diff --git a/src/ask/concepts.ts b/src/ask/concepts.ts
index 0b6b4c64..f99c7d6b 100644
--- a/src/ask/concepts.ts
+++ b/src/ask/concepts.ts
@@ -254,8 +254,8 @@ export const CONCEPTS: readonly Concept[] = [
     aliases: ['compare output', 'compare outputs', 'diff output', 'compare runs', 'what changed between runs', 'compare with previous run'],
     description: '/compare shows what changed between two command outputs (by default the latest and the previous run of the same command; /compare N M picks them by /copy number). Also Actions → Compare output…. Added, removed and changed lines; c copies the diff, r a comparison report.'},
   {id: 'view', label: 'View a file or output', support: 'actionable', open: '/view', covers: ['/view'],
-    aliases: ['view file', 'view a file', 'read a file', 'open a file read only', 'view output', 'read markdown', 'preview markdown', 'view log', 'pager'],
-    description: '/view opens a read-only native viewer: /view README.md for a file (Markdown renders; m shows the source), /view N for completed output N (numbered as /copy), /view alone for the latest output. Also Actions → View output. Find with /, n/N, go to a line with :, w wraps, l numbers, c copies the whole text; for a file, e edits it in /edit and o opens it in your own editor at the line shown (as /open does). less, bat and Glow still work as usual.'},
+    aliases: ['view file', 'view a file', 'read a file', 'open a file read only', 'view output', 'read markdown', 'preview markdown', 'view html', 'preview html', 'view log', 'pager'],
+    description: '/view opens a read-only native viewer: /view README.md for a file (Markdown and HTML render natively, embedded HTML included: never run or loaded, images as their text; m shows the source), /view N for completed output N (numbered as /copy), /view alone for the latest output. Also Actions → View output. Find with /, n/N, go to a line with :, w wraps, l numbers, c copies the whole text; for a file, e edits it in /edit and o opens it in your own editor at the line shown (as /open does). less, bat and Glow still work as usual.'},
   {id: 'edit-file', label: 'Edit a small text file', support: 'actionable', open: '/edit', covers: ['/edit'],
     aliases: ['edit a file', 'edit file', 'edit text file', 'change a config file', 'quick edit', 'small editor', 'native editor', 'edit notes'],
     description: '/edit notes.md opens a small text file (up to 1 MiB, your own, UTF-8) in NMSh\'s editor: type, select with Shift, Ctrl+Z undo, Ctrl+Y redo, Ctrl+F find, Ctrl+C/Ctrl+X copy and cut, Ctrl+S saves (atomically; if the file changed on disk you choose Overwrite, Reload or Keep editing), Esc closes and asks about unsaved changes. e in /view edits the file shown. vim, nano, micro and your $EDITOR still work as usual.'},
diff --git a/src/markdown/blocks.ts b/src/markdown/blocks.ts
index 6ed72200..752c5624 100644
--- a/src/markdown/blocks.ts
+++ b/src/markdown/blocks.ts
@@ -4,7 +4,13 @@
  * replies and GitHub text, not a conformance implementation. Streaming-safe: an unclosed fence runs to the end and
  * is marked open, so a half-arrived reply still reads as code. Bounded in depth and lines; tabs are expected to be
  * expanded by the caller's display scrubber.
+ *
+ * With `html` (`/view`, QA F01), an HTML block (CommonMark's kinds: a raw block such as `<pre>`, a comment, a block-level
+ * tag, or one complete known tag alone on its line) runs to its end and becomes blocks through the HTML converter;
+ * Markdown between HTML blocks reads as Markdown, as on GitHub. Without it, HTML is paragraph text.
  */
+import {htmlToBlocks} from './html.js';
+import {BLOCK, KNOWN} from './markup.js';
 
 export type Alignment = 'left' | 'center' | 'right' | undefined;
 
@@ -133,13 +139,39 @@ function interrupts(line: string, next: string | undefined): boolean {
   return Boolean(line.includes('|') && next !== undefined && delimiterRow(next) && splitRow(line).length === delimiterRow(next)!.length);
 }
 
-export function parseBlocks(source: string): Block[] {
+export function parseBlocks(source: string, options: {html?: boolean} = {}): Block[] {
   const lines = source.replace(/\r\n?/gu, '\n').split('\n');
   if (lines.length > MAX_LINES) lines.length = MAX_LINES;
-  return parseLines(lines, 0);
+  return parseLines(lines, 0, Boolean(options.html));
 }
 
-function parseLines(lines: readonly string[], depth: number): Block[] {
+/** The kind of HTML block a line starts (CommonMark 1, 2, 6, 7), or undefined. */
+function htmlBlockKind(line: string): 1 | 2 | 6 | 7 | undefined {
+  if (indentOf(line) > 3) return undefined;
+  const text = line.trimStart();
+  if (text[0] !== '<') return undefined;
+  if (/^<(?:pre|script|style|textarea)(?:[\s>]|$)/iu.test(text)) return 1;
+  if (text.startsWith('<!--')) return 2;
+  const name = /^<\/?([A-Za-z][A-Za-z0-9-]{0,31})(?=[\s/>]|$)/u.exec(text)?.[1]?.toLowerCase();
+  if (!name) return undefined;
+  if (BLOCK.has(name)) return 6;
+  // One complete tag alone on its line (`<picture>`, `<img …>`): only names HTML has, so `<T>` stays text.
+  return KNOWN.has(name) && text.length <= 4096 && /^<[^<>]*>\s*$/u.test(text) ? 7 : undefined;
+}
+
+/** Index just past an HTML block starting at `start`: kinds 1 and 2 to their end marker, the others to a blank line. */
+function htmlBlockEnd(lines: readonly string[], start: number, kind: 1 | 2 | 6 | 7): number {
+  const closes = kind === 1 ? /<\/(?:pre|script|style|textarea)>/iu : kind === 2 ? /-->/u : undefined;
+  let index = start;
+  if (closes) {
+    for (; index < lines.length; index++) if (closes.test(lines[index]!) && (index > start || closes.test(lines[start]!.slice(4)))) return index + 1;
+    return index;
+  }
+  while (index < lines.length && !blank(lines[index]!)) index++;
+  return index;
+}
+
+function parseLines(lines: readonly string[], depth: number, html = false): Block[] {
   const blocks: Block[] = [];
   if (depth > MAX_DEPTH) {
     const text = lines.join('\n').trim();
@@ -165,6 +197,14 @@ function parseLines(lines: readonly string[], depth: number): Block[] {
       continue;
     }
 
+    const markup = html ? htmlBlockKind(line) : undefined;
+    if (markup) {
+      const end = htmlBlockEnd(lines, index, markup);
+      blocks.push(...htmlToBlocks(lines.slice(index, end).join('\n')));
+      index = end;
+      continue;
+    }
+
     const heading = headingOf(line);
     if (heading) { blocks.push({kind: 'heading', level: heading.level, text: heading.text}); index++; continue; }
 
@@ -181,13 +221,13 @@ function parseLines(lines: readonly string[], depth: number): Block[] {
         else break;
         index++;
       }
-      blocks.push({kind: 'quote', blocks: parseLines(inner, depth + 1)});
+      blocks.push({kind: 'quote', blocks: parseLines(inner, depth + 1, html)});
       continue;
     }
 
     const item = ITEM.exec(line);
     if (item) {
-      const parsed = parseList(lines, index, depth);
+      const parsed = parseList(lines, index, depth, html);
       blocks.push(parsed.block);
       index = parsed.next;
       continue;
@@ -238,7 +278,7 @@ function parseLines(lines: readonly string[], depth: number): Block[] {
   return blocks;
 }
 
