Three steps to turn Claude Code from a chat window into the system everything else sits on top of. Not a feature tour. The install, the prompts that work, and the one file that makes every session 10x better.
The problem nobody talks about
Every time you open Claude Code, you start from zero. No memory of your project. No knowledge of your conventions. No awareness of which files matter and which to ignore. You spend the first five minutes of every session re-explaining who you are and what you're working on.
Multiply that across ten sessions a day. That's fifty minutes of repeated setup. Every single day.
Three steps fix this. Install. Prompts. CLAUDE.md. You set it up once. Claude follows it every time.
Why this changes everything
- Setup once, run forever. The kit you build today saves time on every session for the next two years. It compounds.
- Persistent memory built in.
CLAUDE.mdanchors per-project context so the agent stops forgetting your conventions, your file structure, and your rules. - Open standard. Skills,
CLAUDE.md, and the prompt patterns in this guide work in Claude Code, Cursor, Codex, Copilot, and Gemini CLI. Build the kit once, use it everywhere.
Step 1: install Claude Code
Run this in your terminal:
npm install -g @anthropic-ai/claude-codeThen claude to launch. First run prompts you to log in to Anthropic. Twenty dollars a month for Claude Pro, which bundles Claude Code with it.
Once you're in, you have an AI coding agent in your terminal that can read your files, run commands, edit code, and commit changes. The next two steps turn it from a generic assistant into something that actually knows your project.
Step 2: the 3 prompts I run daily
First:"Read the project, summarize what each module does, flag anything that looks broken." Use this on every new repo. It tells you what Claude actually understands before you ask for changes.
Second:"Refactor X to do Y. Keep the existing tests green. No new dependencies." Most refactors fit this template. The constraints kill most of the bad output.
Third:"Debug this stack trace. Walk me through what you check first." Higher accuracy than asking it to "fix the bug" because it forces stepwise reasoning instead of guessing.
These three cover seventy percent of my actual usage. They work because they give Claude constraints. Without constraints, it guesses. With them, it follows a path.
Step 3: the CLAUDE.md anchor
Drop a file called CLAUDE.md at the root of every repo you work in. What goes in it:
- Project description in three sentences
- The file map (which folders matter)
- Conventions you follow
- Things to never touch (auto-generated files, secret configs, vendored libraries)
- The test command
Claude Code reads this file on every session start. It anchors the agent in your project before you type a word. This single file kills eighty percent of the "why does it keep doing the wrong thing" problem.
Here's a starter template:
# Project Name
One-line description of what this project does.
## Tech Stack
- Framework: Next.js 16 (App Router)
- Language: TypeScript
- Styling: Tailwind CSS v4
## Structure
- `app/`: pages and layouts
- `components/`: shared UI components
- `lib/`: utilities and helpers
## Conventions
- Use server components by default, "use client" only when needed
- Prefer named exports
- No barrel files
## Do Not Touch
- `node_modules/`, `.next/`, `dist/`
- Any `.env` files
- Auto-generated types in `generated/`
## Commands
- `npm run dev`: start dev server
- `npm run build`: production build
- `npm run test`: run testsCustomize this for your project. The more specific you are about conventions and off-limits files, the less time you spend correcting the agent.