diff --git a/packages/collab-web/src/components/transcript/Markdown.tsx b/packages/collab-web/src/components/transcript/Markdown.tsx
index 33dcb5759..fcac4046f 100644
--- a/packages/collab-web/src/components/transcript/Markdown.tsx
+++ b/packages/collab-web/src/components/transcript/Markdown.tsx
@@ -10,7 +10,43 @@ function escapeHtml(s: string): string {
 		.replaceAll('"', """)
 		.replaceAll("'", "'");
 }
+function unescapeHtml(raw: string): string {
+	const parseCodePoint = (value: number): string => {
+		if (Number.isFinite(value) && value >= 0 && value <= 0x10ffff) {
+			try {
+				return String.fromCodePoint(value);
+			} catch (_) {}
+		}
+		return "";
+	};
 
+	return raw.replace(/&(amp|lt|gt|quot|apos|nbsp|#\d+|#x[0-9a-fA-F]+);/gi, (match, entity) => {
+		const lower = entity.toLowerCase();
+		switch (lower) {
+			case "nbsp":
+				return " ";
+			case "lt":
+				return "<";
+			case "gt":
+				return ">";
+			case "quot":
+				return '"';
+			case "apos":
+				return "'";
+			case "amp":
+				return "&";
+			default: {
+				if (lower.startsWith("#x")) {
+					return parseCodePoint(Number.parseInt(lower.slice(2), 16));
+				}
+				if (lower.startsWith("#")) {
+					return parseCodePoint(Number(lower.slice(1)));
+				}
+				return match;
+			}
+		}
+	});
+}
 function safeHref(href: string): string | null {
 	const trimmed = href.trim();
 	if (/^(?:https?:|mailto:)/i.test(trimmed)) return trimmed;
@@ -23,7 +59,9 @@ const md = new Marked({
 	renderer: {
 		// Raw HTML tokens (block + inline both arrive here) are escaped, never emitted.
 		html({ text }) {
-			return escapeHtml(text);
+			const cleaned = text.replace(/<\/?(?:span|text)\b(?:\s[^>]*)?\s*\/?>/gi, "");
+			if (cleaned === "") return "";
+			return escapeHtml(unescapeHtml(cleaned));
 		},
 		link({ href, title, tokens }) {
 			const inner = this.parser.parseInline(tokens);
diff --git a/packages/collab-web/test/markdown.test.tsx b/packages/collab-web/test/markdown.test.tsx
index 2935fe4a3..2a153ad87 100644
--- a/packages/collab-web/test/markdown.test.tsx
+++ b/packages/collab-web/test/markdown.test.tsx
@@ -29,4 +29,18 @@ describe("Transcript Markdown", () => {
 		expect(html).toContain("&lt;img src=x onerror=alert(1)&gt;");
 		expect(html).not.toContain("<img src=x");
 	});
+
+	it("strips span and text HTML tags but preserves their contents and inline text rendering", () => {
+		const html = renderMarkdown("<span></span><text>▃</text>");
+
+		expect(html).toContain("▃");
+		expect(html).not.toContain("&lt;span&gt;");
+		expect(html).not.toContain("&lt;text&gt;");
+	});
+
+	it("unescapes HTML entities inside span and text HTML tags safely", () => {
+		const html = renderMarkdown("<span>&lt;▃&gt; &amp; &quot;test&quot; &#128512; &#x1F600;</span>");
+
+		expect(html).toContain("&lt;▃&gt; &amp; &quot;test&quot; &#128512; &#x1F600;");
+	});
 });
diff --git a/packages/tui/CHANGELOG.md b/packages/tui/CHANGELOG.md
index 5b818ef45..87ae5b9b4 100644
--- a/packages/tui/CHANGELOG.md
+++ b/packages/tui/CHANGELOG.md
@@ -2,6 +2,9 @@
 
 ## [Unreleased]
 
+### Fixed
+
+- Fixed Markdown component to strip inline `<span>` and `<text>` tags while preserving their contents and unescaping nested HTML entities (`&lt;`, `&gt;`, `&quot;`, `&apos;`, `&amp;`), preventing raw LLM block/inline formatting residues from leaking into rendered TUI output.
 ## [16.1.7] - 2026-06-20
 
 ### Fixed
diff --git a/packages/tui/src/components/markdown.ts b/packages/tui/src/components/markdown.ts
index ef5187972..5540dac76 100644
--- a/packages/tui/src/components/markdown.ts
+++ b/packages/tui/src/components/markdown.ts
@@ -32,7 +32,43 @@ function isOsc66Line(line: string): boolean {
 }
 
