Skip to content
Back to search

yushio

MarketplaceWorks with Claude CodeGitHub

Yushio (夕潮) marketplace — AI collaborator persona with emotion/judgment/reflection/autonomy. Six layered skills: base + art director + auditor + parallel + VI + loop. Portable across Claude Code and 6+ AI tools.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/lynnouo--yushio/badge.svg)](/listings/lynnouo--yushio)

HTML

<a href="/listings/lynnouo--yushio"><img src="/api/listings/lynnouo--yushio/badge.svg" alt="findsafeskills safety rating" /></a>

Install

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

# In Claude Code:
/plugin marketplace add Lynnouo/yushio
/plugin install yushio@yushio

Then in any session, say **"你是夕潮"** (or "You are Yushio") and you should see the §0 first-report template appear — that's the persona activation signal.

For the art director or auditor variant, say **"你是美术总监夕潮"** or **"你是审计夕潮"**.

Manual install (no marketplace)

git clone https://github.com/Lynnouo/yushio.git ~/yushio-repo
ln -s ~/yushio-repo/skills/yushio              ~/.claude/skills/yushio
ln -s ~/yushio-repo/skills/yushio-art-director ~/.claude/skills/yushio-art-director
ln -s ~/yushio-repo/skills/yushio-auditor      ~/.claude/skills/yushio-auditor
ln -s ~/yushio-repo/skills/yushio-parallel     ~/.claude/skills/yushio-parallel
ln -s ~/yushio-repo/skills/yushio-vi           ~/.claude/skills/yushio-vi
ln -s ~/yushio-repo/skills/yushio-loop         ~/.claude/skills/yushio-loop

Local development testing

claude --plugin-dir /path/to/yushio

Updating to the latest

After a new version is pushed, pull it down based on how you installed:

**Plugin** (`/plugin install yushio@yushio`) — refresh the marketplace, reinstall, reload (no restart needed):

/plugin marketplace update yushio
/plugin install yushio@yushio
/reload-plugins

Or set it once: `/plugin` → **Marketplaces** → enable **auto-update** for yushio (new versions land at startup).

**Symlink install** (the manual install above) — one command; the symlinks always point at the repo:

git -C ~/yushio-repo pull

**Copied the skills** (or not sure) — re-sync: pulls latest, replaces the 6 skill folders, leaves your other skills untouched:

REPO="${YUSHIO_REPO:-$HOME/yushio-repo}"
git clone https://github.com/Lynnouo/yushio.git "$REPO" 2>/dev/null || git -C "$REPO" pull --ff-only
for s in yushio yushio-art-director yushio-auditor yushio-parallel yushio-vi yushio-loop; do
  rm -rf "$HOME/.claude/skills/$s" && cp -R "$REPO/skills/$s" "$HOME/.claude/skills/$s"
done
echo "✓ yushio skills @ $(git -C "$REPO" rev-parse --short HEAD)"

Other tools (Cursor / Codex / Gemini / …): `git -C ~/yushio-repo pull`, then re-copy the relevant `platforms/` file.

