OmegaOS skill — the build half of the Blueprint OS chain.
/blueprint-osdesigns an OS./stackturns that design into a running app on the canonical stack. Triggers/stackand/omg-stack.
The stack, non negotiable (R-BLUEPRINT-STACK)
| Layer | Techno | Role |
|---|---|---|
| Shell / UI | Stax | The panel grammar. Every OS wears the same navigation |
| Front | Next.js (App Router) | The app |
| Data + realtime | Convex | The single model. Reactive by default |
| Auth | Clerk | Identity, organizations, roles |
| AI | Claude, called from a Convex action | Never from the client |
If a design suggests something else, the design changes, not the stack.
Stripe is opt-in, not part of the default. Pass --stripe when the app actually
charges someone. Most OS products are built and used long before they are sold, and a
billing surface nobody calls is dead code that still demands keys, a webhook endpoint,
and a dashboard setup. Scaffold it when it earns its place.
Stax is pulled every single time
Before any scaffold, stack-new.sh runs the Stax sync — fast-forward only, never
clobbering local commits. The app is always vendored from the current main of
github.com/agentik-os/stax. The commit that was vendored is written into
stax.lock.json at the app root, so a build is always traceable to a Stax revision.
This is the rule, not an optimization: an app scaffolded from a stale Stax drifts from the rest of the family, and the drift is invisible until the panel grammar disagrees.
How to run it
bash ~/.omega/skills/stack/scripts/stack-new.sh <app-name> [--blueprint <dir>] [--dir <parent>] [--stripe]
| Flag | Effect |
|---|---|
--blueprint <dir> |
Read a Blueprint OS folder: phase 09-data/schema.ts becomes the Convex schema, phase 10-stax/panneaux.md drives the panel registry |
--dir <parent> |
Where the app is created. Defaults to ~/Station/SideBusiness |
--stripe |
Opt-in billing. Adds the checkout route, the webhook route, lib/stripe.ts, the stripe dependency, and the Stripe env block |
--no-install |
Scaffold the files, skip npm install (offline / inspection) |
The script is idempotent on a fresh dir and refuses to overwrite an existing app — rerunning it on a live project is a destructive act, so it stops instead (R-DESTRUCT).
What the script does, in order
- Sync Stax —
stax-sync.sh, ff-only. Records the commit. create-next-app— TypeScript, App Router, Tailwind,src/.- Vendor Stax — calls the existing
stax-scaffold.sh(never reimplemented here):panels-core.ts,panels-react.tsx,tokens.css,stax-ui.cssintosrc/stax/. - Convex —
convex/schema.ts(from the blueprint if given, else the canonical pattern),convex/auth.config.tswired to Clerk, an example query and mutation. - Clerk — middleware, provider, and the Convex
ConvexProviderWithClerkbridge. - Stripe — only with
--stripe— checkout route, webhook route, andlib/stripe.ts. The price IDs are left as named placeholders because only the operator can create them. Without the flag, none of this is written andstripeis not installed. .env.example— every key, with who provides it and where to get it.NEEDS-OPERATOR.md— the exact list of what a human must do before the app runs.
Then it prints the list. It never invents a key, never fakes a Stripe ID, and never
claims the app runs before npm run dev has actually been executed.
The Stripe IDs — read this before promising anything
Only relevant when the app was scaffolded with --stripe.
Stripe is the one layer that cannot be fully scaffolded. Products and prices live in
the operator's Stripe account, and their IDs (price_…, prod_…) do not exist until a
human creates them. The scaffold therefore writes placeholders and lists them.
Never guess a price_…. A wrong price ID fails at checkout, in production, on a real
customer. See references/stripe.md for the full setup path and the webhook events
that actually matter.
Building from a blueprint
When --blueprint points at a Blueprint OS folder, the chain is:
| Blueprint phase | Becomes |
|---|---|
09-data/schema.ts |
convex/schema.ts, verbatim if valid |
10-stax/panneaux.md |
The panel registry and the inspector stubs |
06-features/features.md |
The NEEDS-BUILD.md checklist, layer by layer |
08-ia/system-prompts.md |
convex/ai/ action stubs, one per agent |
Gate, enforced mechanically. stack-new.sh runs blueprint-check.sh --gates-only
before it scaffolds anything, and refuses on an unfranchised gate. Building from a
blueprint that failed the parity gate produces a demo whose missing socle is discovered
at delivery, which is exactly the failure mode phase 5 exists to prevent.
--force-gates overrides it. Use it knowing what it costs, and it says so out loud.
Run the full check yourself when you want more than the gates:
bash ~/.omega/skills/blueprint-os/scripts/blueprint-check.sh <blueprint-dir>
Rules of conduct
Pull Stax first, always. No exception, no "it was synced this morning".
Never fabricate an ID or a key. Placeholder plus a line in NEEDS-OPERATOR.md.
Never start a dev server to "verify" (R-TEST) — except here, where the code is
brand-new and not yet deployed. That is the documented exception: one npm run dev,
one HTTP probe, then kill it. Real runtime evidence, not a green install log (L1).
Secrets never enter the repo (R-ENV / L0). .env.local is gitignored; the
.env.example carries names only. Keys live in ~/.omega/secrets/<app>.env.
The panel grammar is not optional. No pages, no modals, no tabs. If the app grows a second navigation mechanic, it stopped being an AgentikOS app.
Reference files
| File | When |
|---|---|
references/convex.md |
Schema, indexes, auth, and calling Claude from an action |
references/clerk.md |
The Clerk plus Convex bridge, and the JWT template that trips everyone |
references/stripe.md |
Products, prices, webhooks, and the IDs the operator must create |
references/env.md |
Every environment variable, who provides it, where it lives |