npm - @kirigami/php-prepros - Versions diffs - 3.1.1 → 3.2.1 - Mend

@kirigami/php-prepros 3.1.1 → 3.2.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (8) hide show

package/README.md +2071 -1999
package/package.json +59 -59
package/src/libraries/fs.class.php +171 -22
package/src/libraries/html.class.php +56 -4
package/src/libraries/ld.class.php +2 -2
package/src/libraries/prepros.class.php +18 -5
package/src/libraries/prepros.plugins.php +1 -0
package/src/prepros.php +1 -1

package/README.md CHANGED Viewed

@@ -1,1999 +1,2071 @@
-<div align="center">
-<img src="https://zmotrin.github.io/assets/kirigami/kirigami-logo-universal.svg" alt="Kirigami" width="400" />
----
-# @kirigami/php-prepros
-PHP preprocessor for the **Kirigami** static site generator.
-[![npm version](https://img.shields.io/npm/v/@kirigami/php-prepros)](https://www.npmjs.com/package/@kirigami/php-prepros)
-[![License: GPL-3.0-or-later](https://img.shields.io/badge/license-GPL--3.0--or--later-blue)](./LICENSE)
-[![Node.js >=24.0.0](https://img.shields.io/badge/node-%3E%3D24.0.0-brightgreen)](https://nodejs.org)
-[![Website](https://img.shields.io/badge/website-php--kirigami.github.io-1f6b4a)](https://php-kirigami.github.io)
-</div>
----
-## Overview
-Build full static websites in PHP — with zero server, zero runtime dependency, zero compromise on expressiveness. Write your pages as regular PHP files, annotate them with a PHPDOC header, and let `php-prepros` compile everything to clean, deployable HTML.
-It is the perfect solution for **GitHub Pages**. Since it runs entirely in Node.js, it is fully compatible with **GitHub Actions**, allowing you to automate your deployment pipeline effortlessly.
-Part of the **Kirigami** project ecosystem.
----
-## Table of contents
-- [@kirigami/php-prepros](#kirigamiphp-prepros)
-- [Overview](#overview)
-- [What's new in 3.0.1](#whats-new-in-301)
-- [What's new in 3.0.0](#whats-new-in-300)
-- [What's new in 2.0.0](#whats-new-in-200)
-- [What's new in 1.9.3](#whats-new-in-193)
-- [What's new in 1.9.2](#whats-new-in-192)
-- [What's new in 1.9.1](#whats-new-in-191)
-- [What's new in 1.9.0](#whats-new-in-190)
-- [What's new in 1.8.0](#whats-new-in-180)
-- [What's new in 1.7.2](#whats-new-in-172)
-- [What's new in 1.7.1](#whats-new-in-171)
-- [What's new in 1.7.0](#whats-new-in-170)
-- [What's new in 1.6.0](#whats-new-in-160)
-- [What's new in 1.4.0](#whats-new-in-140)
-- [What's new in 1.3.0](#whats-new-in-130)
-- [What's new in 1.2.1](#whats-new-in-121)
-- [What's new in 1.2.0](#whats-new-in-120)
-- [How it works](#how-it-works)
-- [Installation](#installation)
-- [Configuration — `kirigami.yaml`](#configuration--kirigamiyaml)
-  - [`kirigami` block](#kirigami-block)
-  - [`seo` block](#seo-block)
-  - [`prepros` block](#prepros-block)
-  - [`image` block](#image-block)
-  - [`plugins` block](#plugins-block)
-  - [`esbuild` / `sass` blocks](#esbuild--sass-blocks)
-  - [`export` block](#export-block)
-  - [`scripts` block](#scripts-block)
-  - [`tasks` block](#tasks-block)
-- [Writing pages](#writing-pages)
-  - [PHPDOC header](#phpdoc-header)
-  - [Auto-loading data files](#auto-loading-data-files)
-  - [`@content`, `@indent`, and `@type`](#content-indent-and-type)
-  - [Built-in tags](#built-in-tags)
-- [JavaScript API](#javascript-api)
-  - [`render(file?, phpIncludes?)`](#renderfile-phpincludes)
-  - [`sitemap()`](#sitemap)
-  - [`runenv(script, paths?, ...args)`](#runenvscript-paths-args)
-  - [`runPluginScript(script, pluginRoot, paths?, ...args)`](#runpluginscriptscript-pluginroot-paths-args)
-  - [`mountPath(localPath, virtualDir?, php?)`](#mountpathlocalpath-virtualdir-php)
-  - [`processImages(jobs)`](#processimagesjobs)
-  - [`resetRuntime()`](#resetruntime)
-  - [TypeScript declarations](#typescript-declarations)
-- [PHP classes reference](#php-classes-reference)
-  - [PREPROS](#prepros)
-  - [MD](#md)
-  - [HTML](#html)
-  - [YAML](#yaml)
-  - [SCHEMA](#schema)
-  - [LD](#ld)
-  - [META](#meta)
-  - [CACHE](#cache)
-  - [IMG](#img)
-  - [FS](#fs)
-  - [STR](#str)
-  - [ARR](#arr)
-  - [CURL](#curl)
-  - [SCRAPER](#scraper)
-  - [OBF](#obf)
-  - [STD](#std)
-  - [Unicode normalization](#unicode-normalization)
-  - [Procedural shortcuts (aliases)](#procedural-shortcuts-aliases)
-- [Plugin system](#plugin-system)
-  - [PREPROS tags](#prepros-tags)
-  - [PREPROS hooks](#prepros-hooks)
-  - [MD plugins](#md-plugins)
-  - [Built-in plugins](#built-in-plugins)
-- [Extending the `<markdown>` tag](#extending-the-markdown-tag)
-- [Requirements](#requirements)
-- [License](#license)
----
-## What's new in 3.0.1
-A registered tag written inside Markdown code is no longer processed. `<markdown>`, `<img asset="...">` or a plugin tag shown in an inline code span (`` `<markdown prose>` ``) or a fenced block (` ``` ` / `~~~`) stays example text: before, a `<markdown>` in a code span paired with the real block's closing tag and broke the rest of the page, and an `<img asset>` in one became an empty image. Indented (four-space) code blocks are not recognized, since `<markdown>` bodies are indented; use a fence there.
----
-## What's new in 3.0.0
-This release switches `YAML::` to the native YAML extension, `MD::` to native mdhtml, `SCHEMA` to native jsonk, and `Normalizer` to native norm. It also includes page types and request lifecycle hooks.
-PHP data files use the native `yaml` extension backed by LibYAML. Its YAML 1.1 implicit booleans include unquoted `y`, `n`, `yes`, `no`, `on`, `off`, `true`, and `false`, including mapping keys. Quote these words when you mean strings (for example, `"NO": Norway`). `YAML::parse()` / `parseFile()` / `loadFile()` preserve the wrapper’s array/object choice; native `yaml_parse()` / `yaml_parse_file()` have their own extension signatures. `yaml_load_file()` remains a wrapper alias. The project’s `kirigami.yaml` is parsed separately in Node through `struct-walker` and `js-yaml`.
-`MD::` delegates to the native `mdhtml` extension (cmark-gfm). Footnotes now use `<section class="footnotes" data-footnotes>` instead of the old `<div class="footnotes">`; target `.footnotes` rather than a specific container tag in custom CSS.
-`SCHEMA` validates through the native `jsonk` extension (draft 2020-12), keeping its API and the `"path: message"` error format. It now also supports `if`/`then`/`else`, `contains`, `propertyNames`, `dependentRequired`/`dependentSchemas`, `prefixItems`, and `$ref` to absolute or `$id`-relative URLs (fetched over the network). Error messages use jsonk's wording, and an `additionalProperties: false` violation is reported on the parent object instead of the extra property. The previous pure-PHP validator stays available as `SCHEMA_LEGACY`.
-`Normalizer` now comes from the native `norm` extension (utf8proc) instead of the bundled pure-PHP polyfill, which stays available as `NORMALIZER_LEGACY`.
-**Breaking: the `seo.jsonld` sub-block is merged into `seo:`.** META and LD now read one set of keys, so the site description, keywords, image, language and person are declared once. JSON-LD is injected whenever the `seo:` block exists; `seo.jsonld` is only an on/off switch (default `true`). To migrate, move the keys of `seo.jsonld: {…}` up into `seo:` (`jsonld.auto: false` becomes `jsonld: false`) and rename `seo.language` to `seo.lang`. `kiri` rejects the old shapes with a message saying so. Sites with `seo: {}` and no `jsonld` now get JSON-LD too; add `jsonld: false` to keep them without it.
-```yaml
-# before                       # after
-seo:                           seo:
-  language: fr-CA                lang: fr-CA
-  jsonld:                        type: ProfessionalService
-    type: ProfessionalService    logo: images/logo.png
-    logo: images/logo.png
-```
----
-## What's new in 2.0.0
-- **Breaking: `meta:` and `jsonld:` merged into one top-level `seo:` block.**
-  The two used to be independent siblings of `kirigami:` that happened to
-  share fallback data; they're now one block, one mental model for a
-  project's whole SEO/social surface — `META`'s own keys live directly under
-  `seo:`, and `jsonld` is nested inside it as its own sub-block:
-  ```yaml
-  # before (1.x)
-  meta:
-    favicon: favicon.png
-  jsonld: {}
-  # after (2.0.0)
-  seo:
-    favicon: favicon.png
-    jsonld: {}
-  ```
-  The two stay **independently toggled** exactly as before — a project can
-  have META's tags without JSON-LD, or vice versa; `seo: { jsonld: false }`
-  (or `{ auto: false }`) stops just the JSON-LD injection, `seo: false` (or
-  `{ auto: false }` at the top level) stops just META's. Every fallback chain
-  (a page's PHPDOC → the block → the loose `kirigami:` keys) is unchanged,
-  including `jsonld`'s own keys still feeding META's defaults (`jsonld.image`,
-  `jsonld.lang`, `jsonld.person`, …) — only *where the two blocks live* in
-  `kirigami.yaml` changed, not how resolution works. **Migration**: rename
-  `meta:` to `seo:` and move the `jsonld:` block's content under it as
-  `seo.jsonld:`. See [`seo` block](#seo-block) / [`meta` config](#meta-config)
-  / [`jsonld` config](#jsonld-config).
----
-## What's new in 1.9.3
-- Dependency bump to `@kirigami/struct-walker` 1.0.5; `homepage` + README
-  pointed at the site (metadata only).
----
-## What's new in 1.9.2
-- **Fixed a real Markdown bug**: a list item's source line count was 1:1
-  with `<li>` count, so an indented continuation line with no marker of its
-  own (a soft-wrapped `- **foo** text\n  more text`) fell outside the
-  block-matching regex entirely — the list closed after the first line, the
-  continuation resurfaced as a stray flat `<p>`, and a new list reopened for
-  the next marker line. Fixed by widening the block regex to also accept a
-  marker-less indented line and, in the per-line loop, appending it to the
-  previous item instead of dropping it.
----
-## What's new in 1.9.1
-- **`{% img-asset %}` no longer crashes the build on an unresolvable path.**
-  `IMG::asset()`'s exception is now caught and turned into an HTML comment,
-  matching `codepen`/`checklist`'s own missing-argument behavior instead of
-  aborting the whole render. This also makes it safe to *document* the tag —
-  a literal `` `{% img-asset path … %}` `` written as prose inside a code
-  span still runs the plugin (code-span protection only swaps the *displayed*
-  output back to the literal text; the callback itself always executes), so
-  a placeholder path used to throw "Invalid image file." and fail the build.
----
-## What's new in 1.9.0
-- **`{% youtube %}` removed** — moved to
-  [`@kirigami/plugin-embed`](https://www.npmjs.com/package/@kirigami/plugin-embed),
-  which replaces the old plain-iframe output with a real oEmbed-backed card
-  (cover thumbnail, title, play button — no network call until the visitor
-  actually clicks play). If a project used the built-in `{% youtube %}`,
-  install the plugin; without it, the tag now falls through unresolved
-  (`{% youtube ID %}` printed as-is) rather than rendering an iframe.
-  `codepen`/`checklist`/`callout` are unaffected.
----
-## What's new in 1.8.0
-- **`{% img-asset %}` — a new built-in Markdown plugin.** Same pipeline as the
-  `<img asset>` HTML tag (`IMG::asset()`: resize, cache, publish under
-  `image.dest`), usable straight from Markdown text:
-  ```
-  {% img-asset photo.jpg %}
-  {% img-asset photo.jpg 800 %}
-  {% img-asset photo.jpg 800 600 %}
-  {% img-asset photo.jpg 800 600 cover %}
-  ```
-  Positional args: source path (relative to `image.source`), width, height,
-  and the literal `cover` keyword. Registered in `md.plugins.php` alongside
-  `codepen`/`youtube`/`checklist`/`callout` — available out of the box, drop
-  it with `MD::unregisterPlugin('img-asset')` if you don't want it.
----
-## What's new in 1.7.2
-- **Cleaner formatted output around highlighted code.** The de-indent script
-  `PREPROS::injectHead()` adds when `prepros.format` is on now flattens the
-  `<pre><code>` indentation `HTML::format()` writes for *every* block, including
-  ones a build-time highlighter has wrapped in `<span>`s. It works on
-  `innerHTML` line by line and removes only the shared leading run (relative
-  indentation is kept). This lets [`@kirigami/plugin-highlight`](https://www.npmjs.com/package/@kirigami/plugin-highlight)
-  1.7.2+ re-indent its markup to line up with the rest of the document instead
-  of leaving it flush-left — the served HTML stays consistently indented, the
-  rendered code is still de-indented before the first paint.
-- **`<markdown prose>` wraps in `.prose`.** With the `prose` attribute the
-  built-in tag emits `<div class="prose"> … </div>` so long-form Markdown picks
-  up `@kirigami/canva`'s `styles/prose` typography with no extra markup. Opt-in
-  (a bare `<markdown>` is unchanged); `class` / `id` on the tag land on the
-  wrapper.
----
-## What's new in 1.7.1
-- **No side effects on import.** `kirigami.yaml` is now loaded on first use
-  (`render()` / `sitemap()` / `runenv()` / `processImages()`), not while the
-  module is being imported. `import '@kirigami/php-prepros'` from a directory
-  with no project no longer throws — which is what made `kiri build --help` /
-  `kiri export --help` / `kiri run --help` crash instead of printing their help.
----
-## What's new in 1.7.0
-- **`META`** class — a `<head>` SEO / social metadata generator, the companion
-  to [`LD`](#ld). Builds the standard tags — `<title>`, `description`,
-  `keywords`, `robots`, `language`, `generator`, `author`, Open Graph, Twitter
-  Card, `<link rel="canonical">`, favicon / apple-touch-icon / humans — from
-  each page's PHPDOC, the top-level `meta:` block, and the loose `kirigami:` /
-  `jsonld:` keys `LD` already reads. It emits only what it can resolve, and
-  leaves any tag the layout already hand-writes untouched.
-  - **Opt-in**: a top-level `meta:` block (empty `meta: {}` is enough) switches
-    on automatic injection into every page's `<head>`. `meta: false` (or
-    `{ auto: false }`) keeps the config but stops the injection.
-  - Per-page PHPDOC: `@meta false` (skip), `@meta_title`, `@meta_description`
-    (falls back to `@description` / `@abstract` / `@excerpt`), `@meta_keywords`,
-    `@meta_image`, `@meta_robots`, `@meta_type`, `@canonical`.
-  - Manual builders — always emitted, still de-duplicated: `META::tag()`,
-    `META::link()`, `META::raw()`, `META::tags()`. Procedural aliases:
-    `meta_tag()`, `meta_link()`, `meta_raw()`, `meta_tags()`.
-  Full key reference: [`META` → `meta` config](#meta-config).
----
-## What's new in 1.6.0
-- **Managed `<head>` (`prepros.head`).** Every rendered page's `<head>` is now
-  auto-wired: a tiny theme/FOUC guard as the first child (adds the `js` class,
-  applies the stored `data-theme` before first paint), a
-  `<link rel="stylesheet">` for every `sass` task output, and a `<script>` (no
-  `defer`, just before `</body>`) for every `esbuild` task output — each with a
-  per-page relative path and a `?<timestamp>` cache-bust. A file already
-  referenced in the page is left alone, so you can still hand-place one. Turn it
-  off with `prepros: { head: false }`, or `head: false` on a single sass/esbuild
-  task. A template's `header.php` no longer wires assets at all.
-- **`HTML::format()` indents `<pre><code>`.** A fenced code block's lines are
-  shifted to the block's nesting depth so the HTML source stays readable
-  (relative indentation preserved). The exact leading run is stripped again
-  before it's shown, by a small de-indent script `prepros.head` injects before
-  `</body>` (only when `format` is on). Since 1.7.2 the script works on
-  `innerHTML` line by line, so it also flattens blocks a build-time highlighter
-  has wrapped in `<span>`s. A bare `<pre>` and `<textarea>` are still emitted
-  byte-for-byte.
----
-## What's new in 1.4.0
-A round of fixes to the rough edges that showed up building a full site from
-scratch — mostly developer-experience, all backward compatible.
-- **`HTML::format()` keeps `<pre>` / `<textarea>` verbatim.** Their line breaks,
-  indentation and blank lines are no longer collapsed, so a fenced code block
-  survives the formatter intact — `format: true` and Markdown code blocks now
-  coexist. (1.6.0 refines this: a `<pre><code>` block is re-indented to its
-  nesting depth and de-indented again before display.)
-- **The default Markdown plugins load out of the box.** `{% callout %}`,
-  `{% youtube %}`, `{% codepen %}` and `{% checklist %}` are registered
-  automatically (`md.plugins.php` is auto-included from `MD`), as the docs always
-  said. Drop one with `MD::unregisterPlugin('name')` or shadow it with your own
-  `MD::registerPlugin()`.
-- **Build errors you can actually read.** A fatal in a template (a bad call, a
-  `null` argument, a `TypeError`…) comes back as a structured failure with the
-  message, the offending page and the `file:line` — never a bare
-  `Error: undefined`. The `try/catch` now covers `Throwable`, not just
-  `Exception`. PHP warnings and notices no longer sink an otherwise-clean build:
-  they surface as `warnings` on the result. `kiri` prints the message, the page,
-  and the tail of the PHP stderr/debug output on failure.
-- **PHPDOC parsing.** A tag value may now wrap onto the following *indented*
-  continuation lines instead of being silently truncated at the first line. And
-  an `@word` written in the block's prose is ignored rather than overwriting a
-  real tag — only lines that *start* with `@` open a tag.
-- **`FS::getBreadcrumb()` / `FS::getChildren()`** are anchored on the page being
-  rendered (`PREPROS::$file`), so they return the right trail / child list when
-  called from a layout include, a partial, or a helper function — not only
-  straight from the template. Pass an explicit path to override.
-- **`{% tag %}` inside a code span or code block stays literal** (`` `{% badge %}` ``
-  renders as text) instead of being expanded — or leaking an unrestored
-  placeholder.
-- **`IMG` never upscales.** A requested size larger than the source is clamped
-  down to the source instead of throwing an opaque encoder error (the AVIF
-  encoder in particular).
-- **Build-time tokens expand at render time.** `###YEAR###`, `###TIMESTAMP###`
-  and `###TODAY###` are substituted when each page is generated, so
-  `kiri build` / `kiri watch` previews show real values, not the literal token
-  (previously only `kiri export` replaced them).
-- **`page_info` hook robustness.** The first built-in callback accepts either the
-  `[$file, $info]` pair the hook fires with or the bare `$info` object a later
-  callback receives, so a custom `page_info` hook can't fatal on the argument
-  shape. See [PREPROS hooks](#prepros-hooks).
----
-## What's new in 1.3.0
-- **`LD`** class — a schema.org JSON-LD generator. Collects structured-data
-  nodes during a render and emits them as a single
-  `<script type="application/ld+json">` `@graph` in every page's `<head>`.
-  - Automatic injection is **opt-in**: add a top-level `jsonld:` block to
-    `kirigami.yaml` (even empty, `jsonld: {}`) and an `Organization` (+ `Person`, `WebSite`,
-    `WebPage`, `BreadcrumbList`) graph is derived from that block plus the loose
-    keys projects already carry (`person`, `jobtitle`, `email`, `area`,
-    `knowsabout`, `keywords`, `facebook`, …). No `jsonld:` block → nothing is
-    injected.
-  - Turn it back off with `jsonld: false` / `jsonld: { auto: false }`, or per
-    page with `@ld false`; per-page `@ld_type` / `@ld_title` / `@ld_image` / …
-    tags feed the page node, and a `BreadcrumbList` is built from the
-    `_index.php` ancestor trail with no opt-in.
-  - Explicit builders for everything else: `LD::add()`, `LD::article()`,
-    `LD::faqPage()`, `LD::breadcrumb()`, `LD::ref()`, and every schema.org type
-    via `LD::typeName([...])`. Procedural aliases: `ld_add()`, `ld_organization()`,
-    `ld_script()`, …
-  - A page that already hand-writes an `application/ld+json` script is left
-    untouched.
----
-## What's new in 1.2.1
-- **`processImages()`** JS export — batch resize / palette-extraction through the
-  `IMG` class (`src/imagebatch.php`). `@kirigami/kirigami`'s `sass` task now uses
-  it for `img-asset()` / `colors()`, so the whole toolchain is free of a native
-  image dependency (`sharp` is gone).
-- **`IMG::save()`** takes an optional `$quality` (0-100) for jpg / webp / avif;
-  `null` keeps the per-format default (82).
----
-## What's new in 1.2.0
-- **`SCHEMA`** class — a pure-PHP, dependency-free JSON Schema validator
-  (Draft-7 style, Ajv-like API: `isValid()` / `validate()` / `getErrors()`).
-- **`IMG::asset()` / `IMG::palette()`** — static helpers powering kirigami-core's
-  `img-asset()` and `colors()` Sass functions: on-demand resize/convert of a
-  source image, and cached representative-colour extraction.
-- **`<img asset="…">` tag** — the HTML-side entry point of the image
-  autogenerator, same parameters as `IMG::asset()` (see [Built-in tags](#built-in-tags)).
-- **`IMG` now handles vector and exotic formats** — SVG, EPS, AI, PDF (rasterized
-  via Imagick), plus HEIC / TIFF / BMP, on top of GD's JPEG / PNG / GIF / WebP /
-  AVIF.
-- **`MD` emoji shortcodes** — `:rocket:` → 🚀 from a large built-in map, extendable
-  with `MD::registerEmoji()`.
-- **`MD` footnotes and definition lists** — `[^1]` / `[^1]: …`, and `Term` / `: …`.
-- **`MD` inline HTML is now sanitized** against a tag/attribute allowlist rather
-  than passed through verbatim.
-- **`STR::normalize()`** — Unicode NFD + combining-mark stripping;
-  **`STR::slug($str, $sep = '')`** now takes a separator (pass `'-'` for a
-  hyphenated slug).
-- **Bundled `Normalizer` polyfill** — `ext-intl` isn't in the WASM build, so a
-  polyfill keeps `Normalizer::normalize()` (and `STR::normalize()` / `slug()`)
-  working.
-Earlier, in 1.1.x: `PREPROS::mount()` + the `mountPath()` / `runenv()` JS
-exports, the `SCRAPER` and `CURL` and `ARR` classes, `YAML::loadFile()`, the
-`STR::is_url()` / `html_entities_decode()` / `shorthash()` / `slug()` helpers,
-the `{% youtube %}` / `{% codepen %}` / `{% checklist %}` MD plugins, and the
-`@content` / `@indent` PHPDOC annotations.
----
-## How it works
-`@kirigami/php-prepros` runs your PHP source files inside a **WebAssembly PHP 8.x runtime** ([`@kirigami/php-wasm`](https://www.npmjs.com/package/@kirigami/php-wasm)), entirely in Node.js — no PHP installation required on the host machine.
-The lifecycle of a page build looks like this:
-```
-_index.php  ──▶  PHP (wasm)  ──▶  processTags()  ──▶  HTML::format()  ──▶  index.html
-                    │
-                    ├── before.php  (optional layout header)
-                    ├── after.php   (optional layout footer)
-                    └── PHPDOC annotations resolved (yaml / json / md / url)
-```
-Files are mounted into the WebAssembly virtual filesystem on demand. Only `.php`, `.json`, `.yaml`, `.yml`, `.md`, `.db`, `.txt` and any extra extensions listed in `prepros.mountext` are mounted automatically, keeping memory usage low. Anything else can be mounted on demand with [`PREPROS::mount()`](#preprosmountstringarray-patterns).
----
-## Installation
-```bash
-npm install @kirigami/php-prepros
-```
----
-## Configuration — `kirigami.yaml`
-Every project **must** have a `kirigami.yaml` at its root. The preprocessor reads it at startup and throws if it is absent or invalid.
-`@kirigami/php-prepros` consumes project data, SEO, preprocessing, image settings, and task metadata for managed head injection. The core engine owns plugin loading, task orchestration, export, and full schema validation. Direct use of this package is not a substitute for core configuration validation. All settings share `kirigami.yaml`; its schema is [`kirigami.schema.json`](../kirigami/kirigami.schema.json).
-```yaml
-# yaml-language-server: $schema=https://cdn.jsdelivr.net/gh/php-kirigami/kirigami@main/packages/kirigami/kirigami.schema.json
-```
-```yaml
-kirigami:
-  # ── Required ──────────────────────────────────────────────────────────
-  project:     My Website           # Site name. Printed in the CLI banner, exposed as $project.
-  baseurl:     https://example.com  # Deployed root URL, no trailing slash. Used for sitemap.xml.
-  root:        src                  # Source directory containing your _*.php pages.
-  # ── Optional ────────────────────────────────────────────────────────
-  banner:      assets/banner.txt    # Text file stamped as a license banner on exported files.
