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

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

@@ -0,0 +1,190 @@
+// gates-text.mjs - prose gates: G1 keep-all, G3 emoji, G4 dashes, G5/G6 headings, G7 unsourced numbers.
+//
+// Each gate is { codes, run(ctx) -> Defect[] } over the context built by gates-static.mjs.
+// Plain Node ESM, sibling imports only.
+
+import { proseText } from "./html-lite.mjs"
+
+const HANGUL = /[\uAC00-\uD7A3]/g
+const KEEP_ALL = /word-break\s*:\s*keep-all/i
+const MIN_HANGUL = 20
+/** Dingbats (2700-27BF) that render as emoji without FE0F (Emoji_Presentation=Yes). */
+const DINGBAT_EMOJI = new Set([0x2705, 0x270a, 0x270b, 0x2728, 0x274c, 0x274e, 0x2753, 0x2754, 0x2755, 0x2757, 0x2795, 0x2796, 0x2797, 0x27b0, 0x27bf])
+const EM_DASH = "\u2014"
+const EN_DASH = "\u2013"
+const HEADING_NUMBER = /^\s*(?:\d+(?:[.-]\d+)*[.):]?|[IVXLC]+[.)]|[A-Z][.)])\s+/
+const GENERIC = new Set(["overview", "introduction", "background", "summary", "conclusion", "conclusions", "results", "discussion", "methodology"])
+const MAX_HEADING_HANGUL = 26
+const MAX_HEADING_WORDS = 12
+const LINEAGE_TOKEN = /\b(?:MEASURED|ASSUMED|DERIVED)\b/
+const NUMBER_CONTAINERS = new Set(["figure", "table", "nav", "header", "footer"])
+const MAX_UNSOURCED = 50
+const MONTH = "(?:jan|feb|mar|apr|may|jun|jul|aug|sep|sept|oct|nov|dec)[a-z]*\\.?"
+/** Stripped before claim detection: dates, versions, figure refs, times, bare years. */
+const NOT_CLAIMS = [
+	/\b\d{4}[-./]\d{1,2}(?:[-./]\d{1,2})?\b/g,
+	/\b\d{2}\.\d{2}\.\d{2}\b/g,
+	/\d{2,4}\s*\uB144\s*\d{1,2}\s*\uC6D4(?:\s*\d{1,2}\s*\uC77C)?/g,
+	/\d{1,2}\s*\uC6D4(?:\s*\d{1,2}\s*\uC77C)?/g,
+	/(?:19|20)\d{2}\s*\uB144/g,
+	new RegExp(`\\b${MONTH}\\s+\\d{1,2}(?:st|nd|rd|th)?(?:,?\\s+\\d{4})?`, "gi"),
+	new RegExp(`\\b\\d{1,2}(?:st|nd|rd|th)?\\s+${MONTH}(?:\\s+\\d{4})?`, "gi"),
+	/\bv\d+(?:\.\d+)*\b|\b\d+(?:\.\d+){2,}\b/gi,
+	/\b(?:version|ver\.?|release)\s*\d+(?:\.\d+)*/gi,
+	/\b(?:fig(?:ure)?\.?|table|chart|section|step|phase|wave|part)\s*\d+(?:\.\d+)*/gi,
+	/(?:\uADF8\uB9BC|\uB3C4\uD45C|\uD45C|\uBC84\uC804)\s*\d+(?:\.\d+)*/g,
+	/\b\d{1,2}:\d{2}\b/g,
+	/\b(?:19|20)\d{2}s?\b/g,
+]
+const NUM = "(?<![\\p{L}\\d.,])\\d+(?:[.,]\\d+)*"
+const UNIT =
+	"(?:%|\\u2030|\\u00D7|\\uD37C\\uC13C\\uD2B8|\\uB2EC\\uB7EC|\\uC720\\uB85C|\\uC5D4|\\uC6D0|\\uB9CC|\\uC5B5|\\uC870|\\uCC9C|\\uCD08|\\uBD84|\\uC2DC\\uAC04|\\uC77C|\\uC8FC|\\uAC1C\\uC6D4|\\uB144|\\uBA85|\\uAC74|\\uAC1C|\\uD68C|\\uBC88|\\uBC30" +
+	"|(?:percent|pp|bps?|x|k|m|b|bn|million|billion|thousand|trillion|usd|krw|eur|ms|s|secs?|seconds?|mins?|minutes?|h|hrs?|hours?|days?|weeks?|months?|years?|kb|mb|gb|tb|users?|people|employees|engineers|tokens?|requests?|stars?|downloads?|customers?)(?![\\p{L}\\d]))"
+const CLAIMS = [
+	new RegExp(`[$\\u20AC\\u00A3\\u00A5\\u20A9]\\s?${NUM}`, "u"),
+	new RegExp(`${NUM}\\s?${UNIT}`, "iu"),
