Running Claude Code in Windows CMD? Three Upgrade Layers to Make It Your Main Workflow
Table of Contents
- You don't need to use it yourself. You only need to forward the block below to your engineer.
- Three things you (the manager) need to know
- 1. CMD is not a tool where installing Claude Code is enough
- 2. With poor tools, you get only a fraction of the AI dividend
- 3. Upgrading is not the engineer's problem. It is the manager's.
- The "upgrade assessment table" to show your engineer
- My own workflow
- What if you get really stuck?
- Advanced: if the engineer wants the technical details directly (optional)
- Layer 1: Windows Terminal
- Layer 2: VS Code + Claude Code
- Layer 3: WSL2 + tmux
You don't need to use it yourself. You only need to forward the block below to your engineer.
After the May 23 session at Feng Chia in Taichung, a manager came up and asked me:
"I had my engineer install Claude Code, but they run it in Windows CMD and it feels clunky. I don't understand the tech myself. How can I help them optimise it?"
The answer is not to go and learn it yourself. It is to copy the block below and send it to your engineer. The AI will do the assessment for them:
I want to optimise my Claude Code workflow on Windows.
Right now I run it in CMD and it feels clunky, but I'm not sure where to start changing things.
Please act as a "Claude Code workflow expert" and first ask me these 5 questions:
1. What do I mainly use Claude Code for?
(A. Writing code B. Organising documents / writing markdown C. Debugging / reading logs D. Running several repos at once)
2. What is my biggest headache right now?
(A. Garbled Chinese characters B. No colours / can't see diffs C. Can't open multiple windows D. Switching windows is slow
E. Poor mouse support F. Other ___)
3. How much time am I willing to spend upgrading my tools?
(A. Under 5 minutes B. Under half an hour C. 1-2 hours D. Willing to spend an afternoon)
4. Do I have VS Code? Yes / No / Not sure
5. Do I have WSL (Windows Subsystem for Linux)? Yes / No / Never heard of it
Based on my answers, pick the most suitable option from the three-layer upgrade plan and recommend it to me:
- Layer 1 (5 minutes): Windows Terminal: replace CMD, get colours, tabs, UTF-8
- Layer 2 (30 minutes): VS Code + Claude Code integration: view diffs, accept/reject with shortcuts
- Layer 3 (1-2 hours): WSL2 + tmux: run several Claude sessions in parallel
Please give me:
- Why this layer (in words I can follow, with an analogy from everyday life)
- Exactly how to install it (step by step, telling me what I will see at each step)
- The difference I should feel once it's installed
I write code but I'm not familiar with terminal configuration, so for every step please tell me what result I will see.
Your engineer pastes this into Claude / Gemini / ChatGPT. The AI assesses the situation with the 5 questions, then gives a customised upgrade path.
Three things you (the manager) need to know
1. CMD is not a tool where installing Claude Code is enough
Your engineer's current situation is like buying a sports car and driving it on a dirt road.
Claude Code is designed to run in a good terminal. On CMD you lose 70% of the experience.
2. With poor tools, you get only a fraction of the AI dividend
My own measurements:
- CMD + Claude Code → engineer productivity +30%
- VS Code integration → +200%
- WSL2 + tmux + multiple sessions → +400%
Same Claude Code, but with the wrong tools the results differ by 13x (+400% vs +30%).
3. Upgrading is not the engineer's problem. It is the manager's.
Engineers are usually busy, and while "it works well enough for now" they will not switch tools on their own. Your role as the manager is to give time, a goal, and room.
Concrete things you can do:
"This Friday from 2 to 3 pm, the whole team spends 1 hour upgrading our Claude Code workflow. After 3 pm, everyone tries one task they used to do in CMD and sees whether it goes faster. At 4 pm we all share what we found."
That is 10 times more effective than engineers slowly figuring it out on their own.