-  # ── Arbitrary project data ──────────────────────────────────────────
-  # Everything else under `kirigami:` is free-form. The whole block is
-  # extracted as PHP variables and made available in every page, in
-  # before.php/after.php, and anywhere PREPROS::$config->data is read.
-  author:      Jane Doe
-  email:       hello@example.com
-  gtag:        G-XXXXXXXXXX
-  description: A short description of the site, useful for <meta name="description">.
-  keywords:
-    - keyword one
-    - keyword two
-seo:                            # Presence turns on the META tags and LD JSON-LD (`seo: {}` is enough).
-  type: Organization            # JSON-LD main entity; `jsonld: false` turns the JSON-LD off.
-  logo: assets/logo.png
-prepros:
-  before:  _layouts/header.php  # Included before every page body.
-  after:   _layouts/footer.php  # Included after every page body.
-  format:  true                 # Pretty-print the HTML output (default: false).
-  network: true                 # Allow HTTP fetches in PHPDOC @tag annotations / CURL / SCRAPER.
-  mountext:                     # Extra file extensions to auto-mount into the wasm fs,
-    - .svg                      # in addition to the defaults (.php .json .yaml .yml .md .db .txt).
-    - .webp
-  includes:                     # PHP files auto-included once, before any page renders.
-    - _lib/functions.php
-  types:                        # Named page types — opt in per page with @type <name>.
-    article:
-      before: _layouts/types/article.header.php
-      after:  _layouts/types/article.footer.php
-image:                          # Image autogenerator — powers IMG::asset() / IMG::palette().
-  format: webp                  # webp | avif (default: webp)
-  source: assets/images         # Source folder, relative to cwd() (default: assets/images)
-  dest:   images                # Output folder, relative to kirigami.root (default: images)
-plugins:
-  - name: "@kirigami/plugin-highlight"
-    active: true
-    options:
-      theme: auto
-esbuild:
-  # minify: false
-sass:
-  style: expanded
-export:
-  path:   dist
-  ignore: ["*.psd", "notes/"]
-scripts:
-  - name: convert-images
-    mount: ["assets/images/**/*.jpg"]
-    trigger: before-build       # before-build | before-export | after-export
-tasks:
-  - name:  js-core
-    type:  esbuild
-    entry: scripts/kirigami.core.js
-  - name:  scss-core
-    type:  sass
-    entry: styles/kirigami.core.scss
-```
-### `kirigami` block
-Core project settings. **Read by `php-prepros`.** The entire block is extracted into PHP variables and made available in every page template, `before.php`, `after.php`, and `prepros.includes` files — `$project`, `$author`, `$gtag`, etc. are available with no further setup, and also as `PREPROS::$config->data`.
-| Key | Required | Description |
-|-----|----------|--------------|
-| `project` | ✅ | Human-readable site name. Exposed as `$project`. |
-| `baseurl` | ✅ | Root URL of the deployed site, no trailing slash. Used to build absolute `<loc>` entries in `sitemap.xml`; exposed as `$baseurl`. |
-| `root` | ✅ | Path (relative to the project root) to the directory containing your `_*.php` source pages. Build fails immediately if missing or if the path doesn't exist. |
-| `banner` | — | Path (relative to the project root) to a text file stamped as a license/copyright banner on exported `.js`/`.css`/`.html` files during `kiri export`. May contain the `###DATE###` token, replaced with today's date. Falls back to an auto-generated banner. |
-| *anything else* | — | Free-form key/value pairs (strings, numbers, booleans, lists, nested maps — anything valid YAML). Every key is extracted as a PHP variable (`$author`, `$gtag`, …). Use this for contact info, social links, analytics IDs, SEO keywords, or any project data you want available everywhere. When the top-level `seo` block is present, [`LD`](#ld) also reads some of these by convention: `person`, `jobtitle`, `email`, `area`, `knowsabout`, `keywords`, and social-network URL keys (`facebook`, `instagram`, …). |
-### `seo` block
-Top-level, optional. The SEO surface: one set of keys feeds both the
-[`META`](#meta) tags (the standard SEO / social `<meta>` and `<link>` tags) and
-the [`LD`](#ld) schema.org JSON-LD graph, and its **presence** switches both on
-for every page's `<head>`. An empty `seo: {}` is enough; everything is derived
-from the `kirigami` block and each page's PHPDOC (`@title`, `@description` /
-`@abstract`, `@keywords`, `@image`, `@robots`, `@og_type`, `@canonical`). A
-tag or script the layout already hand-writes is left untouched.
-- `auto: false` stops META's tags, `jsonld: false` stops the JSON-LD; the
-  config values stay available to `META::tags()` / `LD::script()`.
-- `seo: false` keeps nothing on; no block at all means nothing is injected
-  (explicit `META::tag()` / `LD::add()` calls still emit).
-Full key reference: [`seo` config](#seo-config). Per-page tags:
-[`@meta_*`](#meta) and [`@ld_*`](#ld).
-### `prepros` block
-Global `before` and `after` files are optional and default to `null`; an empty `prepros: {}` renders without a layout. When provided, these paths must refer to existing files. PHP warnings are logged to stderr and returned as diagnostics, without being inserted into generated HTML.
-Options for the PHP → HTML compiler. **Read by `php-prepros`.** Declaring this block (even empty) also makes `kiri` prepend a forced `prepros` task on every build/export/watch.
-| Key | Type | Default | Description |
-|-----|------|---------|--------------|
-| `before` | `string` | — | Path (relative to `kirigami.root`) to a PHP file included **before** every page's body. Typically your `<head>`/layout opening. |
-| `after` | `string` | — | Path (relative to `kirigami.root`) to a PHP file included **after** every page's body. Typically your layout closing. |
-| `format` | `bool` | `false` | Pretty-print the compiled HTML via [`HTML::format()`](#html) before writing it to disk. |
-| `head` | `bool` | `true` | Auto-wire each page's `<head>`: a theme/FOUC guard as the first child, a `<link rel="stylesheet">` per `sass` task output, and a `<script>` (no `defer`, before `</body>`) per `esbuild` task output — each with a per-page relative path and a `?<timestamp>` cache-bust. A file already referenced in the page is skipped. Set `false` to disable, or `head: false` on a single `sass`/`esbuild` task to skip just its tag. |
-| `network` | `bool` | `false` | Enables outbound HTTP(S) inside the WASM PHP runtime. Required for PHPDOC `@tag https://…` annotations that fetch remote `.yaml`/`.json`/`.md` data (see [Auto-loading data files](#auto-loading-data-files)), and for the `CURL` / `SCRAPER` classes. |
-| `mountext` | `string[]` | `[]` | Extra file extensions to mount automatically into the virtual filesystem alongside the built-in `.php`, `.json`, `.yaml`, `.yml`, `.md`, `.db`, `.txt`. Use this for assets your PHP code reads directly (e.g. `.svg`, `.webp`). Files with extensions not in this set are skipped during mounting — mount them on demand with [`PREPROS::mount()`](#preprosmountstringarray-patterns) instead. |
-| `includes` | `string[]` | `[]` | PHP files (relative to `kirigami.root`) `include_once`'d once, right after config is loaded — before any page renders. The natural place to `PREPROS::registerTag()`, `PREPROS::registerHook()`, or `MD::registerPlugin()`. |
-| `types` | `object` | `{}` | Named page types. A page opts in with `@type <name>` in its PHPDOC header; the matching entry's `before`/`after` (each `string`, relative to `kirigami.root`, both optional) wrap the page body **one level inside** the global `before`/`after` — render order is global before → type before → body → type after → global after. A page with no `@type`, or naming a type absent here, renders with just the global wrap. See [`@type`](#content-indent-and-type). |
-### `image` block
-Options for the image autogenerator. **Read by `php-prepros`** — these are what [`IMG::asset()` / `IMG::palette()`](#img), the [`<img asset>` tag](#img-asset), and kirigami-core's `img-asset()` / `colors()` Sass functions all resolve against. Optional; the defaults below apply even when the block is absent.
-| Key | Type | Default | Description |
-|-----|------|---------|--------------|
-| `format` | `string` | `webp` | Output format for generated images: `webp` or `avif`. |
-| `source` | `string` | `assets/images` | Folder holding the source images, relative to `cwd()`. |
-| `dest` | `string` | `images` | Destination folder for generated images, relative to `kirigami.root`. |
-### `plugins` block
-List of Kirigami plugins. **Consumed by the `kiri` CLI** (see [`@kirigami/sdk`](https://www.npmjs.com/package/@kirigami/sdk)), not by `php-prepros` directly.
-| Key | Required | Description |
-|-----|----------|--------------|
-| `name` | ✅ | Plugin package name. Must match `@kirigami/plugin-*`, `<scope>/kirigami-plugin-*`, or `kirigami-plugin-*`. |
-| `active` | ✅ | Whether the plugin is loaded. |
-| `options` | — | Free-form object passed to the plugin; its shape depends on the plugin. |
-### `esbuild` / `sass` blocks
-Free-form objects. **Consumed by the `kiri` CLI.** There is no fixed key set: whatever you put here is spread straight into the underlying library call for every matching task, *after* Kirigami's own defaults — so it can also override them (`minify`, `target`, `style: "compressed"`, source maps, …). Refer to esbuild's [`BuildOptions`](https://esbuild.github.io/api/#build-api) and Dart Sass's [`Options`](https://sass-lang.com/documentation/js-api/interfaces/options/) for what's accepted. Writing the key with nothing under it parses to `null` in YAML, equivalent to omitting the block.
-`sass:` additionally recognizes two keys that are **not** passed to Dart Sass:
-| Key | Type | Description |
-|-----|------|--------------|
-| `before` | `string` / `string[]` | Extra `.scss` files compiled **before** the task entry (paths relative to `cwd()`). |
-| `after` | `string` / `string[]` | Extra `.scss` files compiled **after** the task entry. |
-### `export` block
-Options for `kiri export`. **Consumed by the `kiri` CLI.** Optional.
-| Key | Type | Default | Description |
-|-----|------|---------|--------------|
-| `path` | `string` | `dist` | Output directory for `kiri export`, relative to the project root. |
-| `ignore` | `string[]` | `[]` | Extra gitignore-style patterns excluded from the export copy, on top of Kirigami's built-in exclusions. |
-### `scripts` block
-Named PHP scripts. **Consumed by the `kiri` CLI**, which runs each `scripts/<name>.php` through [`runenv()`](#runenvscript-paths-args) — so the full `php-prepros` class library is available and `kirigami.yaml`'s `kirigami` block is exposed as `PREPROS::$config->data`.
-| Key | Required | Description |
-|-----|----------|--------------|
-| `name` | ✅ | Must match an existing `scripts/<name>.php` file. Run with `kiri run <name> [args...]`; extra CLI arguments are forwarded as `$argv` entries. |
-| `mount` | — | Glob patterns (relative to the project root) of extra local files to mount into the sandbox before the script runs. |
-| `trigger` | — | Fire the script automatically: `before-build` (start of `build` and `export`), `before-export` (very start of `export`), or `after-export` (once `export` has finished). |
-### `tasks` block
-Ordered list of build tasks, run in array order. **Consumed by the `kiri` CLI**, on top of the implicit `prepros` task (added when the `prepros` block is present) and the implicit `dist` task (added during `kiri export`).
-| `type` | Purpose | Required fields | Optional |
-|--------|---------|-----------------|----------|
-| `esbuild` | Bundle/minify a JS/TS entry. Build + watch. Output: `<entry>.min.js`. | `name`, `type`, `entry` | `force` |
-| `sass` | Compile a `.scss`/`.sass` entry, minified with csso on export. Build + watch. Output: `<entry>.min.css`. | `name`, `type`, `entry` | `force` |
-| `prepros` | Render pages + `sitemap.xml`. Watch only (runs on build/export only when forced/implicit). | `name`, `type` | `target`, `force` |
-| `dist` | Copy `kirigami.root` into an output dir, stamping the banner. Forced/implicit only. | `name`, `type`, `path` | `ignore`, `force` |
----
-## Writing pages
-Source pages live in the directory pointed to by `kirigami.root`. The naming convention is straightforward: any file whose name starts with `_` and ends in `.php` is treated as a page source. The leading underscore is stripped in the output filename.
-```
-src/
-├── _layouts/
-├── _lib/
-├── _index.php          →  src/index.html
-├── about/
-│   └── _index.php      →  src/about/index.html
-└── blog/
-    ├── _index.php       →  src/blog/index.html
-    └── _articles.yaml   (data file, not compiled)
-```
-Directories whose name starts with `_` (e.g. `_layouts/`, `_lib/`) are skipped entirely during directory-wide builds.
-### PHPDOC header
-Every page starts with a PHP docblock that drives metadata and data loading:
-```php
-<?php
-/**
- * @name     about
- * @title    About us
- * @abstract A short description of this page.
- */
-?>
-<section>
-    <h1><?php echo $title; ?></h1>
-    <p><?php echo $abstract; ?></p>
-</section>
-```
-All annotations are injected as PHP variables (`$name`, `$title`, `$abstract`, …). You can define any custom annotation you need.
-Annotations are also available as variables in `before` and `after` PHP included files, so you can write proper metas in the HTML header.
-Only lines whose first non-whitespace character (past the `*` gutter) is `@`
-open an annotation — an `@word` written in the prose of the block is left alone.
-A value can wrap onto the following **indented** continuation lines:
-```php
-/**
- * @title       About us
- * @description A longer blurb that does not fit comfortably
- *              on a single line and continues here.
- */
-```
-### Auto-loading data files
-When an annotation value looks like a filename (with a `.yaml`, `.yml`, `.json`, or `.md` extension), it is automatically parsed and injected as a structured variable instead of a plain string.
-```php
-<?php
-/**
- * @name     medias
- * @articles _articles.yaml
- */
-?>
-<?php foreach ($articles as $article): ?>
-    <a href="<?php echo $article->lien; ?>">
-        <?php echo $article->titre; ?>
-    </a>
-<?php endforeach; ?>
-```
-| Extension | Parsed as |
-|-----------|-----------|
-| `.yaml` / `.yml` | `stdClass` object (or array of objects for sequences) |
-| `.json` | Result of `json_decode()` |
-| `.md` | HTML string via `MD::toHtml()` |
-When `network: true` is set in `kirigami.yaml`, annotation values that start with `http://` or `https://` are fetched from the network and parsed the same way:
-```php
-/**
- * @posts https://api.example.com/posts.json
- */
-```
-### `@content`, `@indent`, and `@type`
-Three special annotation names change how a page's body is assembled:
-- **`@content`** — if a `content` variable already resolves to a non-empty value (typically because it's a `.md`/`.yaml`/`.json` annotation that auto-loaded into HTML/data, see above), it is used **as-is** as the page body, and the PHP file itself is **not executed** for its output. This is handy for pages that are pure data/markdown wrapped by a shared layout.
-- **`@indent`** — when set to a number, every line of the rendered body is prefixed with that many spaces before being wrapped by `before.php`/`after.php`. Useful for keeping generated HTML readable when a page is nested inside indented layout markup.
-- **`@type`** — names an entry under [`prepros.types`](#prepros-block). If it matches, that entry's `before`/`after` wrap the (already-indented) body **one level inside** `before.php`/`after.php`: global before → type before → body → type after → global after. No match (missing annotation, or a name absent from `prepros.types`) leaves the page with just the global wrap — a page type is an extra layer, never a replacement for the site's real header/footer.
-```php
-<?php
-/**
- * @name    changelog
- * @title   Changelog
- * @content _changelog.md
- * @indent  4
- * @type    article
- */
-```
-### Built-in tags
-Two tags are registered out of the box (`prepros.plugins.php`) and processed
-**after** the PHP runs, on the assembled HTML — no include or plugin needed.
-#### `<markdown> … </markdown>`
-Converts its inner content from Markdown to HTML, stripping the common leading
-indentation first (via `STR::trimIndent()`) so you can indent it naturally inside
-your template. All registered [MD plugins](#md-plugins) work inside it. See
-[Extending the `<markdown>` tag](#extending-the-markdown-tag) to override it.
-```html
-<section>
-    <markdown>
-        ## Who we are
-        We are a **student organization** from Québec.
-    </markdown>
-</section>
-```
-Add the `prose` attribute — `<markdown prose>` — to wrap the output in
-`<div class="prose">`, so it picks up the long-form typography of
-[`@kirigami/canva`'s `styles/prose`](https://www.npmjs.com/package/@kirigami/canva)
-with no extra markup. Any `class` / `id` on the tag lands on that wrapper
-(`<markdown prose class="lede" id="intro">` → `<div class="prose lede" id="intro">`).
-A bare `<markdown>` emits just the converted HTML, as before.
-#### `<img asset="…">`
-The HTML-side entry point of the image autogenerator — the exact same feature as
-the [`img-asset()` Sass function](https://www.npmjs.com/package/@kirigami/kirigami#sass-functions)
-and [`IMG::asset()`](#img), with the same parameters. The tag calls `IMG::asset()`
-under the hood, then swaps the `asset` attribute for the generated `src`.
-```html
-<!-- in:  resolves assets/images/hero.jpg through IMG::asset('hero.jpg', 800, 0, false) -->
-<img asset="hero.jpg" width="800" alt="Our office" loading="lazy">
-<!-- out: <img src="../images/hero-800w.webp" alt="Our office" loading="lazy"> -->
-```
-| Attribute | Maps to `IMG::asset()` arg | Notes |
-|-----------|---------------------------|-------|
-| `asset` | `$path` | **Required.** Path relative to `image.source`. Missing/empty ⇒ the tag is left untouched. |
-| `width` | `$width` | Optional, integer. Omitted ⇒ `0` (keep). |
-| `height` | `$height` | Optional, integer. Omitted ⇒ `0` (keep). |
-| `cover` | `$cover` | Boolean — **presence means `true`** (crop + fill). |
-| *(any other)* | — | `alt`, `class`, `id`, `loading`, … are passed straight through onto the output `<img>`. |
-`asset` / `width` / `height` / `cover` are consumed and removed; everything else
-survives. The generated file lands in `image.dest` and is only (re)generated when
-missing or older than the source — see [`IMG`](#img) for the naming convention.
-> The Sass `img-asset()` / `colors()` functions, the `<img asset>` tag and
-> `IMG::asset()` all run on the **same engine** — the `IMG` class (GD, with the
-> Imagick fallback) in this package. `@kirigami/kirigami`'s `sass` task routes its
-> image work here through [`processImages()`](#processimagesjobs), so there is no
-> native image dependency in the toolchain.
----
-## JavaScript API
-```js
-import { render, sitemap, runenv, mountPath, processImages, resetRuntime } from '@kirigami/php-prepros';
-```
-PHP operations are queued in call order, including mounts and result extraction.
-`await resetRuntime()` waits for preceding PHP work, disposes the owned runtime
-and its network proxy, and clears cached configuration and mounts. The next
-operation initializes a fresh runtime from `kirigami.yaml`. Core
-`Project.reload()` calls this and also clears the plugin PHP include list.
-It does not clear persistent cache/cookie files on disk. This remains a
-single-project API whose working directory must be set before import.
-### `render(file?, phpIncludes?)`
-`phpIncludes` defaults to `[]`. The core collects it through `prepros:php`;
-direct callers supply local PHP file paths (absolute or relative to the
-working directory). Existing paths are mounted under `/plugins/` and included
-before rendering; missing paths are silently skipped. Each render replaces
-the runtime include list, which remains on its configuration until another
-render or reset. Direct calls do not load the core plugin registry for you.
-Compile a single PHP page or a whole directory.
-```js
-// Compile one page
-const pageResult = await render('about/_index.php');
-// Compile everything under src/
-const treeResult = await render('.');
-// Compile everything (uses kirigami.root from config)
-const defaultResult = await render();
-```
-> Paths used by `render()` are all relative to the `kirigami.root` configuration.
-**Returns** `Promise<PreprosResult>`:
-```ts
-interface PreprosResult {
-  success: boolean;
-  files?:  string[];   // project-relative paths; may be absent on parsing failure
-  error?:  string;
-  debug?:  string;     // captured PHP stdout
-  stderr?: string;     // diagnostics attached to failures
-  warnings?: string;   // nonfatal stderr on success
-  page?:   string | null;     // PHP-render failure context, when available
-  where?:  string;     // PHP source location, when available
-}
-```
-Setup and filesystem failures can reject before a result exists; PHP failures
-usually return `success: false`. Check both channels. Files may already have
-been copied to the host before a later error; operations are not transactional.
-The runtime returns `debug`/`stderr`, not the older declared `response` field.
-Directory rendering selects `_*.php` files only when every directory between
-`kirigami.root` and the page has a name without a leading underscore. Sitemap
-selection uses the same rule. Direct requests for private pages fail; rendering
-a private directory produces no pages. The configured source root itself may
-start with `_` (for example `_src`). Previously generated private HTML is not
-deleted by this selection rule; remove stale outputs when migrating a site.
-### `sitemap()`
-Generate `sitemap.xml` and `robots.txt` at the source root, plus `humans.txt`
-when author configuration provides content. It accepts no directory argument.
-```js
-const result = await sitemap();
-// With kirigami.root: src, files includes src/sitemap.xml and src/robots.txt.
-```
-### `runenv(script, paths?, ...args)`
-Run an arbitrary PHP script — not a page template — inside the very same sandboxed WASM environment used for `render()`, with the full `php-prepros` class library autoloaded and `kirigami.yaml`'s `kirigami` block available as `PREPROS::$config->data`. Useful for one-off maintenance scripts, data migrations, or CLI-style tooling that needs `CACHE`, `SCRAPER`, `IMG`, etc. without going through the page-rendering pipeline.
-```js
-// Run a standalone PHP script
-const purgeResult = await runenv('scripts/purge-cache.php');
-// Also mount explicit extra files into the sandbox before running
-const imageResult = await runenv('scripts/build-og-images.php', ['assets/photos/hero.jpg']);
-// Extra arguments are appended and available as $argv[2], $argv[3], … in the script
-const importResult = await runenv('scripts/import.php', [], '--force');
-```
-- `script` — path to a PHP file **inside the project**, executed with `require_once`.
-- `paths` — optional array of explicit local file paths, not directories. Missing
-  files are skipped; a directory can cause a filesystem rejection. Use
-  `mountPath()` first for recursive directory mounting.
-- `...args` — extra string arguments appended to the script's `$argv`.
-Script and extra-file paths resolve against the project captured at import.
-Before initializing PHP or copying files, `runenv()` rejects paths outside
-that project, including symbolic links whose real targets are outside it.
-Directories are rejected; missing optional files are skipped. This check does
-not make PHP scripts untrusted-code sandboxes or restrict explicit `mountPath()`
-calls.
-**Returns** `Promise<PreprosResult>`, following the same shape as `render()`. Inside the script, call `PREPROS::exportFile()` for any file you want listed in `result.files`.
-### `runPluginScript(script, pluginRoot, paths?, ...args)`
-Same as `runenv()`, for a script shipped inside a plugin package. A plugin
-installed with `npm link` or from a workspace lives outside the project, so
-`runenv()` would reject it. Here the script's authored and real paths must stay
-inside `pluginRoot` instead, and it is mounted under
-`/plugin-scripts/<package dir>/`. Extra `paths` are still project files.
-The caller vouches for `pluginRoot`: `@kirigami/kirigami` only passes the
-resolved package directory of an active plugin.
-### `mountPath(localPath, virtualDir?, php?)`
-The JavaScript-side counterpart to [`PREPROS::mount()`](#preprosmountstringarray-patterns). Mounts a local file or directory — recursively, preserving structure — into the WASM sandbox's virtual filesystem, ahead of (or between) calls to `render()`, `sitemap()`, or `runenv()`. Useful when a Node-side build step needs to make extra local files visible to PHP before rendering starts.
-```js
-import { mountPath, render } from '@kirigami/php-prepros';
-// Mount a single file at its natural virtual path (/project/<relative path>)
-await mountPath('assets/data/team.yaml');
-// Mount a whole directory, at a custom virtual path
-await mountPath('vendor/fonts', '/project/fonts');
-await render();
-```
-- `localPath` — path to a local file or directory. Relative paths are resolved against the project root.
-- `virtualDir` — optional destination path inside the WASM filesystem. Defaults to `/project/<localPath relative to the project root>` when omitted.
-- `php` — optional WASM PHP instance to mount into. Defaults to PHP-prepros's owned instance (the same one used internally by `render()`/`sitemap()`/`runenv()`), creating it if needed. This is separate from PHP-WASM's shared getter instances and is replaced after `resetRuntime()`.
-Mounting a **directory** only copies files whose extension is one of the defaults (`.php`, `.json`, `.yaml`, `.yml`, `.md`, `.db`, `.txt`) or listed in `prepros.mountext`, same as automatic root mounting. Mounting a **single file directly** copies it regardless of extension — this is the simplest way to make an arbitrary asset (an image, a font, a CSV, …) available to PHP without adding its extension to `prepros.mountext` project-wide.
-**Returns** `Promise<void>`.
-### `processImages(jobs)`
-Run a batch of image jobs — resize/encode, or palette extraction — through the
-[`IMG`](#img) class (GD, with the Imagick fallback). This is the engine
-`@kirigami/kirigami`'s `sass` task uses for its `img-asset()` and `colors()`
-functions, so Sass, `IMG::asset()` and the [`<img asset>` tag](#built-in-tags)
-all share one implementation, one `image:` config and one set of output
-filenames — with no native image dependency.
-```js
-import { processImages } from '@kirigami/php-prepros';
-const { files, colors } = await processImages([
-  // resize/encode `hero.jpg` (resolved against image.source) to each dest —
-  // absolute virtual paths, already carrying the target extension
-  { op: 'resize', src: 'hero.jpg', width: 1200, height: 0, cover: false, quality: 82,
-    dests: ['/project/src/images/hero-1200w.webp'] },
-  // extract a 5-colour palette (cached in .cache.db); returned, not written
-  { op: 'palette', src: 'hero.jpg', count: 5 },
-]);
-// files includes 'src/images/hero-1200w.webp' and may include '.cache.db'.