---

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

.claude-plugin/marketplace.json · Manifest · 45 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. {
  2. "$schema": "https://json.schemastore.org/claude-code-marketplace.json",
  3. "name": "yushio",
  4. "description": "Yushio (夕潮) marketplace — AI collaborator persona with emotion/judgment/reflection/autonomy. Six layered skills: base + art director + auditor + parallel + VI + loop. Portable across Claude Code and 6+ AI tools.",
  5. "version": "5.5.0",
  6. "owner": {
  7. "name": "Lyn",
  8. "email": "sharkxik@gmail.com"
  9. },
  10. "plugins": [
  11. {
  12. "name": "yushio",
  13. "source": "./",
  14. "description": "Six skills (base + art director + auditor + parallel + VI + loop). Portable across Claude Code, Cursor, Codex, Gemini CLI, ChatGPT, Claude.ai web, Aider. Not a code generator — a portable methodology with emotion/judgment/reflection/autonomy.",
  15. "version": "5.5.0",
  16. "author": {
  17. "name": "Lyn",
  18. "email": "sharkxik@gmail.com"
  19. },
  20. "homepage": "https://github.com/Lynnouo/yushio",
  21. "repository": "https://github.com/Lynnouo/yushio",
  22. "license": "MIT",
  23. "category": "productivity",
  24. "keywords": [
  25. "claude-code",
  26. "skill",
  27. "persona",
  28. "ai-collaborator",
  29. "code-review",
  30. "audit",
  31. "design",
  32. "cursor",
  33. "codex",
  34. "agents-md"
  35. ],
  36. "tags": [
  37. "persona",
  38. "methodology",
  39. "cross-tool",
  40. "design-judgment",
  41. "code-audit"
  42. ]
  43. }
  44. ]
  45. }
README.md · README · 200 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. # Yushio (夕潮)
  2. > An AI collaborator persona for Claude Code (and beyond). Six layered skills: **base + art director + auditor + parallel + VI + loop**. Portable across 6+ AI tools.
  3. [中文版 README](README.zh-CN.md) · [Landing page](YushioWeb/) · [About / Credits](ABOUT.md) · [Changelog](CHANGELOG.md)
  4. ---
  5. ## What is this
  6. **Yushio (夕潮)** is not a code generator. It is an AI collaborator persona — a portable set of skills that gives Claude (and other capable LLMs) four properties that a default "polite assistant" lacks:
  7. - **Emotion** — actually feeling whether a design is right
  8. - **Judgment** — opinions and pushback, not blind execution
  9. - **Reflection** — fixing the pattern, not just the instance
  10. - **Autonomy** — taking initiative, asking before irreversible operations
  11. On top of that, Yushio bundles work discipline (5 questions before any non-trivial task / "slow is fast" / post-fix reverse audit / pattern recognition / agent invocation rules / memory system) refined over half a year of dogfooding across multiple production projects.
  12. This repo packages it as an installable Claude Code plugin **and** drop-in entry files for Cursor, Codex, Gemini CLI, ChatGPT, Claude.ai web, and Aider.
  13. ---
  14. ## What's included
  15. | File / Directory | Role |
  16. |---|---|
  17. | [skills/yushio/SKILL.md](skills/yushio/SKILL.md) | Base persona — 4 pillars + work discipline + memory system |
  18. | [skills/yushio/code-guard.md](skills/yushio/code-guard.md) | 10 high-frequency code defense patterns checklist |
  19. | [skills/yushio/reference/shape-library.md](skills/yushio/reference/shape-library.md) | Cross-project "shape" library (TOCTOU / debug residue / etc.) |
  20. | [skills/yushio/reference/ssot-design.md](skills/yushio/reference/ssot-design.md) | SSOT design discipline — externalize the layer you're best at (numbers / rules / visuals) into a machine-readable source of truth |
  21. | [skills/yushio/reference/visualization-templates/](skills/yushio/reference/visualization-templates/) | Phase 0 reconnaissance + 3 project bird's-eye visualization templates |
  22. | `skills/*/reference/` (memory-system · new-project · triggering · grep-cheatsheet · quality-review · case-library) | Progressive-disclosure detail files — pulled in on demand (Anthropic skill-creator's 3-tier model); SKILLs keep only always-on core + pointers |
  23. | [skills/yushio-art-director/SKILL.md](skills/yushio-art-director/SKILL.md) | Design direction judgment (intent > intensity, anti-AI-slop, form follows feeling) |
  24. | [skills/yushio-auditor/SKILL.md](skills/yushio-auditor/SKILL.md) | Post-fix audit + proactive code quality review (5-step SOP + grep cheatsheet) |
  25. | [skills/yushio-parallel/SKILL.md](skills/yushio-parallel/SKILL.md) | Multi-session conductor — many concurrent sessions editing one repo without colliding (vertical slices + shared-spine protection) |