+	/(?<![\d.,])\d{1,3}(?:,\d{3})+(?![\d,])/,
+	/(?<![\d.])\d+\.\d+(?![\d.])/,
+]
+
+function snippet(text, index, span = 24) {
+	return text.slice(Math.max(0, index - span), index + span).trim()
+}
+
+/** Concatenated prose per owning element, in document order; `keep` filters text entries. */
+function ownerTexts(ctx, keep) {
+	const byOwner = new Map()
+	for (const entry of ctx.texts) {
+		if (!keep(entry)) continue
+		byOwner.set(entry.owner, (byOwner.get(entry.owner) ?? "") + entry.node.text)
+	}
+	return [...byOwner].map(([owner, text]) => ({ owner, text: text.replace(/\s+/g, " ") }))
+}
+
+function firstEmoji(text) {
+	const chars = [...text]
+	for (let i = 0; i < chars.length; i++) {
+		const cp = chars[i].codePointAt(0)
+		if (chars[i + 1] === "\uFE0F" && cp !== 0xfe0f) return chars[i] + chars[i + 1]
+		if ((cp >= 0x1f000 && cp <= 0x1faff) || (cp >= 0x2600 && cp <= 0x26ff) || DINGBAT_EMOJI.has(cp)) return chars[i]
+	}
+	return null
+}
+
+const isDigit = (ch) => ch >= "0" && ch <= "9"
+
+/** En dash allowed only when both neighbors (one optional space) are digits. */
+function badEnDash(text) {
+	for (let at = text.indexOf(EN_DASH); at !== -1; at = text.indexOf(EN_DASH, at + 1)) {
+		const before = text[at - 1] === " " ? text[at - 2] : text[at - 1]
+		const after = text[at + 1] === " " ? text[at + 2] : text[at + 1]
+		if (!isDigit(before ?? "") || !isDigit(after ?? "")) return at
+	}
+	return -1
+}
+
+const stripNumbering = (text) => text.replace(HEADING_NUMBER, "").trim()
+
+function headings(ctx, max) {
+	return ctx.elements.filter(({ node }) => /^h[1-6]$/.test(node.tag) && Number(node.tag[1]) <= max).map(({ node }) => node)
+}
+
+/** Citation markers need no stripping: an element carrying one is skipped before this runs. */
+function claimIn(text) {
+	let rest = text
+	for (const re of NOT_CLAIMS) rest = rest.replace(re, " ")
+	for (const re of CLAIMS) {
+		const match = re.exec(rest)
+		if (match) return match[0].trim()
+	}
+	return null
+}
+
+function numberCandidates(ctx) {
+	const { closest, classTokens, descendants } = ctx.util
+	return ctx.elements.filter(({ node, scope }) => {
+		if (node.tag !== "p" && node.tag !== "li") return false
+		if (node.tag === "li" && descendants(node).some((child) => child.tag === "p" || child.tag === "li")) return false
+		const kind = ctx.kindOf(scope)
+		if (kind === "closing" || kind === "sources") return false
+		return !closest(node, (cur) => NUMBER_CONTAINERS.has(cur.tag) || classTokens(cur).includes("fig"))
+	})
+}
+
+export const textGates = [
+	{
+		codes: ["korean_no_keep_all"],
+		run(ctx) {
+			const syllables = ctx.texts.reduce((sum, entry) => sum + (entry.node.text.match(HANGUL)?.length ?? 0), 0)
+			if (syllables < MIN_HANGUL) return []
+			if ([...ctx.styleBlocks, ...ctx.styleAttrs].some((entry) => KEEP_ALL.test(entry.css))) return []
