oh-my-opencode 5.0.0-beta.89 → 5.0.0-beta.90

package/packages/omo-codex/plugin/skills/ulw-research/scripts/design-spec.mjs

@@ -0,0 +1,161 @@
+// design-spec.mjs - read the palette and lineage mode out of design-spec.md, and render a
+// design-spec.md draft from a format extract. Nothing is guessed: every value the extract
+// does not carry is written `TODO: ask` and listed under Open questions.
+// Plain Node ESM, node: builtins and siblings only.
+
+import { DESIGN_SPEC_SECTIONS, LINEAGE_MODES } from "./contracts.mjs"
+import { hexColors } from "./css-lite.mjs"
+
+const TODO = "TODO: ask"
+// English and Korean section names (the requester writes specs in either language).
+const PALETTE_HEADING = /^(palette|tokens|colou?rs?|\uD1A0\uD070|\uD314\uB808\uD2B8|\uC0C9\uC0C1|\uCEEC\uB7EC)/i
+const FENCE = /^\s{0,3}(`{3,}|~{3,})/
+const HEADING = /^(#{1,6})\s+(.*?)\s*#*\s*$/
+const TOKEN_DECLARATION = /--[\w-]+\s*:\s*([^;]*)/g
+const FIGURE_STANDARD =
+	"Each figure sits in a fixed-size container with a caption; the image scales to fit with its aspect ratio preserved (object-fit: contain), never stretched, cropped, or spilling out. Every chart carries a title, axis labels, units, and value labels."
+
+/**
+ * Walk markdown lines with fence awareness.
+ * visit(line, { fenced, heading: { level, text } | null })
+ */
+function walkLines(markdown, visit) {
+	let fence = null
+	for (const line of String(markdown ?? "").split(/\r?\n/)) {
+		const fenceMatch = FENCE.exec(line)
+		if (fence) {
+			const closes = fenceMatch && fenceMatch[1][0] === fence[0] && fenceMatch[1].length >= fence.length && line.trim() === fenceMatch[1]
+			if (closes) fence = null
+			else visit(line, { fenced: true, heading: null })
+			continue
+		}
+		if (fenceMatch) {
+			fence = fenceMatch[1]
+			continue
+		}
+		const heading = HEADING.exec(line)
+		visit(line, { fenced: false, heading: heading ? { level: heading[1].length, text: heading[2].replace(/^[\d.\s]+/, "") } : null })
+	}
+}
+
+/**
+ * Palette = hex values of every `--x: #hex` inside any fenced code block, plus every hex
+ * anywhere under a heading named Palette / Tokens / Colors / Colours or their Korean equivalents (subheadings included).
+ * @returns {Set<string>} normalized #rrggbb values
+ */
+export function parsePalette(specMarkdown) {
+	const palette = new Set()
+	let paletteLevel = 0
+	walkLines(specMarkdown, (line, { fenced, heading }) => {
+		if (heading) {
+			if (paletteLevel && heading.level > paletteLevel) return
+			paletteLevel = PALETTE_HEADING.test(heading.text) ? heading.level : 0
+			return
+		}
+		if (fenced) for (const match of line.matchAll(TOKEN_DECLARATION)) for (const hex of hexColors(match[1])) palette.add(hex)
+		if (paletteLevel) for (const hex of hexColors(line)) palette.add(hex)
+	})
+	return palette
+}
+
+/** `Lineage: inline|section` under the Citations heading; absent or unrecognized -> "inline". */
+export function parseLineageMode(specMarkdown) {
+	let citationsLevel = 0
+	let mode = null
+	walkLines(specMarkdown, (line, { fenced, heading }) => {
+		if (mode || fenced) return
+		if (heading) {
+			if (citationsLevel && heading.level > citationsLevel) return
+			citationsLevel = /^citations\b/i.test(heading.text) ? heading.level : 0
+			return
+		}
+		const match = citationsLevel ? /^\s*(?:[-*]\s+)?lineage\s*:\s*(\S+)/i.exec(line) : null
+		if (match && LINEAGE_MODES.includes(match[1].toLowerCase())) mode = match[1].toLowerCase()
+	})
+	return mode ?? "inline"
+}
+
+function entriesOf(value) {
+	if (value instanceof Map) return [...value]
+	return value !== null && typeof value === "object" ? Object.entries(value) : []
+}
+
+function isKnown(value) {
+	if (value === undefined || value === null || value === "" || value === "unknown") return false
+	if (Array.isArray(value)) return value.length > 0
+	return true
+}
+
+function tokenFence(entries, dark) {
+	const lines = entries.map(([name, value]) => `${dark ? "    " : "  "}${name}: ${value};`)
+	const body = dark ? ["@media (prefers-color-scheme: dark) {", "  :root {", ...lines, "  }", "}"] : [":root {", ...lines, "}"]
+	return ["```css", ...body, "```"]
+}
+
+/**
+ * Render a design-spec.md draft. `extract` follows the format-extract shape
+ * ({ origin, bytes, date?, tokens: { light, dark }, fonts, colors, breakpoints, headingNumbering,
+ * citationStyle, lineageMode, figureContainers, skeleton, measure, notes }); every field is optional.
