snflkd/fluent-korean

What it solves

Many AI models, particularly coding agents like Claude Code, often produce "broken machine-Korean" characterized by dropped particles, unnatural sentence endings, and telegraphic noun strings. This happens because these agents are often optimized for brevity to save tokens, which degrades the quality of Korean output and can lead to misunderstandings or even negatively impact the model's reasoning process.

How it works

It provides a set of output-style guidelines (system prompts) that instruct the LLM to prioritize clear, unambiguous, and natural Korean over elegant prose. It functions as a pre-defined rule set added to the system prompt to regulate how the model generates Korean text.

Users can apply these guidelines in several ways:

• Claude Code CLI: Install as a plugin via the marketplace.
• Claude Web/Desktop: Paste the guideline text into personal or project instructions.
• Other Environments: Use the provided Markdown files as plain writing instructions for any LLM.

Who it’s for

Developers and non-developers who use LLMs (especially Claude) and require high-quality, natural Korean output for reports, coding instructions, or general communication.

Highlights

• Two Variants: Includes a version that maintains coding instructions and one without them for non-coding tasks.
• Customizable: Offers optional blocks of text to adjust the AI's tone (e.g., making it more polite, simplifying language for beginners, or forcing all output to be in Korean).
• Flexible Deployment: Works as a Claude Code plugin or as raw text for any AI environment.
• Human-Curated: Developed by a Korean literature major to ensure linguistic accuracy.