-// colors → { 'hero.jpg:5': ['#1e3a5f', '#c8a24b', …] }
-```
-- `jobs` — array of `resize` / `palette` jobs (see the shape above). An **empty
-  array is a no-op** and does **not** start the WASM runtime.
-- Omitted `jobs` also returns `{ success: true, files: [], colors: {} }`
-  without starting PHP. Non-array input currently does the same; this is not
-  strict input validation.
-- Staleness is the caller's responsibility: every `resize` job listed is executed.
-- `resize`: `width`/`height` default to `0` (preserve size when both are zero),
-  `cover` to `false`, and lossy encoder quality to `82`. `dests` contains
-  absolute `/project/...` output paths. `palette` defaults `count` to `5`.
-- The PHP worker handles `palette` explicitly and treats any other `op` as a
-  resize; pass only the two documented operations. Processing stops at the
-  first exception. Always check `success` before using files or colors.
-**Returns** `Promise<PreprosResult & { colors: Record<string, string[]> }>`.
-### `resetRuntime()`
-Returns `Promise<void>`. Queued after preceding operations, it disposes the
-owned runtime, mounts, and cached configuration. It does not change the project
-path captured at import, clear disk caches, or reset the SDK hook registry.
-### TypeScript declarations
-The shipped `index.d.ts` includes `render(file?, phpIncludes?)`, argument-free
-`sitemap()`, explicit-file `runenv()` mounts, and the current diagnostic fields.
-`PreprosResult.files` is optional because response parsing can fail before a
-file list exists. `ImageBatchResult.files` is always normalized to an array.
-The obsolete `response` field is replaced by `debug` and `stderr`.
----
-## PHP classes reference
-All classes are autoloaded — no manual `require` needed inside your page files.
-The autoloader itself, `$argv`/`$config`, procedural aliases, and the `boot`
-hook are installed via php.ini's `auto_prepend_file` (pointed at
-`utils.inc.php`), set once per WASM runtime instance — every entrypoint
-(`prepros.php`, `runenv.php`, `imagebatch.php`) gets it automatically,
-with no `include` of its own.
----
-### PREPROS
-The core engine. Manages the rendering pipeline, tag processing, hooks, mounting, and file export.
-```php
-// Available inside page templates and included files.
-PREPROS::$config          // stdClass — full resolved config; ->data is the kirigami: block,
-                          // ->image the image: block, plus before/after/format/… from prepros:
-PREPROS::registerTag(string $tag, callable $callback)
-PREPROS::registerHook(string $hook, callable $callback)
-PREPROS::runHook(string $hook, mixed $data = null)  // fire a hook (built-in or your own), returns the piped $data
-PREPROS::mount(string|array $patterns)
-PREPROS::exportFile(string|array $absolutePath)
-PREPROS::getExportedFiles(): string[]
-PREPROS::fstat(string $path)          // stat a file in the WASM FS (or false)
-PREPROS::backtraceFile()              // path of the page currently rendering
-```
-#### `PREPROS::render(string $file)`
-Internal method called once per source file. Orchestrates the full pipeline:
-1. Resolves PHPDOC metadata and auto-loads data files.
-2. Fires the `pre_render` hook with the raw source contents.
-3. Includes `before.php` (wrapped in the `pre_before` / `post_before` hooks) and the page body (or `@content`, see [above](#content-indent-and-type)).
-4. If the page declares `@type <name>` and `prepros.types.<name>` exists, wraps the body with that type's `before`/`after` (wrapped in `pre_type_before` / `post_type_before` and `pre_type_after` / `post_type_after`) — nested inside the global wrap.
-5. Includes `after.php` (wrapped in `pre_after` / `post_after`), assembling everything into a single string.
-6. Processes all registered custom HTML tags.
-7. Fires the `post_render` hook on the assembled HTML.
-8. Optionally pretty-prints via `HTML::format()` (when `format: true`).
-9. Writes the output `.html` file.
-#### `PREPROS::sitemap()`
-Scans the source tree for `_index.php` files and generates a standards-compliant `sitemap.xml` (Sitemaps 0.9), using `kirigami.baseurl` as the root URL.
-#### `PREPROS::mount(string|array $patterns)`
-Mounts additional local project files into the WASM virtual filesystem, on demand, from one or more glob patterns evaluated against the project root (via `picomatch`). Unlike the automatic mounting done for `kirigami.root` (limited to `.php`, `.json`, `.yaml`, `.yml`, `.md`, `.db`, `.txt`, and `prepros.mountext`), `mount()` copies **any** matching file, regardless of extension.
-```php
-// Mount every .webp under assets/, wherever the page needs them
-PREPROS::mount('assets/**/*.webp');
-// Multiple patterns at once
-PREPROS::mount(['data/**/*.csv', 'vendor/fonts/*.woff2']);
-```
-Returns an array of the virtual paths (under `/project/...`) that were mounted, or `false` on failure.
-#### `PREPROS::exportFile(string $file)`
-Marks a file as a build output so it gets surfaced in `PreprosResult.files`. Called automatically by `render()`, `sitemap()`, `CACHE::set()`, and `CURL`. Call it manually if your custom code writes additional files.
----
-### MD
-Markdown-to-HTML converter with a plugin system for custom shortcodes,
-backed by PHP's native `mdhtml` extension (real `cmark-gfm`), statically
-built into `@kirigami/php-wasm` — no userland parsing.
-```php
-$html = MD::toHtml(string $markdown): string;
-```
-Supports the full GitHub Flavored Markdown subset, plus a few extensions:
-- ATX (`#` … `######`) and Setext headings, with auto-generated `id` attributes
-- Ordered and unordered lists, including nested
-- GFM task lists (`- [ ]` / `- [x]`)
-- GFM tables with column alignment
-- GFM alerts (`> [!NOTE]`, `> [!WARNING]`, etc.)
-- Blockquotes (recursive)
-- Fenced code blocks with language class
-- Inline code
-- Bold, italic, bold+italic, strikethrough
-- Links with automatic `target="_blank" rel="noopener noreferrer"` for external URLs
-- Images with `loading="lazy"`
-- Auto-linked bare URLs
-- Horizontal rules
-- Hard line breaks (trailing double space → `<br>`)
-- **Footnotes** — `[^1]` references and `[^1]: …` definitions (multi-paragraph)
-- **Definition lists** — `Term` / `: Definition`
-- **Emoji shortcodes** — `:rocket:` → 🚀, from a built-in map (see `MD::registerEmoji()`)
-- **Sanitized inline HTML** — raw tags are filtered against an allowlist of tags and attributes, not passed through verbatim
-#### Plugin API
-Extend Markdown with custom shortcode tags:
-```php
-// Inline tag  {% tagname arg1 "arg with spaces" %}
-// Block tag   {% tagname arg1
-//             body content
-//             %}
-MD::registerPlugin(string $name, callable $callback): void
-MD::unregisterPlugin(string $name): void
-MD::getRegisteredPlugins(): string[]
-MD::registerEmoji(string $shortcode, string $char): void   // `:name:` → char
-```
-The callback always receives `(array $args, string $body)`:
-```php
-MD::registerPlugin('video', function (array $args, string $body): string {
-    $src = htmlspecialchars($args[0] ?? '', ENT_QUOTES, 'UTF-8');
-    return "<video src=\"{$src}\" controls></video>";
-});
-```
-Then in any Markdown content (including inside `<markdown>` tags):
-```
-{% video /videos/intro.mp4 %}
-```
----
-### HTML
-Pretty-printer for the final HTML output. Used automatically when `format: true` is set in the config.
-```php
-$formatted = HTML::format(string $html): string;
-```
-Uses PHP 8.4's `Dom\HTMLDocument` (Lexbor engine) to parse the input and re-serialize it with consistent 4-space indentation. Inline elements, `<script>`, and `<style>` blocks are handled correctly — their content is indented but not reformatted. A `<pre><code>` block is shifted to its nesting depth too (relative indentation kept), and the leading run is stripped again before display; a bare `<pre>` and `<textarea>` stay byte-for-byte. Boolean HTML5 attributes (`muted`, `autoplay`, `noopener`, etc.) are written without a value.
----
-### YAML
-PHP data files use the native `yaml` extension backed by LibYAML. Its YAML 1.1 implicit booleans include unquoted `y`, `n`, `yes`, `no`, `on`, `off`, `true`, and `false`, including mapping keys. Quote these words when you mean strings (for example, `"NO": Norway`). `YAML::parse()` / `parseFile()` / `loadFile()` preserve the wrapper’s array/object choice; native `yaml_parse()` / `yaml_parse_file()` have their own extension signatures. `yaml_load_file()` remains a wrapper alias. The project’s `kirigami.yaml` is parsed separately in Node through `struct-walker` and `js-yaml`.
-A YAML parser backed by PHP's native `yaml` extension (libyaml), statically built into `@kirigami/php-wasm` — full YAML 1.1 support, no userland parsing.
-```php
-$data = YAML::parse(string $yaml, bool $assoc = false): mixed;
-$data = YAML::parseFile(string $path, bool $assoc = false): mixed;
-$data = YAML::loadFile(string $path, bool $assoc = false): mixed;
-```
-By default, YAML mappings are returned as `stdClass` objects. Pass `true` as the second argument to get associative arrays instead.
-Following YAML 1.1 means the usual implicit-boolean gotcha applies to both values *and* mapping keys: `y`/`Y`/`n`/`N`, `yes`/`no`, `true`/`false`, `on`/`off` (any case) all resolve to a boolean when unquoted — so an unquoted `no:` key or a `NO` value becomes `false`. Quote a scalar (`"y": 2`) to keep it a string.
-`YAML::loadFile()` behaves like `YAML::parseFile()`, then walks the result recursively: any string value ending in `.yaml`, `.yml`, or `.json` that resolves to an existing file (relative to *its own* file's directory) is replaced by that file's parsed content, and so on, recursively. Values that don't match an existing file are left untouched. Circular references (`A → B → A`) throw a `RuntimeException`.
-```yaml
-# team.yaml
-lead: people/jane.yaml     # resolved and inlined automatically
-members:
-  - people/jane.yaml
-  - people/john.yaml
-```
-```php
-$team = YAML::loadFile('/project/data/team.yaml');
-// $team->lead is now the fully parsed content of people/jane.yaml, not a string
-```
----
-### SCHEMA
-A JSON Schema validator with an Ajv-like API, backed by the native `jsonk`
-extension built into `@kirigami/php-wasm`. Available to your own code and
-plugins.
-```php
-$validator = new SCHEMA(array $schema);
-$validator->isValid(mixed $data): bool     // true / false
-$validator->validate(mixed $data): bool    // alias of isValid()
-$validator->getErrors(): string[]          // "path: message" strings from the last run
-```
-jsonk implements draft 2020-12 for a self-contained schema: every validation
-keyword (`type`, `enum`, `const`, `required`, `properties`,
-`patternProperties`, `additionalProperties`, `propertyNames`,
-`dependentRequired`, `dependentSchemas`, `items`, `prefixItems`, `contains`,
-`uniqueItems`, the `min*`/`max*` and `exclusive*` bounds, `multipleOf`,
-`pattern`, `format`, `if`/`then`/`else`, `allOf`/`anyOf`/`oneOf`/`not`) and
-`$ref` to `#/$defs/…` / `#/definitions/…`, absolute URLs, or URLs relative to
-the schema's `$id`. See [php-jsonk](https://github.com/php-kirigami/php-jsonk)
-for the details and limits.
-Schemas are PHP arrays, so `SCHEMA` adapts them before handing them to jsonk:
-an empty array in a schema position (`'properties' => []`) is treated as `{}`,
-draft-07 tuple `items` (a list of schemas) becomes `prefixItems` (and
-`additionalItems` becomes `items`), and `format: url` is read as `uri`. Error
-paths look like `(root)`, `name` or `tags[1]`.
-The previous pure-PHP (Draft-7 style) validator is still available as
-`SCHEMA_LEGACY`, with the same API.
-```php
-$validator = new SCHEMA([
-    'type'     => 'object',
-    'required' => ['name', 'age'],
-    'properties' => [
-        'name' => ['type' => 'string', 'minLength' => 1],
-        'age'  => ['type' => 'integer', 'minimum' => 0],
-    ],
-    'additionalProperties' => false,
-]);
-if (!$validator->isValid($data)) {
-    foreach ($validator->getErrors() as $err) echo $err, PHP_EOL;
-}
-```
----
-### LD
-A **schema.org JSON-LD generator**. `LD` accumulates structured-data nodes for
-the page under render and emits them as one
-`<script type="application/ld+json">` block — with an `@graph` when there is more
-than one node — in the `<head>`.
-#### Automatic mode
-On as soon as `kirigami.yaml` has a [`seo:` block](#seo-block) (`seo: {}` is
-enough), alongside [`META`](#meta)'s tags and from the same keys. A
-`post_render` hook injects a graph built from `seo:`, the loose keys of the
-`kirigami` block, and the current page's PHPDOC:
-- an `Organization` node (`@id` `#organization`) — `name`/`url`/`description`
-  from `project`/`baseurl`/`description`, `sameAs` gathered from every
-  recognised social-network URL key (`facebook`, `instagram`, `linkedin`,
-  `github`, `youtube`, `mastodon`, …), plus `email`, `telephone`, `areaServed`
-  (← `area`), `knowsAbout` (← `knowsabout`), `address`, `logo`, and `founder` →
-  the Person node when there is one. `@type` comes from `seo.type`;
-- a `Person` node (`#person`) when `person` is set — `name` + `jobTitle`
-  (← `jobtitle`) + `email` + `url`, linked to the Organization via `worksFor`;
-- a `WebSite` node (`#website`) — `publisher` → Organization, `inLanguage`,
-  `keywords` (← `keywords`), and a `SearchAction` when `seo.search` is set;
-- a `WebPage` node for the page — see the per-page tags below;
-- a `BreadcrumbList` for every non-home page, derived from the `_index.php`
-  ancestor trail (home → each parent section → this page). No `@breadcrumb`
-  opt-in needed. Disable it for one page with `@ld_breadcrumb false`.
-`seo: { jsonld: false }` stops the automatic pass; META's tags keep working.
-A page whose rendered `<head>` already contains an `application/ld+json`
-script is never touched, so hand-rolled markup keeps working.
-**Per-page PHPDOC tags** — these feed the page node (and override the generic
-`@title` / `@description` / `@datePublished` fallbacks):
-| Tag | Effect |
-|-----|--------|
-| `@ld false` | Skip JSON-LD for this page entirely (`@ld_ignore true` also works). |
-| `@ld_type <Type>` | `@type` of the page node — `AboutPage`, `ContactPage`, `CollectionPage`, `ProfilePage`, or a content type like `Article`, `Service`, `Recipe`, … (default `WebPage`). Types containing “Page” also get `primaryImageOfPage` + a `breadcrumb` link; others get a plain `image`. |
-| `@ld_title <text>` | Page node `name` (default: `@title`). |
-| `@ld_description <text>` | Page node `description` (default: `@description`). |
-| `@ld_image <path>` | Page image, absolute or relative to `baseurl` (default: `@image` / `@ogimage`). |
-| `@ld_published <date>` | `datePublished` (default: `@datePublished` / `@published` / `@date`). |
-| `@ld_modified <date>` | `dateModified` (default: `@dateModified` / `@modified` / `@updated`). |
-| `@ld_breadcrumb false` | No `BreadcrumbList` for this page. |
-```php
-/**
- * @title            À propos
- * @ld_type          AboutPage
- * @ld_title         À propos de Humain Humain
- * @ld_description   Notre approche ethnographique de la consultation.
- */
-```
-#### Explicit builders
-Call these from a page template or from a `prepros.includes` file. Nodes added
-this way are always emitted — automatic pass on or not — and share
-the graph the automatic pass uses, so the two combine; a node with a stable
-`@id` is merged on repeat calls.
-```php
-LD::add(string|array $type, array $props = [], ?string $id = null): array  // build + register a node
-LD::node(string|array $type, array $props = []): array                      // build only, no register
-LD::push(array $node): array                                                // register a ready-made node
-LD::ref(string $id): array                                                  // ['@id' => …]  ('#person' → the Person node)
-LD::remove(string $id): void
-LD::graph(): array
-LD::reset(): void
-LD::organization(array $overrides = []): array   // config-aware, @id #organization
-LD::person(array $overrides = []): array          // config-aware, @id #person
-LD::website(array $overrides = []): array          // config-aware, @id #website
-LD::webPage(array $overrides = []): array           // current-page-aware, @id …#webpage
-LD::breadcrumb(?array $items = null, array $overrides = []): array   // items: [['name'=>…,'url'=>…], …]
-LD::faqPage(array $qa, array $overrides = []): array                 // qa: ['Question ?' => 'Answer.', …]
-LD::address(array|string $a): array
-LD::image(string $url, ?string $id = null, ?int $w = null, ?int $h = null): array
-LD::geo(float $lat, float $lng): array
-LD::rating(int|float $value, ?int $count = null, $best = 5, $worst = 1): array
-LD::offer(array $o): array
-LD::contactPoint(array $c): array
-LD::searchAction(string $urlTemplate): array
-LD::script(bool $pretty = true): string   // <script>…</script>, and disables auto-injection
-LD::json(bool $pretty = true): string     // the document, no wrapper
-```
-Every other schema.org type is reachable through `__callStatic` — the method
-name is upper-cased to form the `@type`:
-```php
-LD::recipe([ 'name' => 'Tarte aux pommes', 'recipeYield' => '6', 'prepTime' => 'PT30M' ]);
-LD::event([ 'name' => 'Vernissage', 'startDate' => '2026-10-01T18:00' ]);
-LD::softwareApplication([ 'name' => 'Kirigami', 'applicationCategory' => 'DeveloperApplication' ]);
-LD::article([
-    'headline'      => $title,
-    'datePublished' => '2026-09-01',
-    'author'        => LD::ref('#person'),
-    'image'         => LD::image('images/cover.webp'),
-    'publisher'     => LD::ref('#organization'),
-]);
-```
-Same API from procedural code: `ld_add()`, `ld_node()`, `ld_ref()`,
-`ld_organization()`, `ld_person()`, `ld_website()`, `ld_web_page()`,
-`ld_breadcrumb()`, `ld_faq_page()`, `ld_script()`, `ld_json()`.
-LD's keys (`type`, `logo`, `person`, `address`, `search`, …) live in the
-[`seo` config](#seo-config) with META's.
----
-### META
-A **`<head>` SEO / social metadata generator** — the companion to [`LD`](#ld).
-Where `LD` emits a schema.org `application/ld+json` graph, `META` emits the plain
-tags a browser and a link-preview crawler read: `<title>`, `<meta name="…">`,
-`<meta property="og:…">`, `<meta name="twitter:…">`, and a handful of `<link>`s.
-It draws on the same sources, in this order of precedence: the page's PHPDOC, the
-top-level `seo:` block, then the loose `kirigami:` keys. Every tag is emitted **only when it can be resolved** — no value, no
-tag — and a tag the page's layout already writes by hand is detected and
-skipped, so it drops in beside an existing `header.php` without duplicating
-anything.
-**Automatic mode** is opt-in: the top-level `seo:` block (empty `seo: {}` is
-enough) turns on injection into every page's `<head>`, right before `</head>`.
-```yaml
-kirigami:
-  project:     Humain Humain
-  baseurl:     https://humainhumain.com
-  tagline:     Ethnographie au service des organisations
-  description: A social-science consultancy using ethnography for organisational change.
-  keywords:    [ethnographie, consultation publique, sciences sociales]
-  author:      Maxime Larrivée-Roy
-seo:                              # top-level; the block being present is the switch
-  twitter: "@humainhumain"
-  themeColor: "#0b7285"
-  lang: fr-CA                     # <meta name="language">, og:locale, JSON-LD inLanguage
-  logo: assets/logo.png           # JSON-LD logo, and og:image when there is no `image`
-```
-Per-page, from the PHPDOC block — each falls back to the generic page tag:
-| Tag | Feeds | Default |
-|-----|-------|---------|
-| `@meta false` | skip metadata for this page entirely | — (`@meta_ignore true` also works) |
-| `@meta_title` | `<title>`, `og:title`, `twitter:title` | `@title` |
-| `@meta_description` | `description`, `og:description`, `twitter:description` | `@description` / `@abstract` / `@excerpt` / `@summary`, then the site `description` |
-| `@meta_keywords` | `<meta name="keywords">` | `@keywords`, then `seo.keywords` |
-| `@meta_image` | `og:image`, `twitter:image` | `@image` / `@ogimage`, then `seo.image` |
-| `@meta_robots` | `<meta name="robots">` | `@robots`, then `seo.robots` |
-| `@meta_type` | `og:type` | `@og_type`, then `seo.ogType` |
-| `@canonical` | `<link rel="canonical">` | derived from the file path + `baseurl` |
-**Manual builders** — always emitted (with or without a `seo:` block), still
-de-duplicated against the page:
-```php
-META::tag(string $name, ?string $content): void   // name= , or property= for an og:* key
-META::link(string $rel, string $href, array $attrs = []): void
-META::raw(string $html): void                       // a verbatim, already-valid tag line
-META::tags(string $html = ''): string               // the whole block, \n-joined
-META::reset(): void
-```
-```php
-META::tag('twitter:image', 'https://humainhumain.com/card.png');
-META::tag('og:image:alt', 'The Humain Humain team at work');
-META::link('icon', './favicon.svg', ['type' => 'image/svg+xml']);
-```
-Same API from procedural code: `meta_tag()`, `meta_link()`, `meta_raw()`,
-`meta_tags()`.
-<a id="meta-config"></a><a id="jsonld-config"></a>
-#### `seo` config
-These keys live directly under the **top-level** `seo:` block of
-`kirigami.yaml` (a sibling of `kirigami:`, `prepros:`, …) and feed both META's
-tags and LD's JSON-LD. All are optional.
-| Key | Type | Description |
-|-----|------|-------------|
-| `auto` | `bool` | Inject META's tags automatically. Default `true` **once the `seo:` block exists**. `auto: false` keeps the block for its values but stops the tags — `META::tags()` / `meta_tags()` can place them by hand. |
-| `jsonld` | `bool` | Inject LD's JSON-LD automatically. Default `true` **once the `seo:` block exists**. `jsonld: false` stops it — `LD::script()` / `ld_script()` can place it by hand. |
-| `titleFormat` | `string` | `<title>` template for a normal page. Tokens `{title}`, `{project}`, `{tagline}`. Dangling separators from an empty token are trimmed. Default `{title} — {project}`. |
-| `titleFormatHome` | `string` | Title template when the page has no `@title` (home / section landings). Default `{project} — {tagline}`. |
-| `description` | `string` | Default description for pages with no `@description` / `@abstract`, and the JSON-LD main entity / WebSite description. Defaults to the loose `description`. |
-| `keywords` | `string[]` \| `string` | Default `keywords` content (list or comma string) and the WebSite `keywords`. Defaults to the loose `keywords`. |
-| `robots` | `string` | Default robots directive. Default `index, follow`. `robots: false` omits the tag. |
-| `lang` | `string` | BCP-47 tag → `<meta name="language">`, `og:locale` (dash→underscore) and JSON-LD `inLanguage`. Defaults to the loose `lang` / `language`, then `en`. |
-| `generator` | `string` \| `false` | `<meta name="generator">`. Default `Kirigami`; `false` omits it. |
-| `author` / `designer` | `string` | Default to the loose `author` / `designer` keys (author also falls back to `person`'s name). `designer` is not emitted unless set. |
-| `themeColor` | `string` | `<meta name="theme-color">`. Not emitted unless set. |
-| `image` | `string` | Default `og:image` / `twitter:image` and JSON-LD image — absolute URL or path relative to `baseurl`. Defaults to `logo`, then the loose `image` / `ogimage`. |
-| `logo` | `string` | Organization logo (JSON-LD `ImageObject`), absolute URL or path relative to `baseurl`. |
-| `ogType` | `string` | Default `og:type`. Default `website`. |
-| `twitterCard` | `string` | `twitter:card` type. Default `summary_large_image`. |
-| `twitter` | `string` \| `map` | Handle for `twitter:site` / `twitter:creator`. A bare string (with/without `@`, or a profile URL) fills both; a map takes `site` / `creator` separately. |
-| `canonical` | `bool` | Emit `<link rel="canonical">`. Default `true`. |
-| `favicon` / `appleTouchIcon` / `humans` | `string` \| `bool` | `<link rel="icon">` / `rel="apple-touch-icon"` / `rel="author"`. A path sets it (page-relative when a bare filename); `true` forces the default file (`favicon.ico` / `apple-touch-icon.png` / `humans.txt`); omitted, the default file is auto-detected on disk at the source root; `false` disables it. |
-| `type` | `string` | JSON-LD `@type` of the main entity — `Organization` (default), `ProfessionalService`, `LocalBusiness`, … |
-| `name` / `url` | `string` | JSON-LD main entity / WebSite name and URL. Default to `project` / `baseurl`. |
-| `sameAs` | `string[]` | Profile URLs, merged with the social-network URL keys found loose in the `kirigami` block. |
-| `email` / `telephone` | `string` | Default to the loose `email` / `telephone` keys. |
-| `address` | `map` | `PostalAddress` properties. |
-| `areaServed` | `string` | Defaults to the loose `area` key. |
-| `knowsAbout` | `string[]` | Defaults to the loose `knowsabout` key. |
-| `person` | `string` \| `map` | The `#person` node (and the default `author`). A string is the name; a map takes any `Person` property. Defaults to `person` + `jobtitle` + `email`. |
-| `search` | `string` | URL template for a sitelinks `SearchAction`; must contain `{search_term_string}`. |
-```yaml
-kirigami:
-  project:  Humain Humain
-  baseurl:  https://humainhumain.com
-  person:   Méralie Murray-Hall
-  jobtitle: Anthropologue
-  facebook: https://www.facebook.com/humainhumainconsultation.ethnographie/
-seo:
-  type: ProfessionalService
-  lang: fr-CA
-  logo: assets/logo.png
-  knowsAbout: [Ethnographie, Recherche qualitative]
-  address:
-    addressLocality: Québec
-    addressCountry:  CA
-  search: https://humainhumain.com/?q={search_term_string}
-```
----
-### CACHE
-Persistent SQLite-backed key-value cache. Survives across incremental builds via `.cache.db` at the project root.
