Turbocharging GitHub PR Creation with Python, OpenAI, and GitHub CLI

Developers who push numerous PRs daily can understand the taxing effort it takes to consistently pull-off comprehensive titles and descriptions. After writing great commit messages there should be no need to again write the PR description - it’s essentially the same thing in a different format.

I decided to automate the process of creating PRs. Not only pushing code with gh pr create, but also generating values for title and description.

Using python as a glue to combine Git, GitHub CLI (gh), and OpenAI’s API, I ended up with a simple script.

I chose to put it as ,pr-create command. The comma (,) is just to make sure my script does not clash with some other command.

Script Overview

The complete script is here in my dotfiles.

If you want to use it, make sure openai and yaspin python modules are available. This can be done with a quick pip install openai yaspin

openai is used to receive completion from OpenAI API and yaspin is for a spinner to indicate progress. Just because I wanted a few sparks in my terminal. Adjust to fit your use case!

My script does 3 things:

1. Extracts the git log.
2. Summarizes the changes into a clear, concise title and description using OpenAI’s gpt-3.5-turbo model.
3. Pushes a PR to GitHub using gh cli with the generated title and description.

Extracting the Git Log

To obtain a history of changes, we first need to determine the base commit from which the current branch diverged. This can be done using the git merge-base command:

base_commit = subprocess.check_output(
    "git merge-base HEAD $(git show-ref --verify --quiet refs/heads/main && echo 'main' || echo 'master')",
    shell=True,
    universal_newlines=True
).strip()

Once we have base_commit, then we use it in git log --pretty to retrieve and format the log:

output = subprocess.check_output(
    f'git log --reverse --pretty=format:"* %s%n%w(0,4,4)%b" {base_commit}..HEAD',
    shell=True,
    universal_newlines=True
)

This command also formats each commit as a list item, ensuring commit messages and bodies are properly indented.

Generating the Title and Description with OpenAI

The OpenAI gpt-3.5-turbo model is then used to generate a human-friendly title and description based on the git log. Using OpenAI API is a breeze with the official openai python module:

response = openai.ChatCompletion.create(
    model='gpt-3.5-turbo',
    temperature=0.2,
    n=1,
    messages=[{'role': 'user', 'content': prompt}]
)
output = response.choices[0].message.content.strip().strip('"')

Here, prompt is a string that instructs to generate a PR title or description from the git log. For instance, a prompt for the title might look like this:

prompt = f"You are an expert software engineer preparing the title of a PR. Summarize key changes into a short (max 100 chars) PR title from git log:\n\n{log}"

Here log is the git log obtained in the previous step.

Creating the Pull Request

Finally, we combine generated title and description to create a command for GitHub CLI to create the PR:

command = f'gh pr create --assignee @me --title {title} --body {body}'

Eventually this command is executed in the shell, creating the PR.

Conclusion

By combining Python, OpenAI, and GitHub CLI, I now have a great tool to speed up PR creation. Once i’m ready to push changes to GitHub, all is needed is to run ,pr-create.

After typing hundreds of PR titles and descriptions, this little trick feels like an incredible time saver.

Granted, titles and descriptions aren’t always on-point. Nonetheless, it’s much easier to quickly edit a few logic mistakes in text instead of starting from empty text field.

If you’d like to use or further explore this script, you can find the complete version here in my dotfiles repo.