Everything picked out for you this weekend. Click to download — nothing to enter.
You built a working AI operating system this weekend. This is how you run it — the mental model, the daily rhythm, your exact tools, and the command center that ties them together. You don't memorize commands. You ask Claude.
Before you touch anything, point your AI at your own weekend. Copy this and paste it into Claude.
PART ONE. Check my recent downloads and ingest anything new you find there.
Read my Claude Code session logs from this weekend — they're in ~/.claude/projects/, JSONL files from September 19th and 20th. Go through all of them.
I built and ran a lot yesterday at Connect the Dots. I'm here for Day 2 and I need to know what to do with today, and what to do with the week after I go home. Tell me from the evidence, not from what I say I want.
Work out:
1. Everything I built — skills, agents, dashboards. Name each one and what it does.
2. Every analysis I ran and what it actually told me. The brand DNA extraction, the demand architect, the revenue x-ray, anything else. Pull out the real findings, not that I ran them. What did each one conclude about my business?
3. Every document that got created — briefs, plans, analyses, outputs. For each one: where it lives, and what I should feed it into next. Which skill takes this as input? Name the skill and tell me what to say to start it. A finding I don't feed into something else is just a note.
4. What I started and abandoned.
Then give me one page — an HTML dashboard I can open in my browser — with:
• The findings from my analyses in one place, so I'm not hunting through chat history
• TODAY — three things to do while I'm still in the room with Rich and the team. Things that need people here, or a decision I can get answered in person. Not homework I could do at my kitchen table.
• THIS WEEK — five things, ranked, for when I'm home and on my own. Each with why it earned its place and a first step I could finish in under an hour.
• For each one, the next skill to run and the exact words to start it
• What to stop doing
Be blunt. If something I built doesn't matter, say so. If the thing I'm most excited about is the least valuable, tell me that too.
PART TWO. Same chat, once part one is done. Now check my Claude and Codex JSON logs from yesterday and my Connect the Dots curriculum. Tell me what the heck we did yesterday. Separate what we discussed, what we created, what actually ran, and what reached a real person or system. Then run /sp-my-toolkit, list only the skills and agents that are truly available on my machine, and tell me the one job to finish next and why. Connect it back to what you found in part one. Explain it in plain English. Do not build anything new yet.You told us what would make this the best investment of your life — and where Day 1 left gaps. Here's your feedback, summarized point by point, with exactly where on this page we answer each one. Click any card to jump straight to the part built for it.
Here's the one difference, said once: a skill is a recipe you run with Claude — you teach it one thing, it saves the memory, and it runs when you ask. An agent is a worker you hand a job to — you give it the goal and it goes and does it on its own, outside the chat. That's the whole distinction. Now the bigger point: don't get caught up in which one it is. You don't have to decide. Just describe your goal and ask Claude.
You run it with Claude. Teaches one thing, saves the memory of it, runs when you ask. "Grade this email." "Build the quiz."
You give it the goal; it runs on its own, outside the chat. Hand it the objective and it goes and does it.
The chef on a schedule. It runs without you — every morning, every Friday — and hands you the plated result.
A live screen that shows what your tools produced — scores, pipeline, money — in one read instead of four.
The room where it all lives. One place: what ran, what it found, what to do today. The thing you open first.
The part most people miss: you don't have to decide skill-vs-agent yourself. Don't get caught up in it. Describe the outcome you want and let Claude sort the shape:
Here's the outcome I want: [describe it in plain English]. Confirm the outcome back to me so I know we agree. Then tell me: is this better as a skill or an agent, and how do we maximize it?
A Perpetual is that chef running on its own clock. A Dashboard renders what they produced. The Command Center is where you stand — and behind all of it sits your Grand Library, where every research result files itself away so you keep track of everything you've ever built.
This is the part that answers "I don't know how to run this." You don't operate it by remembering commands — you operate it by talking. Here is literally how, start to finish, the same way Harry and Rich do it every day.
