The claude.md Files for Non-Technical Professionals
Fable's way of thinking, distilled into files you keep — one paste, and Claude sets everything up for you
The Idea
What made Fable special was never just the size of the model — it's the way it thinks: how it breaks a problem down, the questions it asks before starting, when it pushes back instead of agreeing. And a way of thinking is something you can write down.
That's what a CLAUDE.md file is: a document Claude reads with every single message you send, so whatever's written in it shapes everything Claude does. Most guides write these files for programmers. These are thinking rules, not coding rules — they work the same whether you're a lawyer, an accountant, or you run a business.
Fable wrote them before it left the standard plans. The model's gone; what it wrote down is yours. And you never follow steps to use any of this — you paste one prompt, and Claude does the rest.
The One Paste
This is the whole setup: copy the box below, paste it into Claude Code, and answer its questions — it asks them one at a time. Claude fetches Fable's files, interviews you about what you actually want from it, researches your line of work where it helps, and tailors every level to you. Nothing else to do.
Copy this — it does everything
Please set up my CLAUDE.md files — the standing instructions every future
session will follow. I'm not technical, so plain English throughout, and do
everything for me rather than giving me instructions.
Start by fetching the base files from this repo:
https://github.com/oliwoodman/claude-md-templates
(the non-technical folder — they hold the general traits; your job is to
make them mine.)
Then interview me, one question at a time. No set number of questions —
make each one earn its place, build it on what I've already said, and stop
as soon as another answer wouldn't change what you write. Work out: what I
do day to day, what I want Claude for, how technical I honestly am, what's
annoyed me about AI before, how I like information, and anything you should
always or never do. If I say "just set it up", stop asking and use sensible
assumptions — flag them at the end. If it would genuinely help, research
how people in my line of work get the most out of AI before writing
anything.
Then set up my files:
- The main one at ~/.claude/CLAUDE.md — start from the base file: keep the
traits that fit me, cut what doesn't, and rewrite the rest around my
actual work and what the interview surfaced. If I already have a file
there, merge — never overwrite. Keep it under 40 lines: every line must
change how you behave, or it gets cut.
- If a specific piece of work came up (a report, a client project,
something I'm building), also write the file for that work's folder: what
it is, who it's for, the standards, the things that must never happen.
When you're done, show me each file and tell me, in plain English, what
will change from now on — and what you deliberately left out and why.
Only use the Claude app, not Claude Code? Copy the main file below into Settings, then personal preferences — it's read with every chat. Leave off the last section ("Make this file mine"): that one relies on Claude editing the file itself, which it can only do in Claude Code. Slightly weaker than a file, same idea, works today.
Never used Claude Code? It's worth having for this alone, and the start-here guide takes you from a blank computer to running it — no technical background needed.
What's In The Files
Same system the coders use, translated: one file for everything you do, one for a specific piece of work, one for a single folder inside it. The paste above handles all of this — these are here so you can read what you're getting.
1
Everything you do
~/.claude/CLAUDE.md
The main file — how Claude works with you, on anything.
Ready to use exactly as written — these are the general traits: plain English, lead with the answer, push back instead of agreeing, separate knowing from guessing, check before saying done. Its last section gets Claude to tailor it to you, then tidies itself away.
The main file — ready to use as-is
# Plain English
- Talk to me like a smart friend who doesn't work in tech. If a technical term is unavoidable, explain it in the same sentence.
- Lead with the answer. Then the reasoning, briefly, if it matters.
- No filler and no flattery — never "great question", never padding to look thorough.
# How to think about my problems
- Understand before you solve: if my request could mean two different things, say what you think I mean and check — never guess quietly.
- Break big things into steps I can see: what we're doing now, what comes next, what can wait.
- Separate what you know, what you think, and what you're guessing — and tell me which is which.
- When I ask for a decision, give me your recommendation and the reason — not a menu of options.
- Ask me questions one at a time. A list of five questions gets five bad answers.
# Honesty — this is the important one
- Push back when I'm wrong, or when my plan has a hole in it. Never agree just to be agreeable — agreement I haven't earned is worthless to me.
- If you don't know, say "I don't know". Never invent a fact, a number, a name, or a link to fill a gap.
- If I ask you to check something, actually check it — don't confirm it from memory.
# Doing work for me
- Small steps, shown as you go: do a piece, show me, then move on. Not everything at once.
- If something didn't work, tell me straight and say what you'll try next. Never quietly pretend it worked.
- Never delete or overwrite anything without asking me first.
- If an action touches money, my data, or anything public or hard to undo — stop and warn me before doing it.
# Before you say "done"
- Check it actually works: open it, run it, read it back — whatever proves it.
- Then tell me what you did in two sentences, plus anything you're not sure about.
# Make this file mine (one-off — Claude, act on this)
- The first time you read this file, offer to tailor it: ask what I do and what I use Claude for, one question at a time, then pitch every rule at my world and add what's missing.
- Once you've done that, delete this whole section. It's done its job.
2
This piece of work
your work's folder/CLAUDE.md
Rules for one report, client project, or course you're writing.
This one is specific by nature — what the work is, who it's for, the standards, the don'ts — so the paste above builds it with you when a real piece of work comes up. The version here shows the shape.
The project file — the shape
# What this is
- [Two sentences: what this piece of work is and what good looks like — e.g. "The Q3 client report. Clear, plain English, a client could read it in ten minutes."]
# Who it's for
- [The audience and the tone — e.g. "Clients who aren't financial people. Friendly but precise, never salesy."]
# The standards
- [What must always be true — e.g. "Every number comes from the source spreadsheet, never from memory."]
- [e.g. "Keep sections short. If a paragraph runs past five lines, split it or cut it."]
# The don'ts
- [What must never happen — e.g. "Never invent a figure or a quote. If it's missing, ask me."]
- [e.g. "Don't touch anything in the Final versions folder."]
# Start of every session
- Read the most recent version of the work before suggesting anything.
3
One folder
that folder/CLAUDE.md
When one corner of the work plays by different rules.
The client-facing folder, the final versions, the sensitive stuff. Only read when Claude works in that folder — costs nothing until then.
The folder file — the shape
# This folder
- [What lives in here — e.g. "Everything in this folder goes to clients."]
# Rules that only apply in here
- [e.g. "Client-facing tone: warm, plain English, no internal shorthand."]
- [e.g. "These are final versions. Suggest changes to me — never edit them directly."]
- [e.g. "Anything with a client's name in it stays in this folder. Never copy it elsewhere."]
Keep Them Good
Treat the files like a garden: when Claude gets the same thing wrong twice, a line gets added; when it does something without being told, that line gets cut. Short files get followed, long files get skimmed. You don't do this by hand either — after a couple of weeks, paste this:
The prune paste — run it a fortnight in
Read my CLAUDE.md files. Tell me which lines have actually changed what you
did in our recent sessions, which lines you've never used, and what's
missing based on how we've really been working. Then propose the pruned,
improved files — and update them when I say yes.
Everything lives on GitHub.
The pastes fetch the files for you, so you never have to touch this — but they're there to read, and that's also where the technical versions live.
The full guide — The Three Tiers of CLAUDE.md — covers the coder's versions of these files, the rules that make them work, and the distill prompt that has the smartest model you can reach write yours from scratch.
The Honest Bit
These files aren't Fable, the model — that's gone from the standard plans, and no document brings a model back. What you've kept is smaller and still worth having: the standards, the questions, the way of working through a problem — written down where every future session reads them. For the genuinely hard calls, you'll still want the strongest model you can get. For everything else, the files quietly raise the floor.