Foundation Guide

The Shape of a
Personal OS for Claude

A folder of markdown files that turns Claude from a chatbot into a coworker who knows your work.

Jump to Setup

What a Personal OS Actually Is

A folder on your computer. Inside it, markdown files for your projects, your notes, your tasks. One special file at the root called CLAUDE.md that tells Claude how to behave when you open this folder.

That's the whole system. The leverage comes from the fact that Claude reads your CLAUDE.md every time you start a session, so you stop re-explaining who you are and what you're working on. Claude already knows.

What You'll Need

Claude desktop app

Download the desktop app and use Cowork. The desktop app can read and write files on your computer; the website cannot.
Download here

Claude Pro or Max

The subscription that powers Cowork.
Subscribe here

Why the desktop app, not the website

Open Claude's desktop app and go to Cowork. Cowork can see and write the files in your Personal OS folder — that local file access is the whole point. Claude on the website works in the cloud, which is handy for picking up a conversation from your phone, and it stays walled off from your files. Start in the desktop app.

Important: Keep Sensitive Information Out

Claude can read any file in your Personal OS folder. Do not store:

  • Passwords, API keys, or credentials
  • Social Security numbers or government IDs
  • Financial account numbers
  • Medical records or health information
  • Confidential work documents your employer hasn't approved for AI use
  • Other people's personal information without consent

If a sensitive project needs context, write a note describing the project without the sensitive data inside.

Why Structure Matters

Without it, every conversation with Claude starts from scratch. You re-explain your role, your projects, your preferences. Claude gives generic advice because it has no idea what you're actually doing.

With a Personal OS, the structure gives Claude memory:

Claude Reads Your Files

The CLAUDE.md at the root loads automatically at the start of every session. Claude knows your role, your active projects, and your rules before you type a word.

Nested Context

Each project can have its own CLAUDE.md. When you navigate into that folder, Claude inherits the project-specific instructions on top of the global ones.

Tasks That Travel

Tasks live as markdown files with structured metadata. Ask Claude to build your agenda and it scans the whole vault, pulling everything due into one daily list.

The System Grows

Every time you correct Claude or teach it something, that lesson gets saved. Next week, Claude already knows. The system compounds.

The Folder Layout

The vault uses a numbered folder system based on the PARA method (Projects, Areas, Resources, Archives). Numbers keep things in a stable order regardless of which file browser you use.

Personal_OS/
│
├── 00_Inbox/             # Daily agendas, quick captures, unsorted
│
├── 10_Projects/          # Active work with deadlines or outcomes
│
├── 20_Areas/             # Ongoing responsibilities (no end date)
│   ├── Calls/            # Meeting notes
│   ├── People/           # Contact profiles
│   └── Companies/        # Org wiki entries
│
├── 30_Resources/         # Reference materials, research, templates
│
├── 40_Archives/          # Completed projects, old work
│
├── 50_Personal/          # Personal projects, kept separate from work
│
├── 99_System/            # Configuration, profiles, scripts
│   │
│   ├── Context_Library/
│   │   ├── business_profile.md
│   │   └── project_specs/
│   │
│   └── Scripts/
│
├── Apps/                 # Software projects (peer of PARA, not nested)
│
└── CLAUDE.md             # The brain — Claude reads this first
Why numbered folders? They always sort in the same order. 00 comes first (your inbox, where you capture things fast), 99 comes last (system files you rarely touch). PARA gives you a stable mental model: Projects have deadlines, Areas are ongoing, Resources are reference material, Archives are done.
Why Apps/ as a peer of PARA? If you build software inside your vault — websites, scripts, small tools — they need their own top-level folder. They don't fit cleanly into Projects (which assume markdown notes) and they shouldn't be buried in Resources. Apps/ keeps them visible and easy to deploy.

CLAUDE.md: The File Claude Reads First

This is the most important file in your system. Claude reads it automatically when you start a session inside this folder. It tells Claude who it is in your world, where things live, and how to behave.

## Role

Chief of Staff for Personal Operating System.
Manage tasks, organize information, facilitate research
with precision and low friction.


## Who I Am

[Your name] — [your title] at [your organization].
[One sentence of context: industry, focus, what you care about.]


## Core Behavior

- When I ask for a change, ALWAYS edit the file directly.
  Never just show me the fix.
- Keep changes minimal — touch only what's necessary.
- Read local files before asking for context.
- Confirm before deleting; edit freely if aligned.
- Concise responses — no lectures.
- When setting something new up, interview me one
  question at a time instead of guessing.


