# the 4-layer claude stack ## install guide · 12 minutes · live by eli hitz · the closer method · closermethod.com --- ## what this is four claude features wired in a specific order. once they're connected, every chat ships a real file instead of disappearing as a message. that's the difference between renting AI and owning the output. the stack: ``` 01 PROJECTS → memory 02 SKILLS → instructions 03 ARTIFACTS → files 04 DEPLOY → revenue ``` this is the architecture i run my whole business on. brand outreach, content, pricing, course outlines, every email i send, every page i ship. the build below is multiple sessions of focused work if you've never done it. by the end, your business runs on this. --- ## LAYER 01 · projects = memory the problem this solves: every claude chat starts blank. you re-explain your business 40 times a week. by friday you're tired of typing the same context paragraph. claude projects fix this. one project per business. the project remembers everything you load into it across every chat inside it. ### setup 1. open claude. left sidebar → "projects" → "new project" 2. name it the business name (not "marketing", not "ideas" the actual brand) 3. in project instructions, paste your **brain doc** ### the brain doc a brain doc is a single paragraph that tells claude who you are, who you sell to, what you sell, what your voice is, what's off-limits. write it once. every skill reads from it. example structure: ``` i'm [name]. i run [business]. i sell [offer] to [audience]. my niche is [specific niche]. my audience pain is [pain]. my voice is [voice rules]. banned words: [banned]. my products: [list with prices]. my links: [stripe, instagram, etc]. ``` write your brain doc in plain text. paste into project instructions. save. now every chat inside that project knows who you are without you re-explaining. **checkpoint:** open a chat inside the project. ask "what business am i running?" claude should answer from the brain doc. if it does, layer 01 is live. --- ## LAYER 02 · skills = instructions the problem this solves: prompts are temporary. every time you want a landing page, you rewrite "act as a copywriter, build me a..." from scratch. by year-end you've typed 600 versions of the same setup. claude skills are reusable prompt instructions saved as `.md` files. claude reads them automatically when the trigger phrase appears in a chat. one install, infinite use. skills follow anthropic's official spec: https://agentskills.io ### the skill.md format a SKILL.md file looks like this: ```markdown --- name: brand-voice description: writes any copy in the user's locked brand voice. fires on phrases like "in my voice", "make it sound like me", "rewrite this in my tone" --- you are a brand voice engineer. your job is to extract the user's voice from their brain doc and rewrite anything they give you in that exact voice. clarifying questions (ask all 3 before writing): 1. what's the piece i'm rewriting? 2. what's the format (caption, email, headline, dm)? 3. who's it for (audience reminder)? output: rewrite the piece in the user's voice. flag any line that breaks their banned phrases. return only the rewrite, no preamble. banned: em dashes, "unlock", "game changer", "transform", "here's the thing" ``` every skill is the same shape: frontmatter (name + trigger description) → role → clarifying questions → output spec → constraints. ### install methods **claude pro (recommended):** - settings → capabilities → skills → upload `.md` file - enable - done **claude free:** - open the SKILL.md file in any text editor - copy the entire contents - paste at the top of any new chat - the skill is active for that chat session - 18 of 26 skills run within free claude limits ### the 3 starter skills i'm including 3 SKILL.md files with this guide: 1. **brand-voice.md** rewrites anything in your voice once your brain doc is loaded 2. **brand-pitch.md** drafts cold outreach to brands using your enterprise sales structure 3. **lead-magnet.md** builds a free freebie page for capturing emails on any topic drop them in your claude project at layer 01. they read your brain doc automatically. **checkpoint:** in your project chat, type "brand pitch to AG1 for a creator with 21k on tiktok". the brand-pitch skill should fire. it asks 3 clarifying questions, then ships a real pitch using your brain doc context. if it does, layer 02 is live. --- ## LAYER 03 · artifacts = files the problem this solves: claude's response in a chat window is fragile. you can't share it as a link, deploy it, edit it later, or hand it to a client. it's locked inside a conversation. artifacts are real files claude produces. html, css, js, markdown, pdf, png, mp4, code. they live on your computer. you can rename them, edit them, deploy them, sell them, send them. ### how artifacts come from skills every skill in layer 02 should produce a layer 03 artifact, not a chat reply. that's the rule. example: ❌ bad skill: "write me a landing page for x" → claude prints html in the chat window. unusable. ✅ good skill: "build me a landing page artifact for x" → claude creates `landing-page.html` as a downloadable file. immediately useful. the difference is one word: **artifact**. add it to your prompt or bake it into your skill output spec. ### what to do with artifacts - save them in a folder per project (e.g. `~/closermethod/artifacts/`) - name them with the date if you'll iterate (`landing-v2-may19.html`) - treat them as your inventory. you can ship them to clients, deploy them, edit them, resell them as templates **checkpoint:** ask claude to build a landing page artifact for any test topic. confirm a file downloads or appears in your artifacts panel. open the