scrooge-mode

Same answer, fewer tokens — KO-first LLM output-compression skill for Claude Code & Codex. A Korean-native caveman alternative, measured on real session output_tokens.

GitHub

Official source

Start locally with the CLI guide