Let me cook!

[EanHD]

There should be a mode or easy cli interface that is like a wizard to make your spec, then after you set up your agent tool (like github ciopilot, ect) you can let it cook and just implement all your tasks, and you can keep track of the progression until it is done. no constantly having to check if you need approval to run a command or if the llm needs you to say "please continue" or something

[EanHD]

Yeah that would be revolutionary bro.

[TabishB]

@EanHD I believe that's more controlled by the coding agent itself? I believe claude code, codex etc usually have some sort of "yolo" mode that allows you to dangerously skip permissions. We can't really control this from our end unless we go down a layer and wrap ourselves in the execution process too.

YOLO mode mentioned here:
https://www.anthropic.com/engineering/claude-code-best-practices