I've tried out a similar project (claude-code.el).
I use Spacemacs in evil-mode and I found it very frustrating to try and type into the Claude Code text box (often my cursor would be somewhere weird, the terminal emulator just really did not seem to "understand" that I was not in Insert Mode). I wound up deciding that I'd rather just use Claude Code in the terminal. The Claude Code text box is ALSO annoying there, so I often just write out instructions in some file (in emacs) and tell then tell CC to read it.
Does this project have any facilities for authoring prompts in a temporary buffer or something?
I also had the spacemacs golden handcuffs on before I switched to nvim. I worry that while using spacemacs was great for getting me up to speed in a productive emacs environment, it hamstrung my long term understanding and usage of emacs. I'd often have issues like you describe with no idea how to solve it. My white whale was getting all my web dev major modes to respect a .editorconfig.
I have also been using Claude-code.el and agree that the terminal emulators can struggle to integrate well with my regular workflow. What I have been doing is typing my prompt in the scratch buffer or minibuffer and then sending it to Claude with Claude-code-send-command (bound to s in the transient menu). I don’t even need to switch to the Claude code buffer to send it.
I tried it and it has similar problems. Claude Code is not a good "citizen" embedded elsewhere as it wants to control the terminal completely.
I use default emacs keybindings, which are a bit friendlier with this since they're similar to the bash/readline keybindings it uses... but it's still jarring.
I think I basically had the same experience: https://news.ycombinator.com/item?id=44813189