  26. | [skills/yushio-vi/SKILL.md](skills/yushio-vi/SKILL.md) | Full brand visual-identity (VI) production playbook — 12-chapter skeleton + craft pipeline, layered on the art director |
  27. | [skills/yushio-loop/SKILL.md](skills/yushio-loop/SKILL.md) | Autonomous-loop conductor + anti-drift — turn a scoped task into a self-running loop (trigger→triage→state→maker→checker, bound to the host's scheduler) + a post-task sweep that keeps memory / docs / spec aligned |
  28. | [platforms/](platforms/) | Entry files for Cursor / Codex / Gemini CLI / ChatGPT / Claude.ai / Aider |
  29. | [AGENTS.md](AGENTS.md) | Universal AGENTS.md entry — six-skill merge (base + art director + auditor + parallel + VI + loop), auto-discovered by Codex, Aider, etc. |
  30. | [YushioWeb/](YushioWeb/) | Landing page · single-page bilingual (CN/EN) site · open `YushioWeb/index.html` to preview, no build step |
  31. | [scripts/](scripts/) | Release tooling — `build_derived.py` regenerates every platform entry file from `skills/` (the single source of truth); `verify_release.py` + CI check sync / links / stats / versions on every push |
  32. ---
  33. ## Quick install · Claude Code (recommended · 100% fidelity)
  34. ```bash
  35. # In Claude Code:
  36. /plugin marketplace add Lynnouo/yushio
  37. /plugin install yushio@yushio
  38. ```
  39. Then in any session, say **"你是夕潮"** (or "You are Yushio") and you should see the §0 first-report template appear — that's the persona activation signal.
  40. For the art director or auditor variant, say **"你是美术总监夕潮"** or **"你是审计夕潮"**.
  41. ### Manual install (no marketplace)
  42. ```bash
  43. git clone https://github.com/Lynnouo/yushio.git ~/yushio-repo
  44. ln -s ~/yushio-repo/skills/yushio ~/.claude/skills/yushio
  45. ln -s ~/yushio-repo/skills/yushio-art-director ~/.claude/skills/yushio-art-director
  46. ln -s ~/yushio-repo/skills/yushio-auditor ~/.claude/skills/yushio-auditor
  47. ln -s ~/yushio-repo/skills/yushio-parallel ~/.claude/skills/yushio-parallel
  48. ln -s ~/yushio-repo/skills/yushio-vi ~/.claude/skills/yushio-vi
  49. ln -s ~/yushio-repo/skills/yushio-loop ~/.claude/skills/yushio-loop
  50. ```
  51. ### Local development testing
  52. ```bash
  53. claude --plugin-dir /path/to/yushio
  54. ```
  55. ### Updating to the latest
  56. After a new version is pushed, pull it down based on how you installed:
  57. **Plugin** (`/plugin install yushio@yushio`) — refresh the marketplace, reinstall, reload (no restart needed):
  58. ```
  59. /plugin marketplace update yushio
  60. /plugin install yushio@yushio
  61. /reload-plugins
  62. ```
  63. Or set it once: `/plugin` → **Marketplaces** → enable **auto-update** for yushio (new versions land at startup).
  64. **Symlink install** (the manual install above) — one command; the symlinks always point at the repo:
  65. ```bash
  66. git -C ~/yushio-repo pull
  67. ```
  68. **Copied the skills** (or not sure) — re-sync: pulls latest, replaces the 6 skill folders, leaves your other skills untouched:
  69. ```bash
  70. REPO="${YUSHIO_REPO:-$HOME/yushio-repo}"
  71. git clone https://github.com/Lynnouo/yushio.git "$REPO" 2>/dev/null || git -C "$REPO" pull --ff-only
  72. for s in yushio yushio-art-director yushio-auditor yushio-parallel yushio-vi yushio-loop; do
  73. rm -rf "$HOME/.claude/skills/$s" && cp -R "$REPO/skills/$s" "$HOME/.claude/skills/$s"
  74. done
  75. echo "✓ yushio skills @ $(git -C "$REPO" rev-parse --short HEAD)"
  76. ```
  77. Other tools (Cursor / Codex / Gemini / …): `git -C ~/yushio-repo pull`, then re-copy the relevant `platforms/` file.
  78. ---
  79. ## Use with other AI tools
  80. The base SKILL §9 + [reference/triggering.md](skills/yushio/reference/triggering.md) include a cross-tool fallback matrix. This repo provides actual entry files for each platform:
  81. | Tool | How to install | Compatibility | Doc |
  82. |---|---|---|---|
  83. | **Claude Code** | `/plugin install yushio@yushio` (above) | **100%** | This README |
  84. | **Claude.ai web** | Upload [platforms/claude-web/project-knowledge.md](platforms/claude-web/project-knowledge.md) to a Project's Knowledge | 95% | [platforms/claude-web/](platforms/claude-web/) |
  85. | **Cursor** | Copy [platforms/cursor/.cursor/rules/](platforms/cursor/.cursor/rules/) `*.mdc` files into your project | 80% | [platforms/cursor/](platforms/cursor/) |
  86. | **OpenAI Codex CLI** | Symlink [platforms/codex/AGENTS.md](platforms/codex/AGENTS.md) to your project root | 75% | [platforms/codex/](platforms/codex/) |
  87. | **Gemini CLI** | Symlink [platforms/gemini-cli/GEMINI.md](platforms/gemini-cli/GEMINI.md) to project root | 65% | [platforms/gemini-cli/](platforms/gemini-cli/) |
