The AI Consultant · Your AI Working Playbook

The 4 Claude Projects — Working Playbook

Built around Austin Marchese's "4 projects." Tab 1 is the first focus (Board of Advisors); tabs 2–4 are ready for future sessions; Tab 5 is the 30-minute setup foundation; Tab 6 is the build engine — the honest clock on turning an idea into a working tool.

🤝 Your working playbook — we go through it together, live, and it's yours to keep.

1 The Video, in 90 Seconds Austin Marchese · 13 min

Stop watching tutorials — build four compounding Claude projects, each usable today, no coding required. Together they form a personal operating system.

PROJECT 1 · TONIGHT

Board of Advisers

Clone real experts so you can ask a "board" before high-stakes decisions.

PROJECT 2

Niche Command Center

A private tool for a job you do daily — the real win is sharper thinking.

PROJECT 3

AI-Optimised Profile

A site built so AI describes you well to buyers & referrals.

PROJECT 4

Internal Operating System

Knowledge / Skills / Project folders + a CLAUDE.md brain + self-improve loop.

The critical-thinking angle: he flagged a comment about making the board think critically rather than agree — a "devil's-advocate" layer. Built into Prompt 4. Project 1 is the natural start because he already owns a deep sales library to clone from.

2 The Session Plan 60–90 min · hands-on

By the end he'll have a working /ask-the-board skill and will have asked it a real question about a live case. Have Claude Code open and one current case in mind.

0–10 MIN · WARM-UP

Frame the outcome & pick the board

Choose 2–3 sales minds he wants first — people whose full body of work he can feed in (books/courses he owns). Start with 2.

✔ 2–3 advisors shortlisted
10–25 MIN · INGEST, DON'T INTERVIEW

Feed his context — cap questions at 3

Run Prompt 1, but lead by ingesting his exported ChatGPT history so Claude drafts the profile itself and asks only 3 short gap questions. This suits his preference for moving fast — context in, minimal questions.

✔ Profile drafted from his data, not a Q&A
25–50 MIN · CLONE ADVISORS

Ingest each advisor into a persona

Run Prompt 2 per advisor — feed book notes, course summaries, or public links. Claude builds a persona file.

✔ One persona file per member
50–70 MIN · BUILD SKILL

Create the /ask-the-board skill

Run Prompt 3 — loops every advisor, returns each take + a synthesis. Test on a live case.

✔ Working /ask-the-board skill
70–85 MIN · SHARPEN

Add the devil's-advocate layer

Run Prompt 4 — forces challenge & disagreement. Re-ask and compare.

✔ A board that argues back
85–90 MIN · LOCK IN

Save & plan the next add

Save to a board/ folder. Note the next 1–2 advisors and one real decision to bring this week.

✔ A habit + a folder to grow
Zenith link: the "ground it in you" step is the same move behind his Zenith Signature Profile — output gets far better when AI pulls context from his own material instead of guessing.
👥 Approach — he values expediency. He likes to move fast, so the winning pattern is ingest, don't interrogate: drop his ChatGPT export (or existing session notes) so Claude drafts the profile itself and only asks the 2–3 real gaps. Same rule for every "interview me" prompt here — feed the context first, let it ask almost nothing.

3 The Prompts copy · paste · run in order

Run in Claude Code, in order. Replace the [highlighted] parts. Nothing assumes a specific author — he supplies advisors from his own library.

Step 1 — ingest, don't interviewFeed your context, then max 3 questions
You are helping me build a personal "board of advisors" inside Claude for my sales
work. I do NOT want a long interview — I've already got a lot of context in my
exported ChatGPT history. Read that first, then ask me almost nothing.

Step 1 — Read my context.
I'm giving you my exported ChatGPT data (the folder / conversations.json from my
ChatGPT export — see the note below) plus anything else I attach. Read it and pull
out: what I sell, my typical clients and deal sizes, how I talk about my work, my
goals, and my recurring sales challenges.

Step 2 — Draft my profile yourself.
From that, write a concise "Sales Profile" and save it as board/my-profile.md. Fill
in everything you can reasonably infer — do NOT ask me to repeat what's already in
the export.

Step 3 — Ask me a MAXIMUM of 3 short questions.
Only the gaps that genuinely matter and that you couldn't find in my data. Keep them
short and concrete, one at a time. Then finalise the profile.

