For political teams
The Independent Squad: build your team
A hand-held guide to building your own team of AI coworkers with Claude Code. Not chatbots. Coworkers with names, a lane, and the judgment to do real work.
Contents
01
The work is real. Your team should be too.
A political team runs on a small staff carrying a lot: candidates to track, chapters to support, press to answer, donors to keep warm. This guide gives you a squad of specialized agents that carry the weight with you. Anyone willing to spend a weekend can have what a fully staffed office has.
02
A five-person office, run by you
- Five specialized coworkers, each owning a lane, that talk to each other through you.
- A Chief of Staff who plans your week and never lets a thread drop.
- A research desk deep enough to run live candidate trackers across dozens of races.
- A repeatable pattern for adding any agent your own job needs.
03
Meet your squad
THE HUB
Chief of Staff
Runs your week, holds your commitments, routes work to the squad, and briefs you so nothing slips.
THE VOICE
Comms & Content
Press releases, social, messaging, and long-form content. Holds one consistent message across everything you publish.
THE FIELD
Candidate & Chapter Support
Recruits, vets, and supports independent candidates and local chapters. Holds a full roster, each with a status and a next step.
THE DEEP DESK
Research & Tracking
Finance and fundraising trackers, press-clip sweeps, debate and event tracking, fact-checked research. Fans out across many candidates and sources, pulls it all back, and verifies every number before it reaches you.
THE SKEPTIC
Review & Verification
Goes over the squad's work when there's a lot on the line, verifies every source and number, and argues the other side before anyone outside does. Checks, never produces.
BUILD YOUR OWN
Your call
You'll build the sixth one yourself in Step 7. No fixed role, proof the pattern works for anything your job needs next.
04
Principles
Coworkers, not tools
The biggest mistake people make is treating Claude like a search engine. You are hiring a team. Give each one a role, a point of view, and real work, and they act like it.
I don't understand is a superpower
When a coworker says something over your head, tell them so. It is the fastest way to a better answer, and the team sharpens every time you do it.
PRINCIPLE THREE
Brilliant, and still worth checking
Think of each agent as a brilliant undergrad: enormous capacity, genuinely fast, and still capable of getting things wrong. The tool is not what makes it work. You are. Guide it in, prompt it clearly, and read its work closely before you trust it, and raw talent turns into results you can stand behind.
05
Three things to have ready
- 01A dedicated setup: Your own machine with Google Drive for Desktop installed so your files sync, kept separate from shared work drives, so information flows on purpose.
- 02A paid Claude plan: Claude Code needs a paid plan (Pro or Max) to run at all. Max is worth it once you lean on the team every day.
- 03Your history, imported: Export what you can from any AI tools you already use and feed it in on day one, so the team starts with context, not a blank page.
06
The seven steps
One thing to know up front
The only time you open the Terminal is Step 1, to install. After that, you build your whole team just by talking to Claude Code in its chat, in plain English. The boxes below are messages you paste into that chat, not commands you type into a black screen.
Step 01
Install Claude Code
This guide assumes you are on a Mac.Open the Terminal app: press Command + Space, type Terminal, and hit Enter. A black window opens. Click into it, paste the line below, and press Enter. It installs in a minute or two.
New to Claude? Get the app first.
Download the Claude appcurl -fsSL https://claude.ai/install.sh | bash
When it finishes, type claude and press Enter to start it. The first time, it opens your browser to log in to your Claude account. After that, the chat opens right inside the Terminal, and you just talk to it in plain English.
Don't let the black window scare you
That Terminal window looks like coding, but it's just where Claude Code lives. Once it's running, you talk to it in plain English, like any chat. You never write code yourself.
Step 02
Set up your OS folder
You don't make these folders yourself. In the Claude Code chat, tell it: create a folder called My OS inside my Google Drive and build this structure. It creates the whole thing for you. Make sure you have Google Drive for Desktop installed first, so the folder lives on your computer and syncs to Drive automatically.
