🚀 Advanced: Real Domains & Production
Here the “pie” changes shape. You see six complete domains—Tax, Sales, Support, Content, Consulting, Freedom—each with a a different distribution of the 6 layers, and the production patterns that keep an OS alive and safe over time.
How to read: the same six ingredients (layers, in cyan) reappear in every domain—but in different proportions. In the Tax OS, the context reaches ~34%; in Sales, the tools weigh more. The path shows each recipe and how to take the OS to production: secure, backed up, and resistant to decay.
Learning path map
🧾 Tax & Bookkeeping OS
Heavy context, ~34%
📈 Sales OS
Tools + skills at the top
🎧 Support OS
Tickets become playbooks
🎬 Content OS
No-hype title factory
💼 Consulting OS
One folder per client
🗽 Freedom OS & Production
Local, backup, context rot
Detailed content
🧾 Tax & Bookkeeping OS
The knowledge-heavy domain (~34%): a compendium of laws, a pipeline for bank statement PDFs, monthly review, and the skeptical reviewer who chews you out before the accountant does.
The author’s distribution: 34% context, 18% tools, 16% skills, 14% rules, 10% agents, 8% identity.
Shows in practice that the biggest part here is context — not agents.
Context dominates · minimal identity · uneven pie chart.
Scan “the worst audits” + legislation from several provinces to create a compendium for your tax system.
Reversing the worst-case scenario prevents the classification error that turns into an audit.
Audit pains · compendium · gray area vs. black and white.
PDF statements become text/markdown in a folder pipeline to save context window space.
Raw PDFs blow through the context window; converting and synthesizing them is what makes review affordable.
raw → converted → synthesized · token savings.
The skill that takes the statement, categorizes it in a ledger, and prepares it to send to the accountant + spending alerts.
It's the domain's most repeated task — the number one candidate to become a skill.
Categorized ledger · alerts · monthly cadence.
Never mix personal and business; block PII (SIN, account); convert everything to one currency (CAD).
Money and sensitive data call for deterministic hooks, not “soft” rules.
PII guard · single currency · account separation.
A tax strategist and a skeptical reviewer who "chews you out before the accountant does".
Agents are few here (~10%) — they’re only used to judge and review.
Adversarial reviewer · review gate · judgment.
A ready-to-use prompt for Claude Code to build the pipeline for converting PDF statements.
It’s the copy-run that brings the raw → converted → synthesized pipeline to life.
Copy · run · verify the output.
📈 Sales OS
The domain where tools and skills matter most: lead scrapers, custom enrichment, qualification with friction, and anti-spray-and-pray rules.
30% context, 20% tools, 16% skills, 14% agents, 12% rules, 8% identity.
Compared to Tax, the stack is more “tightly coupled”: tool and context lead together.
Tools + context at the top · same shape, new proportions.
Cold email structures, the ICP (ideal customer profile), and lead demographics stored as context.
Without a clear ICP, the scraper collects junk and you burn your reputation.
ICP · cold email · won/lost deals.
Apify Actors (LinkedIn, Google Maps) become a lead database that enters the CRM via CLI.
It's the tools slice (~20%) — the largest after context.
Apify actors · lead database · CRM seamless.
Your own enrichment (instead of paying for Clay) + humanizers that remove the "AI smell" from the copy.
Internalizing the enrichment cuts costs and gives you control over quality.
Custom enrichment · humanize copy · human speak.
Friction points in the skill that force you to rescan the lead until you confirm it’s good.
Don't spend 100 calls on 50 bad leads — friction filters them before the effort.
Intentional friction · verification loop · quality > volume.
SDR + orchestrator + human review; rules: no spray-and-pray, respect opt-outs, brand voice.
Spamming burns your domain and brand; an opt-out becomes a hook that immediately removes the contact from the CRM.
SDR · do not contact · don’t be pushy.
A ready-to-use prompt that qualifies leads against the ICP with friction points and a Yes/No/Maybe verdict.
It’s the copy-run that keeps you from burning calls on cold leads.
Copy · run · check the verdict.
🎧 Support OS
Aggregated tickets become playbooks; triage, drafting, and escalation skills; a skeptical reviewer before the customer; and the hook that requires citing the policy source.
A structure similar to sales: context (tickets) + skills (triage/draft) do the heavy lifting.
Shows how the same format works for a customer service domain.
Tickets as context · response skills · escalation.
Pull Zendesk/HubSpot tickets + help docs; a long thread = a proxy for a complex case.
The longest threads hide the cases most worth turning into playbooks.
Aggregate tickets · long thread = complexity · help docs.
Synthesize long threads into nuggets: "how to handle the client who always brings up X."