+ * @returns {string}
+ */
+export function renderDesignSpec(extract = {}) {
+	const source = extract ?? {}
+	const open = []
+	const field = (label, value, format = String) => {
+		if (isKnown(value)) return format(value)
+		open.push(label)
+		return TODO
+	}
+	const fonts = source.fonts ?? {}
+	const measure = source.measure ?? {}
+	const light = entriesOf(source.tokens?.light)
+	const dark = entriesOf(source.tokens?.dark)
+	const colors = Array.isArray(source.colors) ? source.colors : []
+	const skeleton = Array.isArray(source.skeleton) ? source.skeleton : []
+	const notes = Array.isArray(source.notes) ? source.notes : []
+
+	const sections = {
+		"Extracted from": [
+			`- Source: ${field("Source path or URL", source.origin)}`,
+			`- Bytes: ${field("Source byte count", source.bytes)}`,
+			`- Date: ${source.date ?? new Date().toISOString().slice(0, 10)}`,
+			...notes.map((note) => `- Note: ${note}`),
+		],
+		Tokens: [
+			...(light.length ? tokenFence(light, false) : [`- Light tokens: ${field("Light tokens", null)}`]),
+			...(dark.length ? tokenFence(dark, true) : [`- Dark tokens: ${field("Dark tokens", null)}`]),
+			...(colors.length
+				? ["- Top colors:", ...colors.map((c) => `  - ${c.hex}${c.count ? ` (${c.count} uses${c.roles?.length ? `; ${c.roles.join(", ")}` : ""})` : ""}`)]
+				: [`- Top colors: ${field("Top colors", null)}`]),
+		],
+		Typography: [
+			`- Body font: ${field("Body font", fonts.body)}`,
+			`- Heading font: ${field("Heading font", fonts.heading)}`,
+			`- Mono font: ${field("Mono font", fonts.mono)}`,
+			`- Body size: ${field("Body size", measure.fontSize)}`,
+			`- Line height: ${field("Line height", measure.lineHeight)}`,
+		],
+		Layout: [
+			`- Breakpoints: ${field("Breakpoints", source.breakpoints, (list) => list.map((px) => `${px}px`).join(", "))}`,
+			`- Measure (max width): ${field("Measure", measure.maxWidth)}`,
+		],
+		Structure: [
+			`- Heading numbering: ${field("Heading numbering", source.headingNumbering)}`,
+			skeleton.length ? "- Skeleton:" : `- Skeleton: ${field("Section skeleton", null)}`,
+			...skeleton.map((entry) => `${"  ".repeat(Math.max(1, entry.level))}- ${entry.text}`),
+		],
+		Figures: [
+			`- Containers: ${field("Figure containers", source.figureContainers, (list) => list.join(", "))}`,
+			`- Standard: ${FIGURE_STANDARD}`,
+		],
+		Citations: [
+			`- Style: ${field("Citation style", source.citationStyle)}`,
+			`- Lineage: ${field("Lineage mode", LINEAGE_MODES.includes(source.lineageMode) ? source.lineageMode : null)}`,
+		],
+	}
+	sections["Open questions"] = open.length ? open.map((label) => `- ${label}: ${TODO}`) : ["- None"]
+
+	const out = ["# Design spec", ""]
+	for (const heading of DESIGN_SPEC_SECTIONS) out.push(`## ${heading}`, ...(sections[heading] ?? [`- ${TODO}`]), "")
+	return out.join("\n")
+}

package/packages/omo-codex/plugin/skills/ulw-research/scripts/entities.mjs

@@ -0,0 +1,39 @@
+// entities.mjs - HTML character reference decoding for html-lite.mjs.