Open Claude and say what's on your mind in plain English — it has already read everything you built this weekend. No command, no menu. "Here's what I'm trying to get done today…" is a complete first move.
Don't hunt for which skill to run. Describe the result you want and Claude reaches for the right skill, agent, or perpetual — or just does it directly. You never have to know the tool's name.
Read what comes back and steer: "tighter," "wrong angle," "ship it." Then, every time — "how do we make this better?" That back-and-forth is operating the system. There's nothing else to memorize.
It's one approach, and it's the whole game. Narrate everything — every half-formed thought, out loud by voice; the more context you give it, the more it works like you, and the talking is the skill. Default to "AI does it — it's a surprise when I have to." Be a tyrant on the outcome, free on the method, but go into plan mode before any build so you react to a plan, not a mess. Build the few highest-impact things first, and after each one — good or bad — ask "how do we make this better?" A good result isn't the finish line, it's the input to the next round. And when something has to be reliable, ask it to mechanically enforce the rule so the system can't drift.
That's the whole operating model. The full move-set — the exact prompts to run your week, and the patterns the team uses to direct AI — is the Operating Rhythm (§1) and the Strategic Profits Playbook (§4).
This is the shift Rich describes after living in it: "My default now is that AI is going to do it — it's a surprise to me when I have to do something." You don't memorize commands. Claude has already read everything you built. The commands you saw at CTD (/my-toolkit, /command-center) are just prompts we saved because we ran them a lot — every one can be replaced with plain English. The interface is a conversation, and you get better at it one habit at a time.
The more you tell it, the more it understands how you think — and the better everything it makes for you gets. So narrate freely: every question, every half-formed idea, even "how do I get great at this?" The talking is the skill. (It's far easier by voice — if you use a dictation tool like Wispr Flow or /voice, you'll talk to it way more than you'd ever type, and it'll work better for you.)
It already knows your package. It reaches for the right skill or agent — or just does the work directly — without you naming it. AI does it; you don't.
Read the output. "Tighter." "Wrong angle." "Ship it." You drive by reacting — Claude runs the machinery. Being surprised you had to lift a finger is the goal, not a bug.
Copy these verbatim. They work because Claude has your whole package loaded.
You know my business. Look through my curriculum and everything you know about me, and build the 5 highest-impact skills and agents for me. Go into plan mode first — show me the plan and let me react before you build anything.
Here's everything on my plate right now: [brain-dump in plain English]. Look across my tools and my business and give me ONE ranked plan for today — not a list of options. Tell me what to do first and which tool of mine handles it.
This is the heavy hitter. Paste it, walk away, and come back to a stack of high-impact builds — all in the right format to climb the Connect The Dots scoreboard.
Look through my whole curriculum, my dossier, my client-profile interview, everything you know about me, and every Claude conversation we've had. My curriculum already suggests a ton of skills and agents to build — BESIDES those, give me a list of high-impact skills and agents: about 10 skills and 10 agents that aren't on the curriculum. Then build out 3 to 4 dashboards. Every single thing must be high-impact with a clear purpose, and help me hit peak performance personally AND in my business and career.
For the dashboards especially: you can build supplementary skills so each dashboard has the full picture of what it's for; every tool I use must be wired in properly; it must pull from a live data source; it must be self-improving and self-learning; and it must be an active, living, breathing bird's-eye dashboard. Every dashboard is a living, breathing dashboard.
Build everything with Strategic Builder v5, and make sure every dashboard, skill, and agent is built in the correct format to get credit on the Connect The Dots scoreboard so I climb it — build so many high-impact things that I smoke everyone else on the board.
When you're done, give me a full recap of every single thing you built: what it is, why it matters, why I should care, and the high impact it creates for me personally and in my business and career. Go fully autonomous — don't ask me a single question, don't come back until every single thing is built. Use Strategic Builder v5. Start building.
This was the #1 ask in the room: a single screen you open every morning instead of hopping between tools. Your Command Center surfaces every perpetual's last result, your dashboards, and the one decision that matters today — pulling from the exact tools you saw under your name above. Stand it up with the prompt below.