The "upgrade assessment table" to show your engineer
| Engineer type | Recommended layer | Why |
|---|---|---|
| Uses Claude Code occasionally (< 5 times a week) | Layer 1 (Windows Terminal) | Upgraded in 5 minutes, lowest cost |
| Mainly writes code, uses it every day | Layer 2 (VS Code integration) | Diffs and shortcuts save time |
| Runs several projects / several tasks at once | Layer 3 (WSL2 + tmux) | Parallel working capacity goes up 4x |
For 90% of engineers, Layer 2 is the sweet spot.
My own workflow
All the material for the full 90-minute talk on May 23 (slides, script, the Prompt Injection PDF) was made with Layer 3 (WSL2 + tmux), running 4 Claude sessions in parallel.
Total time: 5 hours. One person + one AI.
But I didn't start with Layer 3 either. I used Layer 1 for 3 months and Layer 2 for half a year before moving up to Layer 3. Upgrading gradually is the right way.
What if you get really stuck?
Two places to ask for help:
- Ultra Lab Discord → go to
#ask-agent. Whatever the engineer asks, the lobster will reply (within 1 minute) - Email: ppc@ultralab.tw → tell me which step your engineer is stuck on, and I will do my best to reply
Advanced: if the engineer wants the technical details directly (optional)
Skip the prompt above and read this instead:
Layer 1: Windows Terminal
Search for "Windows Terminal" in the Microsoft Store and install it. Change the settings: PowerShell as the default, JetBrains Mono 14pt, the One Half Dark colour scheme. Ctrl+Shift+T opens a tab, Alt+Shift+- splits horizontally, Alt+Shift+= splits vertically.
Layer 2: VS Code + Claude Code
In VS Code Extensions, search for "Claude" and install the official extension. Ctrl+Shift+P → "Claude" to start it. You see the code and Claude in the same window, and diffs pop up automatically.
Layer 3: WSL2 + tmux
wsl --install # PowerShell admin
# After rebooting, open Ubuntu
sudo apt install tmux nodejs npm
npm install -g @anthropic-ai/claude-code
tmux new -s work
# Ctrl+B " horizontal split / Ctrl+B % vertical split
PPC's 4-pane workflow: top left repo-A, top right repo-B, bottom left the dev server, bottom right a git status loop.
Write back and tell me how far your engineer has got, or which step they are stuck on.
FAQ
What's wrong with running Claude Code in Windows CMD?
Claude Code is designed to run in a good terminal. On CMD you lose 70% of the experience, like buying a sports car and driving it on a dirt road. The lowest-cost first step is Windows Terminal: about 5 minutes to replace CMD and get colours, tabs and UTF-8.
What are the three upgrade layers for a Claude Code workflow?
Layer 1 is Windows Terminal, about 5 minutes: replace CMD and get colours, tabs and UTF-8. Layer 2 is VS Code + Claude Code integration, about 30 minutes: view diffs and accept/reject with shortcuts. Layer 3 is WSL2 + tmux, about 1 to 2 hours: run several Claude sessions in parallel.
Should I use Windows Terminal, VS Code or WSL2 + tmux for Claude Code?
It depends on how often and how you use it. Occasional use (fewer than 5 times a week) points to Layer 1, Windows Terminal. Mainly writing code and using it every day points to Layer 2, VS Code integration. Running several projects or tasks at once points to Layer 3, WSL2 + tmux. For 90% of engineers Layer 2 is the sweet spot, and upgrading gradually is the right way: the author used Layer 1 for 3 months and Layer 2 for half a year before moving up to Layer 3.
How can a non-technical manager help an engineer optimise Claude Code?
You don't need to learn it yourself. Forward the assessment prompt in this post to your engineer and have them paste it into Claude, Gemini or ChatGPT. The AI asks 5 questions first, then gives a customised upgrade path. The manager's role is to give time, a goal and room, for example by setting aside 1 hour for the whole team to upgrade together, then having everyone try one task they used to do in CMD and share what they found.
How do I run several Claude Code sessions with tmux on WSL2?
Run wsl --install in an admin PowerShell, reboot and open Ubuntu, install tmux, nodejs and npm with apt, install @anthropic-ai/claude-code globally with npm, then start a session with tmux new -s work and split the window into panes. The author's 4-pane layout is repo-A top left, repo-B top right, the dev server bottom left and a git status loop bottom right.