cn-linebreak

Chinese web pages suffer from a stubborn typographic habit: the browser splits lines one character at a time. A lone character gets stranded at the line end, a closing period dangles at the line start, and a product name is cut mid-phrase. On decks, landing pages, dashboards, and long-form copy where reading rhythm matters, these small flaws immediately erode perceived quality. cn-linebreak turns the rules from a Chinese line-breaking repair guide into executable, static checks so you can catch and fix them before shipping.

The tool works in two layers. Audit mode scans for explicit wbr-produced orphan lines, line-start and line-end punctuation violations (CLReq / GB-T 15834 prohibited rules), whether word-break: keep-all actually covers h1, h2, p and other text elements, whether white-space: nowrap leaks onto body copy, whether configured protected phrases are split by wbr, and whether long passages have zero break points, then emits a severity-graded JSON report. Fix mode conservatively inserts wbr at natural pauses such as commas, full stops, semicolons, and enumeration marks, while skipping script, style, pre, code, textarea, .no-break zones, and any hits in a custom protected-phrase dictionary, outputting repaired HTML for human review.

The package is zero-dependency pure JavaScript (CommonJS plus one ESM plugin entry) and runs on Node 18 or later. It also ships as a DeepSeek Harness standard plugin: once installed, the agent can invoke the cn_linebreak_audit tool directly by passing HTML and an audit or fix mode. If you own typographic quality for Chinese copy, need a CI gate for line-breaking problems, or want a lightweight pre-publish check, cn-linebreak is a practical, plug-and-play starting point.