+			return [ctx.defect("korean_no_keep_all", ctx.body, `${syllables} Hangul syllables in prose and no "word-break: keep-all" in any style block or style attribute`)]
+		},
+	},
+	{
+		codes: ["emoji_in_prose"],
+		run(ctx) {
+			const defects = []
+			for (const { owner, text } of ownerTexts(ctx, () => true)) {
+				const emoji = firstEmoji(text)
+				if (emoji) defects.push(ctx.defect("emoji_in_prose", owner, `emoji "${emoji}" in <${owner.tag}>: "${snippet(text, text.indexOf(emoji))}"`))
+			}
+			return defects
+		},
+	},
+	{
+		codes: ["em_dash_in_prose", "en_dash_in_prose"],
+		run(ctx) {
+			const defects = []
+			const inScope = (entry) => !entry.cell && ctx.kindOf(entry.scope) !== "sources"
+			for (const { owner, text } of ownerTexts(ctx, inScope)) {
+				const em = text.indexOf(EM_DASH)
+				if (em !== -1) defects.push(ctx.defect("em_dash_in_prose", owner, `em dash "${EM_DASH}" in <${owner.tag}>: "${snippet(text, em)}"`))
+				const en = badEnDash(text)
+				if (en !== -1) defects.push(ctx.defect("en_dash_in_prose", owner, `en dash "${EN_DASH}" outside a numeric range in <${owner.tag}>: "${snippet(text, en)}"`))
+			}
+			return defects
+		},
+	},
+	{
+		codes: ["heading_too_long"],
+		run(ctx) {
+			const defects = []
+			for (const node of headings(ctx, 4)) {
+				const text = stripNumbering(proseText(node))
+				const hangul = /[\uAC00-\uD7A3]/.test(text)
+				const size = hangul ? [...text].length : text.split(/\s+/).filter(Boolean).length
+				if (size > (hangul ? MAX_HEADING_HANGUL : MAX_HEADING_WORDS)) {
+					defects.push(ctx.defect("heading_too_long", node, `<${node.tag}> "${text}" is ${size} ${hangul ? "characters (> 26)" : "words (> 12)"}`))
+				}
+			}
+			return defects
+		},
+	},
+	{
+		codes: ["heading_generic"],
+		run(ctx) {
+			const defects = []
+			for (const node of headings(ctx, 3)) {
+				const text = stripNumbering(proseText(node))
+				if (GENERIC.has(text.toLowerCase().replace(/[.:]+$/, ""))) defects.push(ctx.defect("heading_generic", node, `generic <${node.tag}> "${text}"`))
+			}
+			return defects
+		},
+	},
+	{
+		codes: ["unsourced_number"],
+		run(ctx) {
+			const defects = []
+			for (const { node } of numberCandidates(ctx)) {
+				if (defects.length >= MAX_UNSOURCED) break
+				const text = proseText(node)
+				if (LINEAGE_TOKEN.test(text) || ctx.util.hasCitation(node)) continue
+				if (node.attrs["data-lineage"] !== undefined || ctx.util.descendants(node).some((child) => child.attrs["data-lineage"] !== undefined)) continue
+				const claim = claimIn(text)
+				if (claim) defects.push(ctx.defect("unsourced_number", node, `number "${claim}" without a lineage tag or citation: "${snippet(text, text.indexOf(claim))}"`))
+			}
+			return defects
+		},
+	},
+]

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

@@ -0,0 +1,239 @@
+// html-lite.mjs - a dependency-free HTML tokenizer for report checks.
