Give Claude a Memory
How to Give Claude a Memory (In About 15 Minutes)
Every time I open Claude, it starts with a blank slate.
No memory of who I am. No idea what I’m working on. No knowledge of decisions I’ve already made, tools I use, or how I like to communicate. So I re-explain. It guesses. I correct it. By the time we’re actually working, five minutes have gone by and I’ve done the cognitive equivalent of briefing a new contractor — every single session.
That’s not Claude being bad. That’s just how it works by default. But a few months ago I realized it doesn’t have to work that way. There’s a fix, and it takes about 15 minutes to set up.
Already set up? The one habit that changes everything: just say save at the end of each session. Claude writes a summary, updates your context, and hands you a ready-to-paste note for next time. That’s it. Skip to what changes after a week →
What You’re Building
A plain-text file called CLAUDE.md that lives in a folder on your computer. When you use Claude’s desktop app (Cowork mode), it reads that file automatically at the start of every session. Before you type a single word, Claude already knows who you are, what you’re working on, and how you like to communicate.
That’s it. One file. Permanent context. No re-explaining.
I built a free starter kit around this idea — templates, a save workflow, a couple of behavioral scripts that make Claude smarter about how you work. The whole thing lives on GitHub and takes about 15 minutes to get running.
What this adds that a plain Claude session doesn’t
- Save & resume — Say “save” at the end of any session. Claude writes a summary, updates your context file, and hands you a paste-ready note for starting the next session exactly where you left off. No re-explaining. No lost progress.
- Project context — Each project gets its own context file. When you mention a project, Claude loads the right details automatically — goals, stakeholders, status, decisions already made. The right context, without you having to dig for it.
- New project wizard — Say “new project” and Claude interviews you, creates the folder, and writes the context file. Setup that used to take 20 minutes of file wrangling takes two minutes of conversation.
- Catch me up — When a long session drifts or you’ve been away, say this. Claude reads your files, tells you where things stand, and waits for your go-ahead before doing anything. Your recovery move when things go sideways.
- Context compaction survival — Every Claude session has a memory limit. When it fills up, a plain session loses its thread. This system keeps working: your context lives in files Claude can always re-read, not just in the conversation window.
- Compound improvement — A plain session knows only what you told it today. This system knows what you told it last week, last month, and last quarter. The longer you use it, the less friction there is.
Want the full picture of what a mature version of this looks like? Read Why I Treat My AI Context Like Infrastructure — the architecture behind the system, why plain text beats everything else, and why the timing matters. Or start with the A Week in the Life walkthrough on GitHub.
Getting Set Up
Go to github.com/johndelay/claude-context-starter-kit.
You don’t need to know anything about GitHub. Click the green Code button at the top of the page, then click Download ZIP. Unzip the file somewhere you’ll find it — your Desktop is fine for now.
Inside you’ll find a handful of files. The one that matters most right now is CLAUDE.md. That’s your context file — the brain you’re about to build.
Open CLAUDE.md in any text editor. (TextEdit on Mac, Notepad on Windows, whatever you use.) You’ll see sections with [bracket placeholders]. Don’t try to fill in everything — that’s a trap. Focus on three sections:
- About Me — Your name, your role, and two or three sentences about your day-to-day. A manager writing “I oversee vendor contracts and operational risk for a mid-size financial firm. A lot of my job is translating between specialists and executives” gives Claude enough to work with immediately.
- Current Projects — Your top two or three active projects. One sentence each: what it is, what done looks like. Don’t overthink it. You can update this any time.
- Communication Preferences — How you like Claude to respond. Concise or detailed? Direct or diplomatic? Documents or chat-length answers? This is the section most people skip and then wonder why Claude’s tone feels off.
// Not sure what “good” looks like? Open CLAUDE-example.md in the kit — it’s a fully filled-in example. Start there.
// Or ask Claude directly: “Can you help me fill in my CLAUDE.md? Interview me for the key sections.” Claude will ask you the right questions and draft the entries for you — no staring at a blank file.
Open the Claude desktop app. If you don’t have it yet, download it from claude.ai — it’s free to install.