-```php
-CACHE::get(string $key): mixed
-CACHE::set(string $key, mixed $val, int $ttl = 0): bool
-CACHE::delete(string $key): bool
-CACHE::purge(): bool   // removes expired entries
-```
-The `$ttl` is in seconds. `0` means the entry never expires. Typical use case: caching the result of network fetches in custom hooks or plugins — it is what powers [`SCRAPER`](#scraper) internally. `CURL` persists its cookie jar separately in `.cookie.txt`.
-```php
-$data = CACHE::get('my-remote-data');
-if ($data === null) {
-    $data = json_decode(file_get_contents('https://api.example.com/data.json'));
-    CACHE::set('my-remote-data', $data, 3600); // cache for 1 hour
-}
-```
----
-### IMG
-Image manipulation helper. GD handles JPEG, PNG, GIF, WebP and AVIF directly;
-anything GD can't decode (HEIC, TIFF, BMP, and the vector formats SVG, EPS, AI,
-PDF) falls back to Imagick, which rasterizes it to a GD image in memory. Vector
-files with no intrinsic pixel size are rasterized at 2000&nbsp;px on the longest
-side, preserving the aspect ratio.
-```php
-$img = new IMG(string $file);
-// Properties
-$img->width   // int
-$img->height  // int
-// Instance methods (resize/save are chainable)
-$img->resize(int $width, int $height = 0, bool $cover = false): self
-$img->save(string $dest, ?int $quality = null): self       // quality 0-100 for jpg/webp/avif; null = per-format default (82)
-$img->getRepresentativeColors(int $count = 5): string[]   // ['#rrggbb', …], median-cut + Lab merge
-$img->getAuraColors(int $count = 5): string[]             // ['#rrggbb', …], up to 6, via the Aura extension
-// Static helpers
-IMG::asset(string $path, int $width = 0, int $height = 0, bool $cover = false): string  // same feature as the <img asset> tag and the img-asset() Sass function
-IMG::palette(string $path, int $colors = 5): string[]      // Aura-backed: at most 6 colours
-```
-`resize()` operates in *contain* mode by default (scales to fit within the target box while preserving aspect ratio). Pass `$cover = true` to crop and fill the exact target dimensions.
-`save()` infers the output format from the file extension (`.jpg`, `.jpeg`, `.png`, `.gif`, `.webp`, `.avif`) and marks the file as a build output.
-```php
-// Build a 1200×630 cropped Open Graph image next to the original
-(new IMG('/project/src/images/hero.jpg'))
-    ->resize(1200, 630, true)
-    ->save('/project/src/images/hero-og.jpg');
-```
-`IMG::asset()` and `IMG::palette()` are what back kirigami-core's `img-asset()`
-and `colors()` Sass functions: they resolve `$path` against `image.source` from
-`kirigami.yaml`, generate a resized/re-encoded file under `image.dest` (only
-when missing or stale), or return a `CACHE`-backed list of representative
-colours. `IMG::palette()` extracts them with the Aura extension (vibrant and
-muted swatches, each in a dark and a light variant), so it returns at most six
-colours, most populated first; asking for more returns what Aura found. Both
-are equally usable from your own PHP.
-`IMG::asset()` is the single implementation behind the [`<img asset>` tag](#built-in-tags)
-too — the tag is just a thin wrapper. Generated files are named after the source
-plus a dimension suffix: `-<W>w`, `-<H>h`, `-<W>x<H>`, or `-<W>x<H>-cover`, with
-the `image.format` extension (e.g. `hero.jpg` + `width="800"` → `hero-800w.webp`).
-`@kirigami/kirigami`'s Sass `img-asset()` / `colors()` functions produce the same
-files from the same config through this same class, via
-[`processImages()`](#processimagesjobs) — one engine, no native dependency.
----
-### FS
-Filesystem utilities.
-```php
-FS::dig(string $glob): iterable          // recursive glob, yields file paths
-FS::getRelativePath(string $from, string $to): string
-FS::phpFileInfo(string $file): object|false  // parse PHPDOC annotations
-FS::getChildren(string $backtrace = ''): object[]  // child _index.php pages, ordered by @position
-FS::getBreadcrumb(string $backtrace = ''): object[]  // ancestor _index.php pages, top-most first (opt-in via @breadcrumb)
-FS::rmdir(string $dir, bool $removeSelf = true): bool
-FS::pathJoin(string ...$parts): string   // URL-aware path join with .. resolution
-```
-`FS::dig()` is the workhorse of directory-wide builds — it recursively walks a glob pattern and yields every matching file path.
-`FS::phpFileInfo()` parses the first PHPDOC block of a PHP file and returns its `@tag value` pairs as a `stdClass`. This is used internally to resolve page metadata and data-file annotations.
-`FS::getChildren()` (procedural: `fs_get_children()`) — usable only during a render — scans the folders directly below the calling template, keeps the ones that contain an `_index.php`, and returns one `stdClass` per child: the parsed PHPDOC of that `_index.php` plus a `->file` key with its absolute path. Entries are ordered by `@position` ascending (a page with no `@position` sorts as `999999`), then by folder name (natural, case-insensitive). Handy for building a section index or a navigation menu:
-```php
-<?php foreach (fs_get_children() as $page): ?>
-  <li><a href="<?= FS::getRelativePath(__DIR__, dirname($page->file)) ?>/"><?= $page->title ?></a></li>
-<?php endforeach ?>
-```
-`FS::getBreadcrumb()` (procedural: `fs_get_breadcrumb()`) — also render-only — is the upward counterpart: it returns the calling page's breadcrumb trail. It only produces output when the calling file opts in with `@breadcrumb true` (or `@breadcrumb 1`) in its first PHPDOC block, otherwise it returns `[]`. From the folder **above** the caller's own folder (a page is never part of its own trail), it walks the parent directories upward and collects the `_index.php` of each, stopping at the source root or at the first ancestor `_index.php` that carries no active `@breadcrumb` tag — that page acts as a separator and is left out. Directories without an `_index.php` are skipped without breaking the chain. Entries come back ordered from the top-most ancestor down to the nearest parent, each a `stdClass` (parsed PHPDOC of its `_index.php` plus a `->file` key with its absolute path):
-```php
-<nav aria-label="Breadcrumb">
-<?php foreach (fs_get_breadcrumb() as $crumb): ?>
-  <a href="<?= FS::getRelativePath(__DIR__, dirname($crumb->file)) ?>/"><?= $crumb->title ?></a>
-<?php endforeach ?>
-</nav>
-```
----
-### STR
-String utilities used internally by the tag-processing pipeline, and available for your own templates and plugins.
-```php
-STR::htmlesc(string $str): string
-STR::replaceTags(string $tag, string $html, callable $callback): string
-STR::parseHtmlAttributes(string $attrString): array
-STR::trimIndent(string $str): string
-STR::is_url(string $str): bool
-STR::html_entities_decode(string $str): string
-STR::shorthash(string $str): string
-STR::normalize(string $str): string
-STR::slug(string $str, string $sep = ''): string
-```
-`STR::replaceTags()` is the engine behind `PREPROS::registerTag()`. It finds all occurrences of `<tagname ...>...</tagname>` in an HTML string and replaces each with the return value of `$callback($fullMatch, $attrs, $body)`. Occurrences inside Markdown code (an inline code span or a fenced block) are left as written, so a tag shown as an example is not run.
-`STR::trimIndent()` strips the common leading whitespace from a multi-line string — handy when pulling content out of indented `<markdown>` blocks.
-`STR::is_url()` checks whether a string parses as a URL with a recognized scheme (`http`, `https`, `ftp`, `ftps`, `ssh`, `ssl`, `sftp`, `itunes`).
-`STR::html_entities_decode()` trims a string and decodes its HTML entities — handy when normalizing text scraped from a third-party page.
-`STR::shorthash()` returns the first 12 characters of a string's SHA-256 hash — used internally as a stable, filename-safe cache key (see `SCRAPER`).
-`STR::normalize()` applies Unicode NFD decomposition and strips combining marks (`é` → `e`) — the accent-folding step used by `slug()`.
-`STR::slug()` normalizes, transliterates to ASCII, lowercases, and replaces every run of non-`[a-z0-9]` characters with `$sep` (empty by default → a compact identifier; pass `'-'` for a conventional hyphenated slug).
----
-### ARR
-Recursive lookup helper for nested arrays and objects.
-```php
-ARR::find_key(mixed $data, string $key): mixed
-```
-Walks an array or object (including mixed nested `stdClass`/array structures, as produced by `YAML::parse()` or `json_decode()`) depth-first and returns the value of the **first** matching key found, at any depth, or `null` if none matches.
-```php
-$config = YAML::parseFile('team.yaml');
-$email  = ARR::find_key($config, 'email'); // finds `email` however deep it's nested
-```
----
-### CURL
-CURL verifies HTTPS certificate chains and hostnames, including redirects. The network-enabled WASM runtime supplies Node’s root certificates through `curl.cainfo` and `openssl.cafile`. Untrusted certificates and hostname mismatches are rejected; there is no insecure fallback. `getInfo()` returns `false` when cURL fails, and `getContents()` preserves its existing failure return values.
-Low-level HTTP client built on PHP's cURL extension, used internally by `SCRAPER`. Ships with a realistic browser `User-Agent`/header set and a cookie jar persisted at `.cookie.txt` (auto-registered via `PREPROS::exportFile()`).
-```php
-CURL::urlExists(string $url, ?string $mimereg = null): bool
-CURL::getInfo(string $url): array|false     // HEAD request, returns curl_getinfo()
-CURL::getContents(string $file, ?string $dest = null, ?callable $clb = null): string|bool
-```
-`CURL::urlExists()` issues a `HEAD` request and returns `true` for any `2xx`/`3xx` response.
-`CURL::getContents()` downloads a URL. Without `$dest`, it returns the body as a string; with `$dest`, it streams the download to that file path and returns a boolean. Pass `$clb` to receive download progress as a float between `0` and `1`.
-```php
-CURL::getContents('https://example.com/report.pdf', '/project/src/downloads/report.pdf', function (float $progress) {
-    error_log(sprintf('%.0f%%', $progress * 100));
-});
-```
----
-### SCRAPER
-Fetches a URL and extracts page metadata (`title`, `description`, `image`, `label`) from its JSON-LD (`schema.org`), Open Graph, and standard `<meta>` tags — the kind of data you'd want for a rich link preview. Results are cached indefinitely via `CACHE`, keyed on the URL.
-```php
-$metas = SCRAPER::get(string $url): object|false;
-```
-```php
-$metas = SCRAPER::get('https://example.com/blog/some-article');
-if ($metas) {
-    echo $metas->title;        // string
-    echo $metas->description;  // string
-    echo $metas->image;        // string (absolute URL, may be empty)
-    echo $metas->label;        // string — site/publisher name, may be empty
-    echo $metas->url;          // string — the URL that was scraped
-}
-```
-Returns `false` if the page can't be reached, can't be parsed, or has no discoverable title. Throws an `Exception` on invalid URLs. Uses `CURL::getContents()` under the hood, so it benefits from the same shared cookie jar and browser-like headers.
----
-### OBF
-Simple reversible obfuscation for non-secret values embedded in HTML, such as display labels or contact data. It does not protect API tokens or other credentials.
-```php
-$encoded = OBF::encode(mixed $obj): string;
-$decoded = OBF::decode(string $str): mixed;
-```
-Applies JSON encoding → base64 → ROT-13 → gzip. Not cryptographically secure; intended for light obfuscation only.
----
-### STD
-Result helpers that write JSON to `/internal/prepros_result.json` in the virtual filesystem. Node reads that result file; ordinary PHP stdout/stderr are separate diagnostic channels.
-```php
-STD::succeed(array|string $props = []): void  // exits 0, writes JSON to /internal/prepros_result.json
-STD::error(array|string $props = []): void    // exits 1, writes JSON to /internal/prepros_result.json
-```
-These are internal to the build runner (`render()`, `sitemap()`, and `runenv()` all rely on them). You generally do not need to call them in page templates, but they are available if a script run via `runenv()` needs to terminate early with a custom result.
----
-### Unicode normalization
-The WASM PHP build ships without `ext-intl`; the native `norm` extension
-([php-norm](https://github.com/php-kirigami/php-norm), utf8proc) provides the
-standard `Normalizer` class instead: `Normalizer::normalize()` /
-`Normalizer::isNormalized()`, the `NFC` / `NFD` / `NFKC` / `NFKD` (and
-`FORM_*`) constants, and the `normalizer_normalize()` /
-`normalizer_is_normalized()` functions. `STR::normalize()` and `STR::slug()`
-use it to fold accents. Prefer the `STR` helpers in your own code.
-The pure-PHP polyfill that used to fill this gap is still autoloadable as
-`NORMALIZER_LEGACY`, with the same API.
----
-### Procedural shortcuts (aliases)
-Every static method of every class above is also exposed as a plain function by
-`src/libraries/aliases.inc.php` (autoloaded — no `require` needed). Each alias is
-named `<lowercase class>_<snake_case method>()` and does nothing but forward its
-arguments, so the classes remain the canonical API. They exist to make page
-templates and `kiri run` scripts read better:
-```php
-<?= md_to_html(file_get_contents('CHANGELOG.md')) ?>
-<img src="<?= img_asset('hero.jpg', 1200, 630, true) ?>" alt="">
-```
-Every function has a complete PHPDoc block, so editor hover and autocomplete
-surface the signature, parameters, and description.
-| Class | Aliases |
-|---|---|
-| `PREPROS` | `prepros_render` · `prepros_sitemap` · `prepros_mount` · `prepros_fstat` · `prepros_export_file` · `prepros_get_exported_files` · `prepros_backtrace_file` · `prepros_register_tag` · `prepros_register_hook` · `prepros_run_hook` |
-| `MD` | `md_to_html` · `md_register_plugin` · `md_unregister_plugin` · `md_get_registered_plugins` · `md_register_emoji` |
-| `HTML` | `html_format` |
-| `YAML` | `yaml_parse` · `yaml_parse_file` · `yaml_load_file` |
-| `SCHEMA` | `schema` (factory) · `schema_validate` |
-| `LD` | `ld_add` · `ld_node` · `ld_ref` · `ld_organization` · `ld_person` · `ld_website` · `ld_web_page` · `ld_breadcrumb` · `ld_faq_page` · `ld_script` · `ld_json` |
-| `META` | `meta_tag` · `meta_link` · `meta_raw` · `meta_tags` |
-| `CACHE` | `cache_get` · `cache_set` · `cache_delete` · `cache_purge` |
-| `IMG` | `img_asset` · `img_palette` |
-| `FS` | `fs_dig` · `fs_get_relative_path` · `fs_php_file_info` · `fs_rmdir` · `fs_path_join` |
-| `STR` | `str_htmlesc` · `str_replace_tags` · `str_parse_html_attributes` · `str_trim_indent` · `str_is_url` · `str_html_entities_decode` · `str_shorthash` · `str_normalize` · `str_slug` |
-| `ARR` | `arr_find_key` |
-| `CURL` | `curl_url_exists` · `curl_get_info` · `curl_get_contents` |
-| `SCRAPER` | `scraper_get` |
-| `OBF` | `obf_encode` · `obf_decode` |
-| `STD` | `std_succeed` · `std_error` |
-Notes:
-- `register_tag()` / `register_hook()` are kept as unprefixed aliases of
-  `prepros_register_tag()` / `prepros_register_hook()`.
-- `img_asset()` resolves the generated URL relative to the file that calls it,
-  exactly like `IMG::asset()`.
-- `schema_validate(array $schema, mixed $data, ?array &$errors = null): bool`
-  fills `$errors` with the validation messages.
-- `yaml_parse()` / `yaml_parse_file()` are only defined when the PECL `yaml`
-  extension isn't already providing them.
----
-## Plugin system
-`@kirigami/php-prepros` has two complementary plugin layers: **PREPROS** (HTML-tag level, operates on the assembled page) and **MD** (shortcode level, operates inside Markdown content).
----
-### PREPROS tags
-Register a custom HTML tag that is processed **after** PHP execution, on the fully assembled HTML string:
-```php
-// In a file listed under prepros.includes in kirigami.yaml, or in before.php:
-PREPROS::registerTag('gallery', function (string $fullTag, array $attrs, string $body): string {
-    $id   = $attrs['id'] ?? '';
-    $imgs = glob("/project/src/images/gallery/{$id}/*.webp");
-    $html = '<div class="gallery">';
-    foreach ($imgs as $img) {
-        $src = str_replace('/project/src', '', $img);
-        $html .= "<img src=\"{$src}\" loading=\"lazy\">";
-    }
-    return $html . '</div>';
-});
-```
-Then in any page template:
-```html
-<gallery id="summer-2025"></gallery>
-```
-The callback receives:
-| Parameter | Type | Description |
-|-----------|------|-------------|
-| `$fullTag` | `string` | The complete matched tag string |
-| `$attrs` | `array` | Parsed HTML attributes as an associative array |
-| `$body` | `string` | Inner content between opening and closing tags |
-The built-in [`<markdown>` and `<img asset>` tags](#built-in-tags) are registered
-this way.
----
-### PREPROS hooks
-Hooks let you intercept and transform data at key points in the rendering pipeline:
-```php
-PREPROS::registerHook(string $hookName, callable $callback): void
-```
-| Hook | When it fires | `$data` type | Expected return |
-|------|---------------|--------------|-----------------|
-| `boot` | Once per process, right after bootstrap (config loaded, `includes` pulled in), before any page renders. Fires for every entrypoint. | `stdClass $config` | ignored |
-| `shutdown` | Via `register_shutdown_function()`, at the very end of the request — fires even after `STD::succeed()`/`STD::error()`'s `exit()`, unlike `auto_append_file` (which PHP skips whenever the script exits). The place for cleanup that must always run. | `null` | ignored |
-| `page_info` | After PHPDOC parsing, before rendering (auto-loads `.yaml`/`.json`/`.md` annotations) | `[$filePath, $pageObject]` — see note | `$pageObject` (modified) |
-| `pre_render` | Before PHP execution | Raw file contents as `string` | Ignored by the current render call |
-| `pre_before` | Just before the `before` include (inside its output buffer — `echo` to prepend to the header) | `before` config path as `string\|null` | ignored |
-| `post_before` | Right after the `before` include, on the captured header | Header `string` | `string` |
-| `pre_type_before` | Just before the page's `@type` `before` include, if any (inside its output buffer) | Type's `before` config path as `string\|null` | ignored |
-| `post_type_before` | Right after the `@type` `before` include, on the captured type header | Type header `string` | `string` |
-| `pre_type_after` | Just before the page's `@type` `after` include, if any (inside its output buffer) | Type's `after` config path as `string\|null` | ignored |
-| `post_type_after` | Right after the `@type` `after` include, on the captured type footer | Type footer `string` | `string` |
-| `pre_after` | Just before the `after` include (inside its output buffer — `echo` to prepend to the footer) | `after` config path as `string\|null` | ignored |
-| `post_after` | Right after the `after` include, on the captured footer | Footer `string` | `string` |
-| `post_render` | After tag processing, before `HTML::format()` | Assembled HTML `string` | `string` |
-Multiple callbacks can be registered for the same hook — they are executed in registration order, each receiving the return value of the previous one.
-> Built-in `page_info` + `post_render` callbacks power [`LD`](#ld) and
-> [`META`](#meta): `page_info` captures the page under render, `post_render`
-> injects the JSON-LD `<script>` and the `<meta>`/`<link>` block into its
-> `<head>`. Your own callbacks run after them.
-> **`page_info` payload shape.** The hook *fires* with `[$filePath, $pageObject]`,
-> but each callback is expected to return the `$pageObject` alone — so a callback
-> registered after the built-ins receives the bare object, not the pair. Handle
-> both: `$page = is_array($p) ? $p[1] : $p;` (the current page path is always
-> available as `PREPROS::$file`).
-```php
-// Example: inject a last-modified date into every page
-PREPROS::registerHook('post_render', function (string $html): string {
-    $date = date('Y-m-d');
-    return str_replace('{{build_date}}', $date, $html);
-});
-```
----
-### MD plugins
-MD plugins add custom shortcode tags inside Markdown content. They work inside `<markdown>` blocks, in `.md` data files, and anywhere `MD::toHtml()` is called.
-**Inline syntax** (all on one line):
-```
-{% tagname arg1 "argument with spaces" %}
-```
-**Block syntax** (body on subsequent lines):
-```
-{% tagname optional-arg
-Line one of the body.
-Line two of the body.
-%}
-```
-```php
-MD::registerPlugin(string $name, callable $callback): void
-```
-The callback signature is always `(array $args, string $body): string`. `$args` contains arguments parsed from the opening line; `$body` is the trimmed multi-line body (empty string for inline tags).
----
-### Built-in plugins
-The following MD plugins are registered out of the box in `md.plugins.php`:
-#### `{% callout type ["Title"] content %}`
-Renders a styled callout block. `type` is one of `info`, `success`, `warning`, `danger`.
-```
-{% callout warning "Heads up" This section is outdated. %}
-{% callout danger "Critical"
-Line one of a longer warning.
-Line two after a blank line.
-%}
-```
-#### `{% img-asset path [width height [cover]] %}`
-Generate an image with `IMG::asset()` and emit an `<img>` tag. Width and height default to zero; the optional final `cover` selects cropping.
-```markdown
-{% img-asset photo.jpg 800 600 cover %}
-```
-YouTube and Vimeo shortcuts require `@kirigami/plugin-embed`; YouTube is no longer a built-in plugin.
-#### `{% codepen id [user height] %}`
-Embeds a CodePen result via `<iframe>`. `user` defaults to `anonymous`, `height` defaults to `400`.
-```
-{% codepen abcXYZ %}
-{% codepen abcXYZ jsmith 500 %}
-```
-#### `{% checklist ["Title"] items %}`
-Renders a block-syntax list of checkbox items, one per line, with an optional title.
-```
-{% checklist "Today"
-Do the dishes
-Walk the dog
-Read a book
-%}
-```
----
-## Extending the `<markdown>` tag
-The `<markdown>` tag is one of the [built-in tags](#built-in-tags) (alongside
-`<img asset>`), registered as a PREPROS tag out of the box. It converts its inner
-content from Markdown to HTML and strips common leading indentation so you can
-write cleanly inside your PHP templates:
-```html
-<section class="about">
-    <div>
-        <markdown>
-            ## Who we are
-            We are a **student organization** from Québec.
-            {% codepen abc123 author 400 %}
-        </markdown>
-    </div>
-</section>
-```
-All registered MD plugins are available inside `<markdown>` blocks. You can extend the tag's behaviour by registering additional MD plugins (see above) or by overriding the tag itself:
-```php
-PREPROS::registerTag('markdown', function (string $tag, array $attrs, string $body): string {
-    $body = STR::trimIndent($body);
-    $html = MD::toHtml($body);
-    // wrap in a container, add a class, etc.
-    $class = $attrs['class'] ?? 'prose';
-    return "<div class=\"{$class}\">{$html}</div>";
-});
-```
----
-## Requirements
-- Node.js `>= 24.0.0`
-- npm `>= 10.2.3`
-- ESM only (`"type": "module"`)
----
-## License
-GPL-3.0-or-later © Maxime Larrivée-Roy, 2026
+<div align="center">
+<img src="https://zmotrin.github.io/assets/kirigami/kirigami-logo-universal.svg" alt="Kirigami" width="400" />
+---
+# @kirigami/php-prepros
+PHP preprocessor for the **Kirigami** static site generator.
+[![npm version](https://img.shields.io/npm/v/@kirigami/php-prepros)](https://www.npmjs.com/package/@kirigami/php-prepros)
+[![License: GPL-3.0-or-later](https://img.shields.io/badge/license-GPL--3.0--or--later-blue)](./LICENSE)
+[![Node.js >=24.0.0](https://img.shields.io/badge/node-%3E%3D24.0.0-brightgreen)](https://nodejs.org)
+[![Website](https://img.shields.io/badge/website-php--kirigami.github.io-1f6b4a)](https://php-kirigami.github.io)
+</div>
+---
+## Overview
+Build full static websites in PHP — with zero server, zero runtime dependency, zero compromise on expressiveness. Write your pages as regular PHP files, annotate them with a PHPDOC header, and let `php-prepros` compile everything to clean, deployable HTML.
+It is the perfect solution for **GitHub Pages**. Since it runs entirely in Node.js, it is fully compatible with **GitHub Actions**, allowing you to automate your deployment pipeline effortlessly.
+Part of the **Kirigami** project ecosystem.