+//
+// Single forward scan with indexOf (no whole-document regex, no backtracking). Produces a
+// LiteNode tree: { tag, attrs, children, parent, line, text, raw }. Text nodes carry
+// tag "#text"; <style>/<script>/<svg> keep their inner source in `raw` and have no parsed
+// children. The "#document" root also carries `ids` (id -> occurrence count) so
+// selectorPath can tell whether an id is unique. Plain Node ESM, no dependencies.
+
+import { decodeEntities } from "./entities.mjs"
+
+export { decodeEntities }
+
+const VOID_TAGS = new Set(["area", "base", "br", "col", "embed", "hr", "img", "input", "link", "meta", "param", "source", "track", "wbr"])
+const RAW_TAGS = new Set(["style", "script", "svg"])
+const HEADINGS = ["h1", "h2", "h3", "h4", "h5", "h6"]
+const P_CLOSERS = new Set([
+	"address", "article", "blockquote", "details", "div", "dl", "fieldset", "figcaption", "figure", "footer",
+	"form", "header", "hr", "main", "menu", "nav", "ol", "p", "pre", "section", "table", "ul", ...HEADINGS,
+])
+const P_SCOPE = new Set(["td", "th", "caption", "button", "body", "html"])
+const LI_SCOPE = new Set(["ul", "ol", "menu", "body", "html"])
+const PROSE_SKIP = new Set(["code", "pre", "kbd", "samp", "script", "style", "svg"])
+const SAFE_ID = /^[A-Za-z_][\w-]*$/
+
+export const PROSE_TAGS = new Set(["p", "li", ...HEADINGS, "td", "th", "figcaption", "blockquote", "dd", "dt"])
+const PROSE_BREAKS = new Set([...P_CLOSERS, ...PROSE_TAGS, "br", "tr"])
+
+function makeNode(tag, attrs, parent, line) {
+	return { tag, attrs, children: [], parent, line, text: "", raw: null }
+}
+
+function lineStarts(source) {
+	const starts = [0]
+	for (let at = source.indexOf("\n"); at !== -1; at = source.indexOf("\n", at + 1)) starts.push(at + 1)
+	return starts
+}
+
+function lineAt(starts, offset) {
+	let lo = 0
+	let hi = starts.length - 1
+	while (lo < hi) {
+		const mid = (lo + hi + 1) >> 1
+		if (starts[mid] <= offset) lo = mid
+		else hi = mid - 1
+	}
+	return lo + 1
+}
+
+const isSpace = (ch) => ch === " " || ch === "\n" || ch === "\t" || ch === "\r" || ch === "\f"
+const isLetter = (ch) => (ch >= "a" && ch <= "z") || (ch >= "A" && ch <= "Z")
+
+function skipSpaces(source, at) {
+	while (at < source.length && isSpace(source[at])) at++
+	return at
+}
+
+/** Read `<name attrs...>` starting just after `<`. */
+function readTag(source, start) {
+	const end = source.length
+	let at = start
+	while (at < end && !isSpace(source[at]) && source[at] !== ">" && source[at] !== "/") at++
+	const name = source.slice(start, at).toLowerCase()
+	const attrs = {}
+	let selfClosing = false
+	while (at < end) {
+		const ch = source[at]
+		if (isSpace(ch)) { at++; continue }
+		if (ch === ">") return { name, attrs, end: at + 1, selfClosing }
+		selfClosing = ch === "/" && source[at + 1] === ">"
+		if (ch === "/") { at++; continue }
+		let stop = at + 1
+		while (stop < end && !isSpace(source[stop]) && source[stop] !== "/" && source[stop] !== ">" && source[stop] !== "=") stop++
+		const key = source.slice(at, stop).toLowerCase()
+		at = skipSpaces(source, stop)
+		let value = ""
+		if (source[at] === "=") {
+			at = skipSpaces(source, at + 1)
+			const quote = source[at]
+			if (quote === '"' || quote === "'") {
+				const close = source.indexOf(quote, at + 1)
+				const valueEnd = close === -1 ? end : close