-function parseList(lines: readonly string[], start: number, depth: number): {block: Extract<Block, {kind: 'list'}>; next: number} {
+function parseList(lines: readonly string[], start: number, depth: number, html: boolean): {block: Extract<Block, {kind: 'list'}>; next: number} {
   const first = ITEM.exec(lines[start]!)!;
   const ordered = /\d/u.test(first[2]!);
   const delimiter = ordered ? first[2]!.slice(-1) : first[2]!;
@@ -272,7 +312,7 @@ function parseList(lines: readonly string[], start: number, depth: number): {blo
     let task: ListItem['task'];
     const taskMatch = /^\[([ xX])\][ \t]+/u.exec(body[0] ?? '');
     if (taskMatch) { task = taskMatch[1] === ' ' ? 'open' : 'done'; body[0] = body[0]!.slice(taskMatch[0].length); }
-    items.push({...(ordered ? {number: Number(match[2]!.slice(0, -1))} : {}), ...(task ? {task} : {}), blocks: parseLines(body, depth + 1)});
+    items.push({...(ordered ? {number: Number(match[2]!.slice(0, -1))} : {}), ...(task ? {task} : {}), blocks: parseLines(body, depth + 1, html)});
   }
   return {block: {kind: 'list', ordered, start: items[0]?.number ?? 1, loose, items}, next: index};
 }
diff --git a/src/markdown/html.ts b/src/markdown/html.ts
new file mode 100644
index 00000000..12f89b76
--- /dev/null
+++ b/src/markdown/html.ts
@@ -0,0 +1,257 @@
+import type {Alignment, Block, ListItem} from './blocks.js';
+import {BLOCK, decodeEntities, DROPPED, imageLabel, RAW_TEXT, readTag, VOID} from './markup.js';
+
+/**
+ * HTML as the Markdown renderer's own blocks (QA F01): what `/view` shows for an HTML file and for HTML inside
+ * Markdown. A supported subset, mapped onto blocks the renderer already draws safely at any width:
+ *   headings, paragraphs and containers (div, section, center, details…), ordered and unordered lists (nested, task
+ *   check boxes), tables (header row, alignment), preformatted code (with its language class), block quotes, rules,
+ *   definition lists; inline emphasis, code, keys, strikethrough, links and images.
+ * Inline content is handed on as normalised inline HTML (escaped text and the few tags the inline parser reads), so
+ * links follow the renderer's http(s)-only rule and an image is its alt text, never fetched or read.
+ *
+ * Nothing is executed or loaded: scripts, styles, templates, embedded and interactive content and the document head
+ * are dropped with their contents, attributes other than a few (href, alt, src for a label, align, start, type,
+ * checked, class for a code language) are ignored, CSS is never applied. The scan is linear; at most MAX_TOKENS
+ * tokens are read and MAX_DEPTH elements nest (deeper content stays, flattened), so hostile markup stays bounded.
+ */
+
+const MAX_TOKENS = 200_000;
+const MAX_DEPTH = 48;
+const MAX_COLUMNS = 32;
+
+type Token = {kind: 'text'; text: string} | {kind: 'open'; name: string; attributes: Record<string, string>; self: boolean} | {kind: 'close'; name: string};
+interface Element {name: string; attributes: Record<string, string>; children: Node[]}
+type Node = string | Element;
+
+function tokenize(source: string): Token[] {
+  const tokens: Token[] = [];
+  const lower = source.toLowerCase();
+  let index = 0;
+  while (index < source.length) {
+    if (tokens.length >= MAX_TOKENS) {
+      tokens.push({kind: 'text', text: ' … (the rest is too long to render here; m shows the source)'});
+      break;
+    }
+    const open = source.indexOf('<', index);
+    if (open < 0) { tokens.push({kind: 'text', text: source.slice(index)}); break; }
+    if (open > index) tokens.push({kind: 'text', text: source.slice(index, open)});
+    if (source.startsWith('<!--', open)) {
+      const end = source.indexOf('-->', open + 4);
+      index = end < 0 ? source.length : end + 3;
+      continue;
+    }
+    if (source[open + 1] === '!' || source[open + 1] === '?') {
+      // A doctype, CDATA section or processing instruction: never content.
+      const end = source.indexOf('>', open);
+      index = end < 0 ? source.length : end + 1;
+      continue;
+    }
+    const tag = readTag(source, open);
+    if (!tag) { tokens.push({kind: 'text', text: '<'}); index = open + 1; continue; }
+    index = tag.end;
+    if (tag.closing) { tokens.push({kind: 'close', name: tag.name}); continue; }
+    tokens.push({kind: 'open', name: tag.name, attributes: tag.attributes, self: tag.selfClosing});
+    if (RAW_TEXT.has(tag.name) && !tag.selfClosing) {
+      // Its content is never markup: skip to its end tag (or the end), unread.
+      const close = lower.indexOf(`</${tag.name}`, index);
+      const end = close < 0 ? -1 : source.indexOf('>', close);
+      index = end < 0 ? source.length : end + 1;
+      tokens.push({kind: 'close', name: tag.name});
+    }
+  }
+  return tokens;
+}
+
+/** Elements an opening tag of the same family closes (an unclosed `<li>`, `<p>`, `<tr>`, a cell). */
+const IMPLIED: Record<string, readonly string[]> = {li: ['li'], p: ['p'], tr: ['tr', 'td', 'th'], td: ['td', 'th'], th: ['td', 'th'], dt: ['dt', 'dd'], dd: ['dt', 'dd']};
+/** Where an implied close stops looking: it never reaches past its own list, table or row. */
+const SCOPE = new Set(['ul', 'ol', 'table', 'tbody', 'thead', 'tfoot', 'tr', 'dl', 'blockquote', 'div', 'body', 'html']);
+
+function build(tokens: readonly Token[]): Element {
+  const root: Element = {name: '#root', attributes: {}, children: []};
+  const stack: Element[] = [root];
+  for (const token of tokens) {
+    const top = stack.at(-1)!;
+    if (token.kind === 'text') { top.children.push(token.text); continue; }
+    if (token.kind === 'close') {
+      // The nearest open element of that name closes, with everything inside it; a stray end tag is ignored.
+      for (let depth = stack.length - 1; depth > 0; depth--) if (stack[depth]!.name === token.name) { stack.length = depth; break; }
+      continue;
+    }
+    const implied = IMPLIED[token.name];
+    if (implied) {
+      for (let depth = stack.length - 1; depth > 0; depth--) {
+        const name = stack[depth]!.name;
+        if (implied.includes(name)) { stack.length = depth; break; }
+        if (SCOPE.has(name)) break;
+      }
+    }
+    const element: Element = {name: token.name, attributes: token.attributes, children: []};
+    stack.at(-1)!.children.push(element);
+    // Past the depth bound an element keeps its place but not its own level: what follows is its sibling.
+    if (!VOID.has(token.name) && !token.self && stack.length < MAX_DEPTH) stack.push(element);
+  }
+  return root;
+}
+
+// ---- Inline -------------------------------------------------------------------------------------------------------
+
+/** Decoded text as the inline parser reads it literally: Markdown punctuation escaped, `&` and `<` as references. */
+function escapeText(text: string): string {
+  return text.replace(/&/gu, '&amp;').replace(/[\\`*_~[\]!]/gu, char => `\\${char}`).replace(/</gu, '&lt;');
+}
+const escapeAttribute = (value: string) => value.replace(/&/gu, '&amp;').replace(/"/gu, '&quot;').replace(/</gu, '&lt;').replace(/>/gu, '&gt;');
+
+/** All text inside, as written (for code and preformatted text). */
+function textContent(nodes: readonly Node[]): string {
+  let out = '';
+  const walk = (list: readonly Node[]) => {
+    for (const node of list) {
+      if (typeof node === 'string') out += node;
+      else if (node.name === 'br') out += '\n';
+      else if (!DROPPED.has(node.name)) walk(node.children);
+    }
+  };
+  walk(nodes);
+  return out;
+}
+
+const STRONG = new Set(['b', 'strong']);
+const EMPHASIS = new Set(['i', 'em', 'cite', 'dfn', 'var']);
+const STRIKE = new Set(['s', 'del', 'strike']);
+const CODE = new Set(['code', 'kbd', 'samp', 'tt']);
+
+/** Inline content as normalised inline HTML: whitespace collapsed, text escaped, only tags the inline parser reads. */
+function inlineText(nodes: readonly Node[]): string {
+  let out = '';
+  for (const node of nodes) {
+    if (typeof node === 'string') { out += escapeText(decodeEntities(node).replace(/\s+/gu, ' ')); continue; }
+    const name = node.name;
+    if (DROPPED.has(name)) continue;
+    if (name === 'br') out += ' ';
+    else if (name === 'img') out += `<img alt="${escapeAttribute(imageLabel(node.attributes))}">`;
+    else if (name === 'input') out += node.attributes.type?.toLowerCase() === 'checkbox' ? ('checked' in node.attributes ? '☑ ' : '☐ ') : '';
+    else if (CODE.has(name)) out += `<code>${escapeAttribute(decodeEntities(textContent(node.children)).replace(/\s+/gu, ' '))}</code>`;
+    else if (STRONG.has(name)) out += `<b>${inlineText(node.children)}</b>`;
+    else if (EMPHASIS.has(name)) out += `<i>${inlineText(node.children)}</i>`;
+    else if (STRIKE.has(name)) out += `<s>${inlineText(node.children)}</s>`;
+    else if (name === 'a' && node.attributes.href) out += `<a href="${escapeAttribute(decodeEntities(node.attributes.href).trim())}">${inlineText(node.children)}</a>`;
+    else if (BLOCK.has(name)) out += ` ${inlineText(node.children)} `;
+    else out += inlineText(node.children);
+  }
+  return out;
+}
+
+const clean = (text: string) => text.replace(/ {2,}/gu, ' ').trim();
+
+// ---- Blocks -------------------------------------------------------------------------------------------------------
+
+function blocksOf(nodes: readonly Node[]): Block[] {
+  const out: Block[] = [];
+  let inline: Node[] = [];
+  const flush = () => {
+    const text = clean(inlineText(inline));
+    if (text) out.push({kind: 'paragraph', text});
+    inline = [];
+  };
+  for (const node of nodes) {
+    if (typeof node !== 'string' && DROPPED.has(node.name)) continue;
+    if (typeof node !== 'string' && BLOCK.has(node.name)) { flush(); out.push(...blockOf(node)); }
+    else inline.push(node);
+  }
+  flush();
+  return out;
+}
+
+function blockOf(element: Element): Block[] {
+  const name = element.name;
+  const heading = /^h([1-6])$/u.exec(name);
+  if (heading) {
+    const text = clean(inlineText(element.children));
+    return text ? [{kind: 'heading', level: Number(heading[1]), text}] : [];
+  }
+  switch (name) {
+    case 'hr': return [{kind: 'rule'}];
+    case 'pre': return [preformatted(element)];
+    case 'blockquote': return [{kind: 'quote', blocks: blocksOf(element.children)}];
+    case 'ul': case 'ol': case 'menu': case 'dir': return [list(element)];
+    case 'table': return table(element);
+    case 'summary': case 'dt': case 'caption': case 'legend': {
+      const text = clean(inlineText(element.children));
+      return text ? [{kind: 'paragraph', text: `<b>${text}</b>`}] : [];
+    }
+    default: return blocksOf(element.children);
+  }
+}
+
+function preformatted(element: Element): Block {
+  const code = element.children.find((node): node is Element => typeof node !== 'string' && node.name === 'code');
+  const classes = `${code?.attributes.class ?? ''} ${element.attributes.class ?? ''}`;
+  const language = /(?:^|\s)(?:language|lang)-([\w+#.-]{1,40})/u.exec(classes)?.[1]?.toLowerCase();
+  const lines = decodeEntities(textContent(element.children)).replace(/\r\n?/gu, '\n').replace(/^\n/u, '').split('\n');
+  while (lines.length > 1 && !lines.at(-1)!.trim()) lines.pop();
+  return {kind: 'code', ...(language ? {language} : {}), lines, closed: true};
+}
+
+function list(element: Element): Block {
+  const ordered = element.name === 'ol';
+  const start = Math.min(1_000_000_000, Math.max(0, Number.parseInt(element.attributes.start ?? '1', 10) || 1));
+  const items: ListItem[] = [];
+  for (const node of element.children) {
+    if (typeof node === 'string') {
+      if (node.trim()) items.push({blocks: [{kind: 'paragraph', text: clean(escapeText(decodeEntities(node).replace(/\s+/gu, ' ')))}]});
+      continue;
+    }
+    if (node.name === 'li') items.push(item(node));
+    // A list written straight inside a list belongs to the item before it.
+    else if ((node.name === 'ul' || node.name === 'ol') && items.length) items.at(-1)!.blocks.push(list(node));
+    else if (!DROPPED.has(node.name)) items.push({blocks: blocksOf([node])});
+  }
+  return {kind: 'list', ordered, start, loose: false, items: items.map((entry, index) => ordered ? {...entry, number: start + index} : entry)};
+}
+
+function item(element: Element): ListItem {
+  const children = [...element.children];
+  // A check box first in the item makes it a task: shown as the renderer's task marker, never as a control.
+  const first = children.findIndex(node => typeof node !== 'string' || node.trim());
+  const box = children[first];
+  let task: ListItem['task'];
+  if (box && typeof box !== 'string' && box.name === 'input' && box.attributes.type?.toLowerCase() === 'checkbox') {
+    task = 'checked' in box.attributes ? 'done' : 'open';
+    children.splice(first, 1);
+  }
+  return {...(task ? {task} : {}), blocks: blocksOf(children)};
+}
+
+function table(element: Element): Block[] {
+  const rows: Array<Array<{text: string; header: boolean; align?: Alignment}>> = [];
+  const before: Block[] = [];
+  const collect = (nodes: readonly Node[]) => {
+    for (const node of nodes) {
+      if (typeof node === 'string') continue;
+      if (node.name === 'caption') before.push(...blockOf(node));
+      else if (node.name === 'thead' || node.name === 'tbody' || node.name === 'tfoot') collect(node.children);
+      else if (node.name === 'tr') {
+        const cells = node.children.filter((cell): cell is Element => typeof cell !== 'string' && (cell.name === 'td' || cell.name === 'th')).slice(0, MAX_COLUMNS);
+        if (cells.length) rows.push(cells.map(cell => ({text: clean(inlineText(cell.children)), header: cell.name === 'th', align: alignment(cell.attributes.align)})));
+      }
+    }
+  };
+  collect(element.children);
+  if (!rows.length) return before;
+  const width = Math.max(...rows.map(row => row.length));
+  const pad = (row: ReadonlyArray<{text: string}>) => Array.from({length: width}, (_, column) => row[column]?.text ?? '');
+  const [header, ...body] = rows;
+  return [...before, {kind: 'table', align: Array.from({length: width}, (_, column) => header![column]?.align), header: pad(header!), rows: body.map(pad)}];
+}
+
+function alignment(value: string | undefined): Alignment {
+  const align = value?.trim().toLowerCase();
+  return align === 'left' || align === 'center' || align === 'right' ? align : undefined;
+}
+
+/** HTML (a document, or a block inside Markdown) as blocks; its inline text is for the inline parser in HTML mode. */
+export function htmlToBlocks(source: string): Block[] {
+  return blocksOf(build(tokenize(source)).children);
+}
diff --git a/src/markdown/inline.ts b/src/markdown/inline.ts
index 1a1c1407..9358f9b5 100644
--- a/src/markdown/inline.ts
+++ b/src/markdown/inline.ts
@@ -1,4 +1,5 @@
 import {safeHyperlinkTarget} from '../output/Hyperlinks.js';
+import {decodeEntities, decodeEntity, DROPPED, imageLabel, KNOWN, RAW_TEXT, readTag} from './markup.js';
 