To anchor you: I sell [life insurance / HNW estate protection] to [my typical client, e.g. high-net-worth families in Canada].
How he feeds his ChatGPT history: in ChatGPT → Settings → Data controls → Export data. He gets an emailed link to a .zip; unzip it and drop the folder (it holds conversations.json and chat.html) into his working folder, then point Claude at it. It's a point-in-time snapshot — if some of it's out of date, that's fine: the 3 questions catch whatever's changed. No export handy? Paul also has months of your coaching-call transcripts to seed it from.
Step 2 — repeat per advisorClone an advisor into a persona
I want to add [ADVISOR NAME] to my board of sales advisors.

Here is the source material I have on them (I'll paste/attach it now):
[paste book notes, course summaries, transcripts, OR their public
YouTube / podcast links — whatever you own or can access]

From this, build a persona file at board/advisors/[advisor-name].md that captures:
- Their core philosophy and how they see selling.
- Their signature frameworks, methods and language (name them).
- The questions they'd ask and the advice they'd give a life-insurance advisor.
- Their tone of voice, so you can answer AS them convincingly.
- What they'd push back on or warn me against.

Read it against my board/my-profile.md so their coaching is specific to MY practice.
Confirm when saved, then I'll give you the next advisor.
Step 3Create the /ask-the-board skill
Create a custom Claude skill called "ask-the-board".

When I run /ask-the-board followed by a question or a case, it should:
1. Load my profile from board/my-profile.md and every persona in board/advisors/.
2. Answer my question IN CHARACTER as each advisor, one at a time — using their
   own frameworks, language and priorities. Label each clearly by name.
3. Where advisors would disagree, show the disagreement — don't blend them.
4. Finish with a "Chair's Synthesis": the strongest combined recommendation,
   the key trade-offs, and the single next action I should take.

Keep each advisor's take tight and practical — this is for real decisions on live
cases, not theory. Build the skill now and then show me how to call it.
Step 4 — the one he flaggedForce the board to think critically
Enhance the /ask-the-board skill to force the board to think critically instead of
just agreeing with me or with each other.

Add these rules to the skill:
- Every advisor must include at least one honest challenge, risk, or blind spot in
  my thinking — not just validation.
- Appoint one advisor each round as "devil's advocate": their job is to argue the
  strongest case AGAINST my proposed approach and against the emerging consensus.
- Before the Chair's Synthesis, add a "What we might be wrong about" section listing
  the biggest risks and the assumptions we haven't tested.
- If the board is converging too easily, say so and stress-test the weak points.

The goal: sharper decisions and real pushback on my cases, especially before I
present to a client. Update the skill and re-run my last question so I can see
the difference.
Pro tip — screenshots count as input. Any prompt he sees in a video, just screenshot and paste — Claude reads it the same as typing. And answering the interview by voice-to-text is ~10× faster.
👥 Approach: the highest-value parts here are the devil's-advocate layer and the persona quality — hard to get right solo, and where his own library becomes a real advantage. That's what makes building it together worthwhile.

2 Niche Command Center Future session · ~30–45 min

A private tool that helps with something he already does daily. For an insurance advisor, the obvious build is a Case & Client-Prep Command Center — model estate/tax scenarios, compare products, and prep the meeting narrative.

Why it lands (his four reasons)

  • He'll actually use it — solves a real case-prep problem, not a hypothetical.
  • Skips analysis paralysis — the problem's already chosen (his own workflow).
  • Sharpens the craft — forces him to map how he actually structures a case.
  • Zero audience pressure — no polish, no marketing; build fast.
Insurance angle: this is where the estate-value / JLTD-style scenario modelling he already does by hand becomes a repeatable tool — "given assets, age, tax exposure → show estate value with vs. without cover, and the meeting talking points."

Mini-plan & prompts

STEP 1

Choose what to build

Let Claude propose options from his past chats, then pick one (e.g. a case analyzer).

STEP 2

Plan the MVP, then build

Plan before building — interview-style planning prompt with guardrails.

STEP 3

Iterate live

Use it immediately; paste/screenshot errors; add features as needed.

Step 1Decide what to build
I want to build a private tool ("command center") that helps me with something I
already do every day in my [life insurance] sales work — just for me, no one else uses it.

