Boris Churnney is the creator of Claude Code. Every so often he drops a tweet with his top tips for how to use it. This is gold. It’s like Alex Hormozi telling you exactly how to build your Skool community to 100,000 members. You learn from the person who made the thing.
The problem? Most people read these tips and have no idea what to do with them. “Do more in parallel” and “spin up different work trees” doesn’t click for everyone. So I, Charles Dove, took all 10 tips, tested them inside my own Claude Code workspace, and I’m going to break down exactly how each one works in practice.
If you’ve been following the Charlie Automates channel, you know I don’t do theory. I do execution. Let’s get into it.
Tip 1: Git Worktrees for Parallel Execution
This is the first tip and it’s a big one. Git worktrees let you run multiple Claude Code instances at the same time. Each one works on a different branch. No file conflicts. No stepping on each other’s toes.
Here’s how it works in practice. You put your app into GitHub. Then you create different branches using worktrees. Each branch handles a different feature. I tested this on my Kanban Pro app. I set up three parallel worktrees:
- Kanban Color Theme for updating the visual palette
- Kanban UI Structure for reworking the board layout
- Kanban Support Queue for cleaning up the support view
All three ran at the same time in separate terminals. Each one had its own plan. Each one executed independently. Then I merged everything together.
The merge can feel slow. That’s the tradeoff. But the build phase? Way faster. Instead of doing three features back to back, I knocked them all out simultaneously. Imagine having a team of three developers working at once. That’s what this feels like.
Pro tip: Open separate terminal windows for each worktree. Use git worktree add to create each one. Name them clearly so you don’t lose track.
Tip 2: Start Every Task in Plan Mode
You’ve probably heard this before. Start every task in plan mode. I do this for everything, not just complex stuff.
Here’s why. When you skip planning and just tell Claude to execute, you get something close to what you want. Maybe 70%. Then you spend the next hour fixing the other 30%. Plan mode forces Claude to outline exactly what it’s going to do before writing a single line of code.
You get to review the development approach, the color theme, the file structure, everything. Then you approve it. Then it executes. The goal is a one-shot success. Get as close to perfect on the first run as possible.
For all three of my parallel worktrees, I started each one in plan mode. The plans were done in minutes. The execution was clean. No surprises.
The takeaway: Plan mode saves you time, tokens, and headaches. Use it every single time.
Tip 3: Invest in Your CLAUDE.md File
This is probably the most important tip on the entire list. Your CLAUDE.md file is how you train Claude Code. It’s the general settings for how Claude interacts with your workspace.
Think of it as your operating manual. You tell it what to do, what not to do, and how you like things done. Then every future session starts with that context already loaded.
Here’s what I include in mine:
- Personal context. Who I am, my social links, my goals, my ICP as a business owner. When Claude communicates with people through plugins (like LinkedIn), it already knows who I am.
- Project context. What I’m working on, whether it’s an app, an n8n workflow, or a Claude plugin. It understands where to place each project.
- Workspace structure. My file tree layout so it knows where everything goes.
- Development preferences. I don’t like it to build without a plan first. So I have it always start with a spec plan or PRD (product requirement document).
- Operating principles. I fed Claude articles from Anthropic themselves, then had it synthesize them into core principles for how I want it to work with me.
Everyone’s CLAUDE.md will look different. That’s the point. Build yours based on how you work. And keep iterating on it. Every time Claude does something you don’t like, tell it to update the CLAUDE.md. This is how you grow your relationship with Claude Code over time.
If you haven’t set yours up yet, run /init and start building.
Tip 4: Create a Shared Skills Repository
This one surprised me. You can create a GitHub repo and dump all your custom skills into one place. Then your team (or anyone you’re teaching) can pull individual skills from it.
I created a repo called “Charlie Automate Skills” and asked Claude to list all my skills and push them there. 78 skills total. Everything from GSD project commands to video analysis workflows.
This is powerful when you start working with other people. Instead of explaining how your custom commands work, you just point them to the repo. They pull what they need. Done.