 /**
  * Inline Markdown for untrusted text (agent replies, GitHub bodies): code spans, emphasis, strong, strikethrough,
@@ -6,6 +7,10 @@ import {safeHyperlinkTarget} from '../output/Hyperlinks.js';
  * module only decides styling. A link target is kept only when it is a safe http(s) URL, so a hostile
  * `[docs](javascript:...)` or `[x](file:///...)` renders as plain text. Parsing is linear and bounded: a span's
  * delimiters never nest more than a few levels and unmatched delimiters stay literal text.
+ *
+ * With `html` (what `/view` shows, QA F01) a few HTML tags are read too: b/strong, i/em, s/del, code/kbd, a (the same
+ * http(s)-only rule), img (its alt text), br, and script-like elements dropped with their content; any other known
+ * element's tags are dropped and its text kept, and an unknown `<word>` stays text. Without it, tags are text.
  */
 
 export interface InlineStyle {
@@ -38,28 +43,22 @@ function safeLink(url: string): string | undefined {
   return target && /^https?:/u.test(target) ? target : undefined;
 }
 
-const INVISIBLE = /[\p{Cc}\p{Cf}\p{Co}\p{Zl}\p{Zp}]/u;
-
-const ENTITIES: Record<string, string> = {amp: '&', lt: '<', gt: '>', quot: '"', apos: "'", nbsp: ' ', copy: '©', reg: '®', mdash: '—', ndash: '–', hellip: '…', rarr: '→', larr: '←', times: '×'};
-
-function decodeEntity(source: string, index: number): {text: string; length: number} | undefined {
-  const match = /^&(?:#(\d{1,7})|#[xX]([0-9a-fA-F]{1,6})|([a-zA-Z]{2,8}));/u.exec(source.slice(index, index + 12));
-  if (!match) return undefined;
-  if (match[3]) { const named = ENTITIES[match[3]]; return named ? {text: named, length: match[0].length} : undefined; }
-  const code = match[1] ? Number(match[1]) : parseInt(match[2]!, 16);
-  if (!code || code > 0x10ffff || (code >= 0xd800 && code <= 0xdfff)) return undefined;
-  const text = String.fromCodePoint(code);
-  // An entity never smuggles in what the display scrubber removes: controls, bidi and other format characters
-  // (direction overrides, zero-width joiners), separators or private-use code points.
-  if (INVISIBLE.test(text)) return undefined;
-  return {text, length: match[0].length};
+export interface InlineOptions {
+  /** Read the HTML tags `/view` renders (see the module comment); otherwise tags are text. */
+  html?: boolean;
 }
 
 /** Raw tokens before emphasis resolution: literal text, finished spans (code, links) and delimiter runs. */
 type Token =
   | {kind: 'text'; text: string}
   | {kind: 'span'; span: InlineSpan[]}
-  | {kind: 'delim'; char: '*' | '_' | '~'; count: number; open: boolean; close: boolean};
+  | {kind: 'delim'; char: '*' | '_' | '~'; count: number; open: boolean; close: boolean}
+  /** An HTML start or end tag that styles what is between (html mode); `target` is a link's refused URL, shown as text. */
+  | {kind: 'tag'; style: 'strong' | 'em' | 'strike' | 'link'; open: boolean; link?: string; target?: string};
+
+const TAG_STYLES: Record<string, 'strong' | 'em' | 'strike' | 'link'> = {b: 'strong', strong: 'strong', i: 'em', em: 'em', cite: 'em', dfn: 'em', var: 'em',
+  s: 'strike', del: 'strike', strike: 'strike', a: 'link'};
+const CODE_TAGS = new Set(['code', 'kbd', 'samp', 'tt']);
 
 /** The text of a bracket pair starting at `start` (`[`), honoring one level of nested brackets and escapes. */
 function bracket(source: string, start: number): number {
@@ -162,8 +161,9 @@ function backtickRuns(source: string): Map<number, number[]> {
   return runs;
 }
 
-function tokenize(source: string): Token[] {
+function tokenize(source: string, options: InlineOptions): Token[] {
   const tokens: Token[] = [];
+  const closes = options.html ? closingTags(source) : () => -1;
   const runs = backtickRuns(source);
   /** Per run length, how many runs are already behind the scan position. */
   const consumed = new Map<number, number>();
@@ -210,6 +210,12 @@ function tokenize(source: string): Token[] {
         index += auto[0].length;
         continue;
       }
+      const tag = options.html ? readTag(source, index) : undefined;
+      if (tag && KNOWN.has(tag.name)) {
+        flush();
+        index = htmlTag(tag, source, closes, tokens);
+        continue;
+      }
     }
     if (char === '!' && source[index + 1] === '[' || char === '[') {
       const image = char === '!';
@@ -222,7 +228,7 @@ function tokenize(source: string): Token[] {
         const url = safeLink(target.url);
         if (image) tokens.push({kind: 'span', span: [{text: label.trim() || 'image', style: {image: true}}]});
         else {
-          const inner = parseInline(label).map(span => ({text: span.text, style: {...span.style, ...(url && !span.style.code ? {link: url} : {})}}));
+          const inner = parseInline(label, options).map(span => ({text: span.text, style: {...span.style, ...(url && !span.style.code ? {link: url} : {})}}));
           // A link whose target was refused still shows its text, and the refused target as text so nothing is hidden.
           tokens.push({kind: 'span', span: url ? inner : [...inner, ...(target.url ? [{text: ` (${target.url.slice(0, 200)})`, style: {}}] : [])]});
         }
@@ -256,13 +262,65 @@ function tokenize(source: string): Token[] {
   return tokens;
 }
 
