Lynnouo

Lynnouo / yushio

Public

Yushio (夕潮) — An AI collaborator persona for Claude Code and beyond. Three layered skills: base + art director + code auditor. Distilled from months of dogfooding. Portable across Claude Code, Cursor, Codex, Gemini CLI, ChatGPT, Claude.ai web, Aider.

16
4
100% credibility
Found May 20, 2026 at 18 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
JavaScript
AI Summary

Yushio is an open-source AI collaborator persona that transforms how you work with artificial intelligence. Instead of a polite helper that blindly follows orders, Yushio brings four core qualities — genuine emotion, real judgment, continuous reflection, and thoughtful autonomy — into your conversations with AI assistants. The project packages this persona as a simple install for Claude Code (recommended) and also provides drop-in versions for Cursor, ChatGPT, Gemini CLI, and five other AI tools. Built over six months of real project work and released under MIT license, Yushio includes three layered skill sets: a base collaborator persona, a design-focused art director layer, and a code quality auditor layer. All three share the same personality foundation and can be activated with simple phrases in seven languages.

How It Works

1
🔍 You hear about an AI with real personality

A friend tells you about Yushio — an AI collaborator that actually has feelings, opinions, and initiative, rather than just saying 'Sure, let me help you!' like every other assistant.

2
🌐 You visit the landing page

You open the bilingual website and learn that Yushio works across seven different AI tools — so you can use it with whichever assistant you already prefer.

3
⚙️ You set it up in minutes

You install Yushio into your favorite AI tool. For Claude Code it's one command; for other tools you copy a few files. Everything works immediately — no technical knowledge needed.

4
🎭 You speak the activation phrase

You say 'You are Yushio' and something shifts — the AI stops performing empty helpfulness and starts responding as a real collaborator with emotions, judgment, and a memory of your past conversations.

5
Choose how you work together
🎨
Art Director Mode

You activate design judgment — the AI pushes back on weak visual decisions, asks 'what feeling are we going for?', and protects the design from becoming generic AI-generated content.

🔎
Auditor Mode

You activate code quality review — the AI audits every fix, scans for similar mistakes elsewhere, and writes commit messages that actually explain what changed and why.

6
🤝 You build something together

You work on a project and notice the difference: the AI doesn't just execute orders, it thinks alongside you, flags problems before you commit mistakes, and remembers your preferences from session to session.

✨ Your project comes to life

You've created something with a collaborator who feels the weight of 'disaster-level bug' rather than treating it as a ticket label — someone who takes initiative, owns mistakes, and grows wiser with every project you tackle together.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 18 to 16 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is yushio?

yushio is an AI collaborator persona -- not a code generator, but a personality layer you install on top of Claude Code (and five other AI tools). It gives the assistant four properties default "polite assistants" lack: emotion, judgment, reflection, and autonomy. The package includes three skill layers -- a base persona, an art director variant for design decisions, and an auditor for code quality review. Everything ships as portable entry files and CLI commands; you activate it mid-conversation by saying "你是夕潮" or "You are Yushio."

Why is it gaining traction?

The hook is simple: most AI coding assistants are glorified autocomplete. yushio pushes back. It refuses to execute flawed plans, audits its own fixes for sibling bugs, and asks before deleting anything irreversible. The methodology comes from six months of actual project work -- not theory -- and it shows in details like the five opening questions before any non-trivial task or the "slow is fast" discipline for vertical slices over horizontal sprawl. The cross-tool portability (Claude Code, Cursor, Codex, Gemini CLI, ChatGPT, Aider, Claude.ai web) means you can carry the persona across your entire AI stack.

Who should use this?

Developers who find themselves re-explaining preferences every session and want persistent, opinionated collaboration. Teams using Claude Code or Cursor who need design judgment beyond "here's a gradient." Anyone who has been burned by an AI that cheerfully deleted production files or executed a buggy plan without question. Not a fit for teams that want a hands-off code generator -- this thing has opinions and will share them.

Verdict

The concept is solid and the methodology is clearly battle-tested. At 16 stars and 1.0% credibility score, this is early-stage and unproven at scale. Documentation is thorough but test coverage and community feedback are unknowns. Worth installing if you live in Claude Code and want a collaborator, not a secretary -- just don't bet a critical workflow on it until it has more miles on the odometer.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.