Based on what you know about my work, propose 5 options I could build that would
save me time or sharpen my case prep this week. For each: what it does, the single
biggest job it removes, and how simple a first version would be.
Then ask me which one to build.
Step 2Plan the MVP before building
We're building: [e.g. a case analyzer that models estate value with vs. without life cover].

Before writing any code, interview me to lock the plan:
- What inputs do I have (assets, ages, tax rates, product options)?
- What outputs do I need to SEE for a client meeting?
- What decisions should it help me make faster?
Keep it to the minimum version that's useful THIS WEEK — no nice-to-haves yet.

When the plan's agreed, build the MVP, then let me start using it and add features
as I hit them.
👥 Approach: a great option if he'd rather build a practical tool than clone advisors — it delivers a working tool AND sharper thinking, and can be the more concrete starting point.

3 AI-Optimised Profile Future session · ~30–45 min

A personal site built so AI — not just Google — describes him well. Increasingly a referral or HNW prospect's first move is to ask an AI "who is this advisor?" This shapes that answer.

Two goals

  • Humans land and think "credible, specialised, I'd trust him with my estate."
  • AI looks him up first — recruiters, referral partners, prospects — and gets the right story.
Insurance angle: position around his niche (e.g. HNW estate protection, JLTD, tax-efficient wealth transfer) so the AI summary reads as a specialist, not a generic agent.

Mini-plan & prompts

  1. Set the page goal — what should a visitor think/feel? Reverse-engineer from that.
  2. Pull inspiration — sites/brands he admires.
  3. Scaffold a Node.js site — so it deploys easily (Cloudflare/Hostinger).
  4. Add an "Ask AI about me" block in the footer — provider links with a preloaded prompt.
  5. Deploy.
BuildScaffold the AI-optimised site
Build me a simple personal website in Node.js for my work as a [life insurance
strategist specialising in HNW estate protection].

First, interview me to fill the copy: who I help, the outcomes I create, proof/
credibility, and the ONE impression I want a visitor to walk away with.

Then scaffold the site optimised for AI discovery:
- Clear, structured "about", "who I help", and "how I work" sections.
- Machine-readable summary metadata so an AI can describe me accurately.
- A footer "Ask AI about me" block with buttons to the main AI providers, each
  opening a chat pre-loaded with a prompt that returns an accurate summary of me.

Keep it clean and credible. We'll deploy it after.
👥 Approach: not the first priority for an advisor mid-pipeline, but strong future-session material — a natural follow-on once he has a Command Center to show off.

4 Internal Operating System Future session · the container

"A bunch of files that take AI from good output → hyper-specific great output." It's the container that holds the board, the command center and his knowledge together — so he never re-explains himself each session.

The structure

  • Knowledge/ — everything Claude should know: product info, objection playbooks, his voice, saved frameworks, meeting notes.
  • Skills/ — repeatable processes (the /ask-the-board skill lives here).
  • Project/ — what he's actively working on (a live case, the command center).
  • CLAUDE.md at root = the brain — tells Claude how to use the folders.
Why it matters for him: his sales knowledge (products, carriers, objection handling, his own winning language) becomes a system that compounds — every case makes the next one sharper.

Mini-plan & prompts

STEP 1

Set up the structure

Create Knowledge / Skills / Project + a CLAUDE.md brain.

STEP 2

Build a /improve system skill

Captures feedback so output gets better every time — the most important part.

STEP 3

Build an /ingest resource skill

Files new articles/transcripts/videos into the right place.

STEP 4 (optional)

Put it on GitHub

Version control — treat his OS as the IP of his practice.

Step 1Set up the folder structure + brain
Set up a personal "internal operating system" for my [life insurance] work in this folder.

Create three folders and a brain file:
- Knowledge/  → things you should always know (my products/carriers, objection
  playbooks, my ideal client, my voice, trusted frameworks).
- Skills/     → my repeatable processes (e.g. ask-the-board).
- Project/    → what I'm actively working on (a live case, a tool).
- CLAUDE.md at the root → explains how to use these folders so I never re-explain.

Interview me briefly to seed the Knowledge folder with the essentials, then create
the structure and the CLAUDE.md.
Step 2The /improve system skill
Create a skill called "improve-system".

