Using Claude on the backlog

[jcrussell]

I really like Forge. I've been using Claude on some personal projects and wanted to see if it could help address the backlog here (and potentially make the project easier for a new maintainer to pick up). I'm still experimenting to see if the output is actually useful.

I started with bugs that seemed fixable with minimal code changes. I am also exploring whether it can create a comprehensive set of unit tests based on the vanilla repo.

If you're interested, I'm working on these branches:

• https://github.com/jcrussell/forge/tree/claude-fixes
• https://github.com/jcrussell/forge/tree/claude-tests

Hopefully, I'll have something valuable by the end of this experiment rather than just a bunch of slop.

[jcrussell]

I made a PR after iterating on the two branches a bit more and then merging them together:

#502

All the changes in lib/ look reasonable with lots of defensive coding changes. I only had it work on bugs/features that required minimal code changes. I'm going to iterate a little more before removing the "draft" label. I'm curious how it will do on the more comprehensive changes (and now there's unit tests) but that's a project for another time.

[jcrussell]

I've closed #502 in favor of smaller PRs. I submitted a new PR (#504) with just the comprehensive unit tests (@jmmaranan, PTAL and kick off tests when you're ready). Once that merges, I plan to ask Claude to make a plan for how to split the claude-fixes branch and submit PRs.

Meanwhile, I'm continuing to add features on this branch:

https://github.com/jcrussell/forge/commits/main/

[jcrussell]

I've been slowly tinkering with e2e tests, see here:

https://github.com/jcrussell/forge/tree/main/tests/e2e

Partial success:

https://github.com/jcrussell/forge/actions/runs/22335726956

Hopefully getting closer to something that is ready to merge.