   Generated commit message:
   "Fix authentication bug and improve error handling"
   
   Options:
     (y) Accept and commit
     (e) Edit message
     (r) Regenerate message
     (q) Abort
   Choose an option [y/e/r/q]:
   

features

• ai-powered commit messages: generates meaningful commit messages based on your actual code changes.
• interactive workflow: accept, edit, regenerate, or abort the commit message.
• error handling: includes checks for staged changes and api errors.

start using gpt_commit today!

like what you see? keep reading and soon you’ll be saving hours each day writing commit messages!

prerequisites

• git: ensure git is installed and initialized in your repository.
• bash: the script is written in bash.
• openai api key: obtain one from openai.
• jq: for json processing.
  • install with homebrew:

    brew install jq
    

  • or with apt:

    sudo apt-get install jq
    

  • or get it at github.com/jqlang/jq

installation

one way to install gpt_commit is as follows:

1. clone the repository:

   git clone https://github.com/strathausen/gpt_commit.git
   

2. navigate to the directory:

   cd gpt_commit
   

3. make the script executable:

   chmod +x gpt_commit
   

4. add the script to your path:

   export PATH="$PATH:$(pwd)"
   

   you might want to add this line to your ~/.bashrc or ~/.zshrc file.

or just copy the script from github and put it whereever you like.

configuration

set your openai api key as an environment variable:

export OPENAI_API_KEY="your-openai-api-key"

what’s next

options to change models or add a custom prompt would be nice. especially on retry it could ask for another prompt on how to improve the message. providing more context to the llm is another idea: what is this codebase about? which commits preceded this one?

initially i had mixed feelings about the whole idea: will this script just contribute to the amount of generated sludge coming from ai lately? then i looked at the commit messages of my team and thought, whatever generated sludge the ai will come up with, it’s almost certainly more helpful than that.

---

check out the repository on github: strathausen/gpt_commit

feel free to contribute or open issues!