+/**
+ * Where `</name` next appears at or after a position. Searches only move forward and each answer is remembered, so a
+ * run of unclosed tags (`<code>` after `<code>`) stays linear.
+ */
+function closingTags(source: string): (name: string, from: number) => number {
+  const lower = source.toLowerCase();
+  const found = new Map<string, number>();
+  return (name, from) => {
+    const known = found.get(name);
+    if (known !== undefined && (known < 0 || known >= from)) return known;
+    const at = lower.indexOf(`</${name}`, from);
+    found.set(name, at);
+    return at;
+  };
+}
+
+/** One HTML tag in html mode: its tokens, and where reading goes on. */
+function htmlTag(tag: NonNullable<ReturnType<typeof readTag>>, source: string, closes: (name: string, from: number) => number, tokens: Token[]): number {
+  const name = tag.name;
+  // An element dropped with its content (a script, a style, a form control) leaves nothing, up to its end tag.
+  if (!tag.closing && !tag.selfClosing && (DROPPED.has(name) || RAW_TEXT.has(name))) {
+    const close = closes(name, tag.end);
+    const end = close < 0 ? -1 : source.indexOf('>', close);
+    return end < 0 ? source.length : end + 1;
+  }
+  if (tag.closing) {
+    const style = TAG_STYLES[name];
+    if (style) tokens.push({kind: 'tag', style, open: false});
+    return tag.end;
+  }
+  if (name === 'br') { tokens.push({kind: 'text', text: ' '}); return tag.end; }
+  if (name === 'img') { tokens.push({kind: 'span', span: [{text: imageLabel(tag.attributes), style: {image: true}}]}); return tag.end; }
+  if (name === 'input' && tag.attributes.type?.toLowerCase() === 'checkbox') { tokens.push({kind: 'text', text: 'checked' in tag.attributes ? '☑ ' : '☐ '}); return tag.end; }
+  if (CODE_TAGS.has(name) && !tag.selfClosing) {
+    // Code is literal: its text up to the end tag, any markup inside dropped, references decoded (safely).
+    const close = closes(name, tag.end);
+    if (close < 0) return tag.end;
+    const end = source.indexOf('>', close);
+    const code = decodeEntities(source.slice(tag.end, close).replace(/<[^<>]{0,200}>/gu, '')).replace(/\s+/gu, ' ');
+    if (code) tokens.push({kind: 'span', span: [{text: code, style: {code: true}}]});
+    return end < 0 ? source.length : end + 1;
+  }
+  const style = TAG_STYLES[name];
+  if (style === 'link') {
+    const target = decodeEntities(tag.attributes.href ?? '').trim();
+    const link = target ? safeLink(target) : undefined;
+    tokens.push({kind: 'tag', style, open: true, ...(link ? {link} : {}), ...(!link && target ? {target: target.slice(0, 200)} : {})});
+  } else if (style) tokens.push({kind: 'tag', style, open: true});
+  // Any other known element: its tags go, its text stays.
+  return tag.end;
+}
+
 /**
  * Resolve emphasis with a simplified CommonMark delimiter stack: a closer pairs with the nearest compatible opener of
  * the same character; two or more on both sides make strong, `~~` makes strikethrough. Unpaired delimiters become text.
  */
 function resolve(tokens: Token[]): InlineSpan[] {
   type Delim = {kind: 'delim'; char: '*' | '_' | '~'; count: number; length: number; open: boolean; close: boolean};
-  type Node = {kind: 'text'; text: string} | {kind: 'span'; span: InlineSpan[]} | Delim;
+  type Node = Exclude<Token, {kind: 'delim'}> | Delim;
   const nodes: Node[] = tokens.map(token => token.kind === 'delim' ? {...token, length: token.count} : token);
   type Wrap = {from: number; to: number; style: 'strong' | 'em' | 'strike'};
   const wraps: Wrap[] = [];
@@ -294,6 +352,9 @@ function resolve(tokens: Token[]): InlineSpan[] {
   }
   const spans: InlineSpan[] = [];
   const active = {strong: 0, em: 0, strike: 0};
+  // HTML tags (html mode) style what they enclose; links nest, the innermost wins.
+  const tagged = {strong: 0, em: 0, strike: 0};
+  const links: Array<{link?: string; target?: string}> = [];
   const starting = new Map<number, Wrap[]>(), ending = new Map<number, Wrap[]>();
   for (const wrap of wraps) {
     starting.set(wrap.from, [...(starting.get(wrap.from) ?? []), wrap]);
@@ -303,7 +364,21 @@ function resolve(tokens: Token[]): InlineSpan[] {
     // Leftover delimiter characters sit outside their pair: close before the closer, open after the opener.
     for (const wrap of ending.get(position) ?? []) active[wrap.style]--;
     const node = nodes[position]!;
-    const style: InlineStyle = {...(active.strong > 0 ? {strong: true} : {}), ...(active.em > 0 ? {em: true} : {}), ...(active.strike > 0 ? {strike: true} : {})};
+    if (node.kind === 'tag') {
+      if (node.style === 'link') {
+        if (node.open) links.push({...(node.link ? {link: node.link} : {}), ...(node.target ? {target: node.target} : {})});
+        else {
+          const closed = links.pop();
+          // A link whose target was refused still shows its text, and the refused target as text so nothing is hidden.
+          if (closed?.target) spans.push({text: ` (${closed.target})`, style: {}});
+        }
+      } else if (node.open) tagged[node.style]++;
+      else if (tagged[node.style] > 0) tagged[node.style]--;
+      continue;
+    }
+    const link = links.at(-1)?.link;
+    const style: InlineStyle = {...(active.strong + tagged.strong > 0 ? {strong: true} : {}), ...(active.em + tagged.em > 0 ? {em: true} : {}),
+      ...(active.strike + tagged.strike > 0 ? {strike: true} : {}), ...(link ? {link} : {})};
     if (node.kind === 'text') spans.push({text: node.text, style});
     else if (node.kind === 'span') for (const span of node.span) spans.push({text: span.text, style: {...style, ...span.style}});
     else if (node.count > 0) spans.push({text: node.char.repeat(node.count), style});
@@ -320,9 +395,9 @@ function resolve(tokens: Token[]): InlineSpan[] {
 }
 
 /** Inline spans for one paragraph's source text. Soft line breaks become spaces; text past the bound stays plain. */
-export function parseInline(source: string): InlineSpan[] {
+export function parseInline(source: string, options: InlineOptions = {}): InlineSpan[] {
   const bounded = source.length > MAX_INLINE ? source.slice(0, MAX_INLINE) : source;
-  const spans = resolve(tokenize(softBreaks(bounded)));
+  const spans = resolve(tokenize(softBreaks(bounded), options));
   if (bounded.length < source.length) spans.push({text: softBreaks(source.slice(MAX_INLINE)), style: {}});
   return spans;
 }
@@ -335,6 +410,6 @@ function softBreaks(text: string): string {
 }
 
