One free download that stands up your entire agency operating system: the memory vault, five working skills, your Google suite, and a command center. It is a folder, not a prompt. Setup takes about ten minutes.
You run an AI agency. A client asks for lead routing, so you build lead routing. Next month they want a reporting flow, so you build that too. Each job ships and then sits on its own, one more thing you maintain by hand.
None of it compounds. You end the quarter with a drawer full of one-off automations and eight browser tabs open to run any of them.
The agencies pulling ahead sell something else. They sell the operating system that produces the automations. Same Claude work under the hood, packaged as one product a client installs and keeps.
What The OS Actually Is
Five parts, wired so they act as one thing.
Claude Code is the engine. It runs the business tasks. A vault of plain markdown is the memory, so the OS remembers what it did and who it did it for. Your Google account is a surface it reads and acts on, not screenshots you paste in. Five starter skills are the hands. And one command center page shows what the OS can do, what is scheduled, and what happened last.
That is the thing you use yourself. It is also the thing you install for a client, or hand to someone on your team so they can run it without you in the room.
Building it from scratch used to eat three weekends. The installer does it in one session.
Why This Is A Folder, Not A Prompt
The first version of this installer was a single markdown file you pasted into a chat. It worked, but a single file has to explain everything at once, so it stays shallow everywhere.
Real Claude skills ship as packages. A lean SKILL.md that conducts. Reference files that carry the depth. Assets that are actual files, not descriptions of files. The agent reads the conductor first and opens each reference exactly when that phase starts, so you get depth without a five-thousand-word wall.
In August 2026 the installer became that package. Here is the whole download:
Copy this.
agentic-os-installer/
├── 00 START HERE.md ← the 60-second install card
├── SKILL.md ← the conductor: five phases, in order
├── references/
│ ├── 00-overview-and-prerequisites.md
│ ├── 01-vault-scaffold.md ← the exact folder taxonomy + house rules
│ ├── 02-starter-skills.md ← what the five skills do, how to run each
│ ├── 03-google-suite.md ← the optional Gmail / Calendar / Drive wiring
│ ├── 04-command-center.md ← the one-page dashboard template
│ ├── 05-troubleshooting.md ← every known snag, with the fix
│ └── about-operations-heroes.md
└── assets/
└── skills/ ← five real, working skills
├── welcome/ ← Step 0: machine check + orientation
├── brain-install/ ← Step 1: your documents become a verified business brain
├── bottleneck-finder/ ← Step 2: a ranked map of what to automate, by ROI
├── skill-forge/ ← Step 3: watch a real workflow, forge a tested agent
└── system-health/ ← Step 5: weekly report on what ran and what broke
The assets/skills/ folder is the part the old version could not have. Those five skills are not examples. They are the working install path a business runs in order: check the machine, build the brain, rank the bottlenecks, forge the agents, keep it all healthy. The installer copies them into the vault it scaffolds, and day one ends with a system that already knows its own next step.
Copy The Conductor File
The download at the bottom of this page grabs the whole folder. If you want to read the brain of it first, or start from a blank chat, this is the file that runs the show.
Copy this.
---
name: agentic-os-installer
description: >-
Stand up a complete Agentic OS for one business in one session: a memory
vault, five working starter skills, an optional Google connection, and a
command center page. Load this when someone says "install my Agentic OS" or
wants to go from an empty folder to a working operating system.
---
# Agentic OS Installer
> **What this does, in one line:** You name a folder and a business. The agent stands up your whole Agentic OS inside it: the memory vault, five working skills, your Google suite if you want it, and one command center page to run it all from.
Agencies keep shipping one-off automations that never add up to anything. This installs the thing that produces the automations instead: an operating system you own, run, and can hand to a client. The install is one session. Your part is about ten minutes of answers.
## The single most important thing to understand
The OS is a folder. Plain markdown files an agent reads, follows, and writes to. The engine (Claude Code today, any agent CLI tomorrow) supplies the intelligence. The folder supplies the memory, the skills, and the rules. That split is why you own it: swap the engine later and the OS survives untouched.
## ⚙️ Before you start (fill these in)
The only user-specific values. Everything else the skill and its reference files walk you through.
| Fill in | What it is | Example |
|---|---|---|
| `BUSINESS_NAME` | What to call this OS so it is branded to you | Bright Studio OS |
| `VAULT_PATH` | The folder where the OS and its memory will live | /Users/you/BrightStudioOS |
| `GOOGLE_ACCOUNT` | Optional. The Google account to connect in Phase 3 | you@brightstudio.com |
> Never paste a password, token, or secret into the chat. Secrets go into the tool's own settings. If one gets pasted anyway, rotate it.
## ✅ What you need
- Claude Code, or any agent CLI that can read files and follow markdown instructions. Codex works too.
- A folder location that is NOT inside iCloud, OneDrive, Dropbox, or Google Drive sync. Cloud sync breaks agent file access.
- Optional: a Google account for Phase 3. Skip it on day one and add it later if you prefer.