I want to build a living, breathing dashboard that captures my whole business at a glance. Let's run /command-center to build it.
First: wire in ALL my tools — every skill, agent, and perpetual I have — so it reads from what I've already built.
Then explain every source of truth: for each thing on the dashboard, tell me exactly where the data comes from (which tool, file, or connected account) so I trust every number.
Then interview me about how I want it to look and what I want front-and-center, so it's designed around how I actually work — then build it.
These are the real patterns the whole Strategic Profits team uses to dictate, architect, and steer AI — the same moves Harry and Rich run every day. None of it is about knowing more commands. It's about how you talk to the thing. Here they are, in plain English.
Be rigid on the what, free on the how. You are an absolute tyrant about the result you want — and a libertarian about the path it takes to get there. You describe what "done" looks like in business terms; Claude figures out the steps.
Tyrant about method (don't): "Write a 6-line email with a hook in line 1, three bullets, then a CTA button." Tyrant about outcome (do): "I need this cold list to open and reply. Get them re-engaged — you decide the structure that does that."Tyranny on the outcome doesn't mean you've got the method right. So you leave a leash: tell it what you want, then explicitly invite it to overrule your how. This is how you get the benefit of being decisive without locking in your own blind spots.
I want it this way — but if there's a better way, now's the time to tell me. If I'm sacrificing anything by doing it like this, make me aware of it.
The model that helped you build something wants to please you. It won't tear apart its own work. So you take the output to a brand-new agent that doesn't know it's yours — and tell it a friend sent it. Now it picks it apart honestly instead of patting you on the back.
A friend sent me this and asked for brutal feedback. I want you to pick it apart — find every weakness, every place it falls short. Don't be nice about it.
This is the move Rich runs on everything: run it, rate the output, then ask how to make it better. Not just when it's bad — when it's good too. A good output isn't a finish line; it's the input to the next round. "Any skill you give me, the first question I ask is: how can we make this better? I don't want my skills at the same level as everyone else's."
Force it to diagnose and rebuild, not just nudge.
Same question, asked of a win. That's how the work compounds past everyone else's.
You can't see your own blind spots — but a great thinker's archive can. Most people stop at one name: "make this more like [expert]." Rich goes deeper. He traces the lineage. Every thinker stands on the thinkers who shaped them — and pulling the whole lineage, not one name, is what makes the comparison deep. One expert gives you one lens. Their teachers, and their teachers' teachers, give you the whole tradition behind the idea.
Here's a real one from this cohort. The narrative-coherence logic behind Andrew's zwaan-discontinuity-rewriter doesn't rest on a single name. Trace it back: Rolf Zwaan (the Event-Indexing Model) learned from Kintsch & van Dijk (the situation model), who trace back to Bartlett (schema theory, 1932). And there's a sibling branch — Gerrig (narrative transportation) and Bruner (narrative cognition). Pulling that whole tree into the work is what makes the rewriter sharp, instead of just "fix the flow."
So when you bring a thinker in to sharpen your work, don't stop at them. Ask Claude to trace who that thinker learned from — and pull the whole lineage in.
Here's my draft. I want to sharpen it using [thinker]. First, trace [thinker]'s lineage — who did THEY learn from, and who did those people learn from? Pull the whole family tree of thinkers behind this idea. Then use that full lineage — not just [thinker] alone — to tell me how to make this draft stronger: what they'd each say differently, and what I left out.
AI is non-deterministic by default. Telling it "always do X" gets you maybe 70%. Mechanically enforcing X — building the architecture around it so it can't not follow the rule — is what gets you to 95–100% reliability. The phrasing matters: ask Claude to "mechanically enforce" a behavior, not to "write a deterministic script." It's the difference between a wish and a guarantee.