 /** The text a reader sees, without styling (search, copy-visible, width measurement). */
-export function plainInline(source: string): string {
-  return parseInline(source).map(span => span.text).join('');
+export function plainInline(source: string, options: InlineOptions = {}): string {
+  return parseInline(source, options).map(span => span.text).join('');
 }
diff --git a/src/markdown/markup.ts b/src/markdown/markup.ts
new file mode 100644
index 00000000..5418750c
--- /dev/null
+++ b/src/markdown/markup.ts
@@ -0,0 +1,136 @@
+/**
+ * The pieces of HTML that `/view` reads (QA F01): character references and single tags, read by bounded linear scans.
+ * Nothing here builds a document, resolves a URL, loads a resource or runs anything. Shared by the HTML converter
+ * (html.ts) and the inline parser, which reads the same tags inside Markdown.
+ */
+
+const ENTITIES: Record<string, string> = {amp: '&', lt: '<', gt: '>', quot: '"', apos: "'", nbsp: ' ', copy: '©', reg: '®', mdash: '—', ndash: '–', hellip: '…', rarr: '→', larr: '←', times: '×'};
+
+const INVISIBLE = /[\p{Cc}\p{Cf}\p{Co}\p{Zl}\p{Zp}]/u;
+
+/** One character reference at `index`, or undefined. */
+export function decodeEntity(source: string, index: number): {text: string; length: number} | undefined {
+  const match = /^&(?:#(\d{1,7})|#[xX]([0-9a-fA-F]{1,6})|([a-zA-Z]{2,8}));/u.exec(source.slice(index, index + 12));
+  if (!match) return undefined;
+  if (match[3]) { const named = ENTITIES[match[3]]; return named ? {text: named, length: match[0].length} : undefined; }
+  const code = match[1] ? Number(match[1]) : parseInt(match[2]!, 16);
+  if (!code || code > 0x10ffff || (code >= 0xd800 && code <= 0xdfff)) return undefined;
+  const text = String.fromCodePoint(code);
+  // An entity never smuggles in what the display scrubber removes: controls, bidi and other format characters
+  // (direction overrides, zero-width joiners), separators or private-use code points.
+  if (INVISIBLE.test(text)) return undefined;
+  return {text, length: match[0].length};
+}
+
+/** Every character reference in `text` decoded (unknown or unsafe ones stay as written). */
+export function decodeEntities(text: string): string {
+  if (!text.includes('&')) return text;
+  let out = '';
+  for (let index = 0; index < text.length;) {
+    const entity = text[index] === '&' ? decodeEntity(text, index) : undefined;
+    if (entity) { out += entity.text; index += entity.length; } else { out += text[index]; index += 1; }
+  }
+  return out;
+}
+
+/** How far one tag may run, and how much of an attribute value is kept. */
+const MAX_TAG = 4096;
+const MAX_ATTRIBUTE = 2048;
+
+export interface Tag {
+  /** Lowercase element name. */
+  name: string;
+  closing: boolean;
+  selfClosing: boolean;
+  /** Lowercase attribute names; raw values (references not decoded), each at most MAX_ATTRIBUTE characters. */
+  attributes: Record<string, string>;
+  /** Index just past the tag's `>`. */
+  end: number;
+}
+
+const NAME_START = /[A-Za-z]/u;
+const NAME = /[A-Za-z0-9-]/u;
+const SPACE = /[\s]/u;
+
+/** The tag starting at `start` (a `<`), or undefined when it is not one (then the `<` is text). */
+export function readTag(source: string, start: number): Tag | undefined {
+  const limit = Math.min(source.length, start + MAX_TAG);
+  let index = start + 1;
+  const closing = source[index] === '/';
+  if (closing) index += 1;
+  if (!NAME_START.test(source[index] ?? '')) return undefined;
+  const nameStart = index;
+  while (index < limit && NAME.test(source[index]!) && index - nameStart < 32) index += 1;
+  const name = source.slice(nameStart, index).toLowerCase();
+  if (index < limit && !SPACE.test(source[index]!) && source[index] !== '>' && source[index] !== '/') return undefined;
+  const attributes: Record<string, string> = Object.create(null) as Record<string, string>;
+  let count = 0;
+  while (index < limit) {
+    while (index < limit && SPACE.test(source[index]!)) index += 1;
+    const char = source[index];
+    if (char === '>') return {name, closing, selfClosing: false, attributes, end: index + 1};
+    if (char === '/' && source[index + 1] === '>') return {name, closing, selfClosing: true, attributes, end: index + 2};
+    if (char === '/') { index += 1; continue; }
+    if (char === undefined || char === '<') return undefined;
+    const keyStart = index;
+    while (index < limit && !SPACE.test(source[index]!) && !'"\'>/=<'.includes(source[index]!)) index += 1;
+    const key = source.slice(keyStart, Math.min(index, keyStart + 64)).toLowerCase();
+    if (index === keyStart) { index += 1; continue; }
+    while (index < limit && SPACE.test(source[index]!)) index += 1;
+    let value = '';
+    if (source[index] === '=') {
+      index += 1;
+      while (index < limit && SPACE.test(source[index]!)) index += 1;
+      const quote = source[index];
+      if (quote === '"' || quote === "'") {
+        // Only within this tag's bound: an unclosed quote never makes one tag scan the rest of the document.
+        let close = index + 1;
+        while (close < limit && source[close] !== quote) close += 1;
+        if (close >= limit) return undefined;
+        value = source.slice(index + 1, close);
+        index = close + 1;
+      } else {
+        const valueStart = index;
+        while (index < limit && !SPACE.test(source[index]!) && source[index] !== '>') index += 1;
+        value = source.slice(valueStart, index);
+      }
+    }
+    // Only the first of a repeated attribute counts, and a hostile tag cannot make the record unbounded.
+    if (!(key in attributes) && count < 64) { attributes[key] = value.slice(0, MAX_ATTRIBUTE); count += 1; }
+  }
+  return undefined;
+}
+
+/** Elements with no content or end tag. */
+export const VOID = new Set(['area', 'base', 'br', 'col', 'embed', 'hr', 'img', 'input', 'keygen', 'link', 'meta', 'param', 'source', 'track', 'wbr']);
+
+/**
+ * Elements dropped with everything in them: code (script), styling (style), templates and fallbacks for scripts,
+ * embedded or interactive content, and the document's metadata. None of it is shown, loaded or run.
+ */
+export const DROPPED = new Set(['script', 'style', 'template', 'noscript', 'iframe', 'object', 'embed', 'applet', 'svg', 'math', 'canvas', 'audio', 'video',
+  'form', 'button', 'select', 'textarea', 'option', 'optgroup', 'datalist', 'head', 'title', 'meta', 'link', 'base', 'frame', 'frameset', 'noframes', 'dialog',
+  'map', 'portal', 'xmp', 'plaintext', 'noembed', 'slot']);
+
+/** Elements whose content is raw text up to their end tag: never read as markup (all of them are dropped). */
+export const RAW_TEXT = new Set(['script', 'style', 'textarea', 'title', 'xmp', 'plaintext', 'noscript', 'iframe', 'noembed', 'noframes', 'template']);
+
+/** Block-level elements (CommonMark's HTML block names, and a few more containers). */
+export const BLOCK = new Set(['address', 'article', 'aside', 'blockquote', 'body', 'caption', 'center', 'colgroup', 'dd', 'details', 'dir', 'div', 'dl', 'dt',
+  'fieldset', 'figcaption', 'figure', 'footer', 'h1', 'h2', 'h3', 'h4', 'h5', 'h6', 'header', 'hgroup', 'hr', 'html', 'legend', 'li', 'main', 'menu', 'nav', 'ol',
+  'p', 'pre', 'search', 'section', 'summary', 'table', 'tbody', 'td', 'tfoot', 'th', 'thead', 'tr', 'ul']);
+
+/** Inline elements read for their meaning (emphasis, code, links, images, breaks) or just for their text. */
+export const INLINE = new Set(['a', 'abbr', 'b', 'bdi', 'big', 'br', 'cite', 'code', 'data', 'del', 'dfn', 'em', 'font', 'i', 'img', 'input', 'ins', 'kbd',
+  'label', 'mark', 'picture', 'q', 'rp', 'rt', 'ruby', 's', 'samp', 'small', 'source', 'span', 'strike', 'strong', 'sub', 'sup', 'time', 'tt', 'u', 'var', 'wbr', 'col', 'track', 'param', 'area', 'keygen']);
+
+/** Names NMSh reads as HTML at all; any other `<word>` stays text (`a generic <T>`). */
+export const KNOWN = new Set([...VOID, ...DROPPED, ...RAW_TEXT, ...BLOCK, ...INLINE]);
+
+/** An image's words: its alt text, else its file name, else "image". */
+export function imageLabel(attributes: Record<string, string>): string {
+  const alt = decodeEntities(attributes.alt ?? '').replace(/\s+/gu, ' ').trim();
+  if (alt) return alt.slice(0, 300);
+  const name = decodeEntities(attributes.src ?? '').split(/[?#]/u)[0]!.split('/').filter(Boolean).pop() ?? '';
+  return name.slice(0, 120) || 'image';
+}
diff --git a/src/markdown/render.ts b/src/markdown/render.ts
index 272a91e9..7fe8aa36 100644
--- a/src/markdown/render.ts
+++ b/src/markdown/render.ts
@@ -6,6 +6,7 @@ import type {SyntaxSgr} from '../input/syntaxTheme.js';
 import {displayWidth, truncateAnsi} from '../util/text.js';
 import {stripTerminalControls} from '../util/terminalControls.js';
 import {parseBlocks, type Alignment, type Block} from './blocks.js';
+import {htmlToBlocks} from './html.js';
 import {highlightCode, type CodeTokenKind} from './highlight.js';
 import {parseInline, plainInline, type InlineSpan, type InlineStyle} from './inline.js';
 
@@ -37,6 +38,8 @@ export interface MarkdownOptions {
   syntax?: SyntaxSgr;
   /** Base text role for prose (agent replies use primary; quieter contexts may pass secondary). */
   tone?: 'primary' | 'secondary';
+  /** Render HTML inside the Markdown (`/view`, QA F01); otherwise tags are shown as text. */
+  html?: boolean;
 }
 
 const RESET = '\u001b[0m';
@@ -54,6 +57,7 @@ interface Context {
   fg: (color: RgbColor) => string;
   tone: RgbColor;
   codeIndex: number;
+  html: boolean;
 }
 
 function glyph(context: Context, nerd: string, safe: string): string {
@@ -158,7 +162,10 @@ function wrapAtoms(atoms: readonly Atom[], width: number, first: string, rest: s
       // Spaces never start a row and never trail one.
       if (empty) continue;
       const next = atoms[index + 1];
-      if (!next || used + 1 + displayWidth(next.text) > width) continue;
+      if (!next) continue;
+      // The next word does not fit after this space: it starts the next row (it may fit without the space, but must
+      // never be glued to the word before it).
+      if (used + 1 + displayWidth(next.text) > width) { push(); continue; }
       ansi += paintAtom(atom); plain += ' '; used += 1;
       continue;
     }
@@ -185,7 +192,7 @@ function wrapAtoms(atoms: readonly Atom[], width: number, first: string, rest: s
 
 function inlineRows(source: string, width: number, context: Context, base: string, first = '', rest = first): Array<{ansi: string; plain: string}> {
   // A hard break inside a paragraph is a new row with the same hanging indent.
-  return source.split('\n').flatMap((line, index) => wrapAtoms(atomsOf(parseInline(line), context, base), width, index === 0 ? first : rest, rest));
+  return source.split('\n').flatMap((line, index) => wrapAtoms(atomsOf(parseInline(line, {html: context.html}), context, base), width, index === 0 ? first : rest, rest));
 }
 