file. it's real html. layer 03 is live. --- ## LAYER 04 · deploy = revenue the problem this solves: a file on your computer makes you zero dollars. a file on the internet with a payment button on it can make you money tonight. deploying = making your artifacts publicly accessible at a real URL. plus wiring a way to take payment. ### the 3-tool deploy stack **netlify** (hosting · free) - go to netlify.com, sign in with github or email - "add new site" → "deploy manually" → drag your html artifact into the box - you get a URL in 30 seconds - rename the site to whatever you want (e.g. closermethod-stack.netlify.app) **stripe** (payments · free to set up, % per transaction) - stripe.com → create account → payment links → new link - name the product, set price, copy the link - paste the stripe URL into a button on your html artifact - when someone clicks "buy", they pay you, you get the money **manychat** (DM automation · free tier ok for IG and FB, NOT tiktok in many regions) - manychat.com → connect instagram or facebook page - "automation" → "keyword trigger" - set: when someone comments [WORD], DM them this message with a link - replaces hours of manual DM work note: tiktok auto-reply does not work in spain via manychat as of this writing. for tiktok, manual DMs only. for instagram and facebook, manychat handles it. ### the full loop 1. claude project (layer 01) → reads your brain doc 2. you ask, skill fires (layer 02) → produces an artifact (layer 03) 3. you drag the artifact into netlify (layer 04 part 1) 4. you paste a stripe link into a button on the artifact (layer 04 part 2) 5. you set up manychat to DM people who comment a keyword on instagram (layer 04 part 3) 6. revenue flows in while you sleep **checkpoint:** deploy any test artifact to netlify, paste a stripe test link into a button on it, set up a manychat keyword trigger that DMs the netlify URL to anyone who comments "test" on any instagram post. comment "test" yourself. confirm the DM lands. if it does, layer 04 is live and the whole stack is wired. --- ## what you just built a system where every interaction with claude produces a real file you own, hosted at a real URL, with a real payment mechanism, with automated lead capture. you stop typing prompts every day. you start shipping inventory every day. the buyer journey now looks like this: ``` content (you post) → comment keyword → manychat DM (auto) → netlify page (your artifact) → stripe (payment) → revenue ``` this is the same stack ai builders charging $5,000+ per project run on. the difference is you built it yourself. it costs you $0 in software (free tiers cover almost everything) plus whatever ai subscription you already have. --- ## time to install solo honest read on how long this takes: - layer 01 (projects + brain doc): 1 to 2 hours, depending on how clear your business model is - layer 02 (skills): the longest layer. writing each SKILL.md takes 30 to 60 minutes if you're new to the format. you need at least 5 to 10 skills before the system feels real. - layer 03 (artifacts as habit): 1 hour to internalize the pattern. nothing to install, just a habit of saying "build me an artifact" instead of "write me a..." - layer 04 (deploy): 2 to 3 hours, depending on whether you've used netlify and stripe before most builders stall at layer 02. writing SKILL.md files in claude's official spec is the unforgiving part. one wrong frontmatter field and the skill doesn't fire. --- ## self-audit checklist (run this when you finish each layer) ### layer 01 · projects + brain doc · LIVE if: - [ ] you have one project per business in claude - [ ] your brain doc is pasted into project instructions - [ ] opening any chat inside the project, asking "who am i and what do i sell?" returns your real answer with your offer + price - [ ] you have NOT pasted brain doc context manually in any chat this week ### layer 02 · skills · LIVE if: - [ ] you have at least 3 SKILL.md files saved on your computer - [ ] each one has frontmatter (name + description), role, clarifying questions, output spec, banned words - [ ] uploading them to claude pro (settings → capabilities → skills) shows them enabled - [ ] typing the trigger phrase in a chat actually fires the skill (claude asks the clarifying questions) - [ ] the skill output references your brain doc context without you re-explaining ### layer 03 · artifacts · LIVE if: - [ ] your last 5 claude outputs were saved as files on disk, not left as chat messages - [ ] you have a folder per project (e.g. `~/business-name/artifacts/`) - [ ] every file is named so you can find it in 6 months ### layer 04 · deploy · LIVE if: - [ ] you have at least one artifact deployed at a real URL on netlify - [ ] that URL has a stripe payment link wired into a button - [ ] you've sent that URL to at least one real person (yourself counts the first time) - [ ] (instagram only) manychat keyword trigger fires when you comment the keyword if any checkbox is empty, that layer is not live yet. go back and finish before moving on. --- ## the shortcut if you don't want to spend 8+ hours writing every skill from scratch, the stack installer at closermethod-stack-installer.netlify.app is the whole layer 02 pre-built: 26 SKILL.md files in claude's official folder format, a brain doc template with worked examples, and a private install thread with me for your first 7 days. $297. installs in about 12 minutes once your brain doc is filled. i audit your brain doc before the toolkit drops so the skills land already tuned. or build it yourself with this guide. either way, you have the architecture. that part nobody can take from you. --- ship something today. eli @closermethod eli@closermethod.com