+//
+// Supports a fixed set of named entities plus decimal and hex numeric references. Unknown or
+// malformed references stay verbatim. Single indexOf scan; plain Node ESM, no dependencies.
+
+const NAMED_ENTITIES = {
+	amp: "&", lt: "<", gt: ">", quot: '"', apos: "'", nbsp: "\u00a0", mdash: "\u2014", ndash: "\u2013",
+	hellip: "\u2026", copy: "\u00a9", reg: "\u00ae", times: "\u00d7", middot: "\u00b7",
+}
+
+function decodeOne(body) {
+	if (Object.hasOwn(NAMED_ENTITIES, body)) return NAMED_ENTITIES[body]
+	if (body[0] !== "#") return null
+	const hex = body[1] === "x" || body[1] === "X"
+	const digits = body.slice(hex ? 2 : 1)
+	if (!(hex ? /^[0-9a-fA-F]{1,6}$/ : /^[0-9]{1,7}$/).test(digits)) return null
+	const cp = Number.parseInt(digits, hex ? 16 : 10)
+	return cp > 0 && cp <= 0x10ffff && (cp < 0xd800 || cp > 0xdfff) ? String.fromCodePoint(cp) : "\ufffd"
+}
+
+/** Decode the supported named entities plus decimal/hex numeric references; unknown ones stay verbatim. */
+export function decodeEntities(text) {
+	const source = String(text)
+	let out = ""
+	let last = 0
+	let at = source.indexOf("&")
+	while (at !== -1) {
+		const semi = source.indexOf(";", at + 1)
+		const decoded = semi !== -1 && semi - at <= 12 ? decodeOne(source.slice(at + 1, semi)) : null
+		if (decoded !== null) {
+			out += source.slice(last, at) + decoded
+			last = semi + 1
+			at = source.indexOf("&", last)
+		} else {
+			at = source.indexOf("&", at + 1)
+		}
+	}
+	return out + source.slice(last)
+}

package/packages/omo-codex/plugin/skills/ulw-research/scripts/entry-guard.mjs

@@ -0,0 +1,21 @@
+import { realpathSync } from "node:fs"
+import { pathToFileURL } from "node:url"
+
+/**
+ * True when the module at `importMetaUrl` is the process entrypoint (`node <script>`).
+ * Compares against the realpath too: Node resolves import.meta.url through symlinks while
+ * process.argv[1] keeps the symlinked spelling, so a plain equality check silently skips the CLI
+ * body when the skill is reached through a symlinked plugin directory.
+ *
+ * @param {string} importMetaUrl the caller's import.meta.url
+ */
+export function isCliEntry(importMetaUrl) {
+	const argv1 = process.argv[1]
+	if (argv1 === undefined) return false
+	if (importMetaUrl === pathToFileURL(argv1).href) return true
+	try {
+		return importMetaUrl === pathToFileURL(realpathSync(argv1)).href
+	} catch {
+		return false
+	}
+}

package/packages/omo-codex/plugin/skills/ulw-research/scripts/format-extract-css.mjs

@@ -0,0 +1,122 @@
+// format-extract-css.mjs - the CSS half of format-extract: tokens, font stacks, ranked colors,
+// breakpoints, and the reading measure. Only values the stylesheet states are returned; a
+// missing value stays undefined so the renderer writes `TODO: ask`.
+// Plain Node ESM, node: builtins and siblings only.
