You may have ended up here by following a link from one of my PRs marked with an Assisted-by trailer. There is a large variance currently in (1) what patch authors may mean when they say "Assisted-by", and (2) what patch reviewers may interpret when they see "Assisted-by". This document describes what "Assisted-by" means for my commits to reduce guesswork and have a productive collaboration. (Unless I've explicitly stated in the commit or PR body that it should be interpreted differently.)
- I used an AI coding agent to help me write some (or rarely, all) of the patch.
- I take full responsibility for the changes in the patch.
- I self-reviewed every line in the patch and understand its logic.
- I made a best effort to verify its fit within the larger design of the project.
- I am capable of discussing the design decisions taken in the proposed changes.
With the advent of coding agents, it's now extremely easy to write code. However, the converse is not true. It's still hard to review someone else's code. Yes, there are AI review assistants that are really good at finding bugs. But they're also often very wrong, and do not replace the need for human review. It's also still the case that fixing bugs once they reach production is expensive. Thus, project maintainers for production software cannot let their guard down and must carefully read all patches before merging.
Therefore, self-review is mandatory. For how can one ask someone else to review a patch they did not themselves review? The onus is on the one proposing the patch to validate what they're proposing.
The level of scrutiny during self-review should be proportional to the level of scrutiny one expects reviewers to exercise for the given proposed change. (And this in turn is commonly proportional to how close to production the proposed change is.)