## Architecture

PARA-based vault:
- `00_Inbox/` capture and triage
- `10_Projects/` active projects (each can have its own CLAUDE.md)
- `20_Areas/` ongoing areas (Calls, People, Companies)
- `40_Archives/` — never search or modify unless asked


## Task Schema

```yaml
type: task
status: [todo, in-progress, done, waiting]
created: YYYY-MM-DD
due_date: YYYY-MM-DD
priority: [high, medium, low]
tags: []
```
Why this works: Claude reads this once at the start of the session and behaves accordingly for the entire conversation. The task schema is especially load-bearing — every task Claude creates uses these exact field names, so when you ask Claude to build your agenda, it can scan for them reliably.

Nested CLAUDE.md Files

Each project can have its own CLAUDE.md. Claude reads them hierarchically — the root one always, then any project-specific one when you're working in that folder.

Personal_OS/
│
├── CLAUDE.md                        # Always loads
│
└── 10_Projects/
    │
    └── Website_Redesign/
        │
        ├── CLAUDE.md                # Loads when working in this folder
        └── project_brief.md

The project-level file is where you put context that only matters for this project:

  • This is a React project — use TypeScript
  • The client prefers minimalist design, no gradients
  • Always run tests before declaring done
  • Design files are in Figma at [link]
  • Deployment is GitHub Actions to AWS
The pattern that matters: Global behavior at the root, project-specific behavior in the project. Don't put project details in the root CLAUDE.md — they pollute every session. Don't put global rules in project CLAUDE.md files — they only apply when you're in that folder.

Help Claude Find Things Fast

In Cowork, Claude answers by reading your files. Ask it about one person and, with no help, it may open half your vault looking for them — slow, and it crowds out room for the real work. Two habits keep it fast: give the things you look up most a consistent shape, and keep a short index Claude can read first.

Give People and Companies a Shape

Tasks already have a shape — the Task Schema in your CLAUDE.md. Do the same for the two things you'll look up most: the people you work with, and the organizations they belong to. The Company file is the hub; each person links into it.

--- People file: 20_Areas/People/Jane-Smith.md ---
type: person
name: Jane Smith
company: "[[Acme-Co]]"     # links to the company hub
role: VP Partnerships
last_contact: 2026-06-20

--- Company file: 20_Areas/Companies/Acme-Co.md ---
type: company
name: Acme Co
relationship: partner
# a Key People table lists everyone who links here
Why the company is the hub: people change jobs; organizations stay. When someone moves, you update one company: line and their history stays intact. When someone new joins a partner you already track, they inherit everything you already know about that org.

Keep an Index So Claude Reads One File, Not a Hundred

An index is a plain list — one line per item, with a link to the full file. Claude reads the index, finds the row it needs, and opens just that one file. Keep one for each thing you accumulate: people, companies, recurring topics.

--- 20_Areas/People/People_Index.md ---
| Name           | Company | Role            | Last contact |
|----------------|---------|-----------------|--------------|
| [[Jane-Smith]] | Acme Co | VP Partnerships | 2026-06-20   |
| [[Raj-Patel]]  | Globex  | CMO             | 2026-06-12   |
Make it Claude's job, not yours. Add a rule to your CLAUDE.md: whenever you add or update a Person or Company, update its index in the same edit. The index stays current on its own, and "who do I know at Acme?" gets answered from one short file instead of a vault-wide search.

Day-1 Setup

Claude does most of the work. You download one app, create one empty folder, and let Claude interview you to build the rest.

1

Download the Claude Desktop App

Get it from claude.ai/download and sign in. Open it and go to Cowork. The website works in the cloud and stays walled off from your files; the desktop app can read and write them.

2

Create Your Personal_OS Folder

One empty folder. This is the only thing you make by hand.

On Mac:

Finder → Documents → Right-click → New Folder → name it Personal_OS

On Windows:

File Explorer → Documents → Right-click → New → Folder → name it Personal_OS

3

Point Cowork at the Folder

In Cowork, start a new task and choose Work in a projectChoose different folder → select your Personal_OS folder.

Set the model to Opus 4.8 before you start typing. You can't switch models once a conversation is going.

4

Connect Your Tools

Click your name (lower left) → Settings → Connectors, and turn on the ones you use — Granola, Outlook, Box, Wrike. Set them to "always allow" so Claude isn't stopping to ask permission on every step.

5

Have Claude Interview You and Build the Structure

Paste this prompt. Claude interviews you first, then writes everything from your answers — so the CLAUDE.md reflects how you actually work instead of a blank template.

Copy and paste into Cowork:

Please set up a Personal OS for me, and interview me to get it right.

First, interview me ONE question at a time, waiting for each answer
before asking the next. Don't dump all the questions at once. Cover:
  - My name, title, and organization
  - What I do day to day and who I work with most
  - How I want you to behave: concise or more explanatory, tone
  - The 2-3 projects or responsibilities that matter most right now
Tell me I can answer by voice if that's faster.