+
+import { customProperties, fontFamilies, hexColors, mediaBreakpoints, resolveVar, scanRules } from "./css-lite.mjs"
+
+const ROLE_ORDER = ["body", "heading", "link", "background", "border", "accent"]
+const TOP_COLORS = 8
+const ABSOLUTE_LENGTH = /^\d+(?:\.\d+)?(?:px|rem|em|ch)$/i
+
+/** Last compound of each comma-separated selector part, pseudo-classes and attributes dropped. */
+function subjects(selector) {
+	return selector.split(",").map((part) => {
+		const compounds = part.trim().split(/\s*[\s>+~]\s*/)
+		return (compounds.at(-1) ?? "").replace(/(?:::?[\w-]+(?:\([^)]*\))?|\[[^\]]*\])+$/g, "").toLowerCase()
+	})
+}
+
+const isBodySubject = (subject) => subject === "body" || subject === "html" || subject === ":root"
+const isHeadingSubject = (subject) => /^h[1-6](?![\w-])/.test(subject)
+const isMonoSubject = (subject) => /^(?:code|pre|kbd|samp)(?![\w-])/.test(subject)
+const isLinkSubject = (subject) => /^a(?![\w-])/.test(subject)
+
+function formatStack(families) {
+	return families.map((family) => (/^-?[A-Za-z_][\w-]*$/.test(family) ? family : JSON.stringify(family))).join(", ")
+}
+
+/** { body?, heading?, mono? } from font-family / font declarations, var() resolved. */
+export function fontStacks(css) {
+	const fonts = {}
+	const entries = fontFamilies(css)
+	const pick = (test) => entries.find((entry) => subjects(entry.selector).some(test))
+	const body = pick(isBodySubject)
+	const heading = pick(isHeadingSubject)
+	const mono = pick(isMonoSubject) ?? entries.find((entry) => entry.families.at(-1)?.toLowerCase() === "monospace")
+	if (body) fonts.body = formatStack(body.families)
+	if (heading) fonts.heading = formatStack(heading.families)
+	if (mono) fonts.mono = formatStack(mono.families)
+	return fonts
+}
+
+function tokenRoles(name) {
+	const roles = []
+	if (/accent|primary|brand/i.test(name)) roles.push("accent")
+	if (/(?:^--|-)bg(?:-|$)|background|surface/i.test(name)) roles.push("background")
+	if (/border|(?:^--|-)rule(?:-|$)/i.test(name)) roles.push("border")
+	return roles
+}
+
+function declarationRoles(selector, prop) {
+	if (prop.startsWith("--")) return tokenRoles(prop)
+	if (prop.startsWith("background")) return ["background"]
+	if (prop.startsWith("border") || prop.startsWith("outline")) return ["border"]
+	if (prop !== "color") return []
+	const parts = subjects(selector)
+	const roles = []
+	if (parts.some(isLinkSubject)) roles.push("link")
+	if (parts.some(isHeadingSubject)) roles.push("heading")
+	if (parts.some(isBodySubject)) roles.push("body")
+	return roles
+}
+
+/**
+ * Top hex colors by use count (ties keep first appearance): each token definition counts once,
+ * and each other declaration counts once per hex it resolves to (var() against light tokens).
+ * Roles are guessed from the selector and property where the color is used.
+ */
+export function rankColors(css, lightTokens) {
+	const table = new Map()
+	for (const rule of scanRules(css)) {
+		for (const { prop, value } of rule.declarations) {
+			const roles = declarationRoles(rule.selector, prop)
+			for (const hex of hexColors(prop.startsWith("--") ? value : resolveVar(value, lightTokens))) {
+				const entry = table.get(hex) ?? { hex, count: 0, roles: new Set() }
+				entry.count++
+				for (const role of roles) entry.roles.add(role)
+				table.set(hex, entry)
+			}
+		}
+	}
+	return [...table.values()]
+		.sort((a, b) => b.count - a.count)
+		.slice(0, TOP_COLORS)
+		.map((entry) => ({ hex: entry.hex, count: entry.count, roles: ROLE_ORDER.filter((role) => entry.roles.has(role)) }))
+}
+
+/** { fontSize?, lineHeight?, maxWidth? }: body size and leading, plus the most used absolute max width. */
+export function readingMeasure(css, lightTokens) {
+	const measure = {}
+	const rules = scanRules(css).filter((rule) => !rule.atRule)
+	for (const rule of rules) {
+		if (!subjects(rule.selector).some(isBodySubject)) continue
+		for (const { prop, value } of rule.declarations) {
+			if (prop === "font-size" && !measure.fontSize) measure.fontSize = resolveVar(value, lightTokens)
+			if (prop === "line-height" && !measure.lineHeight) measure.lineHeight = resolveVar(value, lightTokens)
+		}
+	}
+	const widths = new Map()
+	for (const rule of rules) {
+		for (const { prop, value } of rule.declarations) {
+			if (prop !== "max-width" && prop !== "max-inline-size") continue
+			const resolved = resolveVar(value, lightTokens).trim()
+			if (ABSOLUTE_LENGTH.test(resolved)) widths.set(resolved, (widths.get(resolved) ?? 0) + 1)
+		}
+	}
+	const widest = [...widths].sort((a, b) => b[1] - a[1])[0]
+	if (widest) measure.maxWidth = widest[0]
+	return measure
+}
+
+/** Every CSS-derived Extract field from one concatenated stylesheet. */
+export function cssFacts(css) {
+	const tokens = customProperties(css)
+	return {
+		tokens,
+		fonts: fontStacks(css),
+		colors: rankColors(css, tokens.light),
+		breakpoints: mediaBreakpoints(css),
+		measure: readingMeasure(css, tokens.light),
+	}
+}

package/packages/omo-codex/plugin/skills/ulw-research/scripts/format-extract.mjs

@@ -0,0 +1,229 @@
+// format-extract.mjs - read a pointed-at HTML or Markdown document and return the facts a
+// design-spec.md draft needs (render with design-spec.renderDesignSpec). Nothing is guessed:
+// a value the document does not state is left out, and the renderer writes `TODO: ask`.
