AutoPR

What is this?

A vision of user experience I had in March of 2023, back when OpenAI had just launched its first "ChatGPT API".

As far as I am aware, AutoPR was the first bot to autonomously generate pull requests in response to issues. Like most demos back in the day, it worked about 20% of the time.

What inspired the initial weekend coding sprint was Guardrails, who were prompting LLMs to generate structured data with JSON Schemas, and reasking it when it did not adhere. This was a time before grammar-powered structured data generation.

Enjoy this brief glimpse into the past.

How did it work?

Triggered by adding a label containing AutoPR to an issue, AutoPR would:

1. Plan a fix
2. Write the code
3. Push a branch
4. Open a pull request

Examples

Well-written issues often lead to better results.

• Create a dice rolling bot
• Create a 'Tech Jargon Generator'
• Create a user-friendly weather app
• Write three programming interview challenges
• Replace GPT2FastTokenizer with tiktoken

Limitations

This GitHub Action was in development, and in alpha release.

It still:

• Incorrectly referenced code in other files
• Duplicated lines
• Called functions that don't exist
• Only worked on GitHub (see #46)

Usage

Please see USAGE.md for more information.

Contributing

Please see CONTRIBUTING.md for more information.