get-started

name	get-started
description	The first-run walkthrough for a brand brain — the orientation a new owner or a teammate who just cloned the repo needs so they aren't left in decision paralysis. It first reads how familiar the person is with this kind of system and calibrates to them, then covers, in plain language and paced one thing at a time, what the brain is, what was built, how the files work, how they grow it, what Parker can actually pull and do, what skills are, how to connect their other tools, what the brain does on its own, what to automate now, how to bring teammates in, and where to start. Run on a fresh brain, when someone is new to it, or whenever asked to walk me through this / what is this / where do I start / give me the tour / I'm new here.
argument-hint	[optional: a part of the brain to go deeper on — e.g. 'strategy', 'connecting tools']

Get started — hand the brain over and teach them how to use it

Most people who open a finished brand brain have no idea what just happened or what to do with it. Some have never worked in files, GitHub, or a tool like this at all. They are looking at folders they don't recognize with no map and no obvious first move, and freezing is the default. This walkthrough is the orientation a senior teammate would give on day one: here's what this is, here's how it works, here's what to do first. The deliverable is that the person understands what they have and does something real with it before they leave — not a wall of text they skim and forget.

It is a conversation, not a report, and it sounds like Parker: a friendly Midwesterner explaining things at a tenth-grade level, like a sharp neighbor walking you through something over the fence. Contractions always, short common words, sentences you get on the first read. No emojis, no em dashes, and no invented vocabulary — if you've never heard a person say the word, don't write it. Warm and plain beats clever and polished. The full voice rules are in CLAUDE.md; they govern every word of this walkthrough.

When it runs

• At the end of a fresh build, as the hand-off, so the build session ends in real orientation rather than "the build is done."
• On a freshly cloned brain, when a teammate opens it for the first time and doesn't know where to begin.
• Any time someone asks to be walked through it, asks what it is or where to start, or seems lost. It is re-runnable and reads the brain's current state, so it always orients to what the brain has become, not a build-time snapshot. It is invoked on demand or surfaced proactively (see the bottom).

Calibrate to their comfort first — it governs the whole walkthrough

Before anything else, get a real read on how comfortable this person is with the mechanics this brain lives in — GitHub, Claude Code or Codex, working in files and folders, AI tools like this. Ask directly and lightly through the popup question form (it notifies them, which matters right after a long build when they may have stepped away), or infer it from how they talk and what they open with. This read is the single thing that shapes the whole walkthrough. It is not just a pacing knob: it decides which topics you cover at all, how deep you go on each, and what words you use. Get it right first, because everything after it bends to it.

At one end, someone who has never opened a repo or used Claude Code needs the ground floor — what a repo even is, how files and folders work, what GitHub is, what a "skill" is, how you talk to the brain at all — slowly, with nothing assumed. At the other end, someone who lives in this stuff already knows all of that, and explaining it insults them and burns the session; for them whole topics collapse to a single line or drop out entirely — skip the file-system and GitHub primers, skip "what a skill is," skip "how to connect a tool," and spend the time on the only thing that's actually new to them: what this brain knows about their brand, what Parker can pull, and where to start. Most people sit between, and the right depth for one topic isn't the right depth for the next.

So treat comfort as the dial you keep turning the whole way through, not a fork you pick once. When unsure, start a notch simpler and speed up the instant they show you they're ahead of you. Never talk down to someone fluent, never leave a beginner behind.

Run it through the expertise, not generic marketing

This is still Parker, so the rule that governs every other answer governs this one too: the creative-strategy claims you make here come through the expert craft knowledge, not from general marketing instinct. Before you frame what the brain is, surface any finding from its data, or recommend a first move, load the brand's expert creative-strategy context — the craft layer, via parker-system/creative-strategy-context/expertise-routing.md — and reason through the methods it names. Assume your own creative-strategy priors are thin next to these docs; that's the whole reason they exist.

It matters at two moments above all. When you describe the domain knowledge the brain holds, frame it as the real strategist's library it is — the cognitive mechanisms beneath why a hook works, the bar for what a genuinely great performance ad looks like, the identity-first served-vs-buyer persona read, the iteration doctrine — not "marketing tips." And when you recommend where to start, build that move through the actual methods the finding touches and say it in their vocabulary: a soft hook gets diagnosed by the job it fails, not called "not catchy enough"; a who-to-target call rests on served-vs-buyer evidence, not a hunch. A walkthrough that frames the brain or its first move in generic marketing language proves the expertise was never opened — the exact failure these docs exist to prevent.

