How to give Claude memory with projects
Stop re-explaining yourself every session.
Hey friends 👋
Welcome to another Signal Pro workflow issue.
Two issues ago, we set up context files. Last week, we built Skills. Both made Claude smarter. Yet neither fixed its biggest flaw.
Every Cowork session starts from zero.
You re-upload the same files, re-explain the project scope, re-state what you’ve already covered. Claude is brilliant in the moment, but it has no memory of what happened yesterday.
Projects fix this.
Today I’ll show you:
Why Projects exist and what makes them different from Skills
What’s inside a Cowork project: instructions, context, and scoped memory
Chat projects vs Cowork projects and when to use which
How to create a project (with a full walkthrough of importing from Chat)
How Projects stack with the context files and Skills you’ve already built
Five project templates you can copy for the most common knowledge work setups
Skills vs Projects: the decision that changes your output
In last week’s issue, I laid out where Projects sit in the Claude stack:
“Who am I?” → Context files (about-me.md, voice-and-style.md, working-rules.md)
“How should Claude talk to me?” → Global instructions
“How should this specific task be done?” → Skills
“What am I working on?” → Projects
When you're staring at a new workflow, wondering how to set it up, the line between a Skill and a Project can blur quite quickly.
Here’s how I think about it now.
A Skill is a recipe. It tells Claude how to make that branded slide deck, structure a client email, or correctly format a weekly report. The output is predictable, and the process is the same every time. You write the instructions once and never think about them again.
A Project is a workspace. It holds everything related to an ongoing piece of work. This contains reference documents, previous outputs, specific instructions, and crucially, memory of what’s happened so far. The work evolves, and the context accumulates. Each session builds on the last.
The test is simple. Ask yourself:
“Does this work need memory of what came before?”
If the answer is no, it’s a Skill. If the answer is yes, it’s a Project.
I’m currently working through a consulting engagement. I’ve got transcripts, book excerpts, PDFs. The kind of work where I ask Claude to pull themes from a call transcript I had on Tuesday, then on Thursday, I need it to cross-reference those themes against new material the client shares. That’s a Project. Claude needs to remember what we found last time.
But when I create a branded slide deck from that research? That’s my branded-deck Skill kicking in. It applies the same formatting rules every time, regardless of the content.
The Project provides the what (these specific transcripts and accumulated understanding). The Skill provides the how (format it this way, structure it that way).
They work together. Skills are the tools in the workshop. Projects are the workshop itself.
What’s inside a Cowork project
A project bundles four things into a single workspace.
Instructions tell Claude how to behave for every task in this project. These sit on top of your global instructions and context files. For my consulting project, the instructions specify the analytical framework to use, the preferred terminology, and how to handle contradictions between sources.
Context is the reference material. You can see in the right-hand panel that the project references both a local folder on my computer and the linked project from Chat. Claude reads from both whenever it needs to ground its analysis.
Scheduled tasks let you automate recurring work within the project scope. A Monday morning summary of what’s been analysed so far. A weekly digest of new materials added to the folder. These run automatically while you’re doing other things.
Memory is what makes Projects different from everything else. Claude remembers context from tasks you’ve run inside the project and applies it to future tasks. You can see “Memory” in the Context panel. Memories are stored here over time as you work within the project, which are scoped directly to the project, so nothing leaks.
And all of this lives on your computer. Here’s the actual folder Claude created when I set up this project:
Chat projects vs Cowork projects
If you already use projects in Claude’s web app, you’re probably wondering what the difference is.
Chat projects are for conversation. You upload reference material, ask questions, brainstorm, iterate on ideas. If Claude generates a document, you download it from the chat and save it yourself.
Cowork projects are for execution. You describe an outcome, Claude works through it autonomously, and the finished files land directly in your folder. Cowork projects also have scheduled tasks, local file system access, and scoped memory that’s paired with the ability to act on what it remembers.
It’s not either/or. You can start in Claude chat for early-stage research, then import into Cowork when you’re ready to produce deliverables. The chat project is the thinking. The Cowork project is the doing.
Three ways to create a project
Open Claude Desktop and make sure you’re on the Cowork tab.
Find Projects in the left sidebar. Click it to open the Projects page.
Click + New project in the top right. You’ll see three options:
Start from scratch is for brand new work. You create a new folder, write instructions, and add any files you want Claude to reference. Use this when you’re beginning something that doesn’t have existing materials yet.
Import a project is the chat-to-Cowork path I just described. You’ll see a list of all your existing Chat projects. Select one, and your files and instructions transfer into a new Cowork project.
Use an existing folder wraps a Cowork project around a folder already on your computer. This is the most common option for people who followed the Setup guide. You already have a working folder with context files. Now you add project-scoped instructions, memory, and scheduled tasks on top.
For today’s example, we’re going to import an existing project I’ve created from Claude chat.
Here’s what it looks like when I import my consulting project. You choose the project, name it, and pick a save location on your computer. Notice “Memory is on” at the bottom. Scoped memory is enabled by default.
Click Create, and the project is ready via one-click import.
One important detail: the folder holding your context can be different from the folder where outputs are saved. This is useful when you have a shared reference folder that multiple projects draw from, but each project saves deliverables to a separate location.
How Projects stack with your existing setup
If you followed the Setup and Skills issues, you already have three layers working: context files, global instructions, and custom Skills. Projects add the fourth layer. They all compound.
When you run a task inside a Cowork project, Claude reads everything in this order:
Your global instructions (Settings > Cowork)
Your context files (about-me.md, voice-and-style.md, working-rules.md)
Your project instructions and context files
Any relevant Skills that match the task
Your actual prompt
Each layer adds specificity without replacing the others. Five layers of context working together in a single task. And the project memory means that next time you ask Claude, it will remember the feedback you gave in your previous Cowork session.
Here are the project templates I use across my most common workflows.
Customise these to your role, and you’ll immediately have a workspace that compounds with every session.
Five project templates you can steal
Each template includes the instructions and folder structure you need. Copy them, adjust the specifics for your role, and you’ll have a working project in under 5 minutes.