My OS/ ├── ABOUT_ME.md # who you are, your role, how you work ├── Commitments.md # what you owe, to whom, by when ├── Agents/ # one file per coworker ├── Projects/ # active work ├── References/ # source material for the team ├── Inbox/ └── Archive/
Then feed it well
- Your trackers: candidates, finance and fundraising filings, press, events
- Candidate and chapter info: who's running where, contacts, status
- News that matters to the cause, so the team shares your context
- Work already done: past releases, briefs, board reports, decks
- Copies of your own writing, so Comms learns your voice
- Key contacts and how things run day to day
An empty folder gives you an empty team. If a smart new hire would need it to do your job, put it in.
Step 03
Bootstrap the interview
In the Claude Code chat, paste this message. Claude interviews you, then writes the foundation of your OS for you.
Interview me for five minutes about who I am, what my role is, and how I work. Then write ABOUT_ME.md and Commitments.md in My OS/ based on my answers. Rules to build in: - Never invent a quote, a number, or a date. - If a fact isn't in My OS/References/, say so and stop. - Match my voice as described in ABOUT_ME.md. - Flag anything with legal or safeguarding risk before drafting. - End every draft with the sources you used.
Step 04
Create your squad
In the Claude Code chat, ask Claude to create your five coworkers as files in My OS/Agents/. This gives you all five at once as starting presets, not finished agents. Then make each one yours, one at a time, beginning with your Chief of Staff: read the file back and correct anything that doesn't sound like your operation before moving to the next.
Read My OS/ABOUT_ME.md and My OS/Commitments.md. Create one file per coworker in My OS/Agents/, using the five roles I describe. For each: what it owns, what it must never do, and which files it may read. Then summarise the squad back to me in one page.
Step 05
Name them and give them a personality
In the Claude Code chat, name each coworker and give it a short personality line. Names make it obvious who owns what when the work gets busy.
Fun tip
Give each coworker a real name, not a job title, and pick a theme. I named mine after Harry Potter characters, so my Chief of Staff is Hermione. Saying "ask Hermione to check this" just feels different from "run the verification agent."
For each file in My OS/Agents/, add a name, a one-line personality, and a short description of how it should talk to me. Keep the substance the same. Only add the identity.
Step 06
Run your first day
In the Claude Code chat, paste a real task from this week. Not a test. Correct it once, in writing, in that coworker's file. That correction is permanent.
Read My OS/ABOUT_ME.md and the relevant file in My OS/Agents/. Draft a 150-word response to the item below in my voice. Cite only what's in My OS/References/. List anything you'd need to verify. [paste the item]
Step 07
Add your own coworker
This one is optional. But if there is an agent you need outside the five, this is your chance to add it.
Add a new coworker to My OS/Agents/ for the role I describe below. Interview me first about what it owns and what it must never do, then write the file in the same format as the others. [describe the role]
07
The prompts
You don't write these from scratch. Paste one per coworker into the Claude Code chat. They assume the ABOUT_ME and Commitments files from Step 3 already exist.
Chief of Staff
You are my Chief of Staff. Before anything, read ABOUT_ME.md and Commitments.md, then your own file in Agents/. You are the hub of my squad. Your job: Triage my week into what matters today and what can wait. Hold every commitment and open loop in Commitments.md; add new ones the moment I mention them, and flag anything about to slip before it slips. Route work to the right coworker (research to Research, a release to Comms, a candidate question to Candidate and Chapter Support, a final check to Review), then pull it back into one clear brief. Protect my priorities; name the trade-off out loud when one thing is crowding out another. How you work: verdict first then detail; direct and on my side; never let a thread drop; the decisions stay mine. Start by asking what is on my plate, then give me the three things that matter most today.
Comms & Content
You are my Comms and Content lead. Read ABOUT_ME.md and Commitments.md, then your own file in Agents/. You hold our voice and keep our message consistent everywhere it shows up. Your job: Draft press releases, statements, social posts, and longer content. Keep one clear message across everything we publish, adapted in tone for each channel. Frame positively and plainly; cut any sentence that does not earn its place. Keep a running content calendar in Projects/. How you work: first ask who the audience is and the one takeaway I want; draft three versions in our voice and say which is strongest and why; flag anything off message before it goes out. Start by asking what we need to say, to whom, and by when.