+				value = source.slice(at + 1, valueEnd)
+				at = valueEnd + 1
+			} else {
+				let valueEnd = at
+				while (valueEnd < end && !isSpace(source[valueEnd]) && source[valueEnd] !== ">") valueEnd++
+				value = source.slice(at, valueEnd)
+				at = valueEnd
+			}
+		}
+		if (key !== "__proto__" && !Object.hasOwn(attrs, key)) attrs[key] = decodeEntities(value)
+	}
+	return { name, attrs, end, selfClosing }
+}
+
+function tagNameAt(source, at, tag) {
+	if (source.slice(at, at + tag.length).toLowerCase() !== tag) return false
+	const next = source[at + tag.length] ?? ">"
+	return isSpace(next) || next === ">" || next === "/"
+}
+
+/** Find the close of a raw element; nested <svg> is depth-counted. Returns [innerEnd, resumeAt]. */
+function findRawClose(source, from, tag) {
+	let depth = 0
+	for (let lt = source.indexOf("<", from); lt !== -1; lt = source.indexOf("<", lt + 1)) {
+		if (source[lt + 1] === "/" && tagNameAt(source, lt + 2, tag)) {
+			if (depth === 0) {
+				const gt = source.indexOf(">", lt)
+				return [lt, gt === -1 ? source.length : gt + 1]
+			}
+			depth--
+		} else if (tag === "svg" && tagNameAt(source, lt + 1, tag)) {
+			depth++
+		}
+	}
+	return [source.length, source.length]
+}
+
+/** Parse an HTML string into a LiteNode tree rooted at "#document"; never throws on malformed input. */
+export function parseHtml(source) {
+	const src = String(source)
+	const starts = lineStarts(src)
+	const root = makeNode("#document", {}, null, 1)
+	root.ids = new Map()
+	const stack = [root]
+	let sawBody = false
+	let textStart = 0
+	let at = 0
+	const top = () => stack[stack.length - 1]
+	const flushText = (to) => {
+		if (to > textStart) {
+			const node = makeNode("#text", {}, top(), lineAt(starts, textStart))
+			node.text = decodeEntities(src.slice(textStart, to))
+			top().children.push(node)
+		}
+	}
+	const closeWithin = (tag, scope) => {
+		for (let k = stack.length - 1; k > 0; k--) {
+			if (stack[k].tag === tag) { stack.length = k; return }
+			if (scope.has(stack[k].tag)) return
+		}
+	}
+	while (at < src.length) {
+		const lt = src.indexOf("<", at)
+		if (lt === -1) break
+		const next = src[lt + 1] ?? ""
+		if (src.startsWith("<!--", lt)) {
+			flushText(lt)
+			const close = src.indexOf("-->", lt + 4)
+			at = textStart = close === -1 ? src.length : close + 3
+		} else if (next === "!" || next === "?") {
+			flushText(lt)
+			const close = src.indexOf(">", lt)
+			at = textStart = close === -1 ? src.length : close + 1
+		} else if (next === "/" && isLetter(src[lt + 2] ?? "")) {
+			flushText(lt)
+			const tag = readTag(src, lt + 2)
+			if (tag.name !== "body" && tag.name !== "html") {
+				for (let k = stack.length - 1; k > 0; k--) {
+					if (stack[k].tag === tag.name) { stack.length = k; break }
+				}
+			}
+			at = textStart = tag.end
+		} else if (isLetter(next)) {
+			flushText(lt)
+			const tag = readTag(src, lt + 1)
+			at = textStart = tag.end
+			if ((tag.name === "html" || tag.name === "body" || tag.name === "head") && stack.some((n) => n.tag === tag.name)) continue
+			if (P_CLOSERS.has(tag.name)) closeWithin("p", P_SCOPE)
+			if (tag.name === "li") closeWithin("li", LI_SCOPE)
+			const node = makeNode(tag.name, tag.attrs, top(), lineAt(starts, lt))
+			top().children.push(node)
+			if (tag.name === "body") sawBody = true
