Creating pull requests with GPT3 and random artwork

A friend of mine has been retweeting great paintings from Twitter accounts like @HenryRothwell, which I’ve greatly enjoyed. E.g.

Good morning - I hope you slept like a caddis-fly larvae-stuffed miller's thumb. I'm starting with 'Winter Landscape', Valerius de Saedeleer, oil on canvas, 1930s. pic.twitter.com/hk6RUFh794

— Henry Rothwell (@HenryRothwell) December 18, 2022

In a similar vein, I’ve started embedding painting images in the description of every pull request I open. E.g.

I create pull requests with:

git createpr

where createpr is an alias to hub pull-request with pre-populated options:

# ~/.gitconfig

createpr = !hub pull-request --draft --browse --edit
    --message=\"$(pull-request-title)\"
    --message=\"$(pull-request-summary)\"
    --message=\"$(random-picture)\"

The pull request body is built by joining the --message options separated by a blank line. Each message component has its own script:

• pull-request-title is a Bash script that uses the OpenAI API to summarise the commit messages into a pull request title. I haven’t quite found the right prompt text to work here — that’s a work-in-progress.

• pull-request-summary is a simple Python script that combines all the commit messages and removes the hard line wrapping¹. The output will generally need some editing before it’s suitable (hence the --edit option in the alias) but it’s a good starting point.

• random-picture is a Python script that plucks a randomly selected image from a set of Twitter accounts and prints the appropriate markdown.

This works nicely — the painting images even unfurl elegantly in Slack.