Inside the app, click into Cowork mode. You’ll see an option to select a workspace folder. Select the folder where you just saved CLAUDE.md.
That’s the connection. Cowork reads everything inside that folder at the start of every new session. Your context file is now live.
Start a new session. Before you ask Claude to do anything, type this:
The test prompt
- “What do you know about me and what I’m working on?”
If Claude comes back with your name, your role, and at least one of your current projects — you’re set. That’s the baseline.
If it comes back generic, the most likely cause is [bracket placeholders] still in your file, or CLAUDE.md saved in the wrong folder. The SETUP_GUIDE.md in the kit has a one-minute health check that will tell you exactly what’s missing — or you can paste this prompt directly into Claude:
Run the health check prompt
Paste this into a new session if your test didn’t work:
“Please run a workspace health check. Read my CLAUDE.md and tell me: (1) Is it filled in with my real information? (2) Are the skills folder and Standing Instructions #12 and #13 present? (3) Do my project pointers have matching project context files? Tell me what’s working and what needs attention.”
Claude will read your files and tell you exactly what’s missing or misconfigured. Most setup problems are visible in under 30 seconds this way.
The kit includes a skills/ folder with two behavioral scripts. They add two commands you’ll use constantly once you’re past Day 1:
- “New project” — Claude walks you through setting up a project folder conversationally. Asks about your goal, your stakeholders, your deadline — then creates everything automatically. No manual file wrangling.
- “Catch me up” — When a long session starts feeling off, say this. Claude reads your context files, tells you the current state, and waits for your confirmation before doing anything. Your recovery move when things go sideways.
To activate: copy the skills/ folder into your workspace folder alongside CLAUDE.md. Then add two lines to your Standing Instructions — the exact text is in skills/README.md. Takes about two minutes.
The exact Standing Instructions text
Add these two lines to the Standing Instructions section of your CLAUDE.md:
12. When the user says “new project,” run the new project wizard from skills/new-project.md.
13. When the user says “catch me up,” run the catch-me-up protocol from skills/catch-me-up.md.
The full text is also in skills/README.md inside the kit — just copy and paste.
// Or just ask Claude: “Can you add Standing Instructions 12 and 13 to my CLAUDE.md?” Claude will read the skills README and add the activation lines for you. One message, done.
Claude can help with any of these steps
- “Can you help me fill in my CLAUDE.md? Interview me for the key sections.” — Claude walks you through each section conversationally and drafts the entries. Works great if you’re not sure what to write.
- “Please run a workspace health check.” — Claude reads your files and tells you exactly what’s missing or misconfigured. The fastest way to debug a setup that isn’t working.
- “Can you add Standing Instructions 12 and 13 to my CLAUDE.md?” — Claude reads the skills README and adds the activation lines for you. No manual editing.
- “New project.” — Once skills are active, Claude sets up your first project folder end-to-end: asks about your goal, creates the files, and loads the context. No manual file wrangling.
- “Can you walk me through setting up this workspace?” — If you’re not sure where to start, just ask. Claude will guide you through the whole thing step by step.
What Changes After a Week
The five-minute re-explaining tax at the start of every session disappears.
Claude knows you. When you mention a project, it loads the context automatically. When you ask it to draft something for your manager, it already knows how she likes to receive information. When you come back after three days off, it picks up where you left off.
That’s not magic — it’s a well-maintained context file and a save habit. At the end of each session, say “save” and Claude writes a session summary, updates your context file, and hands you a paste-ready message for starting the next session right where you left off. The system compounds. The longer you use it, the less friction there is.
Want to see the whole picture? I wrote a narrative walkthrough that shows the system end-to-end — what day one actually looks like, how project folders work, what happens when context compacts mid-session. A Week in the Life →
The Resources
- Starter Kit — github.com/johndelay/claude-context-starter-kit — everything you need, free, no account required
- User Guide — A Week in the Life — narrative walkthrough of the system working end-to-end
- Setup Guide —
SETUP_GUIDE.mdinside the kit — first-time checklist and health check if something isn’t working
Fifteen minutes to set up. The payoff runs indefinitely.