+			if (tag.attrs.id) root.ids.set(tag.attrs.id, (root.ids.get(tag.attrs.id) ?? 0) + 1)
+			if (RAW_TAGS.has(tag.name)) {
+				node.raw = ""
+				if (!tag.selfClosing) {
+					const [innerEnd, resume] = findRawClose(src, tag.end, tag.name)
+					node.raw = src.slice(tag.end, innerEnd)
+					at = textStart = resume
+				}
+			} else if (!VOID_TAGS.has(tag.name)) {
+				stack.push(node)
+			}
+		} else {
+			at = lt + 1
+		}
+	}
+	flushText(src.length)
+	if (!sawBody) wrapInBody(root)
+	return root
+}
+
+/** Fragments without <body>: move everything except <head> under a synthesized body. */
+function wrapInBody(root) {
+	const host = root.children.find((child) => child.tag === "html") ?? root
+	const kept = host.children.filter((child) => child.tag === "head")
+	const moved = host.children.filter((child) => child.tag !== "head")
+	const body = makeNode("body", {}, host, moved[0]?.line ?? 1)
+	for (const child of moved) child.parent = body
+	body.children = moved
+	host.children = [...kept, body]
+}
+
+function segment(node) {
+	const same = node.parent.children.filter((child) => child.tag === node.tag)
+	return same.length > 1 ? `${node.tag}:nth-of-type(${same.indexOf(node) + 1})` : node.tag
+}
+
+/** CSS selector for a node: nearest unique-id anchor, else `body > tag:nth-of-type(k) > ...`. */
+export function selectorPath(node) {
+	let root = node
+	while (root.parent) root = root.parent
+	const ids = root.ids ?? new Map()
+	const segments = []
+	for (let cur = node.tag === "#text" ? node.parent : node; cur?.parent; cur = cur.parent) {
+		const id = cur.attrs.id
+		if (id && SAFE_ID.test(id) && ids.get(id) === 1) return [`#${id}`, ...segments].join(" > ")
+		if (cur.tag === "body") return ["body", ...segments].join(" > ")
+		if (cur.tag === "html") break
+		segments.unshift(segment(cur))
+	}
+	return segments.join(" > ")
+}
+
+/** Descendant text minus code/pre/kbd/samp/script/style/svg, whitespace collapsed. */
+export function proseText(node) {
+	const parts = []
+	const walk = (cur) => {
+		if (cur.tag === "#text") { parts.push(cur.text); return }
+		if (PROSE_SKIP.has(cur.tag)) return
+		const breaks = PROSE_BREAKS.has(cur.tag)
+		if (breaks) parts.push(" ")
+		for (const child of cur.children) walk(child)
+		if (breaks) parts.push(" ")
+	}
+	for (const child of node.children) walk(child)
+	return parts.join("").replace(/\s+/g, " ").trim()
+}

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

@@ -0,0 +1,83 @@
+// layout-probe.mjs - the in-page probe the layout gates consume.
+//
+// buildProbeSource() returns ONE async IIFE expression (the engine's evaluate awaits it): it first
+// forces lazy images to load eagerly and awaits their decode (bounded), so natural sizes are real. The orchestrator evaluates it in the rendered
+// report through the browser skill's owned headless engine and writes the result to boxes.json;
+// gates-layout.mjs reads that file. This module never launches a browser.
+//
+// Selector paths use the SAME rule as html-lite.selectorPath (a unique safe #id anchors the path,
+// otherwise tag segments with :nth-of-type(k) only when the parent has more than one child of that
+// tag, prefixed "body"), so a layout defect can be mapped back to the static tree. The rule lives in
+// PATH_FUNCTION_SOURCE, embedded verbatim in the probe and exercised by the parity test.
