One person shipped seven apps to the App Store working with AI agents every day. Three things came out of it. Two are free, one is not. Pick one to open it.
7 live in the App Store
01 / FREE
The map is the seller's real vault: every node a file, every line a reference the session rituals made. Drag it around.
Two ship in the kit. The rest are curated, so you install them from their authors and keep getting their updates.
Read when the task matches, never auto-loaded. A dozen auto-loaded skills is a tax on every session you run.
Get the system free. One email, instant download, no account.
02 / FREE
A key that lands in a chat is burned. Here it never gets there.
Type a made-up key. It goes to the key window, never a chat.
It holds a scoped use-token. It can call your APIs. It cannot read a single key.
The failure mode this design exists for:
Simulation: runs only in this tab, stores nothing, and the ciphertext above is a stage prop. The real mechanic (AES-256-GCM, your own Vercel deployment, browser key window) ships in the free kit and installs itself.
It installs itself: repo, deployment, encrypted store, 14 provider templates. Ends with a quiz you have to pass.
The proxy ships in the free kit. Same download as the vault.
03 / $49
The templates cost nothing because they were never the valuable part. This took five months and seven shipped apps.
Today: 254 learnings, redacted, English index, four chapters. Entries are the German originals; translations follow and your download page always serves the newest build. Project names are removed, because the lesson is the product.
$49one-time, $129 at the next step, 30-day refund
PRICE
You could rebuild the vault structure from screenshots, and the proxy is a pattern, not a secret. Giving those away costs me nothing and saves you a month. What is genuinely hard to get is knowing why each rule exists, and that is the only thing with a price on this page.
the system
Freeone email, instant download, no account
the data
$49one-time, $129 at the next step, 30-day refund
staying current
The data purchase is a snapshot. This is the tap: every new learning I write lands in your vault, in your format, pulled by your own agent with the included sync skill. Category file first, index second, marked as someone else's evidence rather than yours.
$19per month, step 01 of 04, cancel any time
Needs the free system first, which is the point: without a vault there is nothing to sync into.
Q & A
It is written with Claude Code in mind, but the system is plain Markdown and git, so any agent that reads files can run it. The self-installing parts assume an agent that can execute commands; by hand they follow the same written steps.
Two of them are, with their licence text next to them. The other four sets are curated rather than redistributed: the registry tells you what each is for, who wrote it and under which licence, and you install it from the author. That respects their licences and means you get their updates instead of my frozen copy.
All three. It is files in a git repo, nothing installs, and the setup scripts are cross-platform Node. The seller's own vault runs on Windows.
The titles and the index are English, all 254 of them. The entries themselves are the German originals, written at speed during the work. Translating them afterwards would have meant rewriting them, and a rewritten note is no longer evidence. Body translations follow in stages, and the four chapters are the English distillate.
30 days, money back, no questions. Reply to your receipt and a human refunds you, the same one who wrote the vault.
No API key ever enters an AI chat. Keys go into a browser window, get encrypted server-side on a deployment you own, and agents use them through the proxy without seeing them. You can play that mechanic in panel two. The setup ends with a five-question quiz on exactly this model, and passing it is mandatory by design.