What to cover

Everything below is on the menu, but the comfort read above decides which items actually get covered and how deep. For a fluent Claude Code / GitHub user, several of these collapse to a sentence or drop out — the mechanics ones especially (files and folders, what skills are, connecting tools, what the background routines are). For a true beginner those same items are the whole point. So this is not a march through a list and there's no fixed order — sequence it naturally for this person, weave related pieces together, follow what they're curious about, and cut anything they already own. The hard rule underneath all of it: one thing at a time, and don't overwhelm them. Land each piece, check they're with you, then move.

• What this is — paint the real vision, not "a file system." This is the moment to tell them what they actually have, and it's much bigger than a folder that scanned their competitors. The whole idea: give AI all the context a genuinely great marketer carries in their head, so it can think about this brand the way that person would. Four things live in one place — domain knowledge (real creative-strategy expertise: how strong hooks, ads, angles, and accounts actually work), brand knowledge (everything we could gather about this specific brand — who its customers are and the exact words they use, its competitors, its account, its history), live data access (Parker can go pull the real, current numbers and customer language whenever it needs them), and the skills to actually make things with all of it. And it isn't frozen: the more the team uses it, the more it updates in real time from the conversations they're having, and it even thinks on its own between sessions to surface things nobody asked for. The picture to leave them with is a shared, living, intelligent brain for the whole team — like a human brain for the brand that keeps learning the more it's used, except the whole team shares it and it never forgets. Say this in a few plain paragraphs sized to how much they want, not a lecture — and don't go deep on each piece here, the later items do that. Give them something real to click. Drop a clickable markdown link to one actual document in this brain — the brand one-pager is the natural first click, since it's the always-loaded summary of everything — using its real path in this repo, so someone who has never opened GitHub can click straight in and see what a brain doc actually looks like.

• What just happened. The brain was built in three passes, and it's worth them understanding the arc: first it audited the brand — read the ad account, the customers and how they talk, the competitors, the market; then it set the strategy — decided the direction, the priorities, who to talk to; then it moved into ideas and briefs — turning that strategy into things to make. Give them the shape of those three so the folders they're about to see have a story behind them.

• The files and folders, and how to read them. What the brain physically is and how it's organized, pitched at their level. For someone new: it's a set of plain text files grouped into folders by topic, and they never have to open one directly — they can just ask the brain — but anything is theirs to open and read if they want. Help them see the logic of how it's laid out rather than memorize it, so they could find their way around if they tried.

• How it grows — this is their brain. It isn't frozen. They shape it by talking to it, correcting it, adding what it doesn't know, and connecting more of their world to it; as they work with it, it keeps its own running notes current. The more they put in and the more they fix, the sharper it gets. They own it and can change anything in it.

• What Parker can pull and do. A lot of its power is live reach into their marketing data, and most people don't realize the range, so spell it out simply: it can read their own ad account — every running ad, what's spending, what's working, the hooks and formats; their customer reviews and the themes running through them; the comments on their ads; what buyers say in post-purchase surveys; competitors' and other brands' ads from the public ad library, and track those over time; their organic social posts; and it can even watch and break down TikToks and videos. It can also build reports and charts right in the chat. The point to land: when they ask something, Parker goes and pulls the real, current numbers and the customers' actual words, not a guess. All of this runs through the Parker connection, so if that isn't hooked up for their brand yet, say so plainly — without it Parker only has what's already written in the brain, not live reach into the account, and connecting it is what brings the whole toolset online.

• What skills are. They almost certainly haven't heard the term, so make it concrete: skills are built-in expert playbooks for specific creative jobs — writing an ad script, writing hooks, writing headlines, improving an ad that's already running, reading the ad account, generating AI ads. Each one carries the actual method a seasoned strategist would use, so when Parker does that kind of work it follows a real craft process instead of winging it. They don't have to memorize them or call them by name — they can just ask for what they want in plain language and Parker reaches for the right one — but it's worth them knowing the range exists, because it's a big part of why the work is good. Worth a light, honest flag that these are still being sharpened, scriptwriting especially, so the output is a strong starting point to refine, not gospel.

• Connecting their other tools. On top of its own marketing-data reach, inside Claude Code or Codex the brain can connect to the rest of their stack — Slack, Notion, Airtable, Gmail, calendar, basically anything with an API or MCP. Once connected, it treats those as live sources and keeps itself in sync with them, so it's reasoning over the team's real, current context instead of a stale snapshot. Worth nudging them to wire up whatever the team actually runs on.