It doesn’t need to be complicated. Create a repo. Push your skills. Share the link. That’s it.
Tip 5: Enable the Slack MCP for Team Bug Tracking
I’m not using this one myself because our team isn’t big enough yet. But I can see how it’s a game changer for larger teams.
The idea is simple. Connect Slack to Claude Code via the MCP (Model Context Protocol). When someone on your team drops a bug report in a Slack channel, you copy the link, paste it into Claude, and say “fix this.”
It pulls all the context instantly. Claude sees the bug, understands your codebase, and either creates a plan or goes straight to execution mode.
For teams with dedicated bug channels for different projects, this is huge. You could tell Claude to scan an entire channel and build a fix plan for every bug listed. No more digging through Slack threads manually.
If you want to set this up, just ask Claude to find the Slack MCP server for you.
Tip 6: Use “Challenge Mode” to Avoid Technical Debt
This is a mindset shift more than a feature. Most people ask Claude to make changes, get the result, and move on. That’s technique A. It works. But it doesn’t push you.
Challenge mode is technique B. You have Claude grill you on your decisions. It asks questions like:
- Why did you choose this approach over that one?
- What happens if this edge case occurs?
- How does this handle scale?
These are the questions that prevent technical debt. Technical debt is when you build something quick and sloppy, then six months later you’re dealing with problems that could’ve been solved from day one.
There’s also the elegant reset approach. After a bunch of iterations and band-aid fixes, you tell Claude: “Knowing everything you know now, scrap this and implement the elegant solution.” It takes all the wisdom from previous attempts and rebuilds from a clean slate. Every new iteration comes with the knowledge of what failed before.
And then there’s technique C, which is about being specific with your prompts up front. “Add a login” is vague. “Add Google OAuth with JWT sessions, 24-hour expiry, redirect to dashboard” is precise. The more specific your prompt, the higher the quality of the output. Build out a detailed spec before you start. Save yourself the back-and-forth.
Tip 7: Use a Dedicated Terminal (Ghosty)
This one doesn’t apply to me since I use Claude Code inside VS Code. But Boris recommends using Ghosty, a standalone terminal, for a more focused experience.
The benefit is you can customize your terminal environment specifically for Claude Code. Custom status lines that show your current git branch, context window usage, and project state. If you’re 40-50% through your context window before it compacts, you’ll see that at a glance.
You can set up aliases too. Type something like cdd and it boots Claude Code right into your project directory. No extra steps.
I’m not personally a big fan of Ghosty. But if you want that hyperfocused terminal experience, check it out at ghosty.org.
Tip 8: Sub-Agents for Parallel Processing
Right next to the CLAUDE.md file, this is one of the most useful features in Claude Code. Sub-agents let Claude spin up separate instances behind the scenes to handle different tasks simultaneously.
I tested this live. I prompted Claude to spin up three sub-agents:
- Research agent for looking into how dogs became domesticated (random example, but it proves the point)
- App ideas agent for finding the best app opportunities to build in 2026
- Design agent for researching the best color themes for my app
All three ran in parallel. Three local agents, three different tasks, all at once. The results came back and the main context window stayed clean.
This is the key benefit. Sub-agents offload individual tasks so your primary session doesn’t get cluttered. You throw more compute at the problem without burning through your main context window.
You can add this to your CLAUDE.md file too. Tell Claude to always ask if you want to spin up sub-agents for any given request. It becomes part of your standard workflow.
This is probably one of the most underrated features in Claude Code right now. I highly suggest integrating sub-agents into your daily workflow.
Tip 9: Claude Code for Data and Analytics
If you’re using something like Supabase for your database, there’s already an MCP for it. Connect it to Claude Code and you can query your data with plain English.
Instead of opening spreadsheets and running manual queries, you just ask:
- “What are our top 10 products this last month?”
- “Show me weekly users.”
- “Compare our quarter performance.”
