Yomiyasu - Setup Guide¶
Source: nanaism/yomiyasu via skills.sh - 6,330 installs; first seen Oct 8, 2026 (evening sweep)
GitHub: nanaism/yomiyasu (1,768 stars, MIT license; very active - pushed Oct 8, 2026; skill at skills/yomiyasu/SKILL.md, ~66 KB plus a slop-catalog reference)
Category: Writing / Content (Japanese)
Quality Tier: ๐ก Beta (strong community traction; MIT; very active; all sampled verdicts Pass; Japanese-language workflows only, verified Oct 8, 2026)
Yomiyasu (่ชญใฟใใใ, "readability") is an agent skill for refining AI-generated Japanese into natural Japanese. It targets the specific tells of machine-written text: unnatural metaphors, ambiguous subject-predicate relationships, skewed syntax, needless decoration, and copy-style tone. It is built for technical articles, specifications, pull request descriptions, internal reports, and essays or note-style posts.
The skill's governing rule is strict: preserve the original meaning and add no information. Before rewriting a sentence it fixes four properties - the claim being made, the weighting and emphasis, the strength of assertion (definite versus tentative), and the sentence's function (evaluation, explanation, request, plan) - and every other rule applies only within those bounds.
Installation¶
# Recommended
npx skills add nanaism/yomiyasu
# Update later
npx skills update yomiyasu
Claude Code users can install through the plugin marketplace path described in the repository README. The skill definition lives in a single file (skills/yomiyasu/SKILL.md), so manual installation is a one-file copy.
What It Provides¶
| Capability | How |
|---|---|
| Meaning preservation | Four checks (claim, weighting, assertive strength, sentence function) run before and after every edit |
| Sentence-ending discipline | Chooses endings by document stance (advice / rules / explanation) so function and politeness level stay correct, without flattening every ending to one style |
| Paragraph and logic structure | Fixes ambiguous connectors, "preview-only" sentences, mixed topics within a paragraph, and broken subject-predicate agreement |
| Subject and object clarity | Restores who-does-what only from stated context; never invents actors or converts system behavior into human action |
| De-personification and metaphor fixes | Replaces AI-typical metaphorical verbs with everyday words while keeping the original metaphor's connotation |
| Self-labeling and negation | Handles "what matters is..." lead-ins and A-not-B contrasts without erasing emphasis or ranking |
| Terminology and punctuation | Keeps technical terms, identifiers, API names, and enum values intact; tunes sentence length (~30-45 characters) and comma placement |
| Structured output | Every pass reports what changed (ๅคใใใจใใ), what AI-ish phrasing remains (ๆฎใใAIใฃใฝใใจใใ), and points to confirm with the writer (ๆธใๆใซ็ขบใใใใ็น) |
Why This Matters for Hermes Agents¶
Agents generate Japanese constantly - product docs, release notes, replies, marketing copy - and the gap between "grammatically correct" and "reads like a person wrote it" is exactly where output gets rejected. This skill packages a professional-grade refinement pass as a repeatable workflow with explicit guardrails: meaning is preserved, nothing is invented, and the agent flags rather than guesses when context is missing. It gives Japanese output the same anti-slop discipline that English-facing tooling gives English.
Usage¶
| You say | What happens |
|---|---|
| ใใใฎๆ็ซ ใ่ชญใฟใใใใใฆใ (make this easier to read) | Full refinement pass under the meaning-preservation rules |
| ใAIใฃใฝใใใชใใใฆใ (remove the AI feel) | Targets the slop catalog: metaphors, sameness, stiff compounds, copy-style tone |
| ใ่ช็ถใชๆฅๆฌ่ชใซใใฆใ (make it natural Japanese) | Rewrites toward natural phrasing while keeping identifiers and terms |
| ใๆ็ซ ใ่ฑ่ญใใฆใ (de-odorize the text) | The full anti-AI-smell pass for article drafts |
Note: the skill warns it can conflict with other Japanese proofreading skills; if outputs degrade when several are active, disable the similar ones.
Verification¶
# Confirm installation
npx skills list | grep yomiyasu
# Update to the latest rules
npx skills update yomiyasu
The repository publishes versioned releases (see its GitHub releases page) - the rule set is actively evolving.
Security¶
skills.sh verdicts for yomiyasu (verified Oct 8, 2026):
| Skill | Gen Agent Trust Hub | Socket | Snyk |
|---|---|---|---|
| yomiyasu | Pass | Pass | Pass |
The skill is instruction-only: it edits text and reads a bundled reference, with no network calls or command execution in its workflow.
Related¶
- 01coder Agent Skills - Content, Publishing & Security Toolkit Setup - Chinese-language content toolkit with a similar language-specific scope
- Callicrate Skills - AGENTS.md & Documentation Authoring Setup - documentation authoring discipline for agents
- Skills Catalog | Skills Marketplace
Pro Tips¶
- Give it the destination. Whether the text is for a blog, a spec, or a report changes the ending style it chooses; state the purpose and reader when you can.
- Expect flagged uncertainty, not silent guesses. Unknown actor or unclear relationship? It asks (maximum two questions) instead of inventing.
- Do not mix with similar proofreading skills. The README explicitly warns of conflicting instructions between competing Japanese refinement skills.
- Technical identifiers survive. API names, keys, types, enum values, and numbers are preserved; the pass will not mangle code-adjacent text.
- Use the three-part output. The changes list, the remaining-slop list, and the confirmation list are designed to be reviewed quickly by a human editor.
โ Skills Catalog | Skills Marketplace โ Powered by CorpusIQ
Integration with CorpusIQ¶
CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.
This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.