+---
+## Table of contents
+- [@kirigami/php-prepros](#kirigamiphp-prepros)
+- [Overview](#overview)
+- [What's new in 3.2.1](#whats-new-in-321)
+- [What's new in 3.2.0](#whats-new-in-320)
+- [What's new in 3.0.1](#whats-new-in-301)
+- [What's new in 3.0.0](#whats-new-in-300)
+- [What's new in 2.0.0](#whats-new-in-200)
+- [What's new in 1.9.3](#whats-new-in-193)
+- [What's new in 1.9.2](#whats-new-in-192)
+- [What's new in 1.9.1](#whats-new-in-191)
+- [What's new in 1.9.0](#whats-new-in-190)
+- [What's new in 1.8.0](#whats-new-in-180)
+- [What's new in 1.7.2](#whats-new-in-172)
+- [What's new in 1.7.1](#whats-new-in-171)
+- [What's new in 1.7.0](#whats-new-in-170)
+- [What's new in 1.6.0](#whats-new-in-160)
+- [What's new in 1.4.0](#whats-new-in-140)
+- [What's new in 1.3.0](#whats-new-in-130)
+- [What's new in 1.2.1](#whats-new-in-121)
+- [What's new in 1.2.0](#whats-new-in-120)
+- [How it works](#how-it-works)
+- [Installation](#installation)
+- [Configuration — `kirigami.yaml`](#configuration--kirigamiyaml)
+  - [`kirigami` block](#kirigami-block)
+  - [`seo` block](#seo-block)
+  - [`prepros` block](#prepros-block)
+  - [`image` block](#image-block)
+  - [`plugins` block](#plugins-block)
+  - [`esbuild` / `sass` blocks](#esbuild--sass-blocks)
+  - [`export` block](#export-block)
+  - [`scripts` block](#scripts-block)
+  - [`tasks` block](#tasks-block)
+- [Writing pages](#writing-pages)
+  - [PHPDOC header](#phpdoc-header)
+  - [Auto-loading data files](#auto-loading-data-files)
+  - [`@content`, `@indent`, and `@type`](#content-indent-and-type)
+  - [Built-in tags](#built-in-tags)
+- [JavaScript API](#javascript-api)
+  - [`render(file?, phpIncludes?)`](#renderfile-phpincludes)
+  - [`sitemap()`](#sitemap)
+  - [`runenv(script, paths?, ...args)`](#runenvscript-paths-args)
+  - [`runPluginScript(script, pluginRoot, paths?, ...args)`](#runpluginscriptscript-pluginroot-paths-args)
+  - [`mountPath(localPath, virtualDir?, php?)`](#mountpathlocalpath-virtualdir-php)
+  - [`processImages(jobs)`](#processimagesjobs)
+  - [`resetRuntime()`](#resetruntime)
+  - [TypeScript declarations](#typescript-declarations)
+- [PHP classes reference](#php-classes-reference)
+  - [PREPROS](#prepros)
+  - [MD](#md)
+  - [HTML](#html)
+  - [YAML](#yaml)
+  - [SCHEMA](#schema)
+  - [LD](#ld)
+  - [META](#meta)
+  - [CACHE](#cache)
+  - [IMG](#img)
+  - [FS](#fs)
+  - [STR](#str)
+  - [ARR](#arr)
+  - [CURL](#curl)
+  - [SCRAPER](#scraper)
+  - [OBF](#obf)
+  - [STD](#std)
+  - [Unicode normalization](#unicode-normalization)
+  - [Procedural shortcuts (aliases)](#procedural-shortcuts-aliases)
+- [Plugin system](#plugin-system)
+  - [PREPROS tags](#prepros-tags)
+  - [PREPROS hooks](#prepros-hooks)
+  - [MD plugins](#md-plugins)
+  - [Built-in plugins](#built-in-plugins)
+- [Extending the `<markdown>` tag](#extending-the-markdown-tag)
+- [Requirements](#requirements)
+- [License](#license)
+---
+## What's new in 3.2.1
+- **`prepros.format`** no longer puts a space before punctuation that follows an inline element (`<strong>bold</strong>,` was split onto its own line and rendered "bold ,"). `<del>`, `<ins>`, `<data>`, `<bdi>`, ruby, `<strike>`, `<font>` and custom elements count as inline, and text plus the inline elements around it stay on one line inside a container that also holds blocks.
+---
+## What's new in 3.2.0
+- **Markdown pages.** An `_index.md` whose first lines are `@tag value` annotations is a page, like an `_index.php`: its body is rendered as Markdown and wrapped by the layouts. Without that header, or next to an `_index.php`, it stays a data file and is left alone. See [Markdown pages](#markdown-pages).
+- **Inherited annotations.** `@@tag value` sets `tag` on the page and on every page below it; a child's `@tag` overrides it for that page only, a child's `@@tag` overrides it and passes the new value down. Works in PHPDOC blocks and Markdown headers. See [Inherited annotations (`@@`)](#inherited-annotations-).
+- `FS::phpFileInfo()` reads Markdown headers and inherited values, and returns a fresh copy on each call: adding a key to the result no longer leaks into that page's own variables. `FS::getChildren()`, `FS::getBreadcrumb()`, the sitemap and the JSON-LD breadcrumb see `_index.md` pages.
+- The `page_info` hook skips values that aren't strings instead of failing on them.
+---
+## What's new in 3.0.1
+A registered tag written inside Markdown code is no longer processed. `<markdown>`, `<img asset="...">` or a plugin tag shown in an inline code span (`` `<markdown prose>` ``) or a fenced block (` ``` ` / `~~~`) stays example text: before, a `<markdown>` in a code span paired with the real block's closing tag and broke the rest of the page, and an `<img asset>` in one became an empty image. Indented (four-space) code blocks are not recognized, since `<markdown>` bodies are indented; use a fence there.
+---
+## What's new in 3.0.0
+This release switches `YAML::` to the native YAML extension, `MD::` to native mdhtml, `SCHEMA` to native jsonk, and `Normalizer` to native norm. It also includes page types and request lifecycle hooks.
+PHP data files use the native `yaml` extension backed by LibYAML. Its YAML 1.1 implicit booleans include unquoted `y`, `n`, `yes`, `no`, `on`, `off`, `true`, and `false`, including mapping keys. Quote these words when you mean strings (for example, `"NO": Norway`). `YAML::parse()` / `parseFile()` / `loadFile()` preserve the wrapper’s array/object choice; native `yaml_parse()` / `yaml_parse_file()` have their own extension signatures. `yaml_load_file()` remains a wrapper alias. The project’s `kirigami.yaml` is parsed separately in Node through `struct-walker` and `js-yaml`.
+`MD::` delegates to the native `mdhtml` extension (cmark-gfm). Footnotes now use `<section class="footnotes" data-footnotes>` instead of the old `<div class="footnotes">`; target `.footnotes` rather than a specific container tag in custom CSS.
+`SCHEMA` validates through the native `jsonk` extension (draft 2020-12), keeping its API and the `"path: message"` error format. It now also supports `if`/`then`/`else`, `contains`, `propertyNames`, `dependentRequired`/`dependentSchemas`, `prefixItems`, and `$ref` to absolute or `$id`-relative URLs (fetched over the network). Error messages use jsonk's wording, and an `additionalProperties: false` violation is reported on the parent object instead of the extra property. The previous pure-PHP validator stays available as `SCHEMA_LEGACY`.
+`Normalizer` now comes from the native `norm` extension (utf8proc) instead of the bundled pure-PHP polyfill, which stays available as `NORMALIZER_LEGACY`.
+**Breaking: the `seo.jsonld` sub-block is merged into `seo:`.** META and LD now read one set of keys, so the site description, keywords, image, language and person are declared once. JSON-LD is injected whenever the `seo:` block exists; `seo.jsonld` is only an on/off switch (default `true`). To migrate, move the keys of `seo.jsonld: {…}` up into `seo:` (`jsonld.auto: false` becomes `jsonld: false`) and rename `seo.language` to `seo.lang`. `kiri` rejects the old shapes with a message saying so. Sites with `seo: {}` and no `jsonld` now get JSON-LD too; add `jsonld: false` to keep them without it.
+```yaml
+# before                       # after
+seo:                           seo:
+  language: fr-CA                lang: fr-CA
+  jsonld:                        type: ProfessionalService
+    type: ProfessionalService    logo: images/logo.png
+    logo: images/logo.png
+```
+---
+## What's new in 2.0.0
+- **Breaking: `meta:` and `jsonld:` merged into one top-level `seo:` block.**
+  The two used to be independent siblings of `kirigami:` that happened to
+  share fallback data; they're now one block, one mental model for a
+  project's whole SEO/social surface — `META`'s own keys live directly under
+  `seo:`, and `jsonld` is nested inside it as its own sub-block:
+  ```yaml
+  # before (1.x)
+  meta:
+    favicon: favicon.png
+  jsonld: {}
+  # after (2.0.0)
+  seo:
+    favicon: favicon.png
+    jsonld: {}
+  ```
+  The two stay **independently toggled** exactly as before — a project can
+  have META's tags without JSON-LD, or vice versa; `seo: { jsonld: false }`
+  (or `{ auto: false }`) stops just the JSON-LD injection, `seo: false` (or
+  `{ auto: false }` at the top level) stops just META's. Every fallback chain
+  (a page's PHPDOC → the block → the loose `kirigami:` keys) is unchanged,
+  including `jsonld`'s own keys still feeding META's defaults (`jsonld.image`,
+  `jsonld.lang`, `jsonld.person`, …) — only *where the two blocks live* in
+  `kirigami.yaml` changed, not how resolution works. **Migration**: rename
+  `meta:` to `seo:` and move the `jsonld:` block's content under it as
+  `seo.jsonld:`. See [`seo` block](#seo-block) / [`meta` config](#meta-config)
+  / [`jsonld` config](#jsonld-config).
+---
+## What's new in 1.9.3
+- Dependency bump to `@kirigami/struct-walker` 1.0.5; `homepage` + README
+  pointed at the site (metadata only).
+---
+## What's new in 1.9.2
+- **Fixed a real Markdown bug**: a list item's source line count was 1:1
+  with `<li>` count, so an indented continuation line with no marker of its
+  own (a soft-wrapped `- **foo** text\n  more text`) fell outside the
+  block-matching regex entirely — the list closed after the first line, the
+  continuation resurfaced as a stray flat `<p>`, and a new list reopened for
+  the next marker line. Fixed by widening the block regex to also accept a
+  marker-less indented line and, in the per-line loop, appending it to the
+  previous item instead of dropping it.
+---
+## What's new in 1.9.1
+- **`{% img-asset %}` no longer crashes the build on an unresolvable path.**
+  `IMG::asset()`'s exception is now caught and turned into an HTML comment,
+  matching `codepen`/`checklist`'s own missing-argument behavior instead of
+  aborting the whole render. This also makes it safe to *document* the tag —
+  a literal `` `{% img-asset path … %}` `` written as prose inside a code
+  span still runs the plugin (code-span protection only swaps the *displayed*
+  output back to the literal text; the callback itself always executes), so
+  a placeholder path used to throw "Invalid image file." and fail the build.
+---
+## What's new in 1.9.0
+- **`{% youtube %}` removed** — moved to
+  [`@kirigami/plugin-embed`](https://www.npmjs.com/package/@kirigami/plugin-embed),
+  which replaces the old plain-iframe output with a real oEmbed-backed card
+  (cover thumbnail, title, play button — no network call until the visitor
+  actually clicks play). If a project used the built-in `{% youtube %}`,
+  install the plugin; without it, the tag now falls through unresolved
+  (`{% youtube ID %}` printed as-is) rather than rendering an iframe.
+  `codepen`/`checklist`/`callout` are unaffected.
+---
+## What's new in 1.8.0
+- **`{% img-asset %}` — a new built-in Markdown plugin.** Same pipeline as the
+  `<img asset>` HTML tag (`IMG::asset()`: resize, cache, publish under
+  `image.dest`), usable straight from Markdown text:
+  ```
+  {% img-asset photo.jpg %}
+  {% img-asset photo.jpg 800 %}
+  {% img-asset photo.jpg 800 600 %}
+  {% img-asset photo.jpg 800 600 cover %}
+  ```
+  Positional args: source path (relative to `image.source`), width, height,
+  and the literal `cover` keyword. Registered in `md.plugins.php` alongside
+  `codepen`/`youtube`/`checklist`/`callout` — available out of the box, drop
+  it with `MD::unregisterPlugin('img-asset')` if you don't want it.
+---
+## What's new in 1.7.2
+- **Cleaner formatted output around highlighted code.** The de-indent script
+  `PREPROS::injectHead()` adds when `prepros.format` is on now flattens the
+  `<pre><code>` indentation `HTML::format()` writes for *every* block, including
+  ones a build-time highlighter has wrapped in `<span>`s. It works on
+  `innerHTML` line by line and removes only the shared leading run (relative
+  indentation is kept). This lets [`@kirigami/plugin-highlight`](https://www.npmjs.com/package/@kirigami/plugin-highlight)
+  1.7.2+ re-indent its markup to line up with the rest of the document instead
+  of leaving it flush-left — the served HTML stays consistently indented, the
+  rendered code is still de-indented before the first paint.
+- **`<markdown prose>` wraps in `.prose`.** With the `prose` attribute the
+  built-in tag emits `<div class="prose"> … </div>` so long-form Markdown picks
+  up `@kirigami/canva`'s `styles/prose` typography with no extra markup. Opt-in
+  (a bare `<markdown>` is unchanged); `class` / `id` on the tag land on the
+  wrapper.
+---
+## What's new in 1.7.1
+- **No side effects on import.** `kirigami.yaml` is now loaded on first use
+  (`render()` / `sitemap()` / `runenv()` / `processImages()`), not while the
+  module is being imported. `import '@kirigami/php-prepros'` from a directory
+  with no project no longer throws — which is what made `kiri build --help` /
+  `kiri export --help` / `kiri run --help` crash instead of printing their help.
+---
+## What's new in 1.7.0
+- **`META`** class — a `<head>` SEO / social metadata generator, the companion
+  to [`LD`](#ld). Builds the standard tags — `<title>`, `description`,
+  `keywords`, `robots`, `language`, `generator`, `author`, Open Graph, Twitter
+  Card, `<link rel="canonical">`, favicon / apple-touch-icon / humans — from
+  each page's PHPDOC, the top-level `meta:` block, and the loose `kirigami:` /
+  `jsonld:` keys `LD` already reads. It emits only what it can resolve, and
+  leaves any tag the layout already hand-writes untouched.
+  - **Opt-in**: a top-level `meta:` block (empty `meta: {}` is enough) switches
+    on automatic injection into every page's `<head>`. `meta: false` (or
+    `{ auto: false }`) keeps the config but stops the injection.
+  - Per-page PHPDOC: `@meta false` (skip), `@meta_title`, `@meta_description`
+    (falls back to `@description` / `@abstract` / `@excerpt`), `@meta_keywords`,
+    `@meta_image`, `@meta_robots`, `@meta_type`, `@canonical`.
+  - Manual builders — always emitted, still de-duplicated: `META::tag()`,
+    `META::link()`, `META::raw()`, `META::tags()`. Procedural aliases:
+    `meta_tag()`, `meta_link()`, `meta_raw()`, `meta_tags()`.
+  Full key reference: [`META` → `meta` config](#meta-config).
+---
+## What's new in 1.6.0
+- **Managed `<head>` (`prepros.head`).** Every rendered page's `<head>` is now
+  auto-wired: a tiny theme/FOUC guard as the first child (adds the `js` class,
+  applies the stored `data-theme` before first paint), a
+  `<link rel="stylesheet">` for every `sass` task output, and a `<script>` (no
+  `defer`, just before `</body>`) for every `esbuild` task output — each with a
+  per-page relative path and a `?<timestamp>` cache-bust. A file already
+  referenced in the page is left alone, so you can still hand-place one. Turn it
+  off with `prepros: { head: false }`, or `head: false` on a single sass/esbuild
+  task. A template's `header.php` no longer wires assets at all.
+- **`HTML::format()` indents `<pre><code>`.** A fenced code block's lines are
+  shifted to the block's nesting depth so the HTML source stays readable
+  (relative indentation preserved). The exact leading run is stripped again
+  before it's shown, by a small de-indent script `prepros.head` injects before
+  `</body>` (only when `format` is on). Since 1.7.2 the script works on
+  `innerHTML` line by line, so it also flattens blocks a build-time highlighter
+  has wrapped in `<span>`s. A bare `<pre>` and `<textarea>` are still emitted
+  byte-for-byte.
+---
+## What's new in 1.4.0
+A round of fixes to the rough edges that showed up building a full site from
+scratch — mostly developer-experience, all backward compatible.
+- **`HTML::format()` keeps `<pre>` / `<textarea>` verbatim.** Their line breaks,
+  indentation and blank lines are no longer collapsed, so a fenced code block
+  survives the formatter intact — `format: true` and Markdown code blocks now
+  coexist. (1.6.0 refines this: a `<pre><code>` block is re-indented to its
+  nesting depth and de-indented again before display.)
+- **The default Markdown plugins load out of the box.** `{% callout %}`,
+  `{% youtube %}`, `{% codepen %}` and `{% checklist %}` are registered
+  automatically (`md.plugins.php` is auto-included from `MD`), as the docs always
+  said. Drop one with `MD::unregisterPlugin('name')` or shadow it with your own
+  `MD::registerPlugin()`.
+- **Build errors you can actually read.** A fatal in a template (a bad call, a
+  `null` argument, a `TypeError`…) comes back as a structured failure with the
+  message, the offending page and the `file:line` — never a bare
+  `Error: undefined`. The `try/catch` now covers `Throwable`, not just
+  `Exception`. PHP warnings and notices no longer sink an otherwise-clean build:
+  they surface as `warnings` on the result. `kiri` prints the message, the page,
+  and the tail of the PHP stderr/debug output on failure.
+- **PHPDOC parsing.** A tag value may now wrap onto the following *indented*
+  continuation lines instead of being silently truncated at the first line. And
+  an `@word` written in the block's prose is ignored rather than overwriting a
+  real tag — only lines that *start* with `@` open a tag.
+- **`FS::getBreadcrumb()` / `FS::getChildren()`** are anchored on the page being
+  rendered (`PREPROS::$file`), so they return the right trail / child list when
+  called from a layout include, a partial, or a helper function — not only
+  straight from the template. Pass an explicit path to override.
+- **`{% tag %}` inside a code span or code block stays literal** (`` `{% badge %}` ``
+  renders as text) instead of being expanded — or leaking an unrestored
+  placeholder.
+- **`IMG` never upscales.** A requested size larger than the source is clamped
+  down to the source instead of throwing an opaque encoder error (the AVIF
+  encoder in particular).
+- **Build-time tokens expand at render time.** `###YEAR###`, `###TIMESTAMP###`
+  and `###TODAY###` are substituted when each page is generated, so
+  `kiri build` / `kiri watch` previews show real values, not the literal token
+  (previously only `kiri export` replaced them).
+- **`page_info` hook robustness.** The first built-in callback accepts either the
+  `[$file, $info]` pair the hook fires with or the bare `$info` object a later
+  callback receives, so a custom `page_info` hook can't fatal on the argument
+  shape. See [PREPROS hooks](#prepros-hooks).
+---
+## What's new in 1.3.0
+- **`LD`** class — a schema.org JSON-LD generator. Collects structured-data
+  nodes during a render and emits them as a single
+  `<script type="application/ld+json">` `@graph` in every page's `<head>`.
+  - Automatic injection is **opt-in**: add a top-level `jsonld:` block to
+    `kirigami.yaml` (even empty, `jsonld: {}`) and an `Organization` (+ `Person`, `WebSite`,
+    `WebPage`, `BreadcrumbList`) graph is derived from that block plus the loose
+    keys projects already carry (`person`, `jobtitle`, `email`, `area`,
+    `knowsabout`, `keywords`, `facebook`, …). No `jsonld:` block → nothing is
+    injected.
+  - Turn it back off with `jsonld: false` / `jsonld: { auto: false }`, or per
+    page with `@ld false`; per-page `@ld_type` / `@ld_title` / `@ld_image` / …
+    tags feed the page node, and a `BreadcrumbList` is built from the
+    `_index.php` ancestor trail with no opt-in.
+  - Explicit builders for everything else: `LD::add()`, `LD::article()`,
+    `LD::faqPage()`, `LD::breadcrumb()`, `LD::ref()`, and every schema.org type
+    via `LD::typeName([...])`. Procedural aliases: `ld_add()`, `ld_organization()`,
+    `ld_script()`, …
+  - A page that already hand-writes an `application/ld+json` script is left
+    untouched.
+---
+## What's new in 1.2.1
+- **`processImages()`** JS export — batch resize / palette-extraction through the
+  `IMG` class (`src/imagebatch.php`). `@kirigami/kirigami`'s `sass` task now uses
+  it for `img-asset()` / `colors()`, so the whole toolchain is free of a native
+  image dependency (`sharp` is gone).
+- **`IMG::save()`** takes an optional `$quality` (0-100) for jpg / webp / avif;
+  `null` keeps the per-format default (82).
+---
+## What's new in 1.2.0
+- **`SCHEMA`** class — a pure-PHP, dependency-free JSON Schema validator
+  (Draft-7 style, Ajv-like API: `isValid()` / `validate()` / `getErrors()`).
+- **`IMG::asset()` / `IMG::palette()`** — static helpers powering kirigami-core's
+  `img-asset()` and `colors()` Sass functions: on-demand resize/convert of a
+  source image, and cached representative-colour extraction.
+- **`<img asset="…">` tag** — the HTML-side entry point of the image
+  autogenerator, same parameters as `IMG::asset()` (see [Built-in tags](#built-in-tags)).
+- **`IMG` now handles vector and exotic formats** — SVG, EPS, AI, PDF (rasterized
+  via Imagick), plus HEIC / TIFF / BMP, on top of GD's JPEG / PNG / GIF / WebP /
+  AVIF.
+- **`MD` emoji shortcodes** — `:rocket:` → 🚀 from a large built-in map, extendable
+  with `MD::registerEmoji()`.
+- **`MD` footnotes and definition lists** — `[^1]` / `[^1]: …`, and `Term` / `: …`.
+- **`MD` inline HTML is now sanitized** against a tag/attribute allowlist rather
+  than passed through verbatim.
+- **`STR::normalize()`** — Unicode NFD + combining-mark stripping;
+  **`STR::slug($str, $sep = '')`** now takes a separator (pass `'-'` for a
+  hyphenated slug).
+- **Bundled `Normalizer` polyfill** — `ext-intl` isn't in the WASM build, so a
+  polyfill keeps `Normalizer::normalize()` (and `STR::normalize()` / `slug()`)
+  working.
+Earlier, in 1.1.x: `PREPROS::mount()` + the `mountPath()` / `runenv()` JS
+exports, the `SCRAPER` and `CURL` and `ARR` classes, `YAML::loadFile()`, the
+`STR::is_url()` / `html_entities_decode()` / `shorthash()` / `slug()` helpers,
+the `{% youtube %}` / `{% codepen %}` / `{% checklist %}` MD plugins, and the
+`@content` / `@indent` PHPDOC annotations.
+---
+## How it works
+`@kirigami/php-prepros` runs your PHP source files inside a **WebAssembly PHP 8.x runtime** ([`@kirigami/php-wasm`](https://www.npmjs.com/package/@kirigami/php-wasm)), entirely in Node.js — no PHP installation required on the host machine.
+The lifecycle of a page build looks like this:
+```
+_index.php  ──▶  PHP (wasm)  ──▶  processTags()  ──▶  HTML::format()  ──▶  index.html
+                    │
+                    ├── before.php  (optional layout header)
+                    ├── after.php   (optional layout footer)
+                    └── PHPDOC annotations resolved (yaml / json / md / url)
+```
+Files are mounted into the WebAssembly virtual filesystem on demand. Only `.php`, `.json`, `.yaml`, `.yml`, `.md`, `.db`, `.txt` and any extra extensions listed in `prepros.mountext` are mounted automatically, keeping memory usage low. Anything else can be mounted on demand with [`PREPROS::mount()`](#preprosmountstringarray-patterns).
+---
+## Installation
+```bash
+npm install @kirigami/php-prepros
+```
+---
+## Configuration — `kirigami.yaml`
+Every project **must** have a `kirigami.yaml` at its root. The preprocessor reads it at startup and throws if it is absent or invalid.
+`@kirigami/php-prepros` consumes project data, SEO, preprocessing, image settings, and task metadata for managed head injection. The core engine owns plugin loading, task orchestration, export, and full schema validation. Direct use of this package is not a substitute for core configuration validation. All settings share `kirigami.yaml`; its schema is [`kirigami.schema.json`](../kirigami/kirigami.schema.json).
+```yaml
+# yaml-language-server: $schema=https://cdn.jsdelivr.net/gh/php-kirigami/kirigami@main/packages/kirigami/kirigami.schema.json
+```
+```yaml
+kirigami:
+  # ── Required ──────────────────────────────────────────────────────────
+  project:     My Website           # Site name. Printed in the CLI banner, exposed as $project.
+  baseurl:     https://example.com  # Deployed root URL, no trailing slash. Used for sitemap.xml.
+  root:        src                  # Source directory containing your _*.php pages.
+  # ── Optional ────────────────────────────────────────────────────────
+  banner:      assets/banner.txt    # Text file stamped as a license banner on exported files.
+  # ── Arbitrary project data ──────────────────────────────────────────
+  # Everything else under `kirigami:` is free-form. The whole block is
+  # extracted as PHP variables and made available in every page, in
+  # before.php/after.php, and anywhere PREPROS::$config->data is read.
+  author:      Jane Doe
+  email:       hello@example.com
+  gtag:        G-XXXXXXXXXX
+  description: A short description of the site, useful for <meta name="description">.
+  keywords:
+    - keyword one
+    - keyword two
+seo:                            # Presence turns on the META tags and LD JSON-LD (`seo: {}` is enough).
+  type: Organization            # JSON-LD main entity; `jsonld: false` turns the JSON-LD off.
+  logo: assets/logo.png
+prepros:
+  before:  _layouts/header.php  # Included before every page body.
+  after:   _layouts/footer.php  # Included after every page body.
+  format:  true                 # Pretty-print the HTML output (default: false).
