The Foundation · AI Taskforce Onboarding

Build Your
Own Claude OS

Your own AI teammate, set up for how you actually work. Four copy-paste messages. About 30 minutes. No tech skills required.

Onboarding · v1 · 8 July 2026

What This Is 02 / 12

01

What This Actually Is

Claude is an AI teammate that lives in a folder on your computer. Right now, it knows nothing about you or your job — that's normal, and it's what this walkthrough fixes.

You'll copy four messages into Claude, one at a time. Each one builds on the last. By the end, you'll have your own set of AI tools, shaped around your actual role — not a generic template.

Before You Start 03 / 12

Before You Start

1

Ask Keith for the green light

AI Taskforce access and permission to install Claude on your work machine. Don't install anything before he says yes.
2

Install Claude, sign in with your Foundation email

Download, install, sign in. Same as any other app.
3

Turn on Cowork mode, create a folder called "Claude Cowork"

This is the folder Claude will work from — your OS lives here.

Standard Kit 04 / 12

Standard Kit — Everyone Gets Three

01

Grammar Normaliser

Keeps everything you write sounding like The Foundation — professional, on-brand, no exceptions.

⇩ Download Skill

02

Date & Time

Never guesses today's date. Checks the real clock before touching deadlines or dates.

⇩ Download Skill

03

Frontend Slides

Builds decks and one-pagers — on brand automatically. This deck was made with it.

⇩ Download Skill

Non-negotiable — every OS in the building runs the same three. Download all three here, then drop them into Claude's Skills settings before you send Message 2.

How This Works 05 / 12

4

Messages. That's It.

Copy → paste → answer a few plain questions → Claude does the technical building. The prompts carry the detail so you don't have to.

Message 1

Tell Claude
About You

Message 2

Your Full
OS Structure

Message 3

Your Own
Agents

Message 4

Plug Into The
Real Knowledge Base

(later — not today)

Message 1 of 4 06 / 12

Tell Claude About You

Claude interviews you first — your role, your tools, what eats your time. Answer honestly, everything after this is built from your answers.

Ready to pastePROMPT1

“Ask me about my job title and role, what I actually do day-to-day, the tools I use…”

Hi! I'm building my own Claude OS for The Foundation, following the AI Taskforce onboarding. Before we build anything, I want you to get to know me first. Ask me about my job title and role, what I actually do day-to-day, the tools and software I already use, the kinds of documents and files I create or work with most, what eats up too much of my time, what I wish I could just hand off to someone, and whether I mostly work solo or alongside a team on shared work. Ask me one or two questions at a time, not a big list all at once. Once you've got a clear picture, summarise what you've learned back to me in plain English, and check I agree with it before we go any further.

Message 2 of 4 07 / 12

Your Full OS Structure

This is the real build — the same structure behind every Claude OS at The Foundation. The prompt carries all the technical detail; you just paste it.

CLAUDE_OS.md context-map.md permissions-model.md skill-register.md report-catalogue.md CHANGELOG.md agent-governance.md promotion-workflow.md

Ready to paste — the big onePROMPT2

“Build my personal Claude OS — constitution, permissions model, changelog, the works…”

Now build my personal Claude OS from what you've learned about me. Use this exact shape — adapt the specifics to my role, not the structure itself:

00_OS/ — the operating layer. Create:
- CLAUDE_OS.md: a short constitution covering what this system is for, how the layers fit together (knowledge base, OS, skills, agents, scheduled tasks, reports), and the standing rule that you are an operator and adviser, not a decision-maker — I decide, you check, draft, and recommend. Include a source-of-truth order: real evidence first, generated indexes second, OS files third, skill or agent output fourth, your own memory last and never trusted as fact on its own.
- context-map.md: a navigation map of every folder and what lives in it. If it ever disagrees with what's actually there, the real folder wins and this file gets regenerated, never edited from memory.
- permissions-model.md: three tiers. Tier 1 — mechanical, no approval needed. Tier 2 — needs my sign-off before it counts as final. Tier 3 — never without my explicit approval (deleting anything, sending or submitting anything externally, recording something as a confirmed outcome). Ask me for real examples from my own work so the tiers actually fit me, rather than guessing generic ones.
- skill-register.md: every skill installed, what it does, when to use it.
- report-catalogue.md: the standard reports this OS can produce for me, and where each one saves to.
- CHANGELOG.md: this OS layer's own append-only history. Never edit or delete a past entry, only add new ones.