When I run /improve-system after we've landed on a good result (e.g. a client email
or case narrative I liked), it should:
- Ask what made this version good vs. earlier drafts.
- Capture that as a durable rule/preference in the right Knowledge file.
- Confirm what it learned, so next time the output starts closer to my standard.

The point: every time I use it, the system gets sharper at sounding like me and
making the calls I'd make. Build it now.
👥 Approach: the video's most ambitious project — the biggest "wow", and best after he's felt value from the board, as the piece that ties everything together. A strong candidate for a dedicated future session.

Where it all connects

PieceLives in
Board of advisors (personas + skill)Knowledge/ + Skills/
Command center toolProject/
His products, objection playbooks, voiceKnowledge/
The brain that runs itCLAUDE.md
Board → this sessionCommand Center → nextInternal OS → ties it togetherProfile → when ready

5 The 30-Minute Foundation AI Bri · 57-sec short

The video Agostino just sent. Its whole argument: Claude out of the box is "mid" — and that's a setup problem, not a model problem. It has zero memory between sessions, so it doesn't know who he is, what he sells, or how he likes things done. Three moves fix it — the same three that quietly sit underneath everything in tabs ①–④.

Why it matters here: the board, the command center and the Internal OS all assume Claude already knows Agostino. This tab is the 30 minutes that makes that true — do it first and every other prompt on this page lands twice as hard.
MOVE 1

Global Instructions

Role, formats, tone — loaded into every session automatically.

MOVE 2

A CLAUDE.md file

Project-specific context Claude reads on launch. (Same file as Tab ④'s "brain".)

MOVE 3

Chunk your skills

Separate focused files, not one giant doc — e.g. three writing skills, not one.

Set it up — live ~30 min · once

Three short moves. The video is about Claude Cowork, but these apply identically to the Claude Code setup on the rest of this page — same three levers, same payoff.

MOVE 1 · ~5 MIN · GLOBAL INSTRUCTIONS

Tell Claude who he is, once

Settings → profile / global instructions. His role, preferred output format, and tone. Loads on every session so he never re-introduces himself.

✔ Claude opens knowing he's an insurance advisor
MOVE 2 · ~15 MIN · CLAUDE.md

Drop a brain into the working folder

A CLAUDE.md with his products/carriers, his ideal client, his voice and his tools. Claude reads it on launch. This is the exact file Tab ④ builds — start it here.

✔ Project context loads automatically
MOVE 3 · ~10 MIN · CHUNK SKILLS

One focused skill per job

Instead of one giant instructions doc, split into small skills — e.g. discovery-call, objection-handling, case-summary. Each stays sharp because it does one thing.

✔ Skills that don't dilute each other
Bridge to the rest of the page: Move 2 is the CLAUDE.md brain in Tab ④, and Move 3 is why /ask-the-board (Tab ①) lives as its own skill file. This tab isn't a fifth project — it's the floor the other four stand on.

The prompts copy · paste · run

Two copy-paste starters so he leaves the session with the foundation already laid — not a to-do.

Move 1Draft his Global Instructions
Help me write my Global Instructions (the profile that loads into every Claude
session automatically). Interview me briefly, then draft a tight version covering:

- Who I am and what I do: [life insurance advisor, HNW estate protection, Canada].
- My clients and the outcomes I create for them.
- How I want you to respond by default: format, length, and tone
  (e.g. direct, practical, no fluff, plain English not jargon).
- What you should NEVER do (e.g. give generic advice, invent numbers, assume a
  product without asking).

Keep it under a page — it loads every session, so it must be tight and high-signal.
Move 2Start his CLAUDE.md brain
Create a CLAUDE.md file in this folder — the project brain you read on launch so I
never re-explain my work.

Don't interview me from scratch. First read my exported ChatGPT history (the folder /
conversations.json I'm attaching) plus anything else here, and draft as much as you
can yourself. Then ask me a MAXIMUM of 3 short questions to fill real gaps.

Write it with sections for:
- My products & carriers, and the cases I handle most.
- My ideal client and my typical deal.
- My objection playbook and the language that actually works for me.
- My voice: how my emails and client notes should sound.
- My tools and where things live.

This is the same file my "Internal OS" uses — build it so it can grow as we add
the board and my command center.
Move 3Split one doc into focused skills
I want to chunk my Claude skills into separate focused files instead of one giant
instructions doc, so each stays sharp.

