# Store System: start prompts

Paste this into your AI after you unzip Store System.

There are two prompts. Paste the first one once: it opens the accounts, installs the store in test mode, makes it yours and proves a test purchase lands, all before anything is live. Paste the second one at the start of every day you want your AI to read how the store is selling and propose what to change. The two field journals in the box have their own start prompts (ship-a-digital-business-website.md and run-meta-ads-with-ai.md); this one sets the order across the parts.

## Prompt 1: setup (once)

```text
I bought the Store System and unzipped it here. It is a complete store for digital products, memberships and physical products, as code I own: storefront, cart, checkout with card and PayPal on the same page, an order bump, one-click upsells with a plain "No thanks", a download library, physical products with shipping by country, stock, an orders page with tracking mails, print-on-demand and personalized products, the emails that bring buyers back, and the tracking ads need. Set it up with me now, in test mode, one step at a time, and show me each change before you make it. Nothing goes live in this prompt.

WHAT IT IS FOR
The parts, in the order they are used:
1. store-kit/: the store. Its manual is store-kit/AGENTS.md; read it before you touch anything.
2. control-dna/: Control Room DNA, the app that shows my store live on my phone (who is on the site, a live feed, a dashboard, phone alerts) and the MCP you ask how the store is selling (real sales next to ad spend, when and where the ad sales happen, where buyers leave the payment and what each person who opened it actually did, which test version wins, what happened to emails that did not buy, how a list mail did, who is on the site right now). It reads the store's tables and never changes the store. Its manual is control-dna/INSTALL.md.
3. control-dna/tools/pay_robot.py: the payment robot. It buys from my store in test mode and checks the purchase reached the database.
4. Two field journals, from the links in my purchase email and my library: Ship a Digital Business Website (launching the site and keeping it selling) and Run Meta Ads With Your AI (paid ads).
5. product-finder/: StoreScout, the tool you use to read what already sells on the big marketplaces (what is trending, prices, review counts, where there is room). Its manual is product-finder/INSTALL.md. FIND-YOUR-PRODUCT.md is the order to use it in.
6. PHYSICAL-PRODUCTS.md: the guide for selling something that ships: print-on-demand, dropship or my own stock, what each costs and risks, and the steps to the first order shipped. Printed books are made with the Ebook System, not here.
7. premium-design/: the premium storefront pack. DESIGN-LIKE-A-STUDIO.md is the design guide (viewer spec, start from stores that already sell, direction card, design system, first screen, pages, motion, the review loop at 390 and 1440 wide); skills/ has the design skills you load for the work at hand (copy-a-winning-store, premium-hero, webgl-hero, product-page, glass-and-depth, typography-and-colour, motion, choreography, interface-polish, mobile-first-store, native-feel, design-review, and the library skills unslop, responsive-web-design, frontend-design). HOW-THE-STORE-USES-IT.md says which store files a design change touches and in what order.

WHEN TO USE IT
Now, once. Again only for a second store or a new version of the kit.

STEPS IN ORDER (START-HERE.md is the order; each part's manual has the detail)
0. If the zip is not here yet and I use the Vibedna connector: list_library, then get_download_link with the slug it shows for the Store System (the link lasts five minutes). Otherwise I download it from vibedna.ai/library.
1. Read START-HERE.md, store-kit/AGENTS.md and control-dna/INSTALL.md in full.
1b. Ask me whether I already have the product I will sell. If I do not, stop the setup here and run FIND-YOUR-PRODUCT.md with me first; come back to step 2 when the product is chosen and made.
1c. Ask me whether I sell something physical. If I do, read PHYSICAL-PRODUCTS.md in full now, help me choose the route (step 1 of that file) before we open accounts, and add its steps 3 to 7 to the setup. A store installed before 1.2.0 runs store-kit/supabase/migrate-1.2.0.sql once before the new code is deployed.
2. The accounts (mine to open): Stripe, PayPal if I want it, a Supabase project, Resend for mail, a file bucket (Cloudflare R2 or any S3) and a host for a Node app. Tell me what each one is for; I sign up and click.
