Puerto Jiménez, Costa Rica
info@gexpsoftware.com
© 2026 Marcelo Retana
After months of refining my AI-powered development workflow, I'm sharing the exact configuration that turned Claude Code into my most productive coding partner.
Six months ago, I started using Claude Code. Today, I can't imagine coding without it.
But here's the thing — out of the box, it's good. With the right configuration, it's transformative.
After hundreds of hours refining my setup, I've open-sourced everything. This post walks you through why this matters, how it works, and how you can set it up in minutes.
When I first started using Claude Code, I noticed something frustrating:
Sound familiar?
The issue isn't Claude's intelligence — it's context. Without proper configuration, every session starts from zero. According to GitHub's 2024 developer survey, 92% of developers now use AI coding tools, but only a fraction configure them beyond defaults — which explains why so many report inconsistent results (GitHub, 2024).
What if Claude could:
That's exactly what my configuration does.
I've organized everything into a clean structure:
~/.claude/
├── CLAUDE.md # Global rules (your coding DNA)
├── commands/ # Reusable workflows
│ ├── focus.md # Deep work sessions
│ ├── quick-commit.md # Safe git commits
│ └── session-end.md # Capture learnings
└── skills/
├── init-project/ # Auto-setup new projects
└── suggest-automation/# Pattern detection
Plus project templates for Node.js, Next.js, and Python.
This one has saved me countless headaches:
### Git Identity - NEVER IGNORE
| Project | Email |
|---------|-------|
| ~/Work/company/** | work@company.com |
| **ALL other projects** | personal@email.com |
Before ANY git commit: verify email first!
Claude checks this before every commit. No more "wrong email" commits to client repos.
This is where it gets interesting. Claude actively watches for patterns:
When pattern detected, ask:
"I noticed you've [pattern] multiple times.
Want me to create a [command/skill] for this?"
The AI literally improves itself based on how you work.
When you open a new project, Claude:
CLAUDE.mdNo manual setup. Just start coding.
Context pollution kills productivity. The config enforces:
/clear suggestions between tasks/compact when context grows too large/focus - Deep Work ModeStarts a focused session with:
/quick-commit - Safe CommitsNever worry about git identity again:
/session-end - Capture LearningsAt the end of each session:
git clone https://github.com/m-ret/awesome-claude-code.git
cd awesome-claude-code
Option A: Run the install script
./install.sh
Option B: Manual installation
mkdir -p ~/.claude
cp global/CLAUDE.md ~/.claude/CLAUDE.md
cp -r global/commands ~/.claude/
cp -r global/skills ~/.claude/
Edit ~/.claude/CLAUDE.md:
Open any project with Claude Code. It will:
Since implementing this setup:
Stack Overflow's 2024 developer survey found that developers using AI tools with proper configuration report 55% higher satisfaction and significantly fewer "unhelpful" responses compared to those using default settings (Stack Overflow, 2024).
This isn't about making Claude do everything automatically. It's about:
As Andrej Karpathy has noted, the most effective way to use AI coding tools is to treat them as a "pair programmer that needs a good onboarding document" — and that's precisely what a well-configured CLAUDE.md provides (Andrej Karpathy, 2024).
Everything is open source and free:
Repository: github.com/m-ret/awesome-claude-code
Includes:
I'm continuously improving this setup. Coming soon:
Star the repo to stay updated.
Yes. The CLAUDE.md configuration, commands, and skills work identically on macOS, Linux, and WSL on Windows. The directory structure (~/.claude/) is resolved to the home directory on all platforms. The install script handles path differences automatically.
No. The CLAUDE.md file is read once at session start and adds minimal context overhead — typically under 2,000 tokens. The productivity gains from reduced back-and-forth far outweigh the marginal token cost. Most users report fewer total tokens used per task because Claude stops asking clarifying questions it no longer needs.
Absolutely. The setup follows Claude Code's configuration hierarchy: global ~/.claude/CLAUDE.md provides your personal defaults, while project-level CLAUDE.md files can override or extend them for team projects. Your git identity rules, personal commands, and session preferences stay private.
The repository provides a battle-tested structure developed over 50+ projects and hundreds of hours of iteration. It includes features most developers wouldn't think to add — pattern detection triggers, session management rules, auto-initialization for new projects, and git identity protection. You could build it yourself, but this gives you a proven starting point.
Questions or suggestions? Let's talk — I'm always looking to improve the workflow.
If you found this useful, share it with another developer who could benefit. The more people contribute, the better it gets for everyone.