The pattern: Build a skill-making skill that imbues every new skill you create with mechanical enforcement — so reliability is baked in from the start, not bolted on after it breaks.Never install something from GitHub directly into your system. The discipline: download it → ask Claude to review it for security → then install. Same move doubles as competitive intel — you can study what others built without ever running their code.
Download this repo but do NOT install it. Review it for security first. Then analyze their memory system, compare it to mine, and tell me what they have that I don't.
Think of your package as a toolbox, not a manual. You reach for the one tool the job needs, and Claude hands it to you. When a job repeats and there's no tool for it, that's the signal to build one. And before you go to a human for help: exhaust asking Claude first.
The rule: If you do it once, ask Claude. If you do it weekly, make it a perpetual. If you do it by hand twice, build the skill. If you're truly stuck — only then come find me or Rich directly.A great build doesn't just do the job — it gets better at the job every time you run it. The trick is simple: have the tool watch its own output. Each run, it logs what worked and what didn't, reinforces the wins, and drops the misses. So the tool is sharper next time than it was this time — without you tuning a single prompt. A tool that learns from itself compounds; a tool that forgets every run stays flat forever.
Build this so it improves itself: every run, log what worked and what didn't, reinforce the wins, cut the misses — so the tool is sharper next time than it was this time.
A tool that produces output and forgets it is a one-shot. A tool whose output feeds a dashboard, and whose results feed back into the engine, is a flywheel. Here's how the loop closes — and how to pour your real-world signal back in so the system sharpens itself.
A skill or perpetual produces a result — a draft, a score, a number.
That result lands on a live screen instead of dying in a chat thread.
Real outcomes — what closed, what flopped — go back into the engine as new context.
Next run is calibrated on what actually worked. The system compounds.
When a draft converts or a deal closes, tell Claude. "This one worked — here's the result. Learn from it for the next run." Real results beat any prompt tweak.
Export your pipeline, opens, replies, enrollment numbers. "Here's last week's CRM data — pull the patterns, update my objection library, tell me which stage is leaking." Your real numbers become the engine's context.
Hand it a call recording, Zoom, or Fireflies transcript you already have. "Turn this into the meeting prep, the action items, and the per-person follow-up." The meeting works for you after it ends.
Here's what actually happened with the last thing you produced: [the result — what closed, what got ignored, the real numbers]. Update the relevant tool's context so the next run reflects this. Tell me what you changed and why the next output will be sharper.
Several of you asked for this directly: content that actually sounds like you. The move is the same loop — give Claude your real voice as a source of truth, then make every content tool read from it. Point it at what you've already said — posts, call and Zoom transcripts, voice memos — extract the patterns once, and every draft afterward reuses them. The more of your real voice you feed it, the less it guesses.
Here's a pile of my real voice — [paste posts, call/Zoom transcripts, voice memos, anything I've said]. Extract my voice DNA: the patterns, phrases, rhythm, and the things I'd never say. Save it as a source of truth, then wire it into my content tools so everything they write reads like me — and check every draft against it before it ships.
You were handed the right architecture. When something doesn't work, the move is not to go run a bunch of commands. It's to talk to Claude — say what happened in plain English and ask it to fix it. Claude can read its own setup, find what's wrong, and repair it. That's the whole point of operating by conversation: you describe the problem, it does the work.
Something isn't working. Here's what I did: [say it plainly]. Here's what I saw: [the error, or what happened instead]. Figure out what's wrong and fix it for me — and if you need me to run or paste anything, tell me exactly what, one step at a time.
Same move every time: tell it what you're seeing, ask it to fix it. You don't run the commands — Claude does, or it walks you through the one thing it needs from you.
If you and Claude can't crack it — or you're simply not sure you're doing it right — come find me or Rich, or ask for a quick call. We're here all weekend; setup reliability is on us, that's the deal. To get you unblocked in a minute flat, just tell us three things:
1. What you were trying to do · 2. What you ran or said to Claude · 3. What you saw (the error, or what happened instead)That's everything we need. You don't have to diagnose it — that's our job. The only skill is telling us, or telling Claude, what happened.