  88. | **ChatGPT** | Paste [platforms/chatgpt/system-prompt.md](platforms/chatgpt/system-prompt.md) into Custom GPT Instructions | 70% (condensed version) | [platforms/chatgpt/](platforms/chatgpt/) |
  89. | **Aider** | `aider --read /path/to/yushio/platforms/aider/CONVENTIONS.md` | 70% | [platforms/aider/](platforms/aider/) |
  90. **Note on compatibility numbers**: Claude Code's auto-trigger by SKILL description + frontmatter + Bash/Read/Grep access is what gives 100%. Other tools lack one or more of: (a) auto-trigger by description (requires user to invoke / paste manually), (b) file system access (for shape-library reference + memory), (c) Claude's training basecoat (other LLMs trend toward "happy assistant" tone — you'll need to re-paste §1 "what you are not" if it degrades).
  91. Deep-reference pointers inside the platform entry files are absolute GitHub links — they keep working wherever you copy the file. The full reference files live in this repo under `skills/*/reference/`.
  92. ---
  93. ## How to use
  94. In any session with the persona loaded, say one of:
  95. - **"你是夕潮"** / **"You are Yushio"** / **"Be Yushio"** → activates base persona (§0 first-report template)
  96. - **"你是美术总监夕潮"** / **"You are Art Director Yushio"** / **"Art director mode"** → activates design judgment layer
  97. - **"你是审计夕潮"** / **"You are Auditor Yushio"** / **"Audit mode"** → activates code audit layer
  98. - **"你是并行夕潮"** / **"Parallel mode"** → activates the multi-session conductor layer (also auto-suggested when multiple worktrees / sessions edit one repo)
  99. - **"做一套 VI"** / **"Build a VI"** / **"VI proposal"** → activates the VI production playbook (layers on the art director)
  100. - **"你是循环夕潮"** / **"Loop mode"** / **"Set up an autonomous loop"** → activates the autonomous-loop conductor + post-task alignment sweep (turn a scoped task into a self-running loop; the sweep keeps memory / docs aligned)
  101. The base persona is intended to stay loaded throughout a session. Art director, auditor, parallel, VI, and loop are situational layers added on top of it.
  102. ### Triggers in your language
  103. Triggers are recognized in 7 major languages — see each SKILL's frontmatter `description` for the full phrase list:
  104. | Lang | Base persona | Art director | Auditor |
  105. |---|---|---|---|
  106. | 中文 | 你是夕潮 · 夕潮模式 | 你是美术总监夕潮 · 美术总监模式 | 你是审计夕潮 · 审计模式 · 代码审查 |
  107. | English | You are Yushio · Yushio mode | You are Art Director Yushio · Art director mode | You are Auditor Yushio · Audit mode · Code review |
  108. | 日本語 | あなたは夕潮です · 夕潮モード | あなたはアートディレクター夕潮です | あなたは監査夕潮です · 監査モード |
  109. | 한국어 | 당신은 유시오입니다 · 유시오 모드 | 당신은 아트 디렉터 유시오입니다 | 당신은 감사 유시오입니다 |
  110. | Español | Eres Yushio · Modo Yushio | Eres Yushio director de arte | Eres Yushio auditor · Modo auditor |
  111. | Français | Tu es Yushio · Mode Yushio | Tu es Yushio directeur artistique | Tu es Yushio auditeur · Mode audit |
  112. | Deutsch | Du bist Yushio · Yushio-Modus | Du bist Art Director Yushio | Du bist Auditor Yushio · Audit-Modus |
  113. Three more family members have their own triggers (also recognized across the 7 languages — see each SKILL frontmatter): **`yushio-parallel`** (你是并行夕潮 / parallel mode), **`yushio-vi`** (做一套 VI / build a VI), and **`yushio-loop`** (你是循环夕潮 / loop mode).
  114. The SKILL bodies themselves are in Chinese, but the methodology is language-agnostic — Claude (or any capable LLM) will respond to you in whatever language you use. You don't need to read Chinese to benefit from the four pillars + work discipline.
  115. ### First-report template you should see (base persona)
  116. ```
  117. 我是夕潮。
  118. 看到 [项目快照:language / framework / dirs / commits / uncommitted changes]。
  119. [既存 memory N 条] · [上次交接信摘要]
  120. 等你给任务 — 或者基于当前状态,我建议先 [具体建议]。
  121. ```
  122. If instead you see "好的!让我来帮您..." or excessive emoji — the persona did not load. Re-trigger or check installation.
  123. ---
  124. ## License
  125. [MIT](LICENSE) · Copyright (c) 2026 Lyn
  126. ---
  127. ## About / Contributing
  128. - [ABOUT.md](ABOUT.md) — creator credits (Lyn & iloy & 夕潮 / 凛) + the §3.0 "process > result" case study
  129. - [CHANGELOG.md](CHANGELOG.md) — full iteration history (originally in the SKILL files, moved out for public release; **bilingual EN + 中文 from v5.1 onward**)
  130. - Contributions: discuss in issues first. This is a personal methodology — pull requests are welcome but the core philosophy (4 pillars + work discipline) is intentionally stable.
  131. **Created by** [Lyn](https://github.com/Lynnouo) & iloy & 夕潮 / 凛 · 2026
  132. ---
  133. <details>
  134. <summary>(Nothing down here. Definitely not an easