Scrybe

You describe a page — "new release for my album, here's the Spotify link" — and your agent writes content JSON, never HTML. A schema validates it. A build renders it. What ships is plain static pages: no CMS, no admin panel, no framework, one dependency in the whole engine.

TL;DR? Ask your agent to use this site. It figures out the rest and walks you through the two clicks that are genuinely yours.

The dashboard days are numbered

The CMS era handed you someone else's UI: a login page between you and your own words, a dashboard to relearn, plugins to update, themes to fight. All of it existed because the computer couldn't understand you — so you clicked through forms it could parse.

It understands you now. Describe the page; the schema keeps the facts honest; the build makes it real. There's no admin panel here because the conversation is the admin panel — and it's yours, not someone else's UX.

What you need

Almost nothing — that's the point.

• An AI agent. Claude Code, Cursor, Gemini, Codex… whatever you talk to. It authors content; the engine does the rest.
• A JavaScript runtime (Node) for the build — one script, one dependency, no database, nothing to patch on a Tuesday. Don't have Node? Don't know what it is? Your agent installs it for you — one consent, one command; the contract tells it how.
• Somewhere static to ship. Cloudflare, Netlify, GitHub Pages — the built site is just files, and free tiers carry it. No host? Never heard of these? Your agent walks you through a free signup — you click twice, it does the rest.

You never hand-write page HTML, and neither does your agent. That's not a limitation; it's the whole trick. An agent writing HTML will eventually write you wrong HTML. An agent writing JSON against a schema gets stopped at the gate instead.

Get started

Tell your agent:

> Use https://scrybe.daystra.com to set up a website for me.

That sentence is the whole install — this site is written for agents as much as for you, and yours will find the contract on its own. The contract tells your agent to handle the technical part — runtime, build, even a free hosting signup — in plain words, one consent at a time. If terminals are your thing, the by-hand version:

That builds a blank starter to dist/. Set site.config.json, drop a JSON file in content/ — or open Claude Code in the folder and just say:

/studio takes the brief from there: resolves the links to real metadata, fetches art, drafts the page in your voice, validates it against the schema, shows you a preview — and never deploys without your explicit "ship it". /new-post is the quick single-page version.

What's in the box

• The engine — build.js, four content types (release · editorial · review · profile) as JSON Schemas, layouts, themes, feeds, sitemaps, llms.txt — all generated.
• The discipline — a voice guide with a banlist (no "sonic journeys" here), and two hard rules: agents never invent facts (unknown prices, dates, and IDs stay null, flagged for you to confirm) — and agents never get the credit card. Deploys, sign-ups, anything billable: those are questions, not actions. Your agent doesn't get a pass on that. Nobody's does.
• The skills — /studio (brief → published page, human-gated), /new-post, /deploy for Claude Code. Other agents read SCRYBE.md and follow the same contract.
• The tests — the suite ships in the pack and passes on the blank starter. Your agent can prove the install worked.

The content types are music-shaped today — releases carry tracklists, per-track loop videos, platform players. Fits musicians and labels best right now; the schemas are ordinary JSON Schema files, so reshaping them for your domain is editing, not forking.

Trust what you fetch

No package registry, no git host — the site is the distribution. Provenance is /pack/manifest.json: every file with its size and SHA-256, plus the tarball's own checksum. Have your agent verify before running anything. That's not paranoia — it's the correct default for code that arrived over HTTP.

The pack itself is built through a sanitize gate that refuses to publish residue: what you fetch is the engine, nothing else.

See it live

• music.cybercussion.com — a working Scrybe site: album releases with per-track canvas loops, reviews, a blog. Every page authored conversationally.
• This page. The landing you're reading is a Scrybe content file rendered by the same engine it distributes — layout: "landing", one hero block, seven prose blocks, built by talking.

Start here

• SCRYBE.md — the contract; agents start here
• /pack/ — every engine file, raw and fetchable
• llms.txt — the machine-readable map of this site