+//
+// Static input only. fetchHtml downloads exactly the given URL with plain fetch; external
+// stylesheets are never followed (each is recorded as a note). A page that renders its
+// content client-side is the orchestrator's job: render it with the browser skill, save the
+// HTML, and pass the saved file. PDF input is rejected with UnsupportedFormatError.
+// Plain Node ESM, node: builtins and siblings only.
+
+import { readFileSync } from "node:fs"
+import { extname } from "node:path"
+
+import { cssFacts } from "./format-extract-css.mjs"
+import { parseHtml, proseText } from "./html-lite.mjs"
+
+export const UNSUPPORTED_MESSAGE = "unsupported: point at the HTML or Markdown source"
+export const FETCH_TIMEOUT_MS = 15_000
+
+const S_MARKER = /\[S\d+\]/
+const SOURCE_ANCHOR = /\[(?:\uCD9C\uCC98|source(?:\s*\d+)?)\]/i
+const LINEAGE_TOKEN = /\b(?:MEASURED|ASSUMED|DERIVED)\b/
+const MD_FOOTNOTE = /\[\^[^\]\s]+\]/
+const HEADING_NUMBER = /^(\d{1,3}(?:\.\d{1,3})*)\.?(?=\s|$)/
+const FENCE = /^\s{0,3}(`{3,}|~{3,})/
+const MD_HEADING = /^(#{1,6})\s+(.*?)\s*#*\s*$/
+/**
+ * @typedef {{ hex: string, count: number, roles: string[] }} ExtractColor
+ * @typedef {{
+ *   origin: string | undefined, bytes: number,
+ *   tokens: { light: Map<string, string>, dark: Map<string, string> },
+ *   fonts: { body?: string, heading?: string, mono?: string }, colors: ExtractColor[], breakpoints: number[],
+ *   headingNumbering: "double-digit" | "decimal" | "none" | "unknown",
+ *   citationStyle: "bracket-S" | "bracket-source" | "footnote" | "unknown",
+ *   lineageMode: "inline" | "section" | "unknown", figureContainers: string[],
+ *   skeleton: { level: number, text: string }[],
+ *   measure: { fontSize?: string, lineHeight?: string, maxWidth?: string }, notes: string[],
+ * }} Extract
+ */
+
+const HTML_EXTENSIONS = new Set([".html", ".htm", ".xhtml"])
+const MD_EXTENSIONS = new Set([".md", ".markdown"])
+
+export class UnsupportedFormatError extends Error {
+	constructor(detail) {
+		super(UNSUPPORTED_MESSAGE)
+		this.name = "UnsupportedFormatError"
+		this.code = "unsupported"
+		this.detail = detail
+	}
+}
+
+function rejectPdf(source, origin) {
+	if (/\.pdf$/i.test(String(origin ?? ""))) throw new UnsupportedFormatError(`pdf origin: ${origin}`)
+	if (String(source).trimStart().startsWith("%PDF-")) throw new UnsupportedFormatError("pdf bytes")
+}
+
+/** @returns {Extract} */
+function baseExtract(source, origin) {
+	return {
+		origin,
+		bytes: Buffer.byteLength(String(source), "utf8"),
+		tokens: { light: new Map(), dark: new Map() },
+		fonts: {},
+		colors: [],
+		breakpoints: [],
+		headingNumbering: "unknown",
+		citationStyle: "unknown",
+		lineageMode: "unknown",
+		figureContainers: [],
+		skeleton: [],
+		measure: {},
+		notes: [],
+	}
+}
+
+/** h2 numbers ("01", "1.2", or null) -> double-digit when every numbered h2 uses two digits. */
+function classifyNumbering(numbers) {
+	if (!numbers.length) return "unknown"
+	const numbered = numbers.filter(Boolean)
+	if (numbered.length * 2 < numbers.length) return "none"
+	return numbered.every((number) => /^\d{2}$/.test(number)) ? "double-digit" : "decimal"
+}
+
+function headingNumber(text) {
+	return HEADING_NUMBER.exec(text)?.[1] ?? null
+}
+
+function citationFacts(text, hasFootnotes, hasLineageAttribute) {
+	const style = S_MARKER.test(text) ? "bracket-S" : SOURCE_ANCHOR.test(text) ? "bracket-source" : hasFootnotes ? "footnote" : "unknown"