01_Knowledge_Base/knowledge/ — my own knowledge base: CLAUDE.md (its operating rules — note format, ID scheme, folder map), raw/ (source documents — read-only, no agent ever writes here), wiki/ (the actual notes), outputs/ (finished things generated from the wiki), and its own separate CHANGELOG.md. Keep the two changelogs apart: OS changes log to 00_OS/CHANGELOG.md, knowledge content changes log to this one — never mix them up.

02_Skills/ — install grammar-normaliser, date-time, and frontend-slides here, and register each in skill-register.md. Then check the tools I mentioned earlier against Claude's connector registry and tell me what's available — don't install anything without asking first.

03_Agents/ — create agent-governance.md now (every future agent advises, it doesn't decide; its output isn't fact until I approve it; every agent also gets packaged as an installable skill) and promotion-workflow.md (the exact steps for turning an approved agent recommendation into an actual change to my knowledge base — evidence checked, my approval requested, then the change made, then logged). Leave the folder ready for actual agents — that's the next message.

04_Scheduled_Tasks/, 05_Reports/generated/, 06_Outputs/ — empty for now, ready for anything recurring or generated later.

Ask me whenever something is genuinely ambiguous — my permission tiers and my report catalogue especially depend on my real work, don't assume generic answers. Once it's built, show me the folder structure, a one-paragraph summary of each file, and log this whole build as the first entry in 00_OS/CHANGELOG.md.

Message 3 of 4 08 / 12

Build Your Own Agents

This is where your OS stops being generic. Claude proposes specialist helpers built around your actual job — you approve, tweak, or drop each one before anything gets built.

Ready to pastePROMPT3

“Propose two to four custom agents that would genuinely save me time…”

Now use everything you know about me to propose two to four custom agents — specialist helpers, each getting their own instructions file under 03_Agents/, each also packaged as an installable skill per agent-governance.md. For each one, tell me what it does, why it fits my role, what it may do on its own, and what it must never do without asking me first — especially: never treat its own output as settled fact, never delete or send anything, never update my knowledge base without going through promotion-workflow.md. Ask me to approve, adjust, or drop each one before building anything — don't build silently. Once I've approved, build each one properly, register it in skill-register.md, and log the build in 00_OS/CHANGELOG.md.

Message 4 of 4 — Save For Later 09 / 12

Plug Into The Real Knowledge Base

Don't send this one yet. Once Keith or Gary confirms the shared SharePoint knowledge base is live, come back and use it.

Ready to pastePROMPT4

“The Foundation's shared knowledge base is now live on SharePoint. Here's where it lives…”

The Foundation's shared knowledge base is now live on SharePoint. Here's where it lives: [paste the link or path Keith or Gary gives you]. Update my Claude OS so 01_Knowledge_Base/knowledge/ points there instead of my placeholder folder, refresh context-map.md to match, and log the change in 00_OS/CHANGELOG.md. If filing finished work back to SharePoint is part of my role, also set up a simple filing-convention.md the same way Gary's OS has one. Check with me before changing anything else.

What You'll End Up With 10 / 12

What You'll End Up With

→
A real OS folder (00_OS) with its own constitution and changelog
→
A permissions model, so Claude always knows what needs your sign-off
→
Custom agents, built around your job, each one governed and packaged as a skill
→
The same 3 standard skills as everyone else on the Taskforce
→
Ready to plug into the shared knowledge base the day it goes live

Rules Of The Road 11 / 12

Rules Of The Road

It's a teammate, not autopilot — you always make the final call.
It checks real dates and real files. It doesn't guess.
Nothing gets deleted, sent, or submitted without asking you first.
If something's unclear, it asks — it doesn't assume.

Welcome To
The Taskforce.

Questions? Ask Keith or Gary.

Brands Built on Experience. · The Foundation — part of the Pierce Partnership

Build YourOwn Claude OS

What This Actually Is

Before You Start

Standard Kit — Everyone Gets Three

Messages. That's It.

Tell Claude About You

Your Full OS Structure

Build Your Own Agents

Plug Into The Real Knowledge Base

What You'll End Up With

Rules Of The Road

Welcome ToThe Taskforce.

Build Your
Own Claude OS

Welcome To
The Taskforce.