A playbook turns a rare case into a repeatable, consistent response.
Nuggets · edge cases · escalation path.
Email triage, draft replies, and rules for when/how to escalate.
Reduces time spent on repetitive questions without sacrificing quality.
Triage · draft reply · escalation path.
A support lead who closes tickets + a skeptical reviewer who checks the response before it goes out.
The review gate prevents an incorrect response from reaching the customer.
Support lead · devil's advocate · double-check.
No promises outside the playbook; a hook that blocks policy responses without a link; the junior only reads the CRM.
An improper promise becomes a legal risk — so it becomes a hook, not a soft rule.
Cite a source (hook) · junior read-only · no overpromising.
A ready-to-use prompt that classifies tickets by urgency/topic and decides whether to respond, escalate, or ask for more information.
It’s the copy-run that organizes the queue before any response.
Copy · run · check the classification.
🎬 Content OS
Competitor reverse engineering, a persona-based headline factory, repurposing engines, and the no-hype rule that prohibits "insane".
A domain with a better balance of context, skills, and rules.
Shows a case where no single layer dominates.
Balance · creative skills · voice rules.
Analyze the top 10 videos in the niche: save the pacing, angle, and storyline as context.
Competitors' patterns inform your own content.
Top 10 · patterns · trend.
Hundreds of titles with AB tests + audience personas giving feedback.
It's the domain's star skill — generates and tests at scale.
AB test · personas · scale.
Reuse content across multiple formats and channels.
Multiplies the reach of every piece you produce.
Repurpose · multichannel · thumbnail briefs.
Producer (idea→publication) + orchestrator (writing/editing via Descript CLI).
Delivers a care package ready for the human editor.
Producer · orchestrator · care package.
Ban hype/overpromising (e.g., "insane") and require a 30–40s anti-formula hook.
Protects the brand voice from falling into AI clichés.
No-hype · 30–40s hook · anti-cliché.
A ready-to-use prompt that generates titles and tests them against audience personas.
It’s the copy-run that brings the title factory to life.
Copy · run · choose the best title.
💼 Consulting OS
One folder per client, playbooks + global wiki, and the adversarial review gate in the client's voice—sometimes with multiple reviewers.
Context still matters, but rules & hooks carry more weight than usual.
Each client requires its own boundaries — hence the weight of the rules.
Heavy rules · per-client context.
Each client is a folder; CLAUDE.md maps “when I mention X, it’s this folder.”
Organizes scattered information into clear drawers.
Folder per client · mapping in the master file.
Client playbooks that roll up into a global playbook; a wiki that triages meetings.
Every business has patterns—the wiki enriches meeting prep.
Client playbook · wiki · standards.
Pre-call prep, proposal/SOW, and audit — always tailored to the objective, never generic.
A generic audit won’t work for consulting; you need the client’s context.
Nuanced pre-call · SOW · audit.
An agent named after the contact, primed with synthesized meeting notes, reviews how the client would react.
Anticipates objections before sending the brief—sometimes it takes several reviewers.
Adversarial reviewer · customer’s voice · objections.
QuickBooks, ledgers, Stripe links in proposals (without PandaDoc), and checking calendar buffer time.
The right connections eliminate billing and scheduling friction.
Stripe link · scheduling buffer · ledgers.
A ready-to-use prompt that creates the reviewer who responds like the client, based on the synthesized meetings.
It’s the copy-run that brings the adversarial review gate to life.
Copy · run · review the brief.
🗽 Freedom OS & Production Standards
Local models for sensitive data, backup across 3 fronts, context rot and the /tldr that updates CLAUDE.md on its own.
Run local models for sensitive information (e.g., birth certificate) — hardware investment.
Some data should never go into a cloud context.
Local · hardware · sensitive data.
The choice between closed, open, and a combination of both models.
The hybrid balances power and privacy.
Closed · open · hybrid.
Private GitHub (with .gitignore for sensitive files) + SSD + cloud.
Three copies protect the OS from any loss.
3 fronts · .gitignore · redundancy.
Define when context gets stale (laws change in months, not overnight).
Knowing the rate of deterioration keeps you from trusting stale data.
Degradation · cutoff · update.
A slash command that saves the summary and updates CLAUDE.md/rules with a human in the loop.
It's the OS's controlled self-improvement over time.
/tldr · self-improvement · human in the loop.
Weekly crons that distill all conversations into synthesized nuggets.
Keep the substrate fresh without manual work.
Cron · synthesis · nuggets.
A ready-to-use prompt that creates the /tldr post-session summary that updates the OS.
It’s the copy-run that closes the production loop.
Copy · run · review the summary.