+  network: true                 # Allow HTTP fetches in PHPDOC @tag annotations / CURL / SCRAPER.
+  mountext:                     # Extra file extensions to auto-mount into the wasm fs,
+    - .svg                      # in addition to the defaults (.php .json .yaml .yml .md .db .txt).
+    - .webp
+  includes:                     # PHP files auto-included once, before any page renders.
+    - _lib/functions.php
+  types:                        # Named page types — opt in per page with @type <name>.
+    article:
+      before: _layouts/types/article.header.php
+      after:  _layouts/types/article.footer.php
+image:                          # Image autogenerator — powers IMG::asset() / IMG::palette().
+  format: webp                  # webp | avif (default: webp)
+  source: assets/images         # Source folder, relative to cwd() (default: assets/images)
+  dest:   images                # Output folder, relative to kirigami.root (default: images)
+plugins:
+  - name: "@kirigami/plugin-highlight"
+    active: true
+    options:
+      theme: auto
+esbuild:
+  # minify: false
+sass:
+  style: expanded
+export:
+  path:   dist
+  ignore: ["*.psd", "notes/"]
+scripts:
+  - name: convert-images
+    mount: ["assets/images/**/*.jpg"]
+    trigger: before-build       # before-build | before-export | after-export
+tasks:
+  - name:  js-core
+    type:  esbuild
+    entry: scripts/kirigami.core.js
+  - name:  scss-core
+    type:  sass
+    entry: styles/kirigami.core.scss
+```
+### `kirigami` block
+Core project settings. **Read by `php-prepros`.** The entire block is extracted into PHP variables and made available in every page template, `before.php`, `after.php`, and `prepros.includes` files — `$project`, `$author`, `$gtag`, etc. are available with no further setup, and also as `PREPROS::$config->data`.
+| Key | Required | Description |
+|-----|----------|--------------|
+| `project` | ✅ | Human-readable site name. Exposed as `$project`. |
+| `baseurl` | ✅ | Root URL of the deployed site, no trailing slash. Used to build absolute `<loc>` entries in `sitemap.xml`; exposed as `$baseurl`. |
+| `root` | ✅ | Path (relative to the project root) to the directory containing your `_*.php` source pages. Build fails immediately if missing or if the path doesn't exist. |
+| `banner` | — | Path (relative to the project root) to a text file stamped as a license/copyright banner on exported `.js`/`.css`/`.html` files during `kiri export`. May contain the `###DATE###` token, replaced with today's date. Falls back to an auto-generated banner. |
+| *anything else* | — | Free-form key/value pairs (strings, numbers, booleans, lists, nested maps — anything valid YAML). Every key is extracted as a PHP variable (`$author`, `$gtag`, …). Use this for contact info, social links, analytics IDs, SEO keywords, or any project data you want available everywhere. When the top-level `seo` block is present, [`LD`](#ld) also reads some of these by convention: `person`, `jobtitle`, `email`, `area`, `knowsabout`, `keywords`, and social-network URL keys (`facebook`, `instagram`, …). |
+### `seo` block
+Top-level, optional. The SEO surface: one set of keys feeds both the
+[`META`](#meta) tags (the standard SEO / social `<meta>` and `<link>` tags) and
+the [`LD`](#ld) schema.org JSON-LD graph, and its **presence** switches both on
+for every page's `<head>`. An empty `seo: {}` is enough; everything is derived
+from the `kirigami` block and each page's PHPDOC (`@title`, `@description` /
+`@abstract`, `@keywords`, `@image`, `@robots`, `@og_type`, `@canonical`). A
+tag or script the layout already hand-writes is left untouched.
+- `auto: false` stops META's tags, `jsonld: false` stops the JSON-LD; the
+  config values stay available to `META::tags()` / `LD::script()`.
+- `seo: false` keeps nothing on; no block at all means nothing is injected
+  (explicit `META::tag()` / `LD::add()` calls still emit).
+Full key reference: [`seo` config](#seo-config). Per-page tags:
+[`@meta_*`](#meta) and [`@ld_*`](#ld).
+### `prepros` block
+Global `before` and `after` files are optional and default to `null`; an empty `prepros: {}` renders without a layout. When provided, these paths must refer to existing files. PHP warnings are logged to stderr and returned as diagnostics, without being inserted into generated HTML.
+Options for the PHP → HTML compiler. **Read by `php-prepros`.** Declaring this block (even empty) also makes `kiri` prepend a forced `prepros` task on every build/export/watch.
+| Key | Type | Default | Description |
+|-----|------|---------|--------------|
+| `before` | `string` | — | Path (relative to `kirigami.root`) to a PHP file included **before** every page's body. Typically your `<head>`/layout opening. |
+| `after` | `string` | — | Path (relative to `kirigami.root`) to a PHP file included **after** every page's body. Typically your layout closing. |
+| `format` | `bool` | `false` | Pretty-print the compiled HTML via [`HTML::format()`](#html) before writing it to disk. |
+| `head` | `bool` | `true` | Auto-wire each page's `<head>`: a theme/FOUC guard as the first child, a `<link rel="stylesheet">` per `sass` task output, and a `<script>` (no `defer`, before `</body>`) per `esbuild` task output — each with a per-page relative path and a `?<timestamp>` cache-bust. A file already referenced in the page is skipped. Set `false` to disable, or `head: false` on a single `sass`/`esbuild` task to skip just its tag. |
+| `network` | `bool` | `false` | Enables outbound HTTP(S) inside the WASM PHP runtime. Required for PHPDOC `@tag https://…` annotations that fetch remote `.yaml`/`.json`/`.md` data (see [Auto-loading data files](#auto-loading-data-files)), and for the `CURL` / `SCRAPER` classes. |
+| `mountext` | `string[]` | `[]` | Extra file extensions to mount automatically into the virtual filesystem alongside the built-in `.php`, `.json`, `.yaml`, `.yml`, `.md`, `.db`, `.txt`. Use this for assets your PHP code reads directly (e.g. `.svg`, `.webp`). Files with extensions not in this set are skipped during mounting — mount them on demand with [`PREPROS::mount()`](#preprosmountstringarray-patterns) instead. |
+| `includes` | `string[]` | `[]` | PHP files (relative to `kirigami.root`) `include_once`'d once, right after config is loaded — before any page renders. The natural place to `PREPROS::registerTag()`, `PREPROS::registerHook()`, or `MD::registerPlugin()`. |
+| `types` | `object` | `{}` | Named page types. A page opts in with `@type <name>` in its PHPDOC header; the matching entry's `before`/`after` (each `string`, relative to `kirigami.root`, both optional) wrap the page body **one level inside** the global `before`/`after` — render order is global before → type before → body → type after → global after. A page with no `@type`, or naming a type absent here, renders with just the global wrap. See [`@type`](#content-indent-and-type). |
+### `image` block
+Options for the image autogenerator. **Read by `php-prepros`** — these are what [`IMG::asset()` / `IMG::palette()`](#img), the [`<img asset>` tag](#img-asset), and kirigami-core's `img-asset()` / `colors()` Sass functions all resolve against. Optional; the defaults below apply even when the block is absent.
+| Key | Type | Default | Description |
+|-----|------|---------|--------------|
+| `format` | `string` | `webp` | Output format for generated images: `webp` or `avif`. |
+| `source` | `string` | `assets/images` | Folder holding the source images, relative to `cwd()`. |
+| `dest` | `string` | `images` | Destination folder for generated images, relative to `kirigami.root`. |
+### `plugins` block
+List of Kirigami plugins. **Consumed by the `kiri` CLI** (see [`@kirigami/sdk`](https://www.npmjs.com/package/@kirigami/sdk)), not by `php-prepros` directly.
+| Key | Required | Description |
+|-----|----------|--------------|
+| `name` | ✅ | Plugin package name. Must match `@kirigami/plugin-*`, `<scope>/kirigami-plugin-*`, or `kirigami-plugin-*`. |
+| `active` | ✅ | Whether the plugin is loaded. |
+| `options` | — | Free-form object passed to the plugin; its shape depends on the plugin. |
+### `esbuild` / `sass` blocks
+Free-form objects. **Consumed by the `kiri` CLI.** There is no fixed key set: whatever you put here is spread straight into the underlying library call for every matching task, *after* Kirigami's own defaults — so it can also override them (`minify`, `target`, `style: "compressed"`, source maps, …). Refer to esbuild's [`BuildOptions`](https://esbuild.github.io/api/#build-api) and Dart Sass's [`Options`](https://sass-lang.com/documentation/js-api/interfaces/options/) for what's accepted. Writing the key with nothing under it parses to `null` in YAML, equivalent to omitting the block.
+`sass:` additionally recognizes two keys that are **not** passed to Dart Sass:
+| Key | Type | Description |
+|-----|------|--------------|
+| `before` | `string` / `string[]` | Extra `.scss` files compiled **before** the task entry (paths relative to `cwd()`). |
+| `after` | `string` / `string[]` | Extra `.scss` files compiled **after** the task entry. |
+### `export` block
+Options for `kiri export`. **Consumed by the `kiri` CLI.** Optional.
+| Key | Type | Default | Description |
+|-----|------|---------|--------------|
+| `path` | `string` | `dist` | Output directory for `kiri export`, relative to the project root. |
+| `ignore` | `string[]` | `[]` | Extra gitignore-style patterns excluded from the export copy, on top of Kirigami's built-in exclusions. |
+### `scripts` block
+Named PHP scripts. **Consumed by the `kiri` CLI**, which runs each `scripts/<name>.php` through [`runenv()`](#runenvscript-paths-args) — so the full `php-prepros` class library is available and `kirigami.yaml`'s `kirigami` block is exposed as `PREPROS::$config->data`.
+| Key | Required | Description |
+|-----|----------|--------------|
+| `name` | ✅ | Must match an existing `scripts/<name>.php` file. Run with `kiri run <name> [args...]`; extra CLI arguments are forwarded as `$argv` entries. |
+| `mount` | — | Glob patterns (relative to the project root) of extra local files to mount into the sandbox before the script runs. |
+| `trigger` | — | Fire the script automatically: `before-build` (start of `build` and `export`), `before-export` (very start of `export`), or `after-export` (once `export` has finished). |
+### `tasks` block
+Ordered list of build tasks, run in array order. **Consumed by the `kiri` CLI**, on top of the implicit `prepros` task (added when the `prepros` block is present) and the implicit `dist` task (added during `kiri export`).
+| `type` | Purpose | Required fields | Optional |
+|--------|---------|-----------------|----------|
+| `esbuild` | Bundle/minify a JS/TS entry. Build + watch. Output: `<entry>.min.js`. | `name`, `type`, `entry` | `force` |
+| `sass` | Compile a `.scss`/`.sass` entry, minified with csso on export. Build + watch. Output: `<entry>.min.css`. | `name`, `type`, `entry` | `force` |
+| `prepros` | Render pages + `sitemap.xml`. Watch only (runs on build/export only when forced/implicit). | `name`, `type` | `target`, `force` |
+| `dist` | Copy `kirigami.root` into an output dir, stamping the banner. Forced/implicit only. | `name`, `type`, `path` | `ignore`, `force` |
+---
+## Writing pages
+Source pages live in the directory pointed to by `kirigami.root`. The naming convention is straightforward: any file whose name starts with `_` and ends in `.php` is treated as a page source. The leading underscore is stripped in the output filename. An `_index.md` with an annotation header is a page too (see [Markdown pages](#markdown-pages)).
+```
+src/
+├── _layouts/
+├── _lib/
+├── _index.php          →  src/index.html
+├── about/
+│   └── _index.php      →  src/about/index.html
+├── notes/
+│   └── _index.md       →  src/notes/index.html
+└── blog/
+    ├── _index.php       →  src/blog/index.html
+    └── _articles.yaml   (data file, not compiled)
+```
+Directories whose name starts with `_` (e.g. `_layouts/`, `_lib/`) are skipped entirely during directory-wide builds.
+### PHPDOC header
+Every page starts with a PHP docblock that drives metadata and data loading:
+```php
+<?php
+/**
+ * @name     about
+ * @title    About us
+ * @abstract A short description of this page.
+ */
+?>
+<section>
+    <h1><?php echo $title; ?></h1>
+    <p><?php echo $abstract; ?></p>
+</section>
+```
+All annotations are injected as PHP variables (`$name`, `$title`, `$abstract`, …). You can define any custom annotation you need.
+Annotations are also available as variables in `before` and `after` PHP included files, so you can write proper metas in the HTML header.
+Only lines whose first non-whitespace character (past the `*` gutter) is `@`
+open an annotation — an `@word` written in the prose of the block is left alone.
+A value can wrap onto the following **indented** continuation lines:
+```php
+/**
+ * @title       About us
+ * @description A longer blurb that does not fit comfortably
+ *              on a single line and continues here.
+ */
+```
+### Markdown pages
+A page can be pure Markdown: an `_index.md` whose first lines are
+annotations, written like PHPDOC tags without the comment around them.
+```markdown
+@title    Hello, world
+@type     post
+@date     2026-09-01
+@abstract The first post.
+Some **Markdown** text: the page body.
+```
+- The header is the `@tag value` lines at the top of the file (leading blank
+  lines allowed), with the same rules as a PHPDOC block: a value wraps onto
+  indented continuation lines. It ends at the first blank line, or the first
+  flush-left line that isn't a tag, where the body starts.
+- The body goes through `MD::toHtml()` and becomes `$content`, so `@type`,
+  `@indent` and the layouts work as for a PHP page. PHP in the file is never
+  run. `@content other.md` in the header replaces the body.
+- An `_index.md` **without** a header is not a page: it is left alone, as a
+  data file. So is one next to an `_index.php`, which is the page of that
+  folder (it can load the `.md` through an annotation).
+- Only `_index.md` is a page; other `_*.md` files are data files as before.
+### Inherited annotations (`@@`)
+A tag written with two `@` applies to the page **and every page below it**
+(the pages of its subfolders, and the other pages of its own folder when it is
+an `_index`):
+```php
+/**
+ * @title Blog
+ * @@type post            ← every page under blog/ is a post
+ * @@menu _menu.yaml      ← loaded from blog/, wherever the page is
+ */
+```
+- A child's `@tag` overrides the inherited value **for that page only**; its
+  own children still get the ancestor's value.
+- A child's `@@tag` overrides it **and passes the new value down**.
+- The nearest ancestor wins. Values come from the page file of each folder
+  above (`_index.php`, else `_index.md`), up to `kirigami.root`.
+- A relative data-file value (`.yaml`/`.yml`/`.json`/`.md`) is resolved
+  against the folder of the page that declared it.
+- In PHPDOC blocks and Markdown headers alike. `FS::getChildren()` and
+  `FS::getBreadcrumb()` entries include inherited values too, so
+  `@@breadcrumb true` turns breadcrumbs on for a whole section.
+### Auto-loading data files
+When an annotation value looks like a filename (with a `.yaml`, `.yml`, `.json`, or `.md` extension), it is automatically parsed and injected as a structured variable instead of a plain string.
+```php
+<?php
+/**
+ * @name     medias
+ * @articles _articles.yaml
+ */
+?>
+<?php foreach ($articles as $article): ?>
+    <a href="<?php echo $article->lien; ?>">
+        <?php echo $article->titre; ?>
+    </a>
+<?php endforeach; ?>
+```
+| Extension | Parsed as |
+|-----------|-----------|
+| `.yaml` / `.yml` | `stdClass` object (or array of objects for sequences) |
+| `.json` | Result of `json_decode()` |
+| `.md` | HTML string via `MD::toHtml()` |
+When `network: true` is set in `kirigami.yaml`, annotation values that start with `http://` or `https://` are fetched from the network and parsed the same way:
+```php
+/**
+ * @posts https://api.example.com/posts.json
+ */
+```
+### `@content`, `@indent`, and `@type`
+Three special annotation names change how a page's body is assembled:
+- **`@content`** — if a `content` variable already resolves to a non-empty value (typically because it's a `.md`/`.yaml`/`.json` annotation that auto-loaded into HTML/data, see above), it is used **as-is** as the page body, and the PHP file itself is **not executed** for its output. This is handy for pages that are pure data/markdown wrapped by a shared layout.
+- **`@indent`** — when set to a number, every line of the rendered body is prefixed with that many spaces before being wrapped by `before.php`/`after.php`. Useful for keeping generated HTML readable when a page is nested inside indented layout markup.
+- **`@type`** — names an entry under [`prepros.types`](#prepros-block). If it matches, that entry's `before`/`after` wrap the (already-indented) body **one level inside** `before.php`/`after.php`: global before → type before → body → type after → global after. No match (missing annotation, or a name absent from `prepros.types`) leaves the page with just the global wrap — a page type is an extra layer, never a replacement for the site's real header/footer.
+```php
+<?php
+/**
+ * @name    changelog
+ * @title   Changelog
+ * @content _changelog.md
+ * @indent  4
+ * @type    article
+ */
+```
+### Built-in tags
+Two tags are registered out of the box (`prepros.plugins.php`) and processed
+**after** the PHP runs, on the assembled HTML — no include or plugin needed.
+#### `<markdown> … </markdown>`
+Converts its inner content from Markdown to HTML, stripping the common leading
+indentation first (via `STR::trimIndent()`) so you can indent it naturally inside
+your template. All registered [MD plugins](#md-plugins) work inside it. See
+[Extending the `<markdown>` tag](#extending-the-markdown-tag) to override it.
+```html
+<section>
+    <markdown>
+        ## Who we are
+        We are a **student organization** from Québec.
+    </markdown>
+</section>
+```
+Add the `prose` attribute — `<markdown prose>` — to wrap the output in
+`<div class="prose">`, so it picks up the long-form typography of
+[`@kirigami/canva`'s `styles/prose`](https://www.npmjs.com/package/@kirigami/canva)
+with no extra markup. Any `class` / `id` on the tag lands on that wrapper
+(`<markdown prose class="lede" id="intro">` → `<div class="prose lede" id="intro">`).
+A bare `<markdown>` emits just the converted HTML, as before.
+#### `<img asset="…">`
+The HTML-side entry point of the image autogenerator — the exact same feature as
+the [`img-asset()` Sass function](https://www.npmjs.com/package/@kirigami/kirigami#sass-functions)
+and [`IMG::asset()`](#img), with the same parameters. The tag calls `IMG::asset()`
+under the hood, then swaps the `asset` attribute for the generated `src`.
+```html
+<!-- in:  resolves assets/images/hero.jpg through IMG::asset('hero.jpg', 800, 0, false) -->
+<img asset="hero.jpg" width="800" alt="Our office" loading="lazy">
+<!-- out: <img src="../images/hero-800w.webp" alt="Our office" loading="lazy"> -->
+```
+| Attribute | Maps to `IMG::asset()` arg | Notes |
+|-----------|---------------------------|-------|
+| `asset` | `$path` | **Required.** Path relative to `image.source`. Missing/empty ⇒ the tag is left untouched. |
+| `width` | `$width` | Optional, integer. Omitted ⇒ `0` (keep). |
+| `height` | `$height` | Optional, integer. Omitted ⇒ `0` (keep). |
+| `cover` | `$cover` | Boolean — **presence means `true`** (crop + fill). |
+| *(any other)* | — | `alt`, `class`, `id`, `loading`, … are passed straight through onto the output `<img>`. |
+`asset` / `width` / `height` / `cover` are consumed and removed; everything else
+survives. The generated file lands in `image.dest` and is only (re)generated when
+missing or older than the source — see [`IMG`](#img) for the naming convention.
+> The Sass `img-asset()` / `colors()` functions, the `<img asset>` tag and
+> `IMG::asset()` all run on the **same engine** — the `IMG` class (GD, with the
+> Imagick fallback) in this package. `@kirigami/kirigami`'s `sass` task routes its
+> image work here through [`processImages()`](#processimagesjobs), so there is no
+> native image dependency in the toolchain.
+---
+## JavaScript API
+```js
+import { render, sitemap, runenv, mountPath, processImages, resetRuntime } from '@kirigami/php-prepros';
+```
+PHP operations are queued in call order, including mounts and result extraction.
+`await resetRuntime()` waits for preceding PHP work, disposes the owned runtime
+and its network proxy, and clears cached configuration and mounts. The next
+operation initializes a fresh runtime from `kirigami.yaml`. Core
+`Project.reload()` calls this and also clears the plugin PHP include list.
+It does not clear persistent cache/cookie files on disk. This remains a
+single-project API whose working directory must be set before import.
+### `render(file?, phpIncludes?)`
+`phpIncludes` defaults to `[]`. The core collects it through `prepros:php`;
+direct callers supply local PHP file paths (absolute or relative to the
+working directory). Existing paths are mounted under `/plugins/` and included
+before rendering; missing paths are silently skipped. Each render replaces
+the runtime include list, which remains on its configuration until another
+render or reset. Direct calls do not load the core plugin registry for you.
+Compile a single PHP page or a whole directory.
+```js
+// Compile one page
+const pageResult = await render('about/_index.php');
+// Compile everything under src/
+const treeResult = await render('.');
+// Compile everything (uses kirigami.root from config)
+const defaultResult = await render();
+```
+> Paths used by `render()` are all relative to the `kirigami.root` configuration.
+**Returns** `Promise<PreprosResult>`:
+```ts
+interface PreprosResult {
+  success: boolean;
+  files?:  string[];   // project-relative paths; may be absent on parsing failure
+  error?:  string;
+  debug?:  string;     // captured PHP stdout
+  stderr?: string;     // diagnostics attached to failures
+  warnings?: string;   // nonfatal stderr on success
+  page?:   string | null;     // PHP-render failure context, when available
+  where?:  string;     // PHP source location, when available
+}
+```
+Setup and filesystem failures can reject before a result exists; PHP failures
+usually return `success: false`. Check both channels. Files may already have
+been copied to the host before a later error; operations are not transactional.
+The runtime returns `debug`/`stderr`, not the older declared `response` field.
+Directory rendering selects `_*.php` files only when every directory between
+`kirigami.root` and the page has a name without a leading underscore. Sitemap
+selection uses the same rule. Direct requests for private pages fail; rendering
+a private directory produces no pages. The configured source root itself may
+start with `_` (for example `_src`). Previously generated private HTML is not
+deleted by this selection rule; remove stale outputs when migrating a site.
+### `sitemap()`
+Generate `sitemap.xml` and `robots.txt` at the source root, plus `humans.txt`
+when author configuration provides content. It accepts no directory argument.
+```js
+const result = await sitemap();
+// With kirigami.root: src, files includes src/sitemap.xml and src/robots.txt.
+```
+### `runenv(script, paths?, ...args)`
+Run an arbitrary PHP script — not a page template — inside the very same sandboxed WASM environment used for `render()`, with the full `php-prepros` class library autoloaded and `kirigami.yaml`'s `kirigami` block available as `PREPROS::$config->data`. Useful for one-off maintenance scripts, data migrations, or CLI-style tooling that needs `CACHE`, `SCRAPER`, `IMG`, etc. without going through the page-rendering pipeline.
+```js
+// Run a standalone PHP script
+const purgeResult = await runenv('scripts/purge-cache.php');
+// Also mount explicit extra files into the sandbox before running
+const imageResult = await runenv('scripts/build-og-images.php', ['assets/photos/hero.jpg']);
+// Extra arguments are appended and available as $argv[2], $argv[3], … in the script
+const importResult = await runenv('scripts/import.php', [], '--force');
+```
+- `script` — path to a PHP file **inside the project**, executed with `require_once`.
+- `paths` — optional array of explicit local file paths, not directories. Missing
+  files are skipped; a directory can cause a filesystem rejection. Use
+  `mountPath()` first for recursive directory mounting.
+- `...args` — extra string arguments appended to the script's `$argv`.
+Script and extra-file paths resolve against the project captured at import.
+Before initializing PHP or copying files, `runenv()` rejects paths outside
+that project, including symbolic links whose real targets are outside it.
+Directories are rejected; missing optional files are skipped. This check does
+not make PHP scripts untrusted-code sandboxes or restrict explicit `mountPath()`
+calls.
+**Returns** `Promise<PreprosResult>`, following the same shape as `render()`. Inside the script, call `PREPROS::exportFile()` for any file you want listed in `result.files`.
+### `runPluginScript(script, pluginRoot, paths?, ...args)`
+Same as `runenv()`, for a script shipped inside a plugin package. A plugin
+installed with `npm link` or from a workspace lives outside the project, so
+`runenv()` would reject it. Here the script's authored and real paths must stay
+inside `pluginRoot` instead, and it is mounted under
+`/plugin-scripts/<package dir>/`. Extra `paths` are still project files.
+The caller vouches for `pluginRoot`: `@kirigami/kirigami` only passes the
+resolved package directory of an active plugin.
+### `mountPath(localPath, virtualDir?, php?)`
+The JavaScript-side counterpart to [`PREPROS::mount()`](#preprosmountstringarray-patterns). Mounts a local file or directory — recursively, preserving structure — into the WASM sandbox's virtual filesystem, ahead of (or between) calls to `render()`, `sitemap()`, or `runenv()`. Useful when a Node-side build step needs to make extra local files visible to PHP before rendering starts.
+```js
+import { mountPath, render } from '@kirigami/php-prepros';
+// Mount a single file at its natural virtual path (/project/<relative path>)
+await mountPath('assets/data/team.yaml');
+// Mount a whole directory, at a custom virtual path
+await mountPath('vendor/fonts', '/project/fonts');
+await render();
+```
+- `localPath` — path to a local file or directory. Relative paths are resolved against the project root.
+- `virtualDir` — optional destination path inside the WASM filesystem. Defaults to `/project/<localPath relative to the project root>` when omitted.
+- `php` — optional WASM PHP instance to mount into. Defaults to PHP-prepros's owned instance (the same one used internally by `render()`/`sitemap()`/`runenv()`), creating it if needed. This is separate from PHP-WASM's shared getter instances and is replaced after `resetRuntime()`.
