We have a lot of unsolved issues and unmerged PRs, and it is hindering our progress.
I have been experimenting with Claude Code/Codex for generating codes. I think a lot of the refactoring we want to do can be automated through them. Since we are using Codex/Claude to review the PR/commits already, I think this is a good time to have some discussion on how much we want this project to be agent driven.
I propose we first add CLAUDE.md/AGENT.md to guide the coding agents. Since we have CI and pre-commit hook setup, this should serve as a good safeguard for any future PRs.
We then can fix CI/CD easily for PyPi.
The next stage will have Codex/Claude to do a full review on the codebase architecture and design. Spot hot spots that can be optimized away. Code agents are pretty good at profiling and identifying hotspots.
The third will be have the GPU version improved and optimized.
We still need to review changes to make sure the physics is correct.
After we reach a stable version, we can start adding features and scaling things up.
I hope this can bring down our open issues and unmerged PR in a few months.
We have a lot of unsolved issues and unmerged PRs, and it is hindering our progress.
I have been experimenting with Claude Code/Codex for generating codes. I think a lot of the refactoring we want to do can be automated through them. Since we are using Codex/Claude to review the PR/commits already, I think this is a good time to have some discussion on how much we want this project to be agent driven.
I propose we first add CLAUDE.md/AGENT.md to guide the coding agents. Since we have CI and pre-commit hook setup, this should serve as a good safeguard for any future PRs.
We then can fix CI/CD easily for PyPi.
The next stage will have Codex/Claude to do a full review on the codebase architecture and design. Spot hot spots that can be optimized away. Code agents are pretty good at profiling and identifying hotspots.
The third will be have the GPU version improved and optimized.
We still need to review changes to make sure the physics is correct.
After we reach a stable version, we can start adding features and scaling things up.
I hope this can bring down our open issues and unmerged PR in a few months.