← Back to context

Comment by edgyquant

16 hours ago

What are people’s workflows for using eMacs with generative coding? I’m trying to get a workflow where I can generate code and approve it like a PR (maybe leave comments for a next round of generation) but I’m new to eMacs and the ai coding package isn’t working with the latest eMacs

Agent Shell connected via ACP to local Claude Code.

https://github.com/xenodium/agent-shell

  • Can't say enough good things about agent-shell.el! Xenodium has done amazing work there.

    • I'm a huge fan of agent-shell, too. It keeps getting better and better, and it's infinitely customizable, like any good Emacs package.

    • I'll have to try it again. It definitely feels like the right way to do things, but when I tried it it had a couple of show-stopper limitations and a show-stopper bug, so I've been using claude-code-ide.el instead.

I’ve been using gptel, I’ll have it generate org babel blocks I can tangle (write to files) or have it use tools to write and edit files in the project I’m working in, and then review with magit.

I use windows 11, wsl2, WezTerm.

emacs -nw directly on the shell.

its like working on a linux computer, but from windows.

I run vterm inside emacs, and claude code inside vterm. I have special commands for extracting file routes, and flow between dired and magit. Everything runs like magic.

I also have copilot-chat and copilot-mode installed. From time to time I use it to review what claude has done, or to chat a little bit.

  • I’m trying to use a similar workflow, but when I attempt to use Claude code in vterm, I have issues.

    My use case is resuming a session and not being able to scroll back to the past history. Otherwise if I start a new session, it works fine.

    It’s probably related to me being on an old version of vterm, but I’m just wondering if you ever have come across this.

    • I found my answer, I needed to be using page up and page down keys to “scroll” through the claude history - which is apparently required due to Claude capturing my C-v and M-v’s in its “text-box”.

IMO, generative coding on Emacs works nicely because magit is awesome for code review. A good agent integration package is going to have a command you can use to send the current cursor line & file to the agent so you can reference and comment.

Which AI coding package? Agent-shell? Gptel-agent?