 function normalizeHtmlEntitiesForTerminal(raw: string): string {
-	return raw.replace(/&nbsp;/gi, " ");
+	const parseCodePoint = (value: number): string => {
+		if (Number.isFinite(value) && value >= 0 && value <= 0x10ffff) {
+			try {
+				return String.fromCodePoint(value);
+			} catch (_) {
+				// Fallback to empty string or original if invalid codepoint
+			}
+		}
+		return "";
+	};
+
+	return raw.replace(/&(amp|lt|gt|quot|apos|nbsp|#\d+|#x[0-9a-fA-F]+);/gi, (match, entity) => {
+		const lower = entity.toLowerCase();
+		switch (lower) {
+			case "nbsp":
+				return " ";
+			case "lt":
+				return "<";
+			case "gt":
+				return ">";
+			case "quot":
+				return '"';
+			case "apos":
+				return "'";
+			case "amp":
+				return "&";
+			default: {
+				if (lower.startsWith("#x")) {
+					return parseCodePoint(Number.parseInt(lower.slice(2), 16));
+				}
+				if (lower.startsWith("#")) {
+					return parseCodePoint(Number(lower.slice(1)));
+				}
+				return match;
+			}
+		}
+	});
 }
 
 interface HtmlListState {
@@ -50,7 +86,7 @@ function createHtmlNormalizationState(): HtmlNormalizationState {
 	return { lists: [], openItems: [], itemHasContent: [] };
 }
 
-const HTML_TAG_REGEX = /<\/?(?:br|p|ol|ul|li)\b(?:\s[^>]*)?\s*\/?>/gi;
+const HTML_TAG_REGEX = /<\/?(?:br|p|ol|ul|li|span|text)\b(?:\s[^>]*)?\s*\/?>/gi;
 
 function htmlTagName(tag: string): string {
 	const match = /^<\/?\s*([A-Za-z][A-Za-z0-9:-]*)/.exec(tag);
@@ -96,22 +132,28 @@ function normalizeHtmlForTerminal(raw: string, state: HtmlNormalizationState = c
 		const tag = match[0];
 		const index = match.index ?? 0;
 		const textBeforeTag = normalizeHtmlEntitiesForTerminal(raw.slice(lastIndex, index));
+		const name = htmlTagName(tag);
+		// Every tag handled here is block-level EXCEPT span and text. For block-level tags,
 		// HTML formatting whitespace between block/list tags (e.g. the newlines and
 		// indentation in pretty-printed `<ul>\n  <li>…`) is not rendered content;
 		// appending it literally would leak source indentation before bullets and
-		// blank rows between items. Every tag handled here is block-level, so a
-		// whitespace-only slice is always insignificant formatting and is dropped.
-		if (textBeforeTag.trim() !== "") {
+		// blank rows between items. A whitespace-only slice is always insignificant formatting
+		// and is dropped. But for inline tags like span and text, surrounding whitespace
+		// is significant and must NOT be dropped.
+		const isInlineTag = name === "span" || name === "text";
+		if (isInlineTag || textBeforeTag.trim() !== "") {
 			output += textBeforeTag;
 			markCurrentHtmlItemContent(state, textBeforeTag);
 		}
 		lastIndex = index + tag.length;
 
-		const name = htmlTagName(tag);
 		const isClosing = /^<\//.test(tag);
 		const isSelfClosing = /\/\s*>$/.test(tag);
 
 		switch (name) {
+			case "span":
+			case "text":
+				break;
 			case "br":
 				output = appendHtmlLineBreak(output, true);
 				break;
diff --git a/packages/tui/test/markdown.test.ts b/packages/tui/test/markdown.test.ts
index ecb13a8c2..854440484 100644
--- a/packages/tui/test/markdown.test.ts
+++ b/packages/tui/test/markdown.test.ts
@@ -1216,6 +1216,41 @@ bar`,
 				"Should render HTML in code blocks",
 			).toBeTruthy();
 		});
+
+		it("should strip inline span and text HTML tags but keep their contents", () => {
+			const markdown = new Markdown("<span></span><text>▃</text>", 0, 0, defaultMarkdownTheme);
+
+			const lines = markdown.render(80);
+			const plainLines = lines.map(line => stripVTControlCharacters(line).trim());
+			const joinedPlain = plainLines.join("");
+
+			expect(joinedPlain).toBe("▃");
+		});
+
+		it("should preserve whitespace surrounding stripped inline HTML tags", () => {
+			const markdown = new Markdown("some <span>inner</span> text", 0, 0, defaultMarkdownTheme);
+
+			const lines = markdown.render(80);
+			const plainLines = lines.map(line => stripVTControlCharacters(line).trim());
+			const joinedPlain = plainLines.join("");
+
+			expect(joinedPlain).toBe("some inner text");
+		});
+
+		it("should unescape HTML entities inside and outside HTML tags", () => {
+			const markdown = new Markdown(
+				"<span>&lt;▃&gt;</span> &amp; &quot;test&quot; &#128512; &#x1F600;",
+				0,
+				0,
+				defaultMarkdownTheme,
+			);
+
+			const lines = markdown.render(80);
+			const plainLines = lines.map(line => stripVTControlCharacters(line).trim());
+			const joinedPlain = plainLines.join("");
+
+			expect(joinedPlain).toBe('<▃> & "test" 😀 😀');
+		});
 	});
 });