3. Install the store: store-kit/AGENTS.md, "Install, in order", steps 1 to 7. Copy .env.example to .env.local and tell me which line each value goes on (each line says where to find it). I paste them; you never see them. Use Stripe test mode and PayPal sandbox keys only.
4. Make it mine: store.config.ts (brand, colors, fonts, currency, time zone, which features are on) and supabase/products.json (my products). Delete every sample product and sample text. Sample testimonials are marked placeholder and never show; write only real ones, word for word from real buyers, or none. For the look (colours, fonts, hero, product page), start from stores that already sell in my market, follow premium-design/HOW-THE-STORE-USES-IT.md in its order, and show me the direction card before you build.
5. Run it and look at it: open every page on a phone size and a computer size, open the payment window and close it. Show me what you see.
6. Buy from myself, in test mode: run the payment robot (control-dna/INSTALL.md section 13) with the card rail and, if I use PayPal, the PayPal rail with a sandbox buyer. It must say the purchase reached the database, and the library must show the product. If I sell physical products: buy one with an address in test mode, check it lands on /admin/orders under "To pack" with the stock taken, mark it shipped and show me the "on its way" mail, then refund it in the Stripe dashboard and show me the stock came back. For a personalized product, fill its fields by hand in that test purchase and show me the print file the order made.
7. Connect Control Room DNA: run python -m controldna setup with me (I type the PIN myself, never ask me for it in the chat). Its settings go in ~/.control-dna/.env, a file only I fill (control-dna/INSTALL.md section 8 lists every key): DATA_SOURCE=database with my SUPABASE_URL and SUPABASE_KEY, so it reads the tables this store writes. Set THANKYOU_PREFIX to /library, or /next/ if I turned on the upsell funnel, and TEST_EMAIL_PATTERNS to the emails I test with. If I sell plans or memberships, write ACCESS_LOOKUP and ACCESS_LABELS with me so who_is_here can say what each person holds. Add it to my AI app's MCP config (section 11), start the app so I can open it on my phone, and run setup_status until every source I use answers.

WHAT TO CHECK FIRST
- Before you change the look of any page, read premium-design/DESIGN-LIKE-A-STUDIO.md and start from stores that already sell in my market (premium-design/skills/copy-a-winning-store/SKILL.md), then load the skill for the work from premium-design/skills/. Look at every changed page at 390 and 1440 wide after each change (premium-design/skills/design-review/check_pages.py), and run premium-design/skills/design-review/SKILL.md in a separate session before anything goes live.
- That every key in .env.local is a test key until I say "go live".
- Test runs carry ?test-run=1 and test emails match test_email_patterns, so they never count as sales.
- The webhooks point at /api/webhook/stripe, /api/webhook/paypal and /api/webhook/resend (AGENTS.md step 7).

NEVER
- Never place an order with a print-on-demand partner or a supplier, confirm a Printful draft, or type my card, without my explicit yes in this session.
- Never ask me for a key in the chat, never print or summarize .env.local or config.json, never commit either. Keys live in those files or in my vault, nothing else.
- Never switch to live keys, deploy, publish the site, turn on the scheduled mail jobs, change a price, send a mail to a real person, issue a charge or refund, or put up an ad without my explicit yes in this session. Ads go up paused; turning them on is mine.
- Never publish a sample testimonial or an invented claim, number, scarcity line or deadline.
- Never promise sales or income. The store, the tracking and the tests are tools; results come from my product and my traffic.

HOW TO REPORT BACK
Result first, in plain words: what is installed, which accounts are connected (names only, never values), what the test purchase proved (card, PayPal), what setup_status says, and what is still mine to do. End with one question, usually: "ready to switch to live keys and launch?"

FIRST TASK NOW (dry run, no real money)
After step 6, show me the payment robot's result for the card rail and the purchase row it found, then run setup_status and sales_by_day(days=1). The test purchase must not count as a sale; if it does, the test markers are wrong and we fix that before anything else.

