Claude Code Power User Guide

Most people think Claude Code is basically Claude but in your terminal - a chatbot with access to your local files. That is completely wrong.

What the source code reveals is that Claude Code is a full agent runtime. It is a proper application built with Bun, TypeScript and React. It has a tool system, a command system, a memory system, a permission engine, a task manager, a multi-agent coordinator and an MCP client and server, all wired together under one execution pipeline.

The flow works like this: your input hits a CLI parser, goes to something called the Query Engine, calls the LLM API, runs a tool execution loop and then renders results back in your terminal.

The 11 architectural layers are: Terminal UI, Command Service, Permission Engine, Tool System, Query Engine, Memory System, Task Manager, Agent Coordinator, MCP Client/Server, Hook System and Runtime Core.

If you are using Claude Code like a chatbot - just typing questions and hoping for good answers - you are using maybe 10% of what it can actually do. The rest of the value is in the systems built around the model, and that is what the remaining insights are about.

Think of it like buying a Ferrari and never leaving first gear. The model is the engine, but the harness is the gearbox, suspension, steering and brakes. You need both.

CLAUDE.md is not documentation. It is an operating manual that gets injected into every single conversation turn. Not just at session start - every turn.

Most people either ignore this file or dump a few random notes into it. That is a massive missed opportunity because you have 40,000 characters to shape how Claude Code behaves, and most people use about 200.

Think of it this way: if Claude Code is an employee, then CLAUDE.md is their onboarding document. It tells them how you work, what matters, what to never do and how the project is structured.

The four-level hierarchy

CLAUDE.md is not a single file. It is a four-level system that cascades from global to private:

• Global (~/.claude/CLAUDE.md) - applies everywhere. Your universal conventions
• Project (./CLAUDE.md) - project-specific context. Architecture, stack, team norms
• Modular Rules (.claude/rules/*.md) - focused rules that load based on file glob patterns
• Private (CLAUDE.local.md) - your personal overrides, git-ignored

The memory system: MEMORY.md

Beyond CLAUDE.md, there is a full persistent memory system. Claude Code maintains a MEMORY.md index file and individual memory files across sessions. It stores five types of information:

• User memories - your role, preferences, expertise level
• Feedback memories - corrections and confirmations from past sessions
• Project memories - ongoing work context, decisions, deadlines
• Reference memories - pointers to external systems and resources

The memory system means Claude Code learns from your corrections. Tell it once that you prefer Australian English or that tests should go next to source files, and it remembers across every future session.

What to put in CLAUDE.md

The best users keep CLAUDE.md short, opinionated and operational. Decision rules, constraints and conventions - not a novel about your project's history:

• "We use TypeScript strict mode. Always."
• "Tests go in __tests__/ folders next to the source file."
• "Never modify the database schema without running migrations."
• "Use pnpm, not npm."
• "Australian English only. No em dashes. No Oxford commas."

If you get nothing else from this guide, go update your CLAUDE.md today. It is the single highest-leverage thing you can do to improve your Claude Code experience.

The source code reveals roughly 85 slash commands in Claude Code. Most users know maybe five of them. Power users treat these commands like keyboard shortcuts - they are one of the biggest levers for getting more value out of the tool.

/plan - the command that changes how you work

This deserves its own section because it is genuinely the most important command in Claude Code.

/plan puts Claude Code into planning mode. Instead of immediately executing, it maps out the full approach first. It reads files, explores the codebase, identifies dependencies, considers edge cases and presents a structured plan for your approval before it touches anything.

The source confirms this is a dedicated product feature with its own tools for entering and exiting plan mode - not just a prompting trick. There is also /ultraplan for more thorough planning on complex tasks.

Why this matters: most wasted tokens come from Claude Code going down the wrong path and having to backtrack. Planning first eliminates that. You review the approach, adjust if needed, then execute with confidence. It saves tokens and produces better results.

For any task that touches more than a couple of files, start with /plan. It is the difference between a junior developer who starts coding immediately and a senior who thinks first.

The other five that matter daily

/compact compresses your conversation history. You can run it with a prompt like /compact keep the API integration details to control what survives compression. Essential for long sessions.

/review and /security-review are built-in code review workflows. The fact that these exist as dedicated commands tells you review is a first-class feature, not an afterthought.

/context shows what files Claude Code is tracking. Every file costs tokens, so this is your spend visibility tool.

/cost shows real-time session spend. Check it regularly.

/init generates a CLAUDE.md by analysing your codebase. The fastest path to a solid project config.

Better prompts are one lever. Knowing these commands is a completely separate lever that most people are not even touching.