Markdown that edits itself.
Status: published (v0.7.3) and browser-verified — 12 starter templates, 7 light themes, the full document-element set, an in-file editor with a template picker, and a dynamic switch (e.g. exam answer keys). Two packages publish in lockstep:
orz-pagedandorz-paged-browser. See DESIGN.md for the design.
Generate a single .paged.html from Markdown that edits itself in the
browser — a document shown as print pages in any browser (A4/Letter, real margins,
running headers/footers, page numbers), edited and saved in the browser, and
exported to PDF by printing it (no PDF binary, no Puppeteer).
It is the paginated-document sibling of:
- orz-mdhtml — a continuous document (
.md.html) - orz-slides — a reveal.js deck (
.slides.html)
and is authored in orz-markdown, the family's parser.
The orz-md-pdf VS Code extension produces a .md.pdf: a real PDF binary with
the Markdown embedded inside, rendered by Puppeteer. orz-paged keeps the file as
editable HTML in a paged view; a PDF is produced only on demand, by the
browser's own Print → Save as PDF. It borrows orz-md-pdf's document model (as
a spec, reimplemented in TypeScript) and its paged.js pagination — not its PDF
binary, Puppeteer, or untested code.
- Print on light paper. White or light-colored backgrounds only — no dark mode, no screen-only constructs (avoid tabs, spoilers, embedded video).
- Assume internet. Fonts, web images, and math/diagram libraries load from CDN at view/print time; files stay small.
- Curated, not sprawling. A dozen common templates + seven light themes (shared with the orz family) + a clear agent skill, rather than every option — but every element/theme/setting stays independently composable for power users and agents.
Markdown ──orz-markdown──► content HTML
──pipeline-model─► paged HTML (page config, headers/footers, elements, fonts, theme)
──paged.js───────► .pagedjs_page boxes (the visible pages — and what prints)
orz-paged <input.md> [-o out.paged.html] [--inline | --cdn] [--title t]
orz-paged --template <name> [-o out.paged.html] # scaffold + render a starter document
orz-paged --new <name> [-o out.md] # write a starter .md you then edit
orz-paged --list-templates # list the templates
--inline (default) embeds the engine + paged.js + every theme. --cdn instead
references jsDelivr for the published orz-paged-browser engine + theme.
Fonts, web images, and math/diagram libraries always load from CDN (internet
assumed) — so files stay small. The document's {{nyml kind: document}} block
(template:, theme:, …) is the source of truth.
Before the document settings block, authors may add {{nyml kind: meta}} with
title, author, description, license, license_name, license_url,
source, date, and comma-separated keywords. Generation consumes this block
and writes normalized <head> tags plus an #orz-meta JSON island. Those
records survive later in-browser saves and framework updates. Programmatic
callers may pass metadata to buildPagedHtml; host values win field by field.
{{nyml
kind: meta
title: Quarterly Report
author: Your Name
license: CC-BY-4.0
license_url: https://creativecommons.org/licenses/by/4.0/
date: 2026-07-11
}}Clear split: a template owns layout (page size, furniture, which elements appear and where) and a theme owns the look (font, decoration color, element styling) — so the same theme renders the same across every template.
Twelve starter templates scaffold a real document, not just defaults —
article, report, and exam each in a title-page and a title-section
variant, three CV styles (cv classic, cv-modern, cv-elegant), plus
letter, cover-letter, and note. Pick one with --template <name>, or in the
in-file editor from the template picker (the 📄 toolbar button); the picker
drops the starter into the editor (keeping any existing content in a comment).
Seven light themes are vendored from orz-markdown and print-adapted —
light-neat-1/2/3, light-academic-1/2, beige-decent-1/2 — selectable per
document (theme:) or live in the editor (switching swaps font + color + style
together). The default is light-academic-1. They are light only by design
(ink on paper); an explicit font_preset / decoration_color / page_background
still overrides a theme's font / accent / page tint.
A dynamic_choices document setting (a key: value map) drives conditional
content: elements tagged data-show-when="key=value" / data-hide-when="key=value"
are kept or dropped at render time. The headline use is exam answer keys — the
question answers are tagged answer-key=show, so one source prints both the
student copy (answer-key: hide) and the instructor key (show). Flip it in the
source, or live from the editor's answer key dropdown. Define your own keys for
any tailored-variant document.
A {{nyml kind: document}} block sets page size/margins, font, headers/footers,
page numbers, and a template/theme; concrete element blocks such as
{{nyml kind: article-title}}, {{nyml kind: toc}}, and
{{nyml kind: question-mc}} insert document structures. orz-paged ships a
curated subset of this model (the most common
templates/elements) with a clear agent skill — see
DESIGN.md §8. The full model orz-paged draws from is in
docs/document-model.md (a spec reference from
orz-md-pdf).
The package ships an agent skill that teaches an AI agent the document format — page templates, headers/footers, the curated elements (CVs, letters, exams), and the dynamic switch. The quickest way to produce a document is to let an agent do it:
- Any agent — point it at
https://cdn.jsdelivr.net/npm/orz-paged/orz-paged-skills/SKILL.md, then describe what you want. - Claude Code — copy
orz-paged-skills/into~/.claude/skills/orz-paged/from this repo or from the installed npm package.
More install routes: https://markdown.orz.how/agents.html
.paged.html files conform to orz-host-save@1: a platform can embed a
document in an iframe, announce itself with a postMessage handshake, and
receive saves ({ source, html }) instead of the file-system path —
standalone behavior, Export to PDF, and downloads are unchanged, and nothing
activates without the handshake. The canonical spec lives in the orz-mdhtml
repo: PROTOCOL.md.
A .paged.html is self-contained executable HTML: opening one runs the
JavaScript embedded in it (the engine, the editor, and — because the parser allows
raw HTML in the source — potentially anything in the content). The trust model is
the same as running a downloaded program, not opening a PDF.
- Only open or edit files from sources you trust. Anyone can craft a file that looks authentic (same chrome, logo, layout) but contains hostile code. The format has no built-in authenticity — appearance proves nothing.
- What the browser limits. A page can't run native code or read your disk silently (Save uses the File System Access prompt and is scoped to the file you pick). Realistic harm from a hostile file is web-context — phishing, exfiltrating what you type or paste, beaconing — within the browser sandbox.
- The one-click update is opt-in and fixed-source. It only checks for and fetches a new framework after you enter edit mode and click Update, always from the canonical jsDelivr packages over HTTPS (the source is hardcoded in the engine — a tampered file cannot redirect it), and it shows the exact URLs for confirmation first. Clicking Update places trust in npm + jsDelivr for those packages.
- Integrity can't be self-verified. A file cannot prove its own integrity (a forgery would just lie). If you need authenticity, verify it out-of-band — a checksum or signature from the publisher over a trusted channel.
MIT