The One File That Stopped Me From Re-Explaining Myself Every Session
What a CLAUDE.md file actually does, and what building a good one forces you to realize about your own working style.
I noticed I was typing the same three sentences at the start of almost every session. No em dashes. No bullet points. Don't sound like a chatbot. Every single time, like Claude had never met me before, because in a sense it hadn't.
A CLAUDE.md file placed in a project's root gets read automatically at the start of every session. Persistent background context, no re-explaining. I'd known that fact for a while without actually building one, which is its own small lesson about the gap between knowing a tool exists and using it.
I sat down with Cowork and built a personalized one. It now knows the writing rules I'd been repeating manually: no em dashes, no bullet-point-heavy responses, no AI-sounding phrases. It knows I run the LLM Council for high-stakes decisions, so it doesn't need that explained from scratch every time it comes up. It knows my active build pipeline and the live deadlines sitting on top of it.
Here's the part I didn't expect going in. Writing down your own rules is a forcing function. I'd assumed I knew how I worked. Sitting down to actually write it out, in a form specific enough for another system to follow, surfaced rules I'd never consciously articulated. Things I just did, without a name for them, until I had to write the name down.
That's the real value of the file. Not just that it saves you retyping the same three sentences. It makes you notice which of your working habits were invisible even to you.
The scaffold
Here's a starting shape close to what I actually built. Copy it, delete what doesn't apply, fill in the rest.
# CLAUDE.md
## Writing rules
- [e.g., no em dashes, no bullet-heavy responses]
- [tone/style rules you keep repeating]
## Decision frameworks I use
- [e.g., "I run a 5-perspective council for high-stakes calls, ask before
assuming a quick answer"]
## Active context
- Current build(s): [what you're working on right now]
- Live deadlines: [anything time-sensitive]
## How I like to work
- [pacing preferences, how much detail you want by default, anything
about how you want Claude to check in or confirm before big steps]How to set it up
Open a blank file
Any plain text editor works.
Name it CLAUDE.md
Exactly that filename, in your project's root folder.
Write how you actually work
Not how you think you're supposed to work. The specific rules, the recurring context, the things you keep repeating at the start of every session.
Read it back once you're done
You'll probably find at least one rule in there you didn't know you had until you wrote it down.
It's free, works at every Claude plan level, and takes about 30 minutes to set up.
Go deeper
Teaching Claude Your Voice (Without Just Describing It) →
Why showing Claude real examples of how you talk beat trying to describe your own writing style.
Building an External Memory So Claude Stops Forgetting Everything on Reset →
A two-file system for picking up exactly where a session left off, instead of starting from zero.
Get new guides and builds in your inbox
No noise. New artifacts, skills, and lessons when they drop — and early access when agents launch.
Want a full structured path instead of one-off guides?
That's Learn AI with AI — a sequenced course that takes you from zero to building real things, with community and accountability built in.
Check out Learn AI with AI →