Look at the sales jobs I do repeatedly and propose 3–5 small, single-purpose skills
I should create — for example: discovery-call prep, objection-handling, case
summary, follow-up email. For each, tell me the one job it owns and what it should
never try to also do.

Then create the first one with me and show me how to call it.
Anthropic's own line on this: the model is only as good as the context you give it. Thirty minutes here is the difference between "AI is overhyped" and a system that sounds like him and makes his calls.
👥 Approach: if Claude has ever felt generic, this is the honest reason — the setup was skipped. It's 30 minutes that pays back on every session, and it flows straight into Tab ④ (they share the CLAUDE.md). A great low-friction opener if the full board build feels like a lot in one go.

6 Build a Real Tool Allie K Miller · 2-min short

The second video Agostino just sent. A 10-step method for turning an idea into a working tool with Claude Code: brainstorm with Opus 4.8, then hand the build to Fable 5, which orchestrates agents and sub-agents to actually construct and test it while you stay in the loop. This isn't a fifth project — it's the engine that builds Tab ②'s Command Center (or any tool on this page).

⏱ The honest clock — read this first. The video is 2 minutes long, so it feels like a 2-minute job. It isn't. Two minutes of watching compresses roughly 2–3 hours of real elapsed time — but most of that is Fable 5 building in the background while you do something else (Allie checks in from her phone between client calls). Your actual hands-on input is more like 45–60 minutes, spread across the loop. A quick video ≠ a quick build — it means a quick start.
SETUP · steps 1–4

One-time · already done

Paid account, desktop app, Code tab, model set. Handled in Tab ⑤. ~0 min tonight.

DESIGN · steps 5–6

Brainstorm with Opus 4.8

Pick the tool, define what "good" looks like. ~25 min of talking. This is where the thinking happens.

BUILD · steps 7–8

Fable 5 does the work

Say "build it." Fable orchestrates sub-agents to construct and test it. ~1–2 hrs, hands-off.

LOOP · steps 9–10

You test, then iterate

Use it, judge it, add features. Ongoing — the good ones live across sessions.

The 10 steps — with a real time on each ~45–60 min hands-on

Setup (1–4) is a one-off you've effectively done via Tab ⑤ — skip straight to Step 5 tonight. The times below are your active input; Fable's build time runs on top, in the background.

STEPS 1–4 · ~0 MIN TONIGHT (one-time setup)

Paid account → desktop app → Code tab → Opus 4.8

Get the paid Claude plan (the $100/$200 tier gives far more tokens for a build weekend), install the desktop app, open the Code tab (not chat/Cowork), and set the model to Opus 4.8 for the thinking part.

✔ Done once — covered in Tab ⑤
STEP 5 · ~15–20 MIN · BRAINSTORM

Think it through with Opus 4.8

Run the prompt below. Opus proposes 10–20 tools/agents worth building for his practice, ranks them, and reasons out loud. He picks one.

✔ A ranked shortlist + one chosen tool
STEP 6 · ~10 MIN · DEFINE "GOOD"

Decide what success looks like

Still in Opus: what does it do, how should it feel, how fast, and how will he know it worked? Lock the spec before a line is built.

✔ A clear success bar
STEP 7 · ~20–60 MIN (Fable works) · BUILD

Switch to Fable 5 → "build it"

Change the model to Fable 5 and tell it to build. Fable acts as the construction manager — spinning up agents and sub-agents to actually assemble the tool. He doesn't babysit this.

✔ A first working version
STEP 8 · ~10–15 MIN (auto) · TEST

Let cheap sub-agents test it

Have smaller, cheaper models — not Fable 5 — run the testing, so the build doesn't burn premium tokens. Runs largely on its own.

✔ Tested without draining tokens
STEP 9 · ~15 MIN · HIS JUDGEMENT

He stays in the loop

He tries it himself. Is it actually valuable? Does it work the way a real case needs? What's missing? No AI replaces this call.

✔ A human verdict on real use
STEP 10 · ONGOING · ITERATE

Iterate, iterate, iterate

Feed back what's off, add the next feature, refine. The tools that matter get sharper over several passes — often across sessions, not one sitting.

✔ A tool that improves every week
Why this maps to Tab ②: the "what should I build?" answer is exactly the Command Center — a case & client-prep tool. This tab is how you build it; Tab ② is what good looks like for his insurance work. Run them together.

