This post follows the post Four Beautiful Legs. It's separated here for clarity.

This guide is a lightly adapted version of a Google Doc I wrote up for my lab mates to help them get AI coding set up to be a more reliable partner in writing data analysis scripts. It helps to understand a little bit about terminal, but I think you can brute force it by just following along (you'll never regret learning a bit of bash though, I promise)!

How to Set Up and Use Claude Code with SuperClaude

Why I'm Sharing This

I know we've all incorporated chat/AI into coding at varying levels (and I know I've been somewhat annoyingly vocal about what's in the lead), but in the last couple days I've gotten a couple things set up that are performing so much better than what I've used before that I wanted to share.

Right off the bat, this does require a Claude subscription – it's $20/month like ChatGPT, but honestly it's so overwhelmingly better for code at this point that if you haven't switched, I really think it's worth taking the time to.

What Makes This Special

The biggest useful feature they offer is Claude Code. Claude Code is basically an AI agent that lives in your terminal. It's a little weird and takes some getting used to, but it's really powerful.

I've been playing with Claude Code pretty much since it came out. It's always been a bit better than the other agents but not enough to make it worth actually trying to get people to use instead of the others. But recently, one tool came out that bundles a lot of the best improvements together: SuperClaude.

At a zoomed-out level, SuperClaude basically passes any prompt you give it through a bunch of filters that enforce best practices for coding, documentation, and error handling. It sort of takes a prompt you give it and punches it up so it looks like it's coming from somebody that actually knows how to write code at a software engineer level.

Installation Mini-Guide

Step 1: Get Homebrew

To get the rest of this set up, you'll need Homebrew installed. It's easy to set up – just run the line on brew.sh and then follow the instructions in terminal.

Step 2: Install the Basics

In terminal, you'll need to run:

brew install npm
brew install uv

Step 3: Install Claude Code

You can use npm to install Claude Code:

npm install -g @anthropic-ai/claude-code

Step 4: Set Up SuperClaude

To get SuperClaude, run these commands:

uv venv
source .venv/bin/activate
uv pip install SuperClaude

Then:

python3 -m SuperClaude install

Step 5: You're Ready!

Once it's done, you can just run:

claude

Tips on Using It

Getting Started

To start, I recommend making a new directory to practice in. I usually name mine something like claude_projectname_test. Copy anything you want to be a part of the test into this folder (old code you want to fix, data you want to analyze, etc – you can also leave it empty if there's nothing you want to test on).

You'll cd into the folder containing that, and then ultimately the universe is sorta open to you.

I used it to one-shot a game I've been trying to build for a long time recently. It took me an hour.

Writing Your First Prompt

Generally, in a first message I recommend giving it a pretty specific idea of what you want and asking it to make a plan to get there. As an example, when I tested the game I gave it the prompt:

I've really wanted a long time to make a game that is a combo of vampire survivors 
and asteroids. The art should be polygons drawn using code to keep that original 
asteroids polygon based style (but maybe we'll want to replace it later with actual 
sprites idk). But let's make it right? I want an advanced, satisfying, upgrade system. 
Asteroids gameplay. Lots to unlock. I need it to be in something like pygame or 
something else comprehensible and easy to work on.

I also recommend telling it what language you want it in (R, Python, etc – choose one you can debug and read yourself. Using this to generate code you can't read is still pretty ill-advised at this point, especially for science).

After it gives you the plan, you can either give it feedback or tell it to build it. It'll pretty much handle the rest itself.

Where It Really Pulls Away

All of this is pretty true so far of any AI agent, but this next part is where it really starts to pull away.

The Testing Magic

After it's done building the code you asked for, you can ask it to set up a suite of test scripts that log results for bugfixing and troubleshooting. If you have data, you can give it the data. Bonus points if you have an example of what the data should look like before and what you expect after (it can handle figures, so if you have data in and an example PNG of what you want at the end, it can handle that – just drop it in the folder and tell it).

When it finishes writing the test scripts, you can have it test and fix the code itself just by asking. For the game I wrote:

Does this code have everything it needs to run the testing/debugging suite 
completely from terminal?

After it finished addressing this, I just told it to bugfix until the game was working properly. It did, and the game runs great (this is the first time any agent has made it to this stage and not failed).

Fine-Tuning Your Results

You can also give it even more specific commands as you find issues:

It looks like the tests you've made are passing but visually at the end the 
results look bad. Add a feature where you take a screenshot so you can evaluate 
the final result and make aesthetic upgrades.

For Lab Work

For work in the lab, critically, you can use the test script, logs, and commented script Claude has written to easily understand what it wrote, how it works, and actually make sure you are getting what you want in the end.

Other Cool Stuff to Try

• Make it terminal-friendly: Ask it to make your scripts work from the terminal with rich feature support like input/output flags, then tell it to run the whole analysis itself
• Think bigger: Ask it if there's a way to make the code more flexible or broadly applicable to other inputs/data
• Deep debugging: If your code goes through many steps as it runs, ask Claude to pretend it's running through the code and to check how all functions interact and make sure there's no conflicts

Important Notes

When This Works Best

This is incredible for prototyping, data analysis, building tools, and even game development. But remember – you still need to be able to read and understand the language you're asking it to use. This isn't quite at the "generate production code you can't read" stage yet.

Quality Check

Always review what it generates, especially for scientific work. The test scripts and logs it creates are your friends here – they help you actually understand what's happening under the hood.

Troubleshooting

If things aren't working:

• Make sure all the dependencies installed correctly
• Check that your Claude subscription is active
• Verify you're in the right directory
• Look at those logs the test scripts generate
• Just ask Claude to diagnose and fix specific error messages – it's pretty good at that

Final Thoughts

Honestly, this setup has changed how I approach coding projects. What used to take me days of iteration now happens in hours, and the code quality is often better than what I'd write myself. Give it a try – I think you'll be as impressed as I've been.

• Also, I write this blog to learn as much as I write it to document my thoughts and (hopfully) teach. I'm by no means an expert. If you see anything in here that is wrong (or even that could just be improved in accuracy), please don't hesitate to reach out!!