A free step-by-step guide to installing Claude so it works with persistent memory, runs multiple tasks at once, and learns your specific projects over time. No technical background required.
This isn't just a better chatbot. It's a fundamentally different way of working. Understanding the shift helps you use the system correctly.
Claude reads a briefing file before every session. It already knows who you are, what you're building, and what the priorities are.
One command spins up multiple agents simultaneously: one building, one reviewing, one documenting. Like a real team.
Assign tasks from your phone. Claude picks them up on your desktop and works while you're in meetings or on the move.
Skills generated from your codebase make Claude more accurate the longer you use it. It learns your patterns, not generic ones.
Follow every step in order. Check them off as you go. Steps marked Mac or Windows: skip the ones that don't apply to your computer.
This is where you'll do most of your work. The desktop app is different from the website. It supports plugins, memory, and multi-agent features the browser can't do. Go to claude.com/download, click your operating system (Mac or Windows), download the file, and open it to install. Sign in with your Anthropic account when it opens.
The free plan won't work for this setup. Open the Claude app. Your plan is shown in the bottom-left corner. You need Pro ($20/month minimum), Max, Team, or Enterprise. If you're on the free plan, upgrade at claude.ai before continuing.
Git is a free tool used to download files from the internet. Go to git-scm.com/downloads, click your operating system, download the installer, and run it with all the default settings (just keep clicking Next).
git --version and press Enter. If you see a version number, you already have it, so skip to step 4.Node.js is a free tool that lets you install programs from the command line. Go to nodejs.org, click the big green button that says "Recommended For Most Users" (the LTS version), download it, and install with all defaults.
node --version and press Enter. If you see "v20.x.x" or similar, you're good.This is the text window you'll type commands into for the next few steps. It looks plain but it's just a faster way to give your computer instructions.
ECC (Everything Claude Code) is a free set of behavior rules and skills that dramatically improve how Claude works. Copy the command below, paste it into your Terminal window, and press Enter.
git clone https://github.com/affaan-m/everything-claude-code.gitRun these two lines one at a time in Terminal. The first creates a folder, the second copies the rules into it.
mkdir -p ~/.claude/rulescp -r everything-claude-code/rules/common ~/.claude/rules/Run these two lines one at a time in Command Prompt.
mkdir %USERPROFILE%\.claude\rulesxcopy /E /I everything-claude-code\rules\common %USERPROFILE%\.claude\rules\commonClaude Code is a separate tool from the desktop app. It runs inside Terminal and powers the multi-agent features. Paste the line below into Terminal and press Enter.
npm install -g @anthropic-ai/claude-codeIn Terminal, type the word below and press Enter. Claude Code opens inside your Terminal window. It shows a setup screen. Pick option 1 (Dark mode) and press Enter. It signs you in automatically through your browser.
claudeWhile inside Claude Code (after the welcome screen), paste this and press Enter. This connects Claude Code to the ECC library of skills and agents.
/plugin marketplace add https://github.com/affaan-m/everything-claude-codeclaude first (step 10), then try again.Still inside Claude Code, paste this and press Enter. This installs 43 skills, 13 agents, and 31 commands: the full toolkit that makes everything else in this guide possible.
/plugin install ecc@eccThis checks whether everything is configured correctly and gives you a personalized report showing what's working and what still needs attention. Read through the results. It's your starting point.
/harness-auditDownload the Claude app from the App Store (iPhone) or Google Play (Android). Sign in with the same account you use on desktop. This lets you assign tasks from anywhere while Claude works on your desktop in the background.
This is the single most impactful thing in this entire guide. Every agent reads this file before doing anything. Think of it as an employee onboarding document. The more detail you put in, the better every agent performs. Run this command, then open the file it creates and fill it in.
/initIn Terminal, type cd followed by the path to your project folder. Replace the example below with the actual path on your computer.
cd /path/to/your/projectcd ~/projects/my-appcd C:\Users\YourName\projects\my-appThis creates a project-specific briefing file inside your project folder. Fill it with: what the project is, what stage it's at, the tech stack, what's currently in progress, and the next milestone. When agents work on this project, they read both your global file and this one.
/initThis reads your actual project files and builds a skill set Claude uses automatically. Without this, Claude is a smart generalist. With this, it's a specialist in your specific codebase. Run it inside each project folder.
/skill-create --instinctsOpen Claude Desktop → click Cowork at the top → click New Project. Create one project for each area you work on. Point each project to the right folder on your computer. Add a brief description of what that workspace is for. Memory and context persist between sessions within each project.
In Cowork, click the gear icon → Connectors → add Google Drive and Gmail. Claude can now pull context from your real files and emails without you manually uploading anything.
With your desktop app open, send a task from the Claude app on your phone. Claude picks it up and works on it while you're away. Test it immediately so you understand how it works.
Hooks make agents self-managing. They trigger automatically. When a session starts, Claude loads context on its own; when it ends, it writes a summary automatically; before anything ships, a quality check runs automatically. Run this and follow the prompts.
/hook-creatorIn Claude Code, turn on Auto accept edits at the bottom of the chat. This lets agents move without waiting for your approval on every single change. Immediately pair it with a quality gate so nothing problematic slips through.
/quality-gate . --strictThis is the moment everything comes alive. Give Claude a real goal, not a test. It breaks the goal into sub-tasks and runs multiple agents in parallel. You come back to a finished output instead of a conversation thread. Replace the example below with something real for your project.
/ecc:plan "audit the [your project] folder, identify what's incomplete or broken, and produce the top 3 priorities with reasoning" --parallel--parallel flag runs multiple agents simultaneously on different parts of the goal. Watch the output. This is your team operating for the first time.In the Cowork desktop app → Scheduled → create a new task set to run every morning. Claude checks all your projects, summarizes what changed, flags what's blocked, and lists today's priorities, before you open your laptop.
As you use Claude, it observes patterns in your work. These two commands let you see what it's learned and lock those observations into permanent skills that improve every future session.
/instinct-status/evolve/evolve once a week. Add it to a Friday routine: run /instinct-status to see what's been learned, then /evolve to lock it in. Your system gets measurably better at your specific work over time.You're set up. Start each morning with this, replacing the project name with whatever you're focused on. It orients all agents toward today's highest-leverage work based on the actual current state of your project.
/ecc:plan "what are the 3 most important things to accomplish on [your project] today based on current state and priorities" --parallelThese are the actions that compound the most. Do them in order. Each one builds on the last.
/init, and fill it out: what the project is, what stage it's at, the tech stack, what's currently broken, and what the next milestone is./hook-creator. Set up at minimum: session start hook, session end hook, and pre-commit quality gate. Takes 10 minutes and saves hours every week./quality-gate . --strict in the same session./evolve converts those observations into permanent skills. Your system gets measurably better every week./instinct-status to see what's been learned, then /evolve to lock it in.