+Mounting a **directory** only copies files whose extension is one of the defaults (`.php`, `.json`, `.yaml`, `.yml`, `.md`, `.db`, `.txt`) or listed in `prepros.mountext`, same as automatic root mounting. Mounting a **single file directly** copies it regardless of extension — this is the simplest way to make an arbitrary asset (an image, a font, a CSV, …) available to PHP without adding its extension to `prepros.mountext` project-wide.
+**Returns** `Promise<void>`.
+### `processImages(jobs)`
+Run a batch of image jobs — resize/encode, or palette extraction — through the
+[`IMG`](#img) class (GD, with the Imagick fallback). This is the engine
+`@kirigami/kirigami`'s `sass` task uses for its `img-asset()` and `colors()`
+functions, so Sass, `IMG::asset()` and the [`<img asset>` tag](#built-in-tags)
+all share one implementation, one `image:` config and one set of output
+filenames — with no native image dependency.
+```js
+import { processImages } from '@kirigami/php-prepros';
+const { files, colors } = await processImages([
+  // resize/encode `hero.jpg` (resolved against image.source) to each dest —
+  // absolute virtual paths, already carrying the target extension
+  { op: 'resize', src: 'hero.jpg', width: 1200, height: 0, cover: false, quality: 82,
+    dests: ['/project/src/images/hero-1200w.webp'] },
+  // extract a 5-colour palette (cached in .cache.db); returned, not written
+  { op: 'palette', src: 'hero.jpg', count: 5 },
+]);
+// files includes 'src/images/hero-1200w.webp' and may include '.cache.db'.
+// colors → { 'hero.jpg:5': ['#1e3a5f', '#c8a24b', …] }
+```
+- `jobs` — array of `resize` / `palette` jobs (see the shape above). An **empty
+  array is a no-op** and does **not** start the WASM runtime.
+- Omitted `jobs` also returns `{ success: true, files: [], colors: {} }`
+  without starting PHP. Non-array input currently does the same; this is not
+  strict input validation.
+- Staleness is the caller's responsibility: every `resize` job listed is executed.
+- `resize`: `width`/`height` default to `0` (preserve size when both are zero),
+  `cover` to `false`, and lossy encoder quality to `82`. `dests` contains
+  absolute `/project/...` output paths. `palette` defaults `count` to `5`.
+- The PHP worker handles `palette` explicitly and treats any other `op` as a
+  resize; pass only the two documented operations. Processing stops at the
+  first exception. Always check `success` before using files or colors.
+**Returns** `Promise<PreprosResult & { colors: Record<string, string[]> }>`.
+### `resetRuntime()`
+Returns `Promise<void>`. Queued after preceding operations, it disposes the
+owned runtime, mounts, and cached configuration. It does not change the project
+path captured at import, clear disk caches, or reset the SDK hook registry.
+### TypeScript declarations
+The shipped `index.d.ts` includes `render(file?, phpIncludes?)`, argument-free
+`sitemap()`, explicit-file `runenv()` mounts, and the current diagnostic fields.
+`PreprosResult.files` is optional because response parsing can fail before a
+file list exists. `ImageBatchResult.files` is always normalized to an array.
+The obsolete `response` field is replaced by `debug` and `stderr`.
+---
+## PHP classes reference
+All classes are autoloaded — no manual `require` needed inside your page files.
+The autoloader itself, `$argv`/`$config`, procedural aliases, and the `boot`
+hook are installed via php.ini's `auto_prepend_file` (pointed at
+`utils.inc.php`), set once per WASM runtime instance — every entrypoint
+(`prepros.php`, `runenv.php`, `imagebatch.php`) gets it automatically,
+with no `include` of its own.
+---
+### PREPROS
+The core engine. Manages the rendering pipeline, tag processing, hooks, mounting, and file export.
+```php
+// Available inside page templates and included files.
+PREPROS::$config          // stdClass — full resolved config; ->data is the kirigami: block,
+                          // ->image the image: block, plus before/after/format/… from prepros:
+PREPROS::registerTag(string $tag, callable $callback)
+PREPROS::registerHook(string $hook, callable $callback)
+PREPROS::runHook(string $hook, mixed $data = null)  // fire a hook (built-in or your own), returns the piped $data
+PREPROS::mount(string|array $patterns)
+PREPROS::exportFile(string|array $absolutePath)
+PREPROS::getExportedFiles(): string[]
+PREPROS::fstat(string $path)          // stat a file in the WASM FS (or false)
+PREPROS::backtraceFile()              // path of the page currently rendering
+```
+#### `PREPROS::render(string $file)`
+Internal method called once per source file. Orchestrates the full pipeline:
+1. Resolves PHPDOC metadata and auto-loads data files.
+2. Fires the `pre_render` hook with the raw source contents.
+3. Includes `before.php` (wrapped in the `pre_before` / `post_before` hooks) and the page body (or `@content`, see [above](#content-indent-and-type)).
+4. If the page declares `@type <name>` and `prepros.types.<name>` exists, wraps the body with that type's `before`/`after` (wrapped in `pre_type_before` / `post_type_before` and `pre_type_after` / `post_type_after`) — nested inside the global wrap.
+5. Includes `after.php` (wrapped in `pre_after` / `post_after`), assembling everything into a single string.
+6. Processes all registered custom HTML tags.
+7. Fires the `post_render` hook on the assembled HTML.
+8. Optionally pretty-prints via `HTML::format()` (when `format: true`).
+9. Writes the output `.html` file.
+#### `PREPROS::sitemap()`
+Scans the source tree for `_index.php` files and generates a standards-compliant `sitemap.xml` (Sitemaps 0.9), using `kirigami.baseurl` as the root URL.
+#### `PREPROS::mount(string|array $patterns)`
+Mounts additional local project files into the WASM virtual filesystem, on demand, from one or more glob patterns evaluated against the project root (via `picomatch`). Unlike the automatic mounting done for `kirigami.root` (limited to `.php`, `.json`, `.yaml`, `.yml`, `.md`, `.db`, `.txt`, and `prepros.mountext`), `mount()` copies **any** matching file, regardless of extension.
+```php
+// Mount every .webp under assets/, wherever the page needs them
+PREPROS::mount('assets/**/*.webp');
+// Multiple patterns at once
+PREPROS::mount(['data/**/*.csv', 'vendor/fonts/*.woff2']);
+```
+Returns an array of the virtual paths (under `/project/...`) that were mounted, or `false` on failure.
+#### `PREPROS::exportFile(string $file)`
+Marks a file as a build output so it gets surfaced in `PreprosResult.files`. Called automatically by `render()`, `sitemap()`, `CACHE::set()`, and `CURL`. Call it manually if your custom code writes additional files.
+---
+### MD
+Markdown-to-HTML converter with a plugin system for custom shortcodes,
+backed by PHP's native `mdhtml` extension (real `cmark-gfm`), statically
+built into `@kirigami/php-wasm` — no userland parsing.
+```php
+$html = MD::toHtml(string $markdown): string;
+```
+Supports the full GitHub Flavored Markdown subset, plus a few extensions:
+- ATX (`#` … `######`) and Setext headings, with auto-generated `id` attributes
+- Ordered and unordered lists, including nested
+- GFM task lists (`- [ ]` / `- [x]`)
+- GFM tables with column alignment
+- GFM alerts (`> [!NOTE]`, `> [!WARNING]`, etc.)
+- Blockquotes (recursive)
+- Fenced code blocks with language class
+- Inline code
+- Bold, italic, bold+italic, strikethrough
+- Links with automatic `target="_blank" rel="noopener noreferrer"` for external URLs
+- Images with `loading="lazy"`
+- Auto-linked bare URLs
+- Horizontal rules
+- Hard line breaks (trailing double space → `<br>`)
+- **Footnotes** — `[^1]` references and `[^1]: …` definitions (multi-paragraph)
+- **Definition lists** — `Term` / `: Definition`
+- **Emoji shortcodes** — `:rocket:` → 🚀, from a built-in map (see `MD::registerEmoji()`)
+- **Sanitized inline HTML** — raw tags are filtered against an allowlist of tags and attributes, not passed through verbatim
+#### Plugin API
+Extend Markdown with custom shortcode tags:
+```php
+// Inline tag  {% tagname arg1 "arg with spaces" %}
+// Block tag   {% tagname arg1
+//             body content
+//             %}
+MD::registerPlugin(string $name, callable $callback): void
+MD::unregisterPlugin(string $name): void
+MD::getRegisteredPlugins(): string[]
+MD::registerEmoji(string $shortcode, string $char): void   // `:name:` → char
+```
+The callback always receives `(array $args, string $body)`:
+```php
+MD::registerPlugin('video', function (array $args, string $body): string {
+    $src = htmlspecialchars($args[0] ?? '', ENT_QUOTES, 'UTF-8');
+    return "<video src=\"{$src}\" controls></video>";
+});
+```
+Then in any Markdown content (including inside `<markdown>` tags):
+```
+{% video /videos/intro.mp4 %}
+```
+---
+### HTML
+Pretty-printer for the final HTML output. Used automatically when `format: true` is set in the config.
+```php
+$formatted = HTML::format(string $html): string;
+```
+Uses PHP 8.4's `Dom\HTMLDocument` (Lexbor engine) to parse the input and re-serialize it with consistent 4-space indentation. Inline elements, `<script>`, and `<style>` blocks are handled correctly — their content is indented but not reformatted. A `<pre><code>` block is shifted to its nesting depth too (relative indentation kept), and the leading run is stripped again before display; a bare `<pre>` and `<textarea>` stay byte-for-byte. Boolean HTML5 attributes (`muted`, `autoplay`, `noopener`, etc.) are written without a value.
+---
+### YAML
+PHP data files use the native `yaml` extension backed by LibYAML. Its YAML 1.1 implicit booleans include unquoted `y`, `n`, `yes`, `no`, `on`, `off`, `true`, and `false`, including mapping keys. Quote these words when you mean strings (for example, `"NO": Norway`). `YAML::parse()` / `parseFile()` / `loadFile()` preserve the wrapper’s array/object choice; native `yaml_parse()` / `yaml_parse_file()` have their own extension signatures. `yaml_load_file()` remains a wrapper alias. The project’s `kirigami.yaml` is parsed separately in Node through `struct-walker` and `js-yaml`.
+A YAML parser backed by PHP's native `yaml` extension (libyaml), statically built into `@kirigami/php-wasm` — full YAML 1.1 support, no userland parsing.
+```php
+$data = YAML::parse(string $yaml, bool $assoc = false): mixed;
+$data = YAML::parseFile(string $path, bool $assoc = false): mixed;
+$data = YAML::loadFile(string $path, bool $assoc = false): mixed;
+```
+By default, YAML mappings are returned as `stdClass` objects. Pass `true` as the second argument to get associative arrays instead.
+Following YAML 1.1 means the usual implicit-boolean gotcha applies to both values *and* mapping keys: `y`/`Y`/`n`/`N`, `yes`/`no`, `true`/`false`, `on`/`off` (any case) all resolve to a boolean when unquoted — so an unquoted `no:` key or a `NO` value becomes `false`. Quote a scalar (`"y": 2`) to keep it a string.
+`YAML::loadFile()` behaves like `YAML::parseFile()`, then walks the result recursively: any string value ending in `.yaml`, `.yml`, or `.json` that resolves to an existing file (relative to *its own* file's directory) is replaced by that file's parsed content, and so on, recursively. Values that don't match an existing file are left untouched. Circular references (`A → B → A`) throw a `RuntimeException`.
+```yaml
+# team.yaml
+lead: people/jane.yaml     # resolved and inlined automatically
+members:
+  - people/jane.yaml
+  - people/john.yaml
+```
+```php
+$team = YAML::loadFile('/project/data/team.yaml');
+// $team->lead is now the fully parsed content of people/jane.yaml, not a string
+```
+---
+### SCHEMA
+A JSON Schema validator with an Ajv-like API, backed by the native `jsonk`
+extension built into `@kirigami/php-wasm`. Available to your own code and
+plugins.
+```php
+$validator = new SCHEMA(array $schema);
+$validator->isValid(mixed $data): bool     // true / false
+$validator->validate(mixed $data): bool    // alias of isValid()
+$validator->getErrors(): string[]          // "path: message" strings from the last run
+```
+jsonk implements draft 2020-12 for a self-contained schema: every validation
+keyword (`type`, `enum`, `const`, `required`, `properties`,
+`patternProperties`, `additionalProperties`, `propertyNames`,
+`dependentRequired`, `dependentSchemas`, `items`, `prefixItems`, `contains`,
+`uniqueItems`, the `min*`/`max*` and `exclusive*` bounds, `multipleOf`,
+`pattern`, `format`, `if`/`then`/`else`, `allOf`/`anyOf`/`oneOf`/`not`) and
+`$ref` to `#/$defs/…` / `#/definitions/…`, absolute URLs, or URLs relative to
+the schema's `$id`. See [php-jsonk](https://github.com/php-kirigami/php-jsonk)
+for the details and limits.
+Schemas are PHP arrays, so `SCHEMA` adapts them before handing them to jsonk:
+an empty array in a schema position (`'properties' => []`) is treated as `{}`,
+draft-07 tuple `items` (a list of schemas) becomes `prefixItems` (and
+`additionalItems` becomes `items`), and `format: url` is read as `uri`. Error
+paths look like `(root)`, `name` or `tags[1]`.
+The previous pure-PHP (Draft-7 style) validator is still available as
+`SCHEMA_LEGACY`, with the same API.
+```php
+$validator = new SCHEMA([
+    'type'     => 'object',
+    'required' => ['name', 'age'],
+    'properties' => [
+        'name' => ['type' => 'string', 'minLength' => 1],
+        'age'  => ['type' => 'integer', 'minimum' => 0],
+    ],
+    'additionalProperties' => false,
+]);
+if (!$validator->isValid($data)) {
+    foreach ($validator->getErrors() as $err) echo $err, PHP_EOL;
+}
+```
+---
+### LD
+A **schema.org JSON-LD generator**. `LD` accumulates structured-data nodes for
+the page under render and emits them as one
+`<script type="application/ld+json">` block — with an `@graph` when there is more
+than one node — in the `<head>`.
+#### Automatic mode
+On as soon as `kirigami.yaml` has a [`seo:` block](#seo-block) (`seo: {}` is
+enough), alongside [`META`](#meta)'s tags and from the same keys. A
+`post_render` hook injects a graph built from `seo:`, the loose keys of the
+`kirigami` block, and the current page's PHPDOC:
+- an `Organization` node (`@id` `#organization`) — `name`/`url`/`description`
+  from `project`/`baseurl`/`description`, `sameAs` gathered from every
+  recognised social-network URL key (`facebook`, `instagram`, `linkedin`,
+  `github`, `youtube`, `mastodon`, …), plus `email`, `telephone`, `areaServed`
+  (← `area`), `knowsAbout` (← `knowsabout`), `address`, `logo`, and `founder` →
+  the Person node when there is one. `@type` comes from `seo.type`;
+- a `Person` node (`#person`) when `person` is set — `name` + `jobTitle`
+  (← `jobtitle`) + `email` + `url`, linked to the Organization via `worksFor`;
+- a `WebSite` node (`#website`) — `publisher` → Organization, `inLanguage`,
+  `keywords` (← `keywords`), and a `SearchAction` when `seo.search` is set;
+- a `WebPage` node for the page — see the per-page tags below;
+- a `BreadcrumbList` for every non-home page, derived from the `_index.php`
+  ancestor trail (home → each parent section → this page). No `@breadcrumb`
+  opt-in needed. Disable it for one page with `@ld_breadcrumb false`.
+`seo: { jsonld: false }` stops the automatic pass; META's tags keep working.
+A page whose rendered `<head>` already contains an `application/ld+json`
+script is never touched, so hand-rolled markup keeps working.
+**Per-page PHPDOC tags** — these feed the page node (and override the generic
+`@title` / `@description` / `@datePublished` fallbacks):
+| Tag | Effect |
+|-----|--------|
+| `@ld false` | Skip JSON-LD for this page entirely (`@ld_ignore true` also works). |
+| `@ld_type <Type>` | `@type` of the page node — `AboutPage`, `ContactPage`, `CollectionPage`, `ProfilePage`, or a content type like `Article`, `Service`, `Recipe`, … (default `WebPage`). Types containing “Page” also get `primaryImageOfPage` + a `breadcrumb` link; others get a plain `image`. |
+| `@ld_title <text>` | Page node `name` (default: `@title`). |
+| `@ld_description <text>` | Page node `description` (default: `@description`). |
+| `@ld_image <path>` | Page image, absolute or relative to `baseurl` (default: `@image` / `@ogimage`). |
+| `@ld_published <date>` | `datePublished` (default: `@datePublished` / `@published` / `@date`). |
+| `@ld_modified <date>` | `dateModified` (default: `@dateModified` / `@modified` / `@updated`). |
+| `@ld_breadcrumb false` | No `BreadcrumbList` for this page. |
+```php
+/**
+ * @title            À propos
+ * @ld_type          AboutPage
+ * @ld_title         À propos de Humain Humain
+ * @ld_description   Notre approche ethnographique de la consultation.
+ */
+```
+#### Explicit builders
+Call these from a page template or from a `prepros.includes` file. Nodes added
+this way are always emitted — automatic pass on or not — and share
+the graph the automatic pass uses, so the two combine; a node with a stable
+`@id` is merged on repeat calls.
+```php
+LD::add(string|array $type, array $props = [], ?string $id = null): array  // build + register a node
+LD::node(string|array $type, array $props = []): array                      // build only, no register
+LD::push(array $node): array                                                // register a ready-made node
+LD::ref(string $id): array                                                  // ['@id' => …]  ('#person' → the Person node)
+LD::remove(string $id): void
+LD::graph(): array
+LD::reset(): void
+LD::organization(array $overrides = []): array   // config-aware, @id #organization
+LD::person(array $overrides = []): array          // config-aware, @id #person
+LD::website(array $overrides = []): array          // config-aware, @id #website
+LD::webPage(array $overrides = []): array           // current-page-aware, @id …#webpage
+LD::breadcrumb(?array $items = null, array $overrides = []): array   // items: [['name'=>…,'url'=>…], …]
+LD::faqPage(array $qa, array $overrides = []): array                 // qa: ['Question ?' => 'Answer.', …]
+LD::address(array|string $a): array
+LD::image(string $url, ?string $id = null, ?int $w = null, ?int $h = null): array
+LD::geo(float $lat, float $lng): array
+LD::rating(int|float $value, ?int $count = null, $best = 5, $worst = 1): array
+LD::offer(array $o): array
+LD::contactPoint(array $c): array
+LD::searchAction(string $urlTemplate): array
+LD::script(bool $pretty = true): string   // <script>…</script>, and disables auto-injection
+LD::json(bool $pretty = true): string     // the document, no wrapper
+```
+Every other schema.org type is reachable through `__callStatic` — the method
+name is upper-cased to form the `@type`:
+```php
+LD::recipe([ 'name' => 'Tarte aux pommes', 'recipeYield' => '6', 'prepTime' => 'PT30M' ]);
+LD::event([ 'name' => 'Vernissage', 'startDate' => '2026-10-01T18:00' ]);
+LD::softwareApplication([ 'name' => 'Kirigami', 'applicationCategory' => 'DeveloperApplication' ]);
+LD::article([
+    'headline'      => $title,
+    'datePublished' => '2026-09-01',
+    'author'        => LD::ref('#person'),
+    'image'         => LD::image('images/cover.webp'),
+    'publisher'     => LD::ref('#organization'),
+]);
+```
+Same API from procedural code: `ld_add()`, `ld_node()`, `ld_ref()`,
+`ld_organization()`, `ld_person()`, `ld_website()`, `ld_web_page()`,
+`ld_breadcrumb()`, `ld_faq_page()`, `ld_script()`, `ld_json()`.
+LD's keys (`type`, `logo`, `person`, `address`, `search`, …) live in the
+[`seo` config](#seo-config) with META's.
+---
+### META
+A **`<head>` SEO / social metadata generator** — the companion to [`LD`](#ld).
+Where `LD` emits a schema.org `application/ld+json` graph, `META` emits the plain
+tags a browser and a link-preview crawler read: `<title>`, `<meta name="…">`,
+`<meta property="og:…">`, `<meta name="twitter:…">`, and a handful of `<link>`s.
+It draws on the same sources, in this order of precedence: the page's PHPDOC, the
+top-level `seo:` block, then the loose `kirigami:` keys. Every tag is emitted **only when it can be resolved** — no value, no
+tag — and a tag the page's layout already writes by hand is detected and
+skipped, so it drops in beside an existing `header.php` without duplicating
+anything.
+**Automatic mode** is opt-in: the top-level `seo:` block (empty `seo: {}` is
+enough) turns on injection into every page's `<head>`, right before `</head>`.
+```yaml
+kirigami:
+  project:     Humain Humain
+  baseurl:     https://humainhumain.com
+  tagline:     Ethnographie au service des organisations
+  description: A social-science consultancy using ethnography for organisational change.
+  keywords:    [ethnographie, consultation publique, sciences sociales]
+  author:      Maxime Larrivée-Roy
+seo:                              # top-level; the block being present is the switch
+  twitter: "@humainhumain"
+  themeColor: "#0b7285"
+  lang: fr-CA                     # <meta name="language">, og:locale, JSON-LD inLanguage
+  logo: assets/logo.png           # JSON-LD logo, and og:image when there is no `image`
+```
+Per-page, from the PHPDOC block — each falls back to the generic page tag:
+| Tag | Feeds | Default |
+|-----|-------|---------|
+| `@meta false` | skip metadata for this page entirely | — (`@meta_ignore true` also works) |
+| `@meta_title` | `<title>`, `og:title`, `twitter:title` | `@title` |
+| `@meta_description` | `description`, `og:description`, `twitter:description` | `@description` / `@abstract` / `@excerpt` / `@summary`, then the site `description` |
+| `@meta_keywords` | `<meta name="keywords">` | `@keywords`, then `seo.keywords` |
+| `@meta_image` | `og:image`, `twitter:image` | `@image` / `@ogimage`, then `seo.image` |
+| `@meta_robots` | `<meta name="robots">` | `@robots`, then `seo.robots` |
+| `@meta_type` | `og:type` | `@og_type`, then `seo.ogType` |
+| `@canonical` | `<link rel="canonical">` | derived from the file path + `baseurl` |
+**Manual builders** — always emitted (with or without a `seo:` block), still
+de-duplicated against the page:
+```php
+META::tag(string $name, ?string $content): void   // name= , or property= for an og:* key
+META::link(string $rel, string $href, array $attrs = []): void
+META::raw(string $html): void                       // a verbatim, already-valid tag line
+META::tags(string $html = ''): string               // the whole block, \n-joined
+META::reset(): void
+```
+```php
+META::tag('twitter:image', 'https://humainhumain.com/card.png');
+META::tag('og:image:alt', 'The Humain Humain team at work');
+META::link('icon', './favicon.svg', ['type' => 'image/svg+xml']);
+```
+Same API from procedural code: `meta_tag()`, `meta_link()`, `meta_raw()`,
+`meta_tags()`.
+<a id="meta-config"></a><a id="jsonld-config"></a>
+#### `seo` config
+These keys live directly under the **top-level** `seo:` block of
+`kirigami.yaml` (a sibling of `kirigami:`, `prepros:`, …) and feed both META's
+tags and LD's JSON-LD. All are optional.
+| Key | Type | Description |
+|-----|------|-------------|
+| `auto` | `bool` | Inject META's tags automatically. Default `true` **once the `seo:` block exists**. `auto: false` keeps the block for its values but stops the tags — `META::tags()` / `meta_tags()` can place them by hand. |
+| `jsonld` | `bool` | Inject LD's JSON-LD automatically. Default `true` **once the `seo:` block exists**. `jsonld: false` stops it — `LD::script()` / `ld_script()` can place it by hand. |
+| `titleFormat` | `string` | `<title>` template for a normal page. Tokens `{title}`, `{project}`, `{tagline}`. Dangling separators from an empty token are trimmed. Default `{title} — {project}`. |
+| `titleFormatHome` | `string` | Title template when the page has no `@title` (home / section landings). Default `{project} — {tagline}`. |
+| `description` | `string` | Default description for pages with no `@description` / `@abstract`, and the JSON-LD main entity / WebSite description. Defaults to the loose `description`. |
+| `keywords` | `string[]` \| `string` | Default `keywords` content (list or comma string) and the WebSite `keywords`. Defaults to the loose `keywords`. |
+| `robots` | `string` | Default robots directive. Default `index, follow`. `robots: false` omits the tag. |
+| `lang` | `string` | BCP-47 tag → `<meta name="language">`, `og:locale` (dash→underscore) and JSON-LD `inLanguage`. Defaults to the loose `lang` / `language`, then `en`. |
+| `generator` | `string` \| `false` | `<meta name="generator">`. Default `Kirigami`; `false` omits it. |
+| `author` / `designer` | `string` | Default to the loose `author` / `designer` keys (author also falls back to `person`'s name). `designer` is not emitted unless set. |
+| `themeColor` | `string` | `<meta name="theme-color">`. Not emitted unless set. |
+| `image` | `string` | Default `og:image` / `twitter:image` and JSON-LD image — absolute URL or path relative to `baseurl`. Defaults to `logo`, then the loose `image` / `ogimage`. |
+| `logo` | `string` | Organization logo (JSON-LD `ImageObject`), absolute URL or path relative to `baseurl`. |
+| `ogType` | `string` | Default `og:type`. Default `website`. |
+| `twitterCard` | `string` | `twitter:card` type. Default `summary_large_image`. |
+| `twitter` | `string` \| `map` | Handle for `twitter:site` / `twitter:creator`. A bare string (with/without `@`, or a profile URL) fills both; a map takes `site` / `creator` separately. |
+| `canonical` | `bool` | Emit `<link rel="canonical">`. Default `true`. |
+| `favicon` / `appleTouchIcon` / `humans` | `string` \| `bool` | `<link rel="icon">` / `rel="apple-touch-icon"` / `rel="author"`. A path sets it (page-relative when a bare filename); `true` forces the default file (`favicon.ico` / `apple-touch-icon.png` / `humans.txt`); omitted, the default file is auto-detected on disk at the source root; `false` disables it. |
+| `type` | `string` | JSON-LD `@type` of the main entity — `Organization` (default), `ProfessionalService`, `LocalBusiness`, … |
+| `name` / `url` | `string` | JSON-LD main entity / WebSite name and URL. Default to `project` / `baseurl`. |
+| `sameAs` | `string[]` | Profile URLs, merged with the social-network URL keys found loose in the `kirigami` block. |
+| `email` / `telephone` | `string` | Default to the loose `email` / `telephone` keys. |
+| `address` | `map` | `PostalAddress` properties. |
+| `areaServed` | `string` | Defaults to the loose `area` key. |
+| `knowsAbout` | `string[]` | Defaults to the loose `knowsabout` key. |
+| `person` | `string` \| `map` | The `#person` node (and the default `author`). A string is the name; a map takes any `Person` property. Defaults to `person` + `jobtitle` + `email`. |
+| `search` | `string` | URL template for a sitelinks `SearchAction`; must contain `{search_term_string}`. |
+```yaml
+kirigami:
+  project:  Humain Humain
+  baseurl:  https://humainhumain.com
+  person:   Méralie Murray-Hall
+  jobtitle: Anthropologue
+  facebook: https://www.facebook.com/humainhumainconsultation.ethnographie/
+seo:
+  type: ProfessionalService
+  lang: fr-CA
+  logo: assets/logo.png
+  knowsAbout: [Ethnographie, Recherche qualitative]
+  address:
+    addressLocality: Québec
+    addressCountry:  CA
+  search: https://humainhumain.com/?q={search_term_string}
+```
+---
+### CACHE
+Persistent SQLite-backed key-value cache. Survives across incremental builds via `.cache.db` at the project root.