 // ---- Code ---------------------------------------------------------------------------------------------------------
@@ -213,7 +220,7 @@ function tokenSgr(kind: CodeTokenKind, context: Context): string {
 export function codeTokenStyle(kind: CodeTokenKind, options: {syntax?: SyntaxSgr; level?: ColorLevel} = {}): string {
   const level = options.level ?? colorLevel();
   return tokenSgr(kind, {level, safe: false, hyperlinks: false, ...(options.syntax ? {syntax: options.syntax} : {}),
-    fg: color => colorEscape(38, color, level), tone: UI_COLORS.primary, codeIndex: 0});
+    fg: color => colorEscape(38, color, level), tone: UI_COLORS.primary, codeIndex: 0, html: false});
 }
 
 function codeRows(block: Extract<Block, {kind: 'code'}>, width: number, context: Context): MarkdownRow[] {
@@ -261,11 +268,11 @@ function alignCell(text: string, width: number, align: Alignment): string {
 function tableRows(block: Extract<Block, {kind: 'table'}>, width: number, context: Context): MarkdownRow[] {
   const columns = block.header.length;
   const all = [block.header, ...block.rows];
-  const natural = Array.from({length: columns}, (_, column) => Math.max(1, ...all.map(row => displayWidth(plainInline(row[column] ?? '')))));
+  const natural = Array.from({length: columns}, (_, column) => Math.max(1, ...all.map(row => displayWidth(plainInline(row[column] ?? '', {html: context.html})))));
   const separator = glyph(context, ' │ ', ' | ');
   const chrome = 3 * (columns - 1);
   const available = width - chrome;
-  const minimum = Array.from({length: columns}, (_, column) => Math.min(natural[column]!, Math.max(3, Math.min(10, ...[block.header[column] ?? ''].map(cell => displayWidth(plainInline(cell)))))));
+  const minimum = Array.from({length: columns}, (_, column) => Math.min(natural[column]!, Math.max(3, Math.min(10, ...[block.header[column] ?? ''].map(cell => displayWidth(plainInline(cell, {html: context.html})))))));
   if (available < minimum.reduce((a, b) => a + b, 0)) {
     // Too narrow for a grid: each row becomes a short record of "Header: value" lines.
     const rows: MarkdownRow[] = [];
@@ -273,7 +280,7 @@ function tableRows(block: Extract<Block, {kind: 'table'}>, width: number, contex
     block.rows.forEach((row, index) => {
       if (index) rows.push({ansi: '', plain: '', kind: 'blank'});
       row.forEach((cell, column) => {
-        const label = plainInline(block.header[column] ?? '');
+        const label = plainInline(block.header[column] ?? '', {html: context.html});
         const lead = `${subtle}${label}:${RESET} `;
         for (const line of inlineRows(cell || '—', width, context, base, lead, '  ')) rows.push({...line, kind: 'table'});
       });
@@ -375,6 +382,15 @@ function renderBlocks(blocks: readonly Block[], width: number, context: Context,
 
 /** Rows for untrusted Markdown at `columns` cells. */
 export function renderMarkdown(source: string, columns: number, options: MarkdownOptions = {}): MarkdownRow[] {
+  return renderSource(source, columns, options, safe => parseBlocks(safe, {html: Boolean(options.html)}));
+}
+
+/** Rows for an untrusted HTML document at `columns` cells (`/view`, QA F01): its supported subset, never run or loaded. */
+export function renderHtml(source: string, columns: number, options: Omit<MarkdownOptions, 'html'> = {}): MarkdownRow[] {
+  return renderSource(source, columns, {...options, html: true}, htmlToBlocks);
+}
+
+function renderSource(source: string, columns: number, options: MarkdownOptions, parse: (safe: string) => Block[]): MarkdownRow[] {
   const width = Math.max(4, columns);
   const level = options.level ?? colorLevel();
   const context: Context = {
@@ -385,10 +401,11 @@ export function renderMarkdown(source: string, columns: number, options: Markdow
     fg: color => colorEscape(38, color, level),
     tone: options.tone === 'secondary' ? UI_COLORS.secondary : UI_COLORS.primary,
     codeIndex: 0,
+    html: Boolean(options.html),
   };
   // The renderer never trusts its caller to have scrubbed: terminal controls and bidi formatting go, tabs become spaces.
   const safe = source.split('\n').map(line => stripTerminalControls(line.replace(/\t/gu, '    '), Number.MAX_SAFE_INTEGER)).join('\n');
-  const rows = renderBlocks(parseBlocks(safe), width, context, 0);
+  const rows = renderBlocks(parse(safe), width, context, 0);
   while (rows.length && rows.at(-1)!.kind === 'blank') rows.pop();
   // A last line of defence: no row may exceed the width whatever a rule above computed.
   return rows.map(row => displayWidth(row.ansi) > width ? {...row, ansi: truncateAnsi(row.ansi, width)} : row);
diff --git a/src/viewer/ViewerPanel.ts b/src/viewer/ViewerPanel.ts
index e4759eec..6d383a6f 100644
--- a/src/viewer/ViewerPanel.ts
+++ b/src/viewer/ViewerPanel.ts
@@ -1,7 +1,7 @@
 import type {Key} from '../terminal/keys.js';
 import type {SyntaxSgr} from '../input/syntaxTheme.js';
 import {graphemes} from '../input/inputLayout.js';
-import {codeTokenStyle, renderMarkdown, type MarkdownRow} from '../markdown/render.js';
+import {codeTokenStyle, renderHtml, renderMarkdown, type MarkdownRow} from '../markdown/render.js';
 import {highlightCode, type CodeToken} from '../markdown/highlight.js';
 import {colorLevel} from '../presentation/capabilities.js';
 import {renderControlRows} from '../ui/controls.js';
@@ -14,9 +14,10 @@ import {formatBytes, inert, type ViewDocument} from './document.js';
 /**
  * The native read-only viewer (#397): one document in a full-height panel, never the transcript. Drawing is bounded
  * to what is visible: only the lines on screen are laid out, one wrapped line takes at most MAX_WRAP_ROWS rows, syntax
- * roles cover the first HIGHLIGHT_LINES lines and Markdown renders up to RENDER_MARKDOWN_BYTES. Every character is
- * drawn inert: control and format characters (bidi overrides, zero-width marks) are shown as visible marks and tabs
- * as spaces, and Markdown links are links only for the renderer's allowed targets.
+ * roles cover the first HIGHLIGHT_LINES lines and Markdown and HTML render up to RENDER_MARKDOWN_BYTES. Every character
+ * is drawn inert: control and format characters (bidi overrides, zero-width marks) are shown as visible marks and tabs
+ * as spaces, and links are links only for the renderer's allowed targets. HTML (a file, or inside Markdown) is
+ * rendered natively from its supported subset: never run, styled by CSS or loaded (QA F01).
  */
 
 const HIGHLIGHT_LINES = 20_000;
@@ -63,11 +64,16 @@ export type ViewerAction = {kind: 'close'} | {kind: 'copy'; text: string; lines:
 export interface ViewerLook {syntax?: SyntaxSgr; hyperlinks?: boolean}
 
 export function createViewer(document: ViewDocument): ViewerState {
-  const markdown = document.format === 'markdown' && Buffer.byteLength(document.text) <= RENDER_MARKDOWN_BYTES;
-  const note = document.format === 'markdown' && !markdown ? 'Too large to render: showing the Markdown source.' : undefined;
+  const renderable = renders(document);
+  const markdown = renderable && Buffer.byteLength(document.text) <= RENDER_MARKDOWN_BYTES;
+  const note = renderable && !markdown ? `Too large to render: showing the ${sourceName(document)} source.` : undefined;
   return {document, top: 0, sub: 0, left: 0, wrap: true, numbers: !markdown, rendered: markdown, ...(note ? {note} : {})};
 }
 
+/** Markdown and HTML have a rendered view beside their source. */
+const renders = (document: ViewDocument) => document.format === 'markdown' || document.format === 'html';
+const sourceName = (document: ViewDocument) => document.format === 'html' ? 'HTML' : 'Markdown';
+
 // ---- Text ---------------------------------------------------------------------------------------------------------
 
 /** Characters drawn as visible marks: controls (not tab) and invisible format, line and paragraph separators. */
@@ -144,7 +150,8 @@ function gutterWidth(state: ViewerState): number {
 
 function markdownRows(state: ViewerState, width: number, look: ViewerLook): MarkdownRow[] {
   if (state.markdown?.width !== width) {
-    state.markdown = {width, rows: renderMarkdown(state.document.text, width, {...(look.syntax ? {syntax: look.syntax} : {}), hyperlinks: Boolean(look.hyperlinks)})};
+    const options = {...(look.syntax ? {syntax: look.syntax} : {}), hyperlinks: Boolean(look.hyperlinks)};
+    state.markdown = {width, rows: state.document.format === 'html' ? renderHtml(state.document.text, width, options) : renderMarkdown(state.document.text, width, {...options, html: true})};
   }
   return state.markdown.rows;
 }
@@ -393,8 +400,8 @@ export function viewerKey(state: ViewerState, key: Key, columns: number, height:
   else if ((key.kind === 'left' || key.kind === 'right') && !state.wrap && !state.rendered) state.left = Math.max(0, state.left + (key.kind === 'left' ? -HORIZONTAL_STEP : HORIZONTAL_STEP));
   else if (text === 'w' && !state.rendered) { state.wrap = !state.wrap; state.left = 0; state.sub = 0; clampTop(state, width, body, look); }
   else if (text === 'l' && !state.rendered) { state.numbers = !state.numbers; state.sub = 0; }
-  else if (text === 'm' && state.document.format === 'markdown') {
-    if (!state.rendered && Buffer.byteLength(state.document.text) > RENDER_MARKDOWN_BYTES) state.note = 'Too large to render: showing the Markdown source.';
+  else if (text === 'm' && renders(state.document)) {
+    if (!state.rendered && Buffer.byteLength(state.document.text) > RENDER_MARKDOWN_BYTES) state.note = `Too large to render: showing the ${sourceName(state.document)} source.`;
     else { state.rendered = !state.rendered; state.numbers = !state.rendered; state.top = 0; state.sub = 0; state.match = undefined; state.matchCount = undefined; }
   }
   else if (text === '/' || key.kind === 'find') state.input = {kind: 'search', value: state.query ?? ''};
@@ -416,7 +423,7 @@ function detail(document: ViewDocument): string {
     const status = document.origin.exitCode === undefined ? '' : ` · exit ${document.origin.exitCode}`;
     return `output ${document.origin.number}${status} · ${lines}${document.format === 'diff' ? ' · diff' : ''}`;
   }
-  const kind = document.format === 'markdown' ? 'Markdown' : document.format === 'diff' ? 'diff' : document.language ?? 'text';
+  const kind = document.format === 'markdown' ? 'Markdown' : document.format === 'html' ? 'HTML' : document.format === 'diff' ? 'diff' : document.language ?? 'text';
   const ending = document.lineEnding === 'crlf' ? ' · CRLF' : document.lineEnding === 'mixed' ? ' · mixed line endings' : '';
   const link = document.origin.symlink ? ` · → ${inert(document.origin.realPath)}` : '';
   return `${kind} · ${lines} · ${formatBytes(document.origin.bytes)}${ending}${link}`;
@@ -440,7 +447,7 @@ function controls(state: ViewerState, columns: number): string[] {
   if (state.input) return renderControlRows([['Enter', state.input.kind === 'search' ? 'find' : 'go'], ['Esc', 'cancel']], width);
   return renderControlRows([
     ['↑↓', 'scroll'], ['/', 'find'], ...(state.query ? [['n/N', 'next/previous'] as const] : []), [':', 'line'],
-    ...(state.document.format === 'markdown' ? [['m', state.rendered ? 'source' : 'rendered'] as const] : []),
+    ...(renders(state.document) ? [['m', state.rendered ? 'source' : 'rendered'] as const] : []),
     ...(!state.rendered ? [['w', state.wrap ? 'no wrap' : 'wrap'] as const, ['l', 'numbers'] as const] : []),
     ['c', 'copy all'], ...(state.document.origin.kind === 'file' ? [['e', 'edit'] as const, ['o', 'open in editor'] as const] : []), ['Esc', 'close'],
   ], width);
diff --git a/src/viewer/document.ts b/src/viewer/document.ts
index e27db3df..3ef6ef77 100644
--- a/src/viewer/document.ts
+++ b/src/viewer/document.ts
@@ -17,7 +17,8 @@ export const MAX_VIEW_LINES = 200_000;
 const SNIFF_BYTES = 8192;
 const CHUNK = 64 * 1024;
 
