Hi this was a nice read. Thanks for links to your files. While I was exploring the repo I found the commands section. Like the `commit.md` are they also required for the above described approach to work ?
No, they are not required. Only the files linked above are required, the repo has my full Claude Code config, so you might find other files that are not relevant to this post but part of my larger setup. Feel free to try them out :)
I am also finding adversarial operations are productive. That is the only thing that research is telling us is proven to help. E.g. ask Claude to remove redundant tests after writing tests. Try asking Claude about the utility of the instructions you are giving it. Here is my CLAUDE.md. I am writing a very specific kind of system code so it gets very particular about optimization: https://github.com/whatchamacallem/libhatchet/blob/main/CLAUDE.md
Hey seems like your repo has changed some of the agent code and also added conveyer.
Can you please explain what the new workflow is and how is it different from this post?
Thank you!
The core workflow is still the same for me.
But couple of things have changed since I wrote this post:
Claude Code has had new features like dynamic workflows
I have been using this workflow in a loop for get a lot of feature work done
So, I have been using dynamic workflows a lot along with git worktrees for managing parallel feature development.
The conveyor skill mostly implements the same workflow with dynamic workflows in Claude Code and with git worktrees.
Hi this was a nice read. Thanks for links to your files. While I was exploring the repo I found the commands section. Like the `commit.md` are they also required for the above described approach to work ?
No, they are not required. Only the files linked above are required, the repo has my full Claude Code config, so you might find other files that are not relevant to this post but part of my larger setup. Feel free to try them out :)
Thank you!
I am also finding adversarial operations are productive. That is the only thing that research is telling us is proven to help. E.g. ask Claude to remove redundant tests after writing tests. Try asking Claude about the utility of the instructions you are giving it. Here is my CLAUDE.md. I am writing a very specific kind of system code so it gets very particular about optimization: https://github.com/whatchamacallem/libhatchet/blob/main/CLAUDE.md
Yes, 100%. I am already working on my next post that covers how I have been using adversarial subagents.