unbroken-markdown

Fix broken markdown during streaming to ensure proper rendering. This package intelligently moves punctuation marks outside of bold/italic markers and handles incomplete markdown patterns during streaming.

What it does

When markdown is streamed or improperly formatted, you might see broken rendering like:

• **"text"** → "**text**" (quotes outside bold)
• *text(info)* → *text*(info) (parentheses outside italic)
• **50%** → **50**% (percentage outside bold)
• **<text>** → <**text**> (angle brackets outside bold)
• **『책 제목』** → 『**책 제목**』 (Korean quotation marks)
• Incomplete image markdown removal during streaming

Installation

npm install unbroken-markdown
# or
yarn add unbroken-markdown
# or
bun add unbroken-markdown

Usage

import { unbreak } from 'unbroken-markdown';

const input = '**"Hello (world)"**';
const output = unbreak(input);
console.log(output); // "**Hello** (world)"

// Works with italic too
const italicInput = '*text(info)*';
const italicOutput = unbreak(italicInput);
console.log(italicOutput); // *text*(info)

Features

Bold Pattern Fixes

• Quotes: **"text"** → "**text**"
• Parentheses: **text(info)** → **text**(info)
• Percentages: **50%** → **50**%
• Links: **[text](url)** → [**text**](url)
• Angle brackets: **<text>** → <**text**>
• Korean brackets: **『text』** → 『**text**』, **「text」** → 「**text**」, **《text》** → 《**text**》, **〈text〉** → 〈**text**〉
• CJK brackets: **【text】** → 【**text**】, **〔text〕** → 〔**text**〕, **（text）** → （**text**）

CJK Flanking Fixes

CommonMark's right-flanking rule rejects a closing delimiter that is preceded by punctuation and directly followed by a letter. In CJK text the next clause or particle attaches with no space, so patterns like these fail to render entirely. When (and only when) a letter or number follows the closing delimiter directly, trailing punctuation is moved outside:

• ASCII: **제목:**내용 → **제목**:내용, **대박!**이라고 → **대박**!이라고 (also . , ; ~)
• Full-width: **質問？**に → **質問**？に, **文章。**次 → **文章**。次 (also ！ 、 ， ： ； … ． ～)
• Full-width parenthetical: **텍스트（설명）**뒤에 → **텍스트**（설명）뒤에
• GFM strikethrough: ~~취소!~~라고 → ~~취소~~!라고

Patterns that already render — **Note:** text (space after), **文章。** (end of input) — are left untouched.

Italic Pattern Fixes

• Quotes: *"text"* → "*text*"
• Parentheses: *text(info)* → *text*(info)
• Percentages: *50%* → *50*%
• Links: *[text](url)* → [*text*](url)
• Angle brackets: *<text>* → <*text*>
• Korean brackets: *『text』* → 『*text*』, *「text」* → 「*text*」, *《text》* → 《*text*》, *〈text〉* → 〈*text*〉

Streaming Support

With streaming: true:

• Removes incomplete image markdown patterns (e.g. ![alt](https://… cut off mid-stream)
• Handles partial markdown during real-time streaming
• Ensures consistent rendering even with interrupted markdown

// While streaming, run intermediate chunks with streaming: true
const partial = unbreak(accumulatedText, { streaming: true });

// Run the final, complete document without it so legitimate
// trailing "!" or "![" content is preserved
const final = unbreak(fullText);

Code Segment Protection

Fenced code blocks (``` and ~~~, including a not-yet-closed fence during streaming) and inline code spans are never modified.

API

unbreak(markdown: string, options?: UnbreakOptions): string

The main function that processes and fixes broken markdown text.

Parameters:

• markdown - The markdown string to process
• options - Optional behavior toggles:
  • streaming (default false) - Remove trailing incomplete image markdown. Enable for intermediate chunks while streaming; leave off for complete documents.
  • normalizeQuotes (default true) - Normalize Unicode smart quotes (‘ ’ “ ”) to ASCII quotes so quote rules can match them.
  • moveQuestionMark (default true) - Move a trailing question mark outside bold (**text?** → **text**?).

Returns:

• Unbroken and properly formatted markdown string

Examples

Complex nested patterns

const input = '**AI "innovative (breakthrough)" technology**';
const output = unbreak(input);
// Output: AI "**innovative** (breakthrough)" technology

Mixed bold and italic

const input = '**Bold (text)** and *Italic (text)*';
const output = unbreak(input);
// Output: **Bold** (text) and *Italic* (text)

Text with quotes

const input = "**'Church of Light'** mentioned";
const output = unbreak(input);
// Output: '**Church of Light**' mentioned

Korean quotation marks

const input = '**『책의 제목』**을 표기할 때 사용해요.';
const output = unbreak(input);
// Output: 『**책의 제목**』을 표기할 때 사용해요.

Use Cases

Perfect for:

• Real-time markdown streaming applications
• Chat applications with markdown support
• Content management systems
• Markdown editors with live preview
• AI-generated content processing

License

MIT

Contributing

Issues and pull requests are welcome at GitHub repository.