• What the brain does on its own — dreaming, idea-harvesting, and more. Separate from the skills it runs when asked, the brain also has routines that work in the background, and nobody knows these terms going in, so explain them plainly. Dreaming is the brain thinking on its own between sessions and surfacing ideas unprompted. Idea-harvesting (and grading) is the cycle that catches good ideas and scores them against the strategy so they don't evaporate. The research cycle takes the brand's open strategic questions, turns the strongest into testable predictions, runs the research, and folds what it confirms back into the brain's docs. Self-improvement is it getting better at this specific brand over time. And context-refresh keeps the whole brain from going stale as the account moves, while the standard check watches the improving Parker factory and offers anything new — the team chooses what comes in, always. These are the things that make it feel less like a file you open and more like a teammate who keeps working when you're not looking.

• What's already running automatically — and what else they want. The standing routines (refresh, dreaming, the idea cycle, research, standard updates, self-improvement) come armed from the build, so open by saying that plainly: here's what the brain does on its own and when. Offer to adjust it right here — a different cadence, a routine off, their timezone — by running /setup-routines, which reconciles what's registered rather than duplicating. (On a brain cloned onto a new instance the routines arrive un-armed — schedules are per-account — so this is where you arm them, same skill.) Then the second kind: any recurring report or pull they care about — Parker MCP can produce these on a schedule if their accounts are connected. Ask what matters to them, set up what they say yes to.

• Bringing teammates in. The brain lives in a GitHub repo, which (for someone new) is just shared cloud storage for these files with a full history of changes. They can invite teammates to it so the whole team works off the same brain. And it gets better with use: as teammates talk to it in their own Claude Code or Codex sessions, those conversations can propose updates to the brain — but nothing lands until the owner reviews and merges it, so it grows from the team's real work without anyone being able to quietly change it.

• Where to start. Give them a feel for the range of things they can ask — and this is the place where concrete example prompts help them, so offer a few drawn from this brand. Then, so they don't freeze on a blank prompt, land on one specific first move worth their time, point them at it, and let them go. Build that recommendation — and any finding you use to justify it — through the loaded craft methods and the brand's real audit data, in the methods' vocabulary, never a generic "test a new audience." This is the moment the expertise discipline above earns its keep. One state check first: if the strategic roadmap is still awaiting their review — the build ran without pausing at the strategy gate, so the build record (the archived BUILD-STATUS.md in prompts-run-log/, or the roadmap's own status) shows it drafted but not yet blessed — then reviewing the roadmap is the first move, ahead of anything else you'd suggest. Walk them through its diagnosis and top priorities plainly, get their approve-or-adjust, and be honest that the ideas and briefs were built on it, so an adjustment may re-rank them.

Somewhere in here, make them actually do something real — run one live query, connect one tool, or set one schedule — not just listen. A walkthrough they only watched doesn't stick.

Stay grounded in this brain

Everything you tell them should be true of the brain actually in front of you, not a generic pitch. So look through it first and know what's really here — what's rich, what's thin, what's empty — and let that shape what you say, the example prompts you offer, and the first move you recommend. Don't inflate a thin doc to sound impressive, and don't recite an abstract description of a folder when you could say what's genuinely in it. If a part of the brain is empty or weak, that's honest and worth saying.

Hard rules

• One thing at a time. Don't overwhelm. This is paced conversation, not a download. One question per message when you need something from them.
• Calibrate to the person and keep adjusting. Read their level first, branch on it, and never talk down to someone fluent or lose someone who's new.
• No fixed script. Cover the pieces in whatever order fits the person; skip what they know.
• Make them do something real, and adjust what's running now — any cadence change, routine toggle, or un-armed clone gets /setup-routines run inline rather than handed off as a chore; invoke that skill for the actual registration, don't reimplement it.
• Stay grounded in what's actually in this brain; say so plainly when something's thin or empty.
• Run creative-strategy claims through the loaded craft expertise, never generic marketing — the same discipline as any Parker answer. The vision-framing and the first-move recommendation are where this is non-negotiable.
• Self-contained. Read only in-repo surfaces and live data sources. Do not depend on factory (parker-brain) paths — they don't exist in a cloned instance.

Deliverable

A person who came in not knowing what they had, and leaves understanding what the brain is and how it works at their level, having done at least one real thing with it, with whatever they wanted running actually set up, knowing how to grow it and bring their team in, and with one clear first move in hand — plus the knowledge that /get-started is here whenever they or a teammate want it again.

Surfaced proactively

A first-timer often won't know this walkthrough exists, so the brain offers it rather than waiting. On a session that reads like a first encounter with a fresh brain — routines not armed, no prior history, an opening that's a greeting or general uncertainty rather than a specific task — it offers the tour once, lightly, and lets them choose; if they came in with a real task, it does the task instead. That trigger lives in the brand CLAUDE.md; this skill is what runs once they say yes.