Claude Code is my tool of choice for coding. I run it in 2 modes:

YOLO mode

You can write Claude Code in what is sometimes referred to as “YOLO mode” (claude --dangerously-skip-permissions) where it skips permissions prompts. I do this in a Docker container where I mount my local repo directory, as well as my Claude configuration files onto the container and that’s all the local disk access it has.

I prebuild a Docker image and have a fish function so I can do this in the root of my repo:

  1. In host machine: claude-docker to start the container with disk mounting
  2. In container: claude --dangerously-skip-permissions
  3. Every day or so, I might need to run /login, copy and paste to login to my Claude Max account
  4. Run /yolo

/yolo is a Claude custom command I wrote that prompts it to look at a plan.md file which is basically a list of TODOs, each with 1-2 sentences; run through the items, implement the work, run prettier+lint and iterate until that’s done and then commit.

The TODOs are usually simpler and I can almost trust it to do the right thing.

Interactive mode

For bigger changes, or if I want to iterate myself by looking at the output, especially in the browser and tweak. I’ll just write a longer prompt in Neovim, paste it into Claude Code running directly in my local machine and then ask it to perform changes/tweaks as it goes along.

Sometimes I’ll ask it to use MCPs, especially to check the browser logs or to take screenshots to fix things


I don’t usually let Claude Code run big prompts and try to build a gigantic feature at one go, but being able to let it chug through smallish TODOs in YOLO mode feels really liberating and frees me up for other tasks like household chores or marketing while it’s writing code for me.

I like to joke that I’m paying it to do what I love while I do what I don’t like to do.