All projects

CompleteUpdated Added

Repo Steward

A simulated GitHub-style repository with an AI steward that reacts to issues and pull requests, where a deterministic policy engine decides which proposals run automatically, wait for approval, or are blocked.

Visit Repo Steward (opens in a new tab)Visit GitHub Repo (opens in a new tab)

About this project

One topic I have been hearing about more and more is how AI can be used productively in the software engineering process. I found that many of these discussions suffered from the fact that the peers I was talking to often had different visions of hypothetical situations involving AI. While one developer may see an AI-enhanced process going well as they envision it, another developer may see an AI-enhanced process going poorly as they envision it. This is not particularly helpful, because the specifics of how AI is enhancing the process may not be very well defined. Therefore, assessing the potential benefits and risks is a very difficult task.

The purpose of this project was to present one potential outlook on how an AI agent may help in the process of pursuing tickets, reviewing pull requests, and updating documentation based on new code. This project aimed to demonstrate ways in which AI can provide real value and benefit to developers.

Screenshots

The Steward Inbox listing two pending proposals, Remove needs-repro and Mark as ai-candidate, each with a confidence bar, the policy rule that applied and Approve and Reject buttons. (view full size)
Proposals that policy routes to a human wait in the Steward Inbox with the rule that decided it.
Pull request #47 conversation view with commits, a reviewer comment and a Repo Steward panel summarising the fix and the remaining test gap. (view full size)
The pull request steward scenario after analysis of PR #47.

Resources