The prompt from the video copy · paste · run in Opus 4.8

This is the exact prompt Allie shows on screen (reconstructed from the video). Run it first, verbatim, to brainstorm — then use the tailored insurance version underneath to keep it on his world.

Step 5 — verbatimAllie's brainstorm prompt
Based on everything you know about me and anything I've shared, take a careful look
at my life as it stands right now. Consider my goals, my routines, the
responsibilities I carry daily and monthly, and the things I keep intending to do
but never quite get to. Weigh each one by its difficulty, its importance, and who
it affects. Then generate 10-20 AI tools or agents that would be genuinely high
value for me. Some should quietly take on the repetitive or draining work I already
do, and others should be entirely net new: capabilities I'd never be able to hire a
person for, but that would be epic to have if they existed. Evaluate all 10-20
against a clear rubric of impact, effort to build, and how meaningfully each would
change my week, then rank them and walk me through your reasoning. Once I've shared
the rankings, stop and let me know which are worth pursuing. After I decide, give me
a focused build plan, ask me any questions, and then I will use Fable 5 to build
each step.
Step 5 — tailored to himSame prompt, aimed at his practice
Based on everything you know about my work as a [life insurance advisor
specialising in HNW estate protection in Canada], take a careful look at my week as
it actually runs. Consider my goals, the sales and admin I carry daily and monthly,
and the things I keep meaning to systematise but never get to (case prep, follow-up,
objection handling, estate scenario modelling, referral nurture).

Weigh each by difficulty, importance, and impact on my pipeline. Then generate
10-20 AI tools or agents that would be genuinely high value for MY practice — some
that quietly take on repetitive work I already do (e.g. drafting client summaries,
prepping a case, chasing follow-ups), and some entirely net new (capabilities I
couldn't hire for but would be a real edge).

Evaluate all 10-20 against a clear rubric of impact, effort to build, and how much
each would change my week. Rank them and walk me through your reasoning. Then stop
and let me pick. Once I decide, give me a focused build plan and ask me anything you
need — then I'll switch you to Fable 5 to build it.
Step 7 — hand off to buildTell Fable 5 to build it
I've decided to build: [the tool we picked, e.g. an estate-value case
analyzer].

You now have the build plan and my answers above. Build it as a working first
version. Act as the orchestrator: use sub-agents to write and assemble the pieces,
and use smaller/cheaper sub-agents (not yourself) to test it so we don't burn
tokens. When it's ready, tell me how to run it and what you'd want me to check.

Keep it to the minimum that's genuinely useful this week — we'll iterate after I've
used it on a real case.
Token reality: Fable drains ~2× faster than Opus, so use it only for the one big bounded build — not the brainstorm (that's Opus) and not the testing (cheap sub-agents, Step 8). Most weeks he won't touch Fable at all. See the Fable cheat sheet for when it's actually worth it.

$ Which Claude plan? the honest answer: probably not $100

The video pushes $100/$200 to use Fable. But independent deep research says Fable is for one big bounded build, not day-to-day — and it drains tokens ~2× faster than Opus. So for him, the honest call is usually stay put.

PlanCost / moWhat it's forVerdict for him
Pro$20Runs Opus 4.8 — the right tool for ~90% of his week (case prep, emails, the board, the $1,000 prompts).Genuinely enough for day-to-day.
Max 5×$100~5× usage + real Fable headroom. Only pays off if he does one big Fable build in the window.Only if he commits to a real build before July 7.
Max 20×$200~20× usage — heavy back-to-back agentic building.Not for him.
The honest take: no need to rush into $100. If he has one ambitious, bounded build he'll keep (the case analyzer, or cloning a paid app to kill a subscription), upgrade for the window, do it, drop back. If not — stay on $20, run everything on Opus, revisit when he has a real target. From July 8, Fable is metered pay-per-use anyway. See the Fable cheat sheet for the full honest breakdown.

Fast win — his "$1,000 prompts" ~5 min · instant value

Before any building, run this. It surfaces the handful of prompts that would each be worth serious money to his practice — a fast, tangible payoff — a quick win that leads naturally into the bigger builds in this tab.

Run first · in Opus 4.8Find my personal $1,000 prompts
Act as a senior AI strategist who understands high-value sales work. Using
everything you know about me and my business — my exported ChatGPT history, anything
I attach, and what you can infer about a [life insurance / HNW estate protection]
advisor's week — find my personal "$1,000 prompts".