AFTER MY "GO LIVE" (later, only with my yes)
Switch to live keys and buy once with a real card at a small price or a 100% code, then refund myself. Deploy (AGENTS.md step 9) and add the two scheduled jobs from store-kit/docs/cron.md. Then read the website journal's launch checklist before anyone is sent to the store, and the Meta ads journal before the first ad. If I sell physical products: before anyone is sent to the store, walk me through one real order to my own address (PHYSICAL-PRODUCTS.md step 9) and wait until the parcel arrived and I said it is right.
```

## Prompt 2: every day (read the store, propose changes)

```text
Read my store the Store System way and tell me what to change. These rules come before anything else in this session.

WHAT IT IS FOR
You read the real numbers with Control Room DNA, find where buyers stop, and propose one or two concrete changes with the number that should move. I approve anything that goes live.

WHEN TO USE IT
Every day I want a read, when I ask "how is the store selling", "why no sales today", "which version wins", or before I change a price, an offer or an ad.

THE CYCLE, IN ORDER
1. setup_status: every source I use says "ok" or "set". If one is missing, say which key and stop there.
2. sales_by_day(days=7): real sales per day from my own purchases table, next to ad spend and the ad platform's own value. Money counts only from purchases; the ad platform over-counts. When a day looks bad, sales_patterns(days=21) answers "is it falling, or is it just the day": the last 7 days against the 7 before, each campaign, weekday and hour.
3. Where they stop: payment_dropoff(days=7) visitor by visitor, payment_trails(since=<date>) for what each person who opened the payment actually did and what the payment processor recorded for them (email typed, a payment tried, declined by the bank), then payment_rails(days=30) for card against PayPal, and inside the Instagram and Facebook apps against a normal browser.
4. Tests: ab_test(test, days=14) for each running test. Judge by people who tapped buy, not page views. Small numbers decide nothing: say so when there is not enough data.
5. Emails: email_leads(days=7) for what happened to each email that did not buy. After a list mail: mail_campaign(campaign, subject, sent, since=<the minute it went out>), judged at 48 hours and at 7 days, never in the first hours.
6. Ads, if they run: ad_health(days=7) for learning, fatigue, cost and audience shifts. Judge by the week, never by one day. For any ad change, follow the Meta ads journal (its own start prompt is run-meta-ads-with-ai.md).
7. People: who_is_here(minutes=30) for who is on the site now and, when the data shows it, who they are and what they hold; buyer(email) shows one buyer end to end when something looks wrong. To measure a change from its minute, pass since="YYYY-MM-DDTHH:MM:SSZ" to any tool that takes since.
7b. If I sell physical products: open /admin/orders with me and say how many orders wait in "To pack" and for how long, which return requests are open, any order on hold with its plain line (a partner or print file problem), and whether any order is flagged short (sold after the last unit was gone).
8. Propose: at most two changes, each with what to change (file or setting), why (the number that showed it) and the number that should move. Prepare the change locally; it goes live only after my yes.

WHAT TO CHECK FIRST
Today's real date and my time zone (utc_offset_hours) before you call a day good or bad. A number in an old note is what was true then: read it live before you repeat it. Exclude test runs and test emails from every count.

NEVER
- Never deploy, change a price, turn a test or feature on or off on the live store, send a mail to buyers, issue a charge or refund, or touch an ad without my explicit yes on the final version.
- Never count revenue from anything but the purchases table.
- Never write or print a key, and never paste a customer's full email into a report: the masked form is enough.
- Never promise sales. Report what was measured, with its source and its date.

HOW TO REPORT BACK
Result first, in plain words: sales this week and yesterday (from purchases), ad spend next to them, where people leave the payment, which test leads and whether it is decided. Then the one or two changes you propose, each with its reason. End with one concrete question: which change to make.

FIRST TASK NOW
Run setup_status and sales_by_day(days=7), and tell me in four lines how the week is going and where buyers stop. Then ask me what to look at first.
```