-export type DocumentFormat = 'text' | 'markdown' | 'diff';
+/** html: rendered natively, never run or loaded (QA F01). */
+export type DocumentFormat = 'text' | 'markdown' | 'html' | 'diff';
 
 export type DocumentOrigin =
   | {kind: 'file'; path: string; realPath: string; symlink: boolean; bytes: number; mtimeMs: number}
@@ -63,11 +64,12 @@ export function displayPath(path: string, home = homedir()): string {
 const FILE_NAMES: Record<string, string> = {makefile: 'makefile', gnumakefile: 'makefile', dockerfile: 'dockerfile', containerfile: 'dockerfile',
   '.bashrc': 'shell', '.zshrc': 'shell', '.zshenv': 'shell', '.profile': 'shell', '.bash_profile': 'shell', 'config.fish': 'fish', justfile: 'makefile'};
 
-/** How a file reads: Markdown and diffs by their extension, syntax roles by extension or well-known name. */
+/** How a file reads: Markdown, HTML and diffs by their extension, syntax roles by extension or well-known name. */
 export function formatForPath(path: string): {format: DocumentFormat; language?: string} {
   const name = basename(path).toLowerCase();
   const extension = extname(name).slice(1);
   if (extension === 'md' || extension === 'markdown' || extension === 'mdx') return {format: 'markdown', language: 'markdown'};
+  if (extension === 'html' || extension === 'htm' || extension === 'xhtml') return {format: 'html', language: languageId('html') ?? 'markup'};
   if (extension === 'diff' || extension === 'patch') return {format: 'diff', language: 'diff'};
   const language = languageId(FILE_NAMES[name] ?? extension);
   return {format: 'text', ...(language ? {language} : {})};
diff --git a/tests/markdownRender.test.ts b/tests/markdownRender.test.ts
index 36395c31..69016659 100644
--- a/tests/markdownRender.test.ts
+++ b/tests/markdownRender.test.ts
@@ -220,3 +220,15 @@ test('security: pathological lines parse and render in linear time', () => {
     assert.ok(Date.now() - started < 1500, `${JSON.stringify(input.slice(0, 20))}… took ${Date.now() - started} ms`);
   }
 });
+
+test('wrapping never glues two words: a word that fits only without the space before it starts the next row', () => {
+  // Found while testing /view's HTML (QA F01); present before it in every Markdown surface.
+  assert.deepEqual(renderMarkdown('aaaa bbbb', 8, {level: 'none'}).map(row => row.plain), ['aaaa', 'bbbb']);
+  assert.deepEqual(renderMarkdown('aaaa bbbb', 9, {level: 'none'}).map(row => row.plain), ['aaaa bbbb']);
+  const rows = renderMarkdown('[CI status](https://github.com/raiseCatError/notMyShell/actions) **A calmer terminal** for *zsh*, Bash and Fish.', 80, {level: 'truecolor'});
+  assert.deepEqual(rows.map(row => row.plain), ['CI status (https://github.com/raiseCatError/notMyShell/actions) A calmer', 'terminal for zsh, Bash and Fish.']);
+  for (const columns of [10, 17, 23, 40]) {
+    const text = 'one two three four five six seven eight nine ten';
+    assert.equal(renderMarkdown(text, columns, {level: 'none'}).map(row => row.plain).join(' '), text, `${columns}`);
+  }
+});
diff --git a/tests/viewHtml.test.ts b/tests/viewHtml.test.ts
new file mode 100644
index 00000000..32e45f74
--- /dev/null
+++ b/tests/viewHtml.test.ts
@@ -0,0 +1,198 @@
+import test from 'node:test';
+import assert from 'node:assert/strict';
+import {readFileSync} from 'node:fs';
+import {renderHtml, renderMarkdown} from '../src/markdown/render.js';
+import {formatForPath, type ViewDocument} from '../src/viewer/document.js';
+import {createViewer, renderViewer, viewerKey} from '../src/viewer/ViewerPanel.js';
+import {displayWidth, stripAnsi} from '../src/util/text.js';
+
+/**
+ * QA F01 (v0.22): `/view` showed a README's embedded HTML (`<picture>`, `<img>`, `<div>`) literally. HTML is now
+ * rendered natively, as terminal rows, for HTML files and for HTML inside Markdown that `/view` shows: a supported
+ * subset (headings, paragraphs, lists, tables, code, quotes, rules, inline emphasis, code, links and images) mapped onto
+ * the Markdown renderer's own blocks. Nothing is executed, fetched or read: scripts, styles, embedded content and event
+ * handlers are dropped, CSS is ignored, an image is its alt text, links follow the renderer's http(s)-only rule, and
+ * the parser is linear and bounded in depth, tags and attribute length. Agent replies and GitHub bodies keep showing
+ * tags as text: HTML is rendered only where `/view` asks for it.
+ */
+
+const plain = (rows: ReadonlyArray<{plain: string}>) => rows.map(row => row.plain).join('\n');
+const markdown = (source: string, columns = 80) => renderMarkdown(source, columns, {level: 'truecolor', glyphs: 'nerd', html: true});
+const html = (source: string, columns = 80, hyperlinks = false) => renderHtml(source, columns, {level: 'truecolor', glyphs: 'nerd', hyperlinks});
+
+const README = `<div align="center">
+
+<picture>
+  <source media="(prefers-color-scheme: dark)" srcset="docs/logo-dark.png">
+  <img alt="NMSh logo" src="docs/logo-light.png" width="240">
+</picture>
+
+# notMyShell
+
+<p align="center">
+  <a href="https://github.com/raiseCatError/notMyShell/actions"><img src="https://img.shields.io/badge/CI-passing-green" alt="CI status"></a>
+  <b>A calmer terminal</b> for <i>zsh</i>, Bash and Fish.
+</p>
+
+</div>
+
+Press <kbd>Ctrl</kbd>+<kbd>C</kbd> to stop.<br>Next line.
+
+- one
+- two`;
+
+test('F01: a README\'s embedded HTML renders: images as alt text, inline styles, Markdown between the blocks unchanged', () => {
+  const rows = markdown(README, 120);
+  const text = plain(rows);
+  for (const tag of ['<div', '</div>', '<picture', '<source', '<img', '<p ', '</p>', '<a ', '<b>', '<i>', '<kbd>', '<br>']) assert.ok(!text.includes(tag), `${tag} is not shown`);
+  assert.match(text, /\[image: NMSh logo\]/u);
+  assert.match(text, /\[image: CI status\]/u);
+  assert.ok(rows.some(row => row.kind === 'heading' && row.plain === 'notMyShell'), 'Markdown between HTML blocks still renders');
+  assert.match(text, /A calmer terminal for zsh, Bash and Fish\./u);
+  assert.match(text, /Press Ctrl\+C to stop\. Next line\./u);
+  assert.match(text, /• one\n• two/u);
+  const styled = rows.map(row => row.ansi).join('\n');
+  assert.match(styled, /\u001b\[1m[^\n]*A/u, '<b> is bold');
+});
+
+test('F01: an HTML document renders its structure: headings, paragraphs, nested lists, tables, code, quotes and rules', () => {
+  const rows = html(`<!DOCTYPE html><html><head><title>T</title><style>h1{color:red}</style></head><body>
+    <h1>Guide &amp; notes</h1>
+    <p>First   paragraph
+       with <strong>strong</strong>, <em>em</em>, <code>a &lt; b</code> and <del>old</del>.</p>
+    <ol start="3"><li>three<ul><li>nested</li></ul></li><li>four</li></ol>
+    <table><thead><tr><th align="left">Key</th><th align="right">Value</th></tr></thead>
+      <tbody><tr><td>a</td><td>1</td></tr><tr><td>b | c</td><td>2</td></tr></tbody></table>
+    <pre><code class="language-ts">const x = 1 &lt; 2;
+  return x;</code></pre>
+    <blockquote><p>Quoted</p></blockquote>
+    <hr>
+    <ul><li><input type="checkbox" checked disabled> done</li><li><input type="checkbox" disabled> open</li></ul>
+    <details><summary>More</summary><p>Hidden text</p></details>
+  </body></html>`);
+  const text = plain(rows);
+  assert.ok(rows.some(row => row.kind === 'heading' && row.plain === 'Guide & notes'));
+  assert.match(text, /First paragraph with strong, em, a < b and old\./u, 'whitespace collapses; entities decode');
+  assert.match(text, /3\. three\n {3}◦ nested\n4\. four/u);
+  assert.ok(rows.some(row => row.kind === 'table' && /Key/u.test(row.plain) && /Value/u.test(row.plain)));
+  assert.ok(rows.some(row => row.kind === 'table' && /b \| c/u.test(row.plain)), 'a cell keeps its own pipe');
+  assert.ok(rows.some(row => row.kind === 'code' && /const x = 1 < 2;/u.test(row.plain)));