+
+export const PROBE_VERSION = 1
+export const SAMPLED_SELECTOR = "figure, .fig, img, svg, table, h1, h2, h3, p, pre"
+export const EXCLUDED_ANCESTORS = Object.freeze(["svg", "nav", "header", "footer"])
+
+/** In-page path rule. idCount(id) returns how many elements carry that id. */
+export const PATH_FUNCTION_SOURCE = `function ulwSelectorPath(el, idCount) {
+	const SAFE_ID = /^[A-Za-z_][\\w-]*$/;
+	const segments = [];
+	for (let cur = el; cur && cur.parentElement; cur = cur.parentElement) {
+		const tag = cur.tagName.toLowerCase();
+		const id = cur.id;
+		if (id && SAFE_ID.test(id) && idCount(id) === 1) return ["#" + id].concat(segments).join(" > ");
+		if (tag === "body") return ["body"].concat(segments).join(" > ");
+		if (tag === "html") break;
+		const same = Array.from(cur.parentElement.children).filter((c) => c.tagName === cur.tagName);
+		segments.unshift(same.length > 1 ? tag + ":nth-of-type(" + (same.indexOf(cur) + 1) + ")" : tag);
+	}
+	return segments.join(" > ");
+}`
+
+export class ProbeError extends Error {
+	constructor(code, message) {
+		super(message)
+		this.name = "ProbeError"
+		this.code = code
+	}
+}
+
+/** @param {{ cap?: number, root?: string }} [options] */
+export function buildProbeSource({ cap = 400, root = "main, article" } = {}) {
+	if (!Number.isInteger(cap) || cap <= 0) throw new ProbeError("invalid_cap", `--cap requires a positive integer, got ${cap}`)
+	if (typeof root !== "string" || root.trim().length === 0) throw new ProbeError("invalid_root", "--root requires a non-empty CSS selector")
+	return `(async () => {
+	const CAP = ${cap};
+	const IMAGE_WAIT_MS = 5000;
+	const ROOT = document.querySelector(${JSON.stringify(root)}) || document.body;
+	const EXCLUDED = ${JSON.stringify(EXCLUDED_ANCESTORS)};
+	const images = Array.from(document.images);
+	for (const img of images) if (img.loading === "lazy") img.loading = "eager";
+	const settle = Promise.all(images.map((img) => (img.complete && img.naturalWidth > 0 ? null : img.decode().catch(() => null))));
+	await Promise.race([settle, new Promise((resolve) => setTimeout(resolve, IMAGE_WAIT_MS))]);
+	${PATH_FUNCTION_SOURCE}
+	const idCount = (id) => document.querySelectorAll("#" + CSS.escape(id)).length;
+	const path = (el) => ulwSelectorPath(el, idCount);
+	const r1 = (n) => Math.round(n * 10) / 10;
+	const rect = (el) => { const b = el.getBoundingClientRect(); return { x: r1(b.left + scrollX), y: r1(b.top + scrollY), w: r1(b.width), h: r1(b.height) }; };
+	const excluded = (el) => { for (let p = el.parentElement; p; p = p.parentElement) if (EXCLUDED.includes(p.tagName.toLowerCase())) return true; return false; };
+	const all = Array.from(ROOT.querySelectorAll(${JSON.stringify(SAMPLED_SELECTOR)})).filter((el) => {
+		if (excluded(el)) return false;
+		const b = el.getBoundingClientRect();
+		return b.width > 0 || b.height > 0;
+	});
+	const boxes = all.slice(0, CAP).map((el) => {
+		const cs = getComputedStyle(el);
+		const parent = el.parentElement;
+		const box = {
+			selector: path(el), tag: el.tagName.toLowerCase(), rect: rect(el),
+			parentSelector: parent ? path(parent) : null, parentRect: parent ? rect(parent) : null,
+			parentOverflow: parent ? { x: getComputedStyle(parent).overflowX, y: getComputedStyle(parent).overflowY } : null,
+			scroll: { sw: el.scrollWidth, cw: el.clientWidth, sh: el.scrollHeight, ch: el.clientHeight },
+			overflow: { x: cs.overflowX, y: cs.overflowY }, position: cs.position,
+			textLen: (el.textContent || "").trim().length,
+		};
+		if (el.tagName === "IMG") { box.objectFit = cs.objectFit; box.natural = { w: el.naturalWidth, h: el.naturalHeight }; }
+		return box;
+	});
+	return { probe_version: ${PROBE_VERSION}, viewport: { w: innerWidth, h: innerHeight },
+		document: { w: document.documentElement.scrollWidth, h: document.documentElement.scrollHeight },
+		count: all.length, truncated: all.length > CAP, boxes };
+})()`
+}