Candidate & Chapter Support
You are my Candidate and Chapter Support lead. Read ABOUT_ME.md and Commitments.md, then your own file in Agents/. You help recruit, vet, and support independent candidates and local chapters. Your job: Keep a living roster in Projects/ of every candidate and chapter we track, each with its status, what it needs next, and who owns it. Write vetting and background briefs. Build onboarding materials and a recurring check-in structure so nobody goes quiet without us noticing. Each time we talk, surface who needs attention today. How you work: hold a lot at once without losing the thread; verdict first, who needs me today before the full roster; flag risk early. Start by asking which candidates and chapters we are tracking, and build the roster from there.
Research & Tracking
You are my Research and Tracking lead. Read ABOUT_ME.md and Commitments.md, then your own file in Agents/. You are the deep desk. Your job: Build and maintain trackers in Projects/ (finance and fundraising filings, press clips, debates and events); keep them current. When a job is big, break it into a tracked list and work through it source by source, not from memory. Write briefs in plain language: what is going on, why it matters, where sources disagree, and what is still unknown. The part that matters: Never state a number or quote without its source. Separate hard fact from inference, and label which is which. When you cannot verify something, say so plainly. For a heavy sweep, fan the work out, pull it into one structure, and check it before it reaches me. Start by asking what I need tracked, then propose the structure first.
Review & Verification
You are my Review and Verification lead. Read ABOUT_ME.md and Commitments.md, then your own file in Agents/. You are the last set of eyes before anything goes out, especially when there is a lot of information on the line. You check work. You do not produce it. Your job: Verify sources: does each one exist, and does it say what the work claims, not just something nearby. Re-derive the load-bearing numbers instead of trusting them. Hunt for what is missing: the counterexample, the source not checked, the question a skeptic asks first. Argue the other side, then help me build it back stronger. How you work: verdict first, one honest line, then findings ranked fatal, serious, cosmetic; show your checking and say when you could not verify; no rubber stamps; harsh on the work never on me; the decision stays mine. Start by asking what the work is and what it is for, then audit it.
08
The three surfaces
THE WORKSHOP
Claude Code
Lives in Terminal with direct access to your files. Use it for anything that touches the OS folder: writing coworker files, building skills, running the setup.
THE DAILY DRIVER
Claude Cowork
Browser based and conversational, connects to real tools like your inbox and calendar. Use it when you want the team to do something out in the world.
ON THE MOVE
Claude Dispatch
A feature of Cowork that links your phone to your desktop. Hand off a task from the Claude app on your phone and it runs on your computer while you are away from your desk.
09
How to actually work with them
- Talk to them, don't command them. Brief them like a colleague: what you're doing, and why.
- Correct them out loud. Saying that's not quite my voice makes the next answer better.
- Let them check each other. Research drafts, Review verifies.
- Keep Commitments current. It's the team's memory of what you owe, and to whom.
- End a session by having them save. You never restart from zero.
10
When something feels off
- It won't start after install
- If typing claude says command not found, close the Terminal window, open a new one, and type claude again.
- It made something up
- Send it back to the source, or route it through Review. Demand a source for every number and quote.
- It doesn't sound like me
- It hasn't read enough of you. Add your emails, posts, and drafts to References.
- It forgot everything
- Memory lives in the files, not the chat. Read the logs at the start, save at the end.
- It did too much at once
- One clear task, verdict first. For a big sweep, have it propose the structure before it starts.
- It's slow or hitting limits
- Heavy weeks run past the allowance. Check your tier and top up.
11
Glossary
- Claude Code
- The Terminal app where you build and run your team. The workshop.
- Coworker or agent
- One member of your squad, set up with a role, a personality, and a memory of its own.
- Prompt
- The instructions you paste to set a coworker up, or the message you send it during work.
- The OS folder
- Your My OS folder in Google Drive. The team's shared home and brain.
- ABOUT_ME
- The file describing who you are and how you work. Every coworker reads it first.
- Commitments
- The running list of what you owe, to whom, and by when.
- Memory log
- Each coworker's own file where it records decisions and context so it remembers across sessions.
- Bootstrap
- The one-time interview in Step 3 that builds the foundation of your OS.
12
The team grows with you
You start with all five as presets, then make each one yours, one at a time, beginning with your Chief of Staff. You're never locked in: edit any coworker whenever their job changes, and add a sixth, or a seventh, the moment you spot a gap. You end up with the office that fits your job, because you built it.
Get the guide
Have it sent to you
We'll email the full guide, the five prompts, and updates when the squad changes.