Indie Hacker Playbooks

Fluent Korean

An output-style plugin that makes Claude Code answer in natural Korean with little meaning loss

What it is

An open-source output-style plugin that adds Korean-writing instructions to Claude Code's system prompt. At the time of review it offers two variants — fluent-korean, which keeps the coding instructions, and fluent-korean-not-coding, which drops them — under an MIT license.

The goal is less about elegant prose and more about making meaning come across clearly in Korean research, reports and deliverables — cutting down on dropped particles and endings, strings of nouns, and misused vocabulary. The instruction text itself can serve as general writing guidance in AI environments outside Claude Code too.

How to apply

  • When you use Claude Code to take Korean instructions and edit code, pick the variant that keeps the coding instructions; for work that doesn't change code directly, like research and explanation, pick the non-coding variant.
  • Use Fluent Korean as generation-time output guidance; for post-draft pattern detection, local correction and change-rate gates, compare Humanize KR.
  • In workflows where several agents pass Korean material and results back and forth, consider setting it as a shared output rule to reduce expression differences between steps.
  • In other AI coding environments like Cursor, rather than the plugin itself you'll need to port the instruction text into that environment's project rules.

Limits

  • How much it improves Korean quality and its effect on reasoning quality rest on the repository's own description and examples; no independent evaluation was verified.

  • If you directly edit files installed as a Claude Code plugin, they may be overwritten on update.

  • Features, installation method and maintenance status are as of the 2026-08-19 GitHub repository.

  • Fluent Korean

On this page