+	let lineage = "unknown"
+	if (hasLineageAttribute || LINEAGE_TOKEN.test(text) || S_MARKER.test(text)) lineage = "inline"
+	else if (SOURCE_ANCHOR.test(text)) lineage = "section"
+	return { style, lineage }
+}
+
+function walk(node, visit) {
+	for (const child of node.children) {
+		visit(child)
+		walk(child, visit)
+	}
+}
+
+function classesOf(node) {
+	return String(node.attrs.class ?? "").split(/\s+/).filter(Boolean)
+}
+
+/** Previous element sibling whose whole text is a heading number (a `.sec-no` style eyebrow). */
+function eyebrowNumber(heading) {
+	const siblings = heading.parent.children.filter((child) => child.tag !== "#text" || child.text.trim())
+	const previous = siblings[siblings.indexOf(heading) - 1]
+	if (!previous || previous.tag === "#text" || /^h[1-6]$/.test(previous.tag)) return null
+	const text = proseText(previous)
+	return HEADING_NUMBER.test(text) && headingNumber(text) === text.replace(/\.$/, "") ? headingNumber(text) : null
+}
+
+/** Nearest `<figure>` or fig/chart-classed ancestor of an img or svg. */
+function figureContainer(node) {
+	for (let cur = node.parent; cur && cur.tag !== "body"; cur = cur.parent) {
+		const classes = classesOf(cur)
+		if (cur.tag === "figure" || classes.some((name) => /fig|chart/i.test(name))) return cur
+	}
+	return null
+}
+
+function rankByCount(names) {
+	const counts = new Map()
+	for (const name of names) counts.set(name, (counts.get(name) ?? 0) + 1)
+	return [...counts].sort((a, b) => b[1] - a[1]).map(([name]) => name)
+}
+
+/** @returns {Extract} throws UnsupportedFormatError on PDF input. */
+export function extractFromHtml(source, { origin } = {}) {
+	rejectPdf(source, origin)
+	const extract = baseExtract(source, origin)
+	const root = parseHtml(source)
+	const styles = []
+	const headings = []
+	const containers = []
+	let hasFootnotes = false
+	let hasLineageAttribute = false
+	let scripts = 0
+	walk(root, (node) => {
+		if (node.tag === "style") styles.push(node.raw ?? "")
+		else if (node.tag === "script") scripts++
+		else if (node.tag === "link" && /(?:^|\s)stylesheet(?:\s|$)/i.test(node.attrs.rel ?? "")) {
+			extract.notes.push(`skipped external stylesheet: ${node.attrs.href ?? "(no href)"}`)
+		} else if (/^h[1-6]$/.test(node.tag)) headings.push(node)
+		else if (node.tag === "img" || node.tag === "svg") {
+			const container = figureContainer(node)
+			if (container) containers.push(...classesOf(container))
+		} else if (node.tag === "sup" && node.children.some((child) => child.tag === "a" && /^#/.test(child.attrs.href ?? ""))) hasFootnotes = true
+		else if (node.attrs?.role === "doc-noteref") hasFootnotes = true
+		if (node.attrs && Object.hasOwn(node.attrs, "data-lineage")) hasLineageAttribute = true
+	})
+	const css = styles.join("\n")
+	for (const match of css.matchAll(/@import\s+(?:url\()?\s*["']?([^"')\s;]+)/gi)) extract.notes.push(`skipped @import: ${match[1]}`)
+	if (styles.length > 1) extract.notes.push(`${styles.length} <style> blocks concatenated`)
+	if (css.trim()) Object.assign(extract, cssFacts(css))
+	if (!extract.fonts.heading && extract.fonts.body) extract.notes.push("no heading font-family rule; headings inherit the body stack")
+
+	const body = root.children.find((child) => child.tag === "html")?.children.find((child) => child.tag === "body")