A $1,000 prompt is one that, if I ran it regularly, would be worth roughly $1,000+
to me: it wins or saves a deal, removes hours of work, sharpens a client
conversation, or unlocks something I couldn't do alone.

Do this:
1. Map where the real money and time sit in my work — prospecting, discovery, case
   prep, objections, closing, follow-up, referrals, admin.
2. Generate 8-12 candidate prompts I could run against MY situation. For each: the
   exact prompt text I'd paste, the job it does, and roughly why it's worth $1,000+.
3. Rank them by value-per-effort and tell me the top 3 to start using this week.
4. For the #1, write the full, ready-to-run version tailored to me — no placeholders
   I have to guess at; ask me at most 2 short questions if you truly need them.

Keep it concrete and specific to insurance sales — no generic "write a blog post"
filler. I want prompts a top advisor would guard.
Where context plugs in: if you feed his ChatGPT export (or notes from past sessions) into this prompt as context, the output gets sharply personal — it stops guessing and starts naming his actual clients, products and bottlenecks.
Let the board decide what to build: he already has his /ask-the-board skill from Tab ①. Once this prompt (or the Step 5 brainstorm) gives a shortlist of tools/prompts, run it past the board — "Board, which of these should I build first and why?" His cloned advisors weigh in, the devil's-advocate stress-tests it, and he picks with conviction. The board isn't just for sales calls; it's his build-decision engine too.
👥 Approach: a 2-minute video can make the build feel like a 2-minute job, so set the honest expectation up front: "the video's the trailer; the build runs in the background while you're on a client call." Open with the $1,000 prompts for a quick win, then work through the brainstorm (Steps 5–6) together — that's where the guidance adds most — and let Fable do the heavy lifting so progress feels fast. Only look at the $100 plan once the value is clear, not before.

7 Publish It — free, on Cloudflare one-time setup · ~5 min

Anything he builds — a dashboard, a client-facing tool, an AI-optimised profile, even this playbook — can go live on the internet for free at https://his-project.pages.dev. No domain, no credit card, unlimited visitors.

The simple version: the only thing he does by hand is grab two Cloudflare keys (5 minutes, all in the browser). After that he never touches a command — he just tells Claude "deploy this with the pages builder" and it publishes for him. Set it up once tonight and it's done forever.
STEP 1

Free account

Sign up at dash.cloudflare.com — no domain, no card.

STEP 2

Account ID

Copy it from Workers & Pages (or the dashboard URL).

STEP 3

API token

One permission: Pages → Edit. Copy it once.

DONE

Hand to Claude

Save the two values; Claude deploys from then on.

Get his two keys browser only · Windows-fine

Both values come straight from the Cloudflare website — nothing to install for this part, so his Windows setup is no obstacle.

STEP 1 · CREATE ACCOUNT

Free Cloudflare account

Go to dash.cloudflare.com/sign-up, confirm the email. Skip adding a domain or card — stop at the dashboard.

✔ Logged into Cloudflare
STEP 2 · ACCOUNT ID

Copy the Account ID

Left sidebar → Workers & Pages. The Account ID is on the right (a long letters-and-numbers string). Or grab it from the dashboard URL: dash.cloudflare.com/THIS-LONG-ID/…. Copy it somewhere safe.

✔ Account ID saved
STEP 3 · API TOKEN

Create a Pages token

Go to dash.cloudflare.com/profile/api-tokensCreate Token → bottom, Custom tokenGet started. Set one permission row: Account → Cloudflare Pages → Edit. Account Resources → Include → his account. Continue → Create. Copy the token now — it's shown once.

✔ Token copied (treat like a password)
STEP 4 · SAVE THEM

Give the two values to the builder

In the agostino-pages-builder folder, run bash setup-creds.sh and paste each when asked (it saves & tests them). Or paste them into a cloudflare.env file. One time only.

✔ Builder now publishes to his account
Security: that token can only manage Pages sites — it can't touch billing, DNS, or anything else. Still, treat it like a password and don't paste it anywhere public. Lost it? Just make a new one.

Publishing a site from then on

Once the keys are saved, going live is one line — or just ask Claude to do it.