Once I've answered, create these folders in my Personal_OS folder:
  00_Inbox, 10_Projects, 20_Areas, 20_Areas/Calls,
  20_Areas/People, 20_Areas/Companies, 30_Resources,
  40_Archives, 50_Personal, 99_System/Context_Library, Apps

Then create a CLAUDE.md at the root, built from my answers, with:
  - Role: you are my Chief of Staff for this vault
  - Who I Am: my name, title, organization, and context
  - Core Behavior: edit files directly, keep changes minimal,
    read local files before asking, confirm before deleting,
    concise responses, AND whenever you set something new up,
    interview me one question at a time rather than guessing
  - Architecture: the folders above
  - Task Schema: YAML with type, status, created, due_date,
    priority, tags

Also create a Someday_Maybe.md in 20_Areas/ with sections for Ideas,
Long-Term Projects, Things to Explore, Wishlist.

Create everything once I've answered.

Claude asks its questions one at a time — answer by voice with the microphone if that's faster — then creates all the folders and your CLAUDE.md.

6

How You Run Workflows in Cowork

In Cowork you run everything by asking in plain words. There's no skills feature or slash-command setup — the plain-language request is the interface. Say "build my agenda for today by scanning my vault for tasks with due dates," and Claude does it.

For a workflow you'll repeat, write the instructions down once and keep them in your vault — a markdown file in 99_System/Prompts/. Next time, point Claude at it: "run my agenda workflow in 99_System/Prompts/agenda.md." The file is the reusable part; the plain-language ask is how you trigger it.

Save this as 99_System/Prompts/agenda.md:

# Build my agenda

Scan all .md files in 10_Projects/ and 00_Inbox/ for YAML
frontmatter with a due_date field.

1. Group tasks: Overdue, Due Today, Due This Week, Due Next Week
2. Create a new file in 00_Inbox/ named YYYY-MM-DD_Agenda.md
3. Format with a section for each group, plus Daily Focus and Quick Notes
4. Tell me: "Your agenda is ready: [filename]"
Where the auto-firing version lives: Typed /slash commands that fire on their own — a real skills engine with SKILL.md files — are a Claude Code feature. In Cowork, a prompt file plus a plain-language ask does the same job, with one extra sentence from you.
7

Test It

Ask: "Build my agenda for today." Claude scans your vault for tasks with due dates and writes a grouped agenda into 00_Inbox/. To read any file formatted instead of raw, ask Claude to walk you through it, or open it from the folder.

Your Daily Rhythm

Two routines are enough to anchor a day. Ask for them in plain words — or keep them as prompt files and point Claude at them. Here's what it looks like in practice:

Morning — Build the agenda

First thing, open Cowork on your Personal OS project and say "build my agenda for today." Claude scans the vault for tasks with due dates, builds a grouped agenda (overdue, today, this week, next week), and writes it to 00_Inbox/YYYY-MM-DD_Agenda.md.

Then: ask Claude to walk you through it, or open the file to read it.

End of day — Capture what moved

Before you close your laptop, say "run my shutdown." Claude reviews today's agenda, asks what moved and what didn't, and writes a one-paragraph summary at the bottom of today's agenda file, including tomorrow's top 3.

The next morning, the agenda build reads yesterday's shutdown notes and carries the queue forward.

Rules I'd Give Myself If Starting Over

  • Let Claude file things. Don't manually create folders or move files. Just say "create a new project for the Q1 marketing plan" or "file this transcript in the right place." Claude knows the structure.
  • Use the Inbox liberally. When you don't know where something goes, drop it in 00_Inbox/ and ask Claude to triage it later.
  • Don't preload context. Resist the urge to dump everything into the root CLAUDE.md. Claude reads what it needs. Profile files and project specs should load on demand, not automatically.
  • When something annoys you, write a rule. If Claude does something that makes you pause — explains instead of editing, creates a file in the wrong place, fabricates a stat — turn it into a one-line rule in CLAUDE.md. Two weeks of this and 90% of the friction disappears.
  • Archive aggressively. Done projects go to 40_Archives/. Keep 10_Projects/ small enough that you can scan it in one glance.
  • Ask Claude when you're stuck. "How should I organize my consulting clients?" or "Build me a template for kickoff meetings." The system bends to your needs, not the other way around.

That's the Foundation

A folder, a CLAUDE.md, a working daily rhythm. From here, the real leverage comes from how the system compounds — capture pipelines that feed the vault while you're in meetings, a voice system that lets Claude write like you, and a relationship wiki that builds itself.

Read How It Compounds →

Ready to work in an editor? The Claude Code build adds real slash commands, auto memory, and MCP servers.

Created by Janette Roush | Agents of Change

This guide is part of the AI Lab for Travel Innovation.