+		?? root.children.find((child) => child.tag === "body")
+		?? root
+	const text = proseText(body)
+	if (scripts && text.length < 200) {
+		extract.notes.push(`little static text with ${scripts} <script> tags: if the page renders client-side, render it with the browser skill and pass the saved HTML`)
+	}
+	const citation = citationFacts(text, hasFootnotes, hasLineageAttribute)
+	extract.citationStyle = citation.style
+	extract.lineageMode = citation.lineage
+	const h2 = headings.filter((node) => node.tag === "h2")
+	extract.headingNumbering = classifyNumbering(h2.map((node) => headingNumber(proseText(node)) ?? eyebrowNumber(node)))
+	extract.skeleton = headings
+		.filter((node) => node.tag === "h1" || node.tag === "h2")
+		.map((node) => ({ level: Number(node.tag[1]), text: proseText(node) }))
+	extract.figureContainers = rankByCount(containers)
+	return extract
+}
+
+/** Markdown carries structure only: skeleton, heading numbering, citation style. @returns {Extract} */
+export function extractFromMarkdown(source, { origin } = {}) {
+	rejectPdf(source, origin)
+	const extract = baseExtract(source, origin)
+	const prose = []
+	let fence = null
+	for (const line of String(source).split(/\r?\n/)) {
+		const fenceMatch = FENCE.exec(line)
+		if (fence) {
+			if (fenceMatch && fenceMatch[1][0] === fence[0] && fenceMatch[1].length >= fence.length && line.trim() === fenceMatch[1]) fence = null
+			continue
+		}
+		if (fenceMatch) {
+			fence = fenceMatch[1]
+			continue
+		}
+		prose.push(line.replace(/`[^`]*`/g, ""))
+		const heading = MD_HEADING.exec(line)
+		if (heading && heading[1].length <= 2) extract.skeleton.push({ level: heading[1].length, text: heading[2] })
+	}
+	const text = prose.join("\n")
+	extract.citationStyle = citationFacts(text, MD_FOOTNOTE.test(text), false).style
+	extract.headingNumbering = classifyNumbering(extract.skeleton.filter((entry) => entry.level === 2).map((entry) => headingNumber(entry.text)))
+	extract.notes.push("markdown source: tokens, fonts, colors, breakpoints, figures, and measure are not carried by Markdown")
+	return extract
+}
+
+/** Dispatch on the file extension: .html/.htm/.xhtml, .md/.markdown; anything else (PDF included) is unsupported. @returns {Extract} */
+export function extractFromPath(filePath) {
+	const extension = extname(filePath).toLowerCase()
+	if (HTML_EXTENSIONS.has(extension)) return extractFromHtml(readFileSync(filePath, "utf8"), { origin: filePath })
+	if (MD_EXTENSIONS.has(extension)) return extractFromMarkdown(readFileSync(filePath, "utf8"), { origin: filePath })
+	throw new UnsupportedFormatError(`extension ${extension || "(none)"}: ${filePath}`)
+}
+
+/** GET exactly `url` (plain fetch, 15 s timeout, no browser) and return the body text. */
+export async function fetchHtml(url) {
+	if (/\.pdf$/i.test(new URL(url).pathname)) throw new UnsupportedFormatError(`pdf url: ${url}`)
+	const response = await fetch(url, { signal: AbortSignal.timeout(FETCH_TIMEOUT_MS) })
+	const type = response.headers.get("content-type") ?? ""
+	if (!response.ok || /pdf/i.test(type)) {
+		await response.body?.cancel()
+		if (!response.ok) throw new Error(`fetch failed: HTTP ${response.status} for ${url}`)
+		throw new UnsupportedFormatError(`pdf response: ${url}`)
+	}
+	const text = await response.text()
+	if (text.trimStart().startsWith("%PDF-")) throw new UnsupportedFormatError(`pdf bytes: ${url}`)
+	return text
+}