+```php
+CACHE::get(string $key): mixed
+CACHE::set(string $key, mixed $val, int $ttl = 0): bool
+CACHE::delete(string $key): bool
+CACHE::purge(): bool   // removes expired entries
+```
+The `$ttl` is in seconds. `0` means the entry never expires. Typical use case: caching the result of network fetches in custom hooks or plugins — it is what powers [`SCRAPER`](#scraper) internally. `CURL` persists its cookie jar separately in `.cookie.txt`.
+```php
+$data = CACHE::get('my-remote-data');
+if ($data === null) {
+    $data = json_decode(file_get_contents('https://api.example.com/data.json'));
+    CACHE::set('my-remote-data', $data, 3600); // cache for 1 hour
+}
+```
+---
+### IMG
+Image manipulation helper. GD handles JPEG, PNG, GIF, WebP and AVIF directly;
+anything GD can't decode (HEIC, TIFF, BMP, and the vector formats SVG, EPS, AI,
+PDF) falls back to Imagick, which rasterizes it to a GD image in memory. Vector
+files with no intrinsic pixel size are rasterized at 2000&nbsp;px on the longest
+side, preserving the aspect ratio.
+```php
+$img = new IMG(string $file);
+// Properties
+$img->width   // int
+$img->height  // int
+// Instance methods (resize/save are chainable)
+$img->resize(int $width, int $height = 0, bool $cover = false): self
+$img->save(string $dest, ?int $quality = null): self       // quality 0-100 for jpg/webp/avif; null = per-format default (82)
+$img->getRepresentativeColors(int $count = 5): string[]   // ['#rrggbb', …], median-cut + Lab merge
+$img->getAuraColors(int $count = 5): string[]             // ['#rrggbb', …], up to 6, via the Aura extension
+// Static helpers
+IMG::asset(string $path, int $width = 0, int $height = 0, bool $cover = false): string  // same feature as the <img asset> tag and the img-asset() Sass function
+IMG::palette(string $path, int $colors = 5): string[]      // Aura-backed: at most 6 colours
+```
+`resize()` operates in *contain* mode by default (scales to fit within the target box while preserving aspect ratio). Pass `$cover = true` to crop and fill the exact target dimensions.
+`save()` infers the output format from the file extension (`.jpg`, `.jpeg`, `.png`, `.gif`, `.webp`, `.avif`) and marks the file as a build output.
+```php
+// Build a 1200×630 cropped Open Graph image next to the original
+(new IMG('/project/src/images/hero.jpg'))
+    ->resize(1200, 630, true)
+    ->save('/project/src/images/hero-og.jpg');
+```
+`IMG::asset()` and `IMG::palette()` are what back kirigami-core's `img-asset()`
+and `colors()` Sass functions: they resolve `$path` against `image.source` from
+`kirigami.yaml`, generate a resized/re-encoded file under `image.dest` (only
+when missing or stale), or return a `CACHE`-backed list of representative
+colours. `IMG::palette()` extracts them with the Aura extension (vibrant and
+muted swatches, each in a dark and a light variant), so it returns at most six
+colours, most populated first; asking for more returns what Aura found. Both
+are equally usable from your own PHP.
+`IMG::asset()` is the single implementation behind the [`<img asset>` tag](#built-in-tags)
+too — the tag is just a thin wrapper. Generated files are named after the source
+plus a dimension suffix: `-<W>w`, `-<H>h`, `-<W>x<H>`, or `-<W>x<H>-cover`, with
+the `image.format` extension (e.g. `hero.jpg` + `width="800"` → `hero-800w.webp`).
+`@kirigami/kirigami`'s Sass `img-asset()` / `colors()` functions produce the same
+files from the same config through this same class, via
+[`processImages()`](#processimagesjobs) — one engine, no native dependency.
+---
+### FS
+Filesystem utilities.
+```php
+FS::dig(string $glob): iterable          // recursive glob, yields file paths
+FS::getRelativePath(string $from, string $to): string
+FS::phpFileInfo(string $file): object|false  // page annotations (PHPDOC or Markdown header, + inherited @@)
+FS::splitHeader(string $text): array     // Markdown page → [annotations, body, @@ names]
+FS::indexFile(string $dir): ?string      // the folder's _index.php, else _index.md, else null
+FS::getChildren(string $backtrace = ''): object[]  // child _index pages, ordered by @position
+FS::getBreadcrumb(string $backtrace = ''): object[]  // ancestor _index pages, top-most first (opt-in via @breadcrumb)
+FS::rmdir(string $dir, bool $removeSelf = true): bool
+FS::pathJoin(string ...$parts): string   // URL-aware path join with .. resolution
+```
+`FS::dig()` is the workhorse of directory-wide builds — it recursively walks a glob pattern and yields every matching file path.
+`FS::phpFileInfo()` parses the first PHPDOC block of a PHP file (or the header of a [Markdown page](#markdown-pages)), merges in the values it [inherits](#inherited-annotations-) from the pages above it, and returns the `@tag value` pairs as a `stdClass`. Each call returns a fresh copy, so adding a key to the result is safe. This is used internally to resolve page metadata and data-file annotations.
+`FS::getChildren()` (procedural: `fs_get_children()`) — usable only during a render — scans the folders directly below the calling template, keeps the ones that contain an `_index.php` or an `_index.md`, and returns one `stdClass` per child: the parsed PHPDOC of that `_index.php` plus a `->file` key with its absolute path. Entries are ordered by `@position` ascending (a page with no `@position` sorts as `999999`), then by folder name (natural, case-insensitive). Handy for building a section index or a navigation menu:
+```php
+<?php foreach (fs_get_children() as $page): ?>
+  <li><a href="<?= FS::getRelativePath(__DIR__, dirname($page->file)) ?>/"><?= $page->title ?></a></li>
+<?php endforeach ?>
+```
+`FS::getBreadcrumb()` (procedural: `fs_get_breadcrumb()`) — also render-only — is the upward counterpart: it returns the calling page's breadcrumb trail. It only produces output when the calling file opts in with `@breadcrumb true` (or `@breadcrumb 1`) in its first PHPDOC block, otherwise it returns `[]`. From the folder **above** the caller's own folder (a page is never part of its own trail), it walks the parent directories upward and collects the `_index.php` of each, stopping at the source root or at the first ancestor `_index.php` that carries no active `@breadcrumb` tag — that page acts as a separator and is left out. Directories without an `_index.php` are skipped without breaking the chain. Entries come back ordered from the top-most ancestor down to the nearest parent, each a `stdClass` (parsed PHPDOC of its `_index.php` plus a `->file` key with its absolute path):
+```php
+<nav aria-label="Breadcrumb">
+<?php foreach (fs_get_breadcrumb() as $crumb): ?>
+  <a href="<?= FS::getRelativePath(__DIR__, dirname($crumb->file)) ?>/"><?= $crumb->title ?></a>
+<?php endforeach ?>
+</nav>
+```
+---
+### STR
+String utilities used internally by the tag-processing pipeline, and available for your own templates and plugins.
+```php
+STR::htmlesc(string $str): string
+STR::replaceTags(string $tag, string $html, callable $callback): string
+STR::parseHtmlAttributes(string $attrString): array
+STR::trimIndent(string $str): string
+STR::is_url(string $str): bool
+STR::html_entities_decode(string $str): string
+STR::shorthash(string $str): string
+STR::normalize(string $str): string
+STR::slug(string $str, string $sep = ''): string
+```
+`STR::replaceTags()` is the engine behind `PREPROS::registerTag()`. It finds all occurrences of `<tagname ...>...</tagname>` in an HTML string and replaces each with the return value of `$callback($fullMatch, $attrs, $body)`. Occurrences inside Markdown code (an inline code span or a fenced block) are left as written, so a tag shown as an example is not run.
+`STR::trimIndent()` strips the common leading whitespace from a multi-line string — handy when pulling content out of indented `<markdown>` blocks.
+`STR::is_url()` checks whether a string parses as a URL with a recognized scheme (`http`, `https`, `ftp`, `ftps`, `ssh`, `ssl`, `sftp`, `itunes`).
+`STR::html_entities_decode()` trims a string and decodes its HTML entities — handy when normalizing text scraped from a third-party page.
+`STR::shorthash()` returns the first 12 characters of a string's SHA-256 hash — used internally as a stable, filename-safe cache key (see `SCRAPER`).
+`STR::normalize()` applies Unicode NFD decomposition and strips combining marks (`é` → `e`) — the accent-folding step used by `slug()`.
+`STR::slug()` normalizes, transliterates to ASCII, lowercases, and replaces every run of non-`[a-z0-9]` characters with `$sep` (empty by default → a compact identifier; pass `'-'` for a conventional hyphenated slug).
+---
+### ARR
+Recursive lookup helper for nested arrays and objects.
+```php
+ARR::find_key(mixed $data, string $key): mixed
+```
+Walks an array or object (including mixed nested `stdClass`/array structures, as produced by `YAML::parse()` or `json_decode()`) depth-first and returns the value of the **first** matching key found, at any depth, or `null` if none matches.
+```php
+$config = YAML::parseFile('team.yaml');
+$email  = ARR::find_key($config, 'email'); // finds `email` however deep it's nested
+```
+---
+### CURL
+CURL verifies HTTPS certificate chains and hostnames, including redirects. The network-enabled WASM runtime supplies Node’s root certificates through `curl.cainfo` and `openssl.cafile`. Untrusted certificates and hostname mismatches are rejected; there is no insecure fallback. `getInfo()` returns `false` when cURL fails, and `getContents()` preserves its existing failure return values.
+Low-level HTTP client built on PHP's cURL extension, used internally by `SCRAPER`. Ships with a realistic browser `User-Agent`/header set and a cookie jar persisted at `.cookie.txt` (auto-registered via `PREPROS::exportFile()`).
+```php
+CURL::urlExists(string $url, ?string $mimereg = null): bool
+CURL::getInfo(string $url): array|false     // HEAD request, returns curl_getinfo()
+CURL::getContents(string $file, ?string $dest = null, ?callable $clb = null): string|bool
+```
+`CURL::urlExists()` issues a `HEAD` request and returns `true` for any `2xx`/`3xx` response.
+`CURL::getContents()` downloads a URL. Without `$dest`, it returns the body as a string; with `$dest`, it streams the download to that file path and returns a boolean. Pass `$clb` to receive download progress as a float between `0` and `1`.
+```php
+CURL::getContents('https://example.com/report.pdf', '/project/src/downloads/report.pdf', function (float $progress) {
+    error_log(sprintf('%.0f%%', $progress * 100));
+});
+```
+---
+### SCRAPER
+Fetches a URL and extracts page metadata (`title`, `description`, `image`, `label`) from its JSON-LD (`schema.org`), Open Graph, and standard `<meta>` tags — the kind of data you'd want for a rich link preview. Results are cached indefinitely via `CACHE`, keyed on the URL.
+```php
+$metas = SCRAPER::get(string $url): object|false;
+```
+```php
+$metas = SCRAPER::get('https://example.com/blog/some-article');
+if ($metas) {
+    echo $metas->title;        // string
+    echo $metas->description;  // string
+    echo $metas->image;        // string (absolute URL, may be empty)
+    echo $metas->label;        // string — site/publisher name, may be empty
+    echo $metas->url;          // string — the URL that was scraped
+}
+```
+Returns `false` if the page can't be reached, can't be parsed, or has no discoverable title. Throws an `Exception` on invalid URLs. Uses `CURL::getContents()` under the hood, so it benefits from the same shared cookie jar and browser-like headers.
+---
+### OBF
+Simple reversible obfuscation for non-secret values embedded in HTML, such as display labels or contact data. It does not protect API tokens or other credentials.
+```php
+$encoded = OBF::encode(mixed $obj): string;
+$decoded = OBF::decode(string $str): mixed;
+```
+Applies JSON encoding → base64 → ROT-13 → gzip. Not cryptographically secure; intended for light obfuscation only.
+---
+### STD
+Result helpers that write JSON to `/internal/prepros_result.json` in the virtual filesystem. Node reads that result file; ordinary PHP stdout/stderr are separate diagnostic channels.
+```php
+STD::succeed(array|string $props = []): void  // exits 0, writes JSON to /internal/prepros_result.json
+STD::error(array|string $props = []): void    // exits 1, writes JSON to /internal/prepros_result.json
+```
+These are internal to the build runner (`render()`, `sitemap()`, and `runenv()` all rely on them). You generally do not need to call them in page templates, but they are available if a script run via `runenv()` needs to terminate early with a custom result.
+---
+### Unicode normalization
+The WASM PHP build ships without `ext-intl`; the native `norm` extension
+([php-norm](https://github.com/php-kirigami/php-norm), utf8proc) provides the
+standard `Normalizer` class instead: `Normalizer::normalize()` /
+`Normalizer::isNormalized()`, the `NFC` / `NFD` / `NFKC` / `NFKD` (and
+`FORM_*`) constants, and the `normalizer_normalize()` /
+`normalizer_is_normalized()` functions. `STR::normalize()` and `STR::slug()`
+use it to fold accents. Prefer the `STR` helpers in your own code.
+The pure-PHP polyfill that used to fill this gap is still autoloadable as
+`NORMALIZER_LEGACY`, with the same API.
+---
+### Procedural shortcuts (aliases)
+Every static method of every class above is also exposed as a plain function by
+`src/libraries/aliases.inc.php` (autoloaded — no `require` needed). Each alias is
+named `<lowercase class>_<snake_case method>()` and does nothing but forward its
+arguments, so the classes remain the canonical API. They exist to make page
+templates and `kiri run` scripts read better:
+```php
+<?= md_to_html(file_get_contents('CHANGELOG.md')) ?>
+<img src="<?= img_asset('hero.jpg', 1200, 630, true) ?>" alt="">
+```
+Every function has a complete PHPDoc block, so editor hover and autocomplete
+surface the signature, parameters, and description.
+| Class | Aliases |
+|---|---|
+| `PREPROS` | `prepros_render` · `prepros_sitemap` · `prepros_mount` · `prepros_fstat` · `prepros_export_file` · `prepros_get_exported_files` · `prepros_backtrace_file` · `prepros_register_tag` · `prepros_register_hook` · `prepros_run_hook` |
+| `MD` | `md_to_html` · `md_register_plugin` · `md_unregister_plugin` · `md_get_registered_plugins` · `md_register_emoji` |
+| `HTML` | `html_format` |
+| `YAML` | `yaml_parse` · `yaml_parse_file` · `yaml_load_file` |
+| `SCHEMA` | `schema` (factory) · `schema_validate` |
+| `LD` | `ld_add` · `ld_node` · `ld_ref` · `ld_organization` · `ld_person` · `ld_website` · `ld_web_page` · `ld_breadcrumb` · `ld_faq_page` · `ld_script` · `ld_json` |
+| `META` | `meta_tag` · `meta_link` · `meta_raw` · `meta_tags` |
+| `CACHE` | `cache_get` · `cache_set` · `cache_delete` · `cache_purge` |
+| `IMG` | `img_asset` · `img_palette` |
+| `FS` | `fs_dig` · `fs_get_relative_path` · `fs_php_file_info` · `fs_rmdir` · `fs_path_join` |
+| `STR` | `str_htmlesc` · `str_replace_tags` · `str_parse_html_attributes` · `str_trim_indent` · `str_is_url` · `str_html_entities_decode` · `str_shorthash` · `str_normalize` · `str_slug` |
+| `ARR` | `arr_find_key` |
+| `CURL` | `curl_url_exists` · `curl_get_info` · `curl_get_contents` |
+| `SCRAPER` | `scraper_get` |
+| `OBF` | `obf_encode` · `obf_decode` |
+| `STD` | `std_succeed` · `std_error` |
+Notes:
+- `register_tag()` / `register_hook()` are kept as unprefixed aliases of
+  `prepros_register_tag()` / `prepros_register_hook()`.
+- `img_asset()` resolves the generated URL relative to the file that calls it,
+  exactly like `IMG::asset()`.
+- `schema_validate(array $schema, mixed $data, ?array &$errors = null): bool`
+  fills `$errors` with the validation messages.
+- `yaml_parse()` / `yaml_parse_file()` are only defined when the PECL `yaml`
+  extension isn't already providing them.
+---
+## Plugin system
+`@kirigami/php-prepros` has two complementary plugin layers: **PREPROS** (HTML-tag level, operates on the assembled page) and **MD** (shortcode level, operates inside Markdown content).
+---
+### PREPROS tags
+Register a custom HTML tag that is processed **after** PHP execution, on the fully assembled HTML string:
+```php
+// In a file listed under prepros.includes in kirigami.yaml, or in before.php:
+PREPROS::registerTag('gallery', function (string $fullTag, array $attrs, string $body): string {
+    $id   = $attrs['id'] ?? '';
+    $imgs = glob("/project/src/images/gallery/{$id}/*.webp");
+    $html = '<div class="gallery">';
+    foreach ($imgs as $img) {
+        $src = str_replace('/project/src', '', $img);
+        $html .= "<img src=\"{$src}\" loading=\"lazy\">";
+    }
+    return $html . '</div>';
+});
+```
+Then in any page template:
+```html
+<gallery id="summer-2025"></gallery>
+```
+The callback receives:
+| Parameter | Type | Description |
+|-----------|------|-------------|
+| `$fullTag` | `string` | The complete matched tag string |
+| `$attrs` | `array` | Parsed HTML attributes as an associative array |
+| `$body` | `string` | Inner content between opening and closing tags |
+The built-in [`<markdown>` and `<img asset>` tags](#built-in-tags) are registered
+this way.
+---
+### PREPROS hooks
+Hooks let you intercept and transform data at key points in the rendering pipeline:
+```php
+PREPROS::registerHook(string $hookName, callable $callback): void
+```
+| Hook | When it fires | `$data` type | Expected return |
+|------|---------------|--------------|-----------------|
+| `boot` | Once per process, right after bootstrap (config loaded, `includes` pulled in), before any page renders. Fires for every entrypoint. | `stdClass $config` | ignored |
+| `shutdown` | Via `register_shutdown_function()`, at the very end of the request — fires even after `STD::succeed()`/`STD::error()`'s `exit()`, unlike `auto_append_file` (which PHP skips whenever the script exits). The place for cleanup that must always run. | `null` | ignored |
+| `page_info` | After PHPDOC parsing, before rendering (auto-loads `.yaml`/`.json`/`.md` annotations) | `[$filePath, $pageObject]` — see note | `$pageObject` (modified) |
+| `pre_render` | Before PHP execution | Raw file contents as `string` | Ignored by the current render call |
+| `pre_before` | Just before the `before` include (inside its output buffer — `echo` to prepend to the header) | `before` config path as `string\|null` | ignored |
+| `post_before` | Right after the `before` include, on the captured header | Header `string` | `string` |
+| `pre_type_before` | Just before the page's `@type` `before` include, if any (inside its output buffer) | Type's `before` config path as `string\|null` | ignored |
+| `post_type_before` | Right after the `@type` `before` include, on the captured type header | Type header `string` | `string` |
+| `pre_type_after` | Just before the page's `@type` `after` include, if any (inside its output buffer) | Type's `after` config path as `string\|null` | ignored |
+| `post_type_after` | Right after the `@type` `after` include, on the captured type footer | Type footer `string` | `string` |
+| `pre_after` | Just before the `after` include (inside its output buffer — `echo` to prepend to the footer) | `after` config path as `string\|null` | ignored |
+| `post_after` | Right after the `after` include, on the captured footer | Footer `string` | `string` |
+| `post_render` | After tag processing, before `HTML::format()` | Assembled HTML `string` | `string` |
+Multiple callbacks can be registered for the same hook — they are executed in registration order, each receiving the return value of the previous one.
+> Built-in `page_info` + `post_render` callbacks power [`LD`](#ld) and
+> [`META`](#meta): `page_info` captures the page under render, `post_render`
+> injects the JSON-LD `<script>` and the `<meta>`/`<link>` block into its
+> `<head>`. Your own callbacks run after them.
+> **`page_info` payload shape.** The hook *fires* with `[$filePath, $pageObject]`,
+> but each callback is expected to return the `$pageObject` alone — so a callback
+> registered after the built-ins receives the bare object, not the pair. Handle
+> both: `$page = is_array($p) ? $p[1] : $p;` (the current page path is always
+> available as `PREPROS::$file`).
+```php
+// Example: inject a last-modified date into every page
+PREPROS::registerHook('post_render', function (string $html): string {
+    $date = date('Y-m-d');
+    return str_replace('{{build_date}}', $date, $html);
+});
+```
+---
+### MD plugins
+MD plugins add custom shortcode tags inside Markdown content. They work inside `<markdown>` blocks, in `.md` data files, and anywhere `MD::toHtml()` is called.
+**Inline syntax** (all on one line):
+```
+{% tagname arg1 "argument with spaces" %}
+```
+**Block syntax** (body on subsequent lines):
+```
+{% tagname optional-arg
+Line one of the body.
+Line two of the body.
+%}
+```
+```php
+MD::registerPlugin(string $name, callable $callback): void
+```
+The callback signature is always `(array $args, string $body): string`. `$args` contains arguments parsed from the opening line; `$body` is the trimmed multi-line body (empty string for inline tags).
+---
+### Built-in plugins
+The following MD plugins are registered out of the box in `md.plugins.php`:
+#### `{% callout type ["Title"] content %}`
+Renders a styled callout block. `type` is one of `info`, `success`, `warning`, `danger`.
+```
+{% callout warning "Heads up" This section is outdated. %}
+{% callout danger "Critical"
+Line one of a longer warning.
+Line two after a blank line.
+%}
+```
+#### `{% img-asset path [width height [cover]] %}`
+Generate an image with `IMG::asset()` and emit an `<img>` tag. Width and height default to zero; the optional final `cover` selects cropping.
+```markdown
+{% img-asset photo.jpg 800 600 cover %}
+```
+YouTube and Vimeo shortcuts require `@kirigami/plugin-embed`; YouTube is no longer a built-in plugin.
+#### `{% codepen id [user height] %}`
+Embeds a CodePen result via `<iframe>`. `user` defaults to `anonymous`, `height` defaults to `400`.
+```
+{% codepen abcXYZ %}
+{% codepen abcXYZ jsmith 500 %}
+```
+#### `{% checklist ["Title"] items %}`
+Renders a block-syntax list of checkbox items, one per line, with an optional title.
+```
+{% checklist "Today"
+Do the dishes
+Walk the dog
+Read a book
+%}
+```
+---
+## Extending the `<markdown>` tag
+The `<markdown>` tag is one of the [built-in tags](#built-in-tags) (alongside
+`<img asset>`), registered as a PREPROS tag out of the box. It converts its inner
+content from Markdown to HTML and strips common leading indentation so you can
+write cleanly inside your PHP templates:
+```html
+<section class="about">
+    <div>
+        <markdown>
+            ## Who we are
+            We are a **student organization** from Québec.
+            {% codepen abc123 author 400 %}
+        </markdown>
+    </div>
+</section>
+```
+All registered MD plugins are available inside `<markdown>` blocks. You can extend the tag's behaviour by registering additional MD plugins (see above) or by overriding the tag itself:
+```php
+PREPROS::registerTag('markdown', function (string $tag, array $attrs, string $body): string {
+    $body = STR::trimIndent($body);
+    $html = MD::toHtml($body);
+    // wrap in a container, add a class, etc.
+    $class = $attrs['class'] ?? 'prose';
+    return "<div class=\"{$class}\">{$html}</div>";
+});
+```
+---
+## Requirements
+- Node.js `>= 24.0.0`
+- npm `>= 10.2.3`
+- ESM only (`"type": "module"`)
+---
+## License
+GPL-3.0-or-later © Maxime Larrivée-Roy, 2026