## What you are building (in plain terms)
Five parts, wired as one system. The vault is the memory: every client, note, and run written where the agent can read it tomorrow. The five starter skills are the hands. Google is the surface the OS acts on. The command center is the single page that shows all of it. Git is the time machine underneath.
```
engine (Claude Code / Codex)
│ reads + writes
▼
VAULT_PATH/
├── 000 OS/skills/ ← five starter skills + every skill you forge later
├── 000 OS/Command Center.md ← the one page that shows the whole system
├── 000 OS/reports/ ← health reports land here
├── 150 Clients/ ← one folder per client
├── 200 Notes/ ← the business brain
└── 210 Runs/ ← one line per agent run. This is how it remembers.
```
Running cost: your existing agent subscription. The OS itself is markdown. It costs nothing to keep and nothing to move.
## The phases. Follow them in order.
Read a phase's reference file right before you start that phase, not all at once.
1. **Scaffold the vault.** Create `VAULT_PATH`, the folder taxonomy, a house-rules file branded to `BUSINESS_NAME`, and a git history so nothing is ever lost. Read `references/01-vault-scaffold.md`.
2. **Install the starter skills.** Copy the five skills from this package's `assets/skills/` into `000 OS/skills/`. Verify the agent sees all five. Read `references/02-starter-skills.md`.
3. **Connect the Google suite (optional).** Wire Gmail, Calendar, and Drive for `GOOGLE_ACCOUNT` so the OS reads real threads and files instead of pasted screenshots. Read `references/03-google-suite.md`.
4. **Build the command center.** One markdown page that lists the skills, what is scheduled, and the latest results. Read `references/04-command-center.md`.
5. **First run and handoff.** Run the welcome skill for the machine check, write the first run-log line, and start the business brain when the owner is ready. Read `references/00-overview-and-prerequisites.md` for what a good day one looks like.
## How to actually run a session
The whole install fits in one sitting. Announce each phase in one line before you start it. Ask before creating anything outside `VAULT_PATH`. If a session gets interrupted, note where you stopped in `210 Runs/` and resume from that phase next time. Every phase is safe to re-run.
## Order matters. A few hard-won rules.
- Vault before skills. The skills reference folders that must already exist.
- Skills before Google. The OS should work end to end offline first, so a connector problem never blocks the install.
- Never run two businesses in one vault. Point the installer at a fresh `VAULT_PATH` per client. Each install stays clean and ready to hand over.
- Log every run. One line in `210 Runs/YYYY-MM-DD.md`. The health check and the command center read it. An OS that stops logging stops learning.
## Troubleshooting
`references/05-troubleshooting.md` is the field guide. Every failure this install is known to produce is in there with the cause and the fix. Consult it the moment anything looks wrong rather than guessing.
How To Install It: 2 Ways
Claude Code. Unzip the download and move the folder into ~/.claude/skills/. Open Claude Code and say: "Use the Agentic OS Installer skill. Install my Agentic OS at ~/MyAgencyOS for Bright Studio." Your folder, your name.
Claude web or desktop. Keep the zip. Go to Settings, then Capabilities, and upload it as a skill. Say the same sentence in a new chat.
The agent takes it from there. It scaffolds the vault, copies in the five skills, offers the Google connection, writes the command center, and runs the first machine check. Your part is about ten minutes of answers.
Done. The OS exists.
What You Get Back
01. The vault. Folders for the OS, your clients, your notes, and a run log. Every session writes one line to the log, so the OS remembers the last client call instead of starting cold each morning.
02. Five working skills. Welcome checks the machine. Brain-install turns your documents into a verified business brain. Bottleneck-finder ranks what to automate by ROI. Skill-forge watches a real workflow and builds a tested agent for it. System-health reports weekly on what ran and what broke.
03. Your Google suite, connected. Gmail, Calendar, and Drive wired in for the account you name, with one rule baked in: the OS reads freely, drafts freely, and a human sends. Optional on day one.
04. The command center. One page that lists the skills, the schedule, the latest runs, and the current health status. The screen you use, the product you demo, and the handoff you give your team.
05. A git history. Every change committed from day zero. Nothing is ever lost, every mistake is reversible, and the whole OS can be zipped and handed over whole.
Honest: Where This Stops
The memory is empty on day one. That is correct, not broken. It fills as the OS does real work, so the first week looks quieter than the second.
The command center is a markdown page, not an app. That is deliberate. One honest page beats an empty dashboard. Build the real thing once the OS has enough running to justify it.
Each client install is its own vault and its own Google account. Do not run two agencies through one OS. Point the installer at a fresh folder per client and every deployment stays clean and ready to hand over.
And it does not lock you to one model. The OS runs on Claude Code or Codex, and the folder survives an engine swap untouched. The operating system is yours either way.
Do One Thing This Week
Download the folder, point the installer at an empty directory, and run the first two phases only. Vault plus skills. Twenty minutes, no Google, no client yet. Just watch the OS assemble itself in front of you.
If you do not like what it builds, you spent twenty minutes. If you do, you have the first version of the thing your best competitors are already selling.