+  assert.ok(rows.some(row => row.kind === 'code' && /^\s*return x;/u.test(row.plain.replace(/^.*?│ ?/u, ''))) || /return x;/u.test(text));
+  assert.ok(rows.some(row => row.kind === 'quote' && /Quoted/u.test(row.plain)));
+  assert.ok(rows.some(row => row.kind === 'rule'));
+  assert.match(text, /☑ done\n☐ open/u);
+  assert.match(text, /More\n*Hidden text/u);
+  assert.doesNotMatch(text, /color:red|<title>|\bT\b\n/u, 'head and styles are not content');
+});
+
+test('F01 security: nothing executes, loads or styles: scripts, styles, handlers, embeds and unsafe links are dropped', () => {
+  const rows = html(`<p onclick="alert(1)">Safe <span style="color:red">text</span></p>
+    <script>alert('x')</script><style>p{display:none}</style><template><p>tmpl</p></template>
+    <noscript>ns</noscript><iframe src="https://evil.example/"></iframe><object data="x.swf"></object><embed src="x">
+    <svg onload="alert(1)"><text>svgtext</text></svg><math><mi>x</mi></math><form action="https://evil"><input value="secret"><button>Go</button></form>
+    <img src="x" onerror="alert(1)" alt="pic">
+    <p><a href="javascript:alert(1)">js</a> <a href="file:///etc/passwd">file</a> <a href="data:text/html,x">data</a> <a href="https://ok.example/">ok</a></p>`, 80, true);
+  const text = plain(rows);
+  for (const banned of ["alert('x')", 'tmpl', 'svgtext', 'secret', 'Go', 'display', 'onclick', 'onerror', 'evil', 'x.swf', '\nns']) assert.ok(!text.includes(banned), `${banned} not shown`);
+  // A refused link keeps its text and shows its target as inert text, as in Markdown: nothing is hidden or followed.
+  assert.match(text, /js \(javascript:alert\(1\)\) file \(file:\/\/\/etc\/passwd\)/u);
+  assert.match(text, /Safe text/u);
+  assert.match(text, /\[image: pic\]/u);
+  const styled = rows.map(row => row.ansi).join('\n');
+  const links = [...styled.matchAll(/\u001b\]8;;([^\u0007\u001b]*)/gu)].map(match => match[1]).filter(Boolean);
+  assert.deepEqual(links, ['https://ok.example/'], 'only the http(s) link is a link');
+});
+
+test('F01 security: terminal controls and invisible characters never come through text, entities or attributes', () => {
+  const rows = html(`<p>A\u001b[31mred\u001b]52;c;aGk=\u0007 &#27;[2J &#x1b;]0;t&#7; &#x202e;gpj.exe x​y</p>
+    <img alt="\u001b[2Jalt" src="a"><a href="https://ok.example/\u001b]8;;evil">t</a>`, 60, true);
+  for (const row of rows) {
+    assert.doesNotMatch(row.plain, /[\u0000-\u0008\u000b-\u001f\u007f-\u009f‮]/u, JSON.stringify(row.plain));
+    assert.doesNotMatch(row.ansi, /\u001b\[2J|\u001b\]52|\u001b\]0;/u, 'no screen, clipboard or title control survives');
+  }
+});
+
+test('F01 limits: deep nesting, many tags, long attributes and unclosed markup stay bounded and never throw', () => {
+  const cases = [
+    '<div>'.repeat(100_000) + 'deep' + '</div>'.repeat(100_000),
+    '<b>'.repeat(50_000) + 'x',
+    `<img alt="${'a'.repeat(200_000)}">`,
+    '<p>' + '<span>a</span>'.repeat(60_000),
+    '<table>' + '<tr><td>x</td></tr>'.repeat(20_000),
+    '<ul>' + '<li>'.repeat(30_000) + 'item',
+    '<p unterminated="never closed',
+    '<!-- comment never closed',
+    '<a href="https://x/">' + '['.repeat(50_000),
+  ];
+  for (const source of cases) {
+    const started = performance.now();
+    const rows = html(source, 40);
+    assert.ok(performance.now() - started < 4000, `bounded: ${source.slice(0, 30)}`);
+    for (const row of rows.slice(0, 200)) assert.ok(displayWidth(row.ansi) <= 40);
+  }
+  assert.match(plain(html('<div>'.repeat(100_000) + 'deep' + '</div>'.repeat(100_000))), /deep/u, 'content past the depth bound is still shown');
+  // Unclosed quotes and tags never make each tag rescan the rest: whole documents stay linear.
+  for (const [source, render] of [["<a x='".repeat(170_000), html], [('<a x="' + 'y'.repeat(4000)).repeat(250), html],
+    ['x <code>a '.repeat(6_000), markdown], ['x <script> '.repeat(6_000), markdown]] as const) {
+    const started = performance.now();
+    render(source, 80);
+    assert.ok(performance.now() - started < 2500, `linear: ${source.slice(0, 12)}…`);
+  }
+});
+
+test('F01: rendering HTML reads no files and makes no requests (the module imports nothing that could)', () => {
+  const source = readFileSync(new URL('../src/markdown/html.ts', import.meta.url), 'utf8');
+  const imports = [...source.matchAll(/^import .* from '([^']+)';$/gmu)].map(match => match[1]!);
+  for (const name of imports) assert.ok(!/^node:|^(?:fs|net|http|https|child_process|dns|tls)$/u.test(name), `html.ts imports ${name}`);
+});
+
+test('F01: agent replies and GitHub bodies are unchanged: without the HTML option, tags stay text', () => {
+  const text = plain(renderMarkdown('Use a <div> here, or <b>this</b>.\n\n<p align="center">x</p>', 80, {level: 'truecolor'}));
+  assert.match(text, /Use a <div> here, or <b>this<\/b>\./u);
+  assert.match(text, /<p align="center">x<\/p>/u);
+  // An unknown tag-like word in /view's Markdown also stays text.
+  assert.match(plain(markdown('A generic <T> and <placeholder> stay.')), /A generic <T> and <placeholder> stay\./u);
+});
+
+test('F01: narrow widths fit, and NO_COLOR and Safe glyphs keep structure without color', () => {
+  for (const columns of [20, 40, 80]) {
+    for (const row of html('<h2>Title that is fairly long</h2><table><tr><th>Name</th><th>Description</th></tr><tr><td>a</td><td>a long cell that wraps</td></tr></table><ul><li>item</li></ul>', columns)) {
+      assert.ok(displayWidth(row.ansi) <= columns, `${columns}: ${row.plain}`);
+    }
+  }
+  const safe = renderHtml('<ul><li>x</li></ul><hr><blockquote>q</blockquote><code>c</code>', 40, {level: 'none', glyphs: 'safe'});
+  assert.match(plain(safe), /- x/u);
+  assert.match(plain(safe), /`c`/u, 'code keeps backticks without color');
+  for (const row of safe) assert.doesNotMatch(row.ansi, /\u001b\[[0-9;]*[34]8;/u, 'no color');
+});
+
+function htmlDocument(text: string, path = '/tmp/page.html'): ViewDocument {
+  return {title: path, origin: {kind: 'file', path, realPath: path, symlink: false, bytes: text.length, mtimeMs: 0}, lines: text.split('\n'), text,
+    format: formatForPath(path).format, language: formatForPath(path).language, notes: [], lineEnding: 'lf', complete: true};
+}
+
+test('F01 viewer: .html opens rendered; m shows the source and back; search finds rendered text; narrow fits', () => {
+  assert.equal(formatForPath('/x/index.html').format, 'html');
+  assert.equal(formatForPath('/x/page.HTM').format, 'html');
+  assert.equal(formatForPath('/x/README.md').format, 'markdown');
+  const state = createViewer(htmlDocument('<h1>Page</h1>\n<p>Hello <b>world</b></p>\n<script>alert(1)</script>'));
+  assert.equal(state.rendered, true);
+  let screen = renderViewer(state, 60, 20).map(row => stripAnsi(row)).join('\n');
+  assert.match(screen, /Page/u);
+  assert.match(screen, /Hello world/u);
+  assert.doesNotMatch(screen, /alert|<p>/u);
+  viewerKey(state, {kind: 'text', value: 'm'}, 60, 20);
+  assert.equal(state.rendered, false);
+  screen = renderViewer(state, 60, 20).map(row => stripAnsi(row)).join('\n');
+  assert.match(screen, /<p>Hello <b>world<\/b><\/p>/u, 'source mode shows the file as it is');
+  viewerKey(state, {kind: 'text', value: 'm'}, 60, 20);
+  assert.equal(state.rendered, true);
+  viewerKey(state, {kind: 'text', value: '/'}, 60, 20);
+  for (const char of 'world') viewerKey(state, {kind: 'text', value: char}, 60, 20);
+  viewerKey(state, {kind: 'enter'}, 60, 20);
+  assert.equal(state.matchCount, 1);
+  for (const row of renderViewer(state, 24, 12)) assert.ok(displayWidth(row) <= 24);
+  // Markdown in /view renders its embedded HTML too.
+  const readme = createViewer({...htmlDocument('# T\n\n<p align="center"><img alt="Logo" src="x.png"></p>', '/tmp/README.md')});
+  assert.match(renderViewer(readme, 60, 20).map(row => stripAnsi(row)).join('\n'), /\[image: Logo\]/u);
+});