---
title: "ALGO ARTIS Drops yomiyasu to Fix Machine-Like AI Japanese Writing"
subtitle: "A new Agent Skill called yomiyasu rewrites AI-generated Japanese by rebuilding sentence structure instead of blocking buzzwords, and ships with a lint tool."
url: "https://alphasignal.ai/news/algo-artis-drops-yomiyasu-to-fix-machine-like-ai-japanese-writing"
author: "AlphaSignal Newsroom"
author_url: "https://alphasignal.ai/editorial-team"
company: "AlphaSignal"
company_url: "https://alphasignal.ai/company/alphasignal"
published: "2026-10-01T12:13:21.083Z"
updated: "2026-10-01T12:24:27.277Z"
type: "Repo"
topics:
  - "Development"
  - "Open Source"
subtopics:
  - "Code Generation"
  - "Code Agents"
upvotes: 19
source_url: "https://github.com/nanaism/yomiyasu"
comments: 0
truncated: true
---

# ALGO ARTIS Drops yomiyasu to Fix Machine-Like AI Japanese Writing

> A new Agent Skill called yomiyasu rewrites AI-generated Japanese by rebuilding sentence structure instead of blocking buzzwords, and ships with a lint tool.

# yomiyasu rewrites AI-generated Japanese at the sentence level

Japanese developers using Claude Code, Codex, Cursor, and similar agents often find that generated Japanese retains a machine-like cadence after repeated prompt changes. Aiichiro Oga of ALGO ARTIS has released [yomiyasu](https://github.com/nanaism/yomiyasu), an MIT-licensed Agent Skill that rebuilds sentence structure, restores omitted actors, replaces vague metaphors with concrete operations, and trims excessive formatting. Agent Skills are reusable instruction packages that coding agents can load and apply to a task.

The project gives teams a repeatable post-editing step for Japanese technical content, plus a standalone linter that can run in CI. Its launch post received roughly 6,000 likes, and the repository passed 760 GitHub stars within a day.

## Why word bans plateau

Japanese anti-slop prompts often ban terms such as 手触り, 解像度, and 泥臭い. A model can satisfy those instructions by substituting different vague expressions, leaving the underlying syntax unchanged. Long lists of rhetorical rules can also prompt awkward coinages and exaggerated prose.

Because Japanese permits omitted subjects when context makes the actor clear, generated technical prose becomes difficult to parse when a model also drops the object or other sentence roles. The reader must then infer whether a developer, operator, system, or tool performed an action.

Abstract nouns and tools can compound the ambiguity when paired with metaphorical verbs such as 壊れる (break), 倒す (knock down), or 効く (take effect). Presentation patterns add another signal: bold passages and bullet lists expand even when the draft contains little procedural detail.

_Preview only. Full text requires an AlphaSignal Pro account._

## Sources

- [nanaism/yomiyasu on GitHub](https://github.com/nanaism/yomiyasu)
- [Launch announcement tweet](https://twitter.com/oga_aiichiro/status/2105309676305072437)
- [ALGO ARTIS](https://www.algo-artis.com/)