TerminalPublish a folder of files
bash scripts/deploy.sh my-first-site ./example-site

…and it prints a live link like https://my-first-site.pages.dev. Or, in Claude Code, just say:

Or — hand it to ClaudeNo terminal needed
Build me a [simple dashboard / one-page site] for [what it's for], then deploy it
live with the pages builder in the agostino-pages-builder folder. Give me the
public pages.dev link when it's done.
Windows note: the two keys are browser-only, so Step 1–3 are painless on Windows. To run deploys himself he'd need Node.js (nodejs.org, LTS) and Git Bash — but the smoother path for him is to let Claude Code do the deploy, which sidesteps the terminal entirely. Cost stays £0: the free tier is 100 sites, unlimited visitors.
👥 Setup note: the full kit is ready — cloudflare-builder-package/agostino-pages-builder/ (credential guide, setup-creds.sh, deploy scripts, SKILL.md, example site). Walk through the 5-minute key grab together, then do one example deploy so a live URL appears — a great milestone — and he owns publishing from then on. His credentials stay on his machine, kept separate from any other Cloudflare account.

8 How to Use It the everyday basics

The plain-English "which button, what happens" guide — so he can drive it himself between our sessions. No jargon, nothing to memorise.

📸 Starter version. These are the universal steps — your exact screens are being added to each one, so it shows precisely what you see on your machine.

A Open Claude the right way the two things that matter

1 · THE TAB

Pick Code, not just chat

Top-left of the Claude desktop app there are tabs. Chat/Cowork is for talking; Code is where you actually build things and save files. For everything on this playbook, click Code.

✔ You're in the build workspace
2 · THE MODEL

Choose the right brain

There's a model dropdown. Two you'll use: Opus 4.8 for ~everything (thinking, writing, day-to-day), and Fable 5 only for one big build (see the Fable cheat sheet). When in doubt → Opus 4.8.

✔ Opus 4.8 selected for normal work

B Run a prompt & feed it context how to actually get value

1 · PASTE & SEND

Copy a prompt, paste, hit enter

Every prompt on this site has a Copy button. Copy it, paste into the box at the bottom, replace the [highlighted bits], send. That's it.

✔ Claude starts working
2 · GIVE IT YOUR STUFF

Attach files or your ChatGPT export

Drag a file (or his exported ChatGPT folder) into the chat, or use the attach button. Claude reads it — that's how it stops guessing and gets specific to him. Screenshots work too: paste an image and it reads it.

✔ Claude answers about HIS world
3 · TALK, DON'T TYPE

Use voice for the long answers

When Claude asks him questions, he can dictate instead of typing — roughly 10× faster and less of a chore. Great for the "tell me about your business" answers — quicker to say than type.

✔ Faster, less friction

C When you hit your session limit what just happened today

On the $20 plan there's a cap on how much you can do in a rolling window. Hit it during a big session and Claude pauses you for a while — totally normal, nothing broken, nothing lost.

  • It resets on a clock — usually about an hour or two. Your work and files are safe; just pick up where you left off when it lifts.
  • Mid-build when it hit? No problem — the files already written stay on your computer, and any page that already deployed is still live on Cloudflare. A half-finished project just means "finish it next session" — nothing rolls back.
  • Right now: take the break, then continue. Nothing to fix.
  • If it keeps happening during our build sessions, that's the one real case for the $100 (Max) plan — far higher limits — but only worth it while actively building. Day-to-day, $20 is fine. See the plan breakdown.
  • Stretch your limit: stay on Opus for normal work (Fable burns through it ~2× faster), and let the cheap testing run on smaller models.
Today in plain terms: nothing went wrong — an intense co-building hour on $20 simply hits the ceiling. Wait for the reset (~1 hr) and carry on.

D Save & publish your work so it doesn't disappear

  • Files save automatically in the folder you're working in — Claude Code writes real files to his computer, not just chat messages.
  • To put something online (a tool, a dashboard, a site), use Tab ⑦ · Publish It — one line, or just ask Claude to deploy it.
  • Golden rule: if something works and he likes it, tell Claude "save this so we don't lose it" — it'll write it to a file.
👥 Note: this tab is his between-sessions confidence — one screen + one screenshot per step. The "session limit" section is deliberately reassuring: hitting a wall with no explanation is frustrating for anyone, so framing it as normal, with the reset clock, keeps it stress-free.