- “Find customers at churn risk who haven’t logged in but still have active subscriptions.”
Claude queries the database, finds the high-risk profiles, and gives you actionable data. From there you can build marketing plans, outreach campaigns, or retention strategies.
Screw spreadsheets. This is how data analysis should work in 2026.
Tip 10: Learn with Diagrams
I referenced this earlier in the video. Go to /config in Claude Code, scroll down to output style, and set it to “explain” instead of default.
Now every time Claude builds something, it explains what it did with visual diagrams. You’re not just reading walls of text. You’re seeing the architecture, the data flow, the component relationships.
I updated my CLAUDE.md to require diagrams with every explanation. So when I ask for a plan to build a static website, I don’t just get a list of steps. I get a visual diagram showing how everything connects.
This is extremely powerful for learning. If you’re trying to understand what Claude actually built for you, diagrams make it click instantly. Visual learners, this one’s for you.
The Big Picture: How These Tips Work Together
Here’s the thing. These 10 tips aren’t isolated tricks. They’re an operating system for how you use Claude Code.
- Start with CLAUDE.md to set your foundation
- Use plan mode before every build
- Run parallel worktrees for speed
- Spin up sub-agents for research and side tasks
- Challenge your own decisions to avoid technical debt
- Connect your tools (Slack, Supabase, etc.) via MCP
- Learn visually with diagrams and explain mode
- Share your skills through a centralized repo
Every tip feeds into the next. Your CLAUDE.md gets better over time as you iterate. Your sub-agents get more useful as you define clearer instructions. Your parallel execution gets smoother as you learn the worktree workflow.
This is the iteration process. It’s not about getting everything perfect on day one. It’s about building a system that gets better every time you use it.
FAQ
What is a CLAUDE.md file and why does it matter?
CLAUDE.md is a markdown file that lives in your project. It tells Claude Code how to behave in your workspace. Your preferences, your project structure, your rules. Claude reads it at the start of every session. The better your CLAUDE.md, the better your results.
Do I need to be a developer to use Git worktrees?
Not really. The concept is simple. You create separate branches that run in parallel. Claude handles the git commands. You just need to open multiple terminal windows and tell each one what to work on. Start small with two worktrees and scale up.
How many sub-agents can I run at once?
There’s no hard limit that I’ve hit. I ran three in my demo. You could run more depending on your subscription tier and what you’re working on. The key is to give each sub-agent a clear, focused task.
Is Ghosty terminal required for Claude Code?
No. Claude Code works perfectly in VS Code’s built-in terminal, which is what I use. Ghosty is an option for people who want a standalone terminal with more customization. Try it if you want. Skip it if you don’t.
What’s the best way to start implementing these tips?
Start with your CLAUDE.md file. That’s the foundation for everything else. Once that’s solid, practice using plan mode on your next project. Then try one parallel worktree. Build from there. Don’t try to implement all 10 at once.
How does challenge mode help with technical debt?
When Claude challenges your decisions before building, it catches bad patterns early. Instead of building something quick that breaks later, you think through edge cases up front. It’s like having a senior developer review your approach before you write code.
Can I use Claude Code for data analytics without Supabase?
Yes. If your database has an MCP server available, you can connect it. Supabase is just one example. Check if your database provider has an MCP integration. If not, you can still export data and have Claude analyze it directly.
Ready to Level Up Your Claude Code Skills?
These 10 tips from the creator himself are a starting point. The real skill is in the iteration. Using Claude Code every day, updating your CLAUDE.md, refining your prompts, and building systems that compound over time.
If you want to go deeper, I break down Claude Code workflows, AI automation strategies, and real builds every week on my YouTube channel @charlieautomates.
Want hands-on coaching? Work with me 1-on-1 and I’ll help you build your Claude Code workflow from scratch.
Want to learn with a community? Join CC Strategic AI on Skool where we share skills, workflows, and help each other ship faster.
Need AI automation for your business? Book a call with CC Strategic and let’s talk about what we can build together.