⌘Development
◎Finance
✦Design

My Claude Guide

⚙Tools✦AI✎Guide

Why Claude over ChatGPT?

Claude is an ever-learning, adaptive tool that can be taught skills and given abilities via plugins and custom instructions. Although it has been a long time since I've used ChatGPT, Claude blew me out of the water from the start and has continued to grow and improve over time.

The difference isn't just quality of output — it's the ecosystem. Claude Code turns your terminal into a full development environment where an AI agent can plan, build, test, and deploy alongside you. That shift from “chatbot” to “collaborator” is what sets it apart.

The Essentials

Plugins

These are the plugins I consider essential for any serious Claude Code setup:

  • Superpowers — gives Claude superpowers when it comes to planning and building. It can deploy multiple agents to complete tasks in parallel, brainstorm designs before writing code, and follow structured workflows that keep projects on track.
  • Code Simplifier — reviews changed code for reuse, quality, and efficiency, then fixes any issues it finds.
  • Code Review — automated code review that catches bugs, logic errors, and security issues before they ship.
  • Context7 — retrieves up-to-date documentation and code examples for any library, so Claude always has current information instead of relying on training data.
  • Playwright — allows Claude to access the web, look things up, interact with pages, and gather information. I recommend the CLI version over the MCP server.

There are plenty of GitHub repos and plugin marketplaces with community-built additions. I'd suggest exploring — people have made some fantastic stuff.

CLAUDE.md — Your Permanent Rulebook

CLAUDE.md is where you define the permanent rules for how Claude behaves. There's a global one (applies to every project) and a per-project one (applies only to that codebase). Claude itself can help you format and refine these files.

Here are some of the rules I run with:

Notifications

Message me on Discord whenever a task is completed. Claude sends a webhook with the project name, my original prompt, and what was done — so I get pinged even if I'm away from the terminal.

Task Management

  1. 1.Plan First: Write plan to tasks/todo.md with checkable items
  2. 2.Verify Plan: Check in before starting implementation
  3. 3.Track Progress: Mark items complete as you go
  4. 4.Explain Changes: High-level summary at each step
  5. 5.Document Results: Add review to tasks/todo.md
  6. 6.Capture Lessons: Update tasks/lessons.md after corrections

Core Principles

  • Simplicity First: Make every change as simple as possible. Impact minimal code.
  • No Laziness: Find root causes. No temporary fixes. Senior developer standards.
  • Minimal Impact: Changes should only touch what’s necessary. Avoid introducing bugs.

MCP Servers & CLI Integrations

Via MCP (Model Context Protocol) servers and CLI integrations, Claude can control your apps — GitHub, Google Calendar, Gmail, and more. This lets Claude help you manage your life, not just your code.

One important distinction: prefer the CLI over MCP when the option is available.MCP servers fill up your “context” — Claude's working memory. The more context you have stored, the worse the output quality becomes. CLIs do the same job without bloating context.

Use /clear often between different tasks, or separate tasks into different Claude terminals. This keeps each conversation focused and high-quality.

My Claude Code Setup

With so many little projects that I'm building and testing, I use Warp as my terminal. It lets me have different projects in different tabs, and tons of terminals open on the same project at the same time.

Previously, I would use VS Code for more advanced tasks where I needed to get a deep understanding of the code and edit it myself — but I had multiple VS Code windows open at once, one for every project. On Warp, I simply have 6 tabs, each with 4 Claude Code terminals for my smaller side projects. Things like a Python script I wrote to cycle multiple API keys to avoid hitting rate limits.

Tips & Best Practices

  • “Use AskUserQuestion until clarity is achieved”

    Put this in your prompt and Claude will keep asking questions until it fully understands what you want before writing a single line of code.

  • claude --dangerously-skip-permissions

    Claude won't ask permission for every step — it just does it. This sounds scary, but in the grand scheme it's necessary if you want to build quickly and efficiently.

  • claude --chrome

    Claude can control your browser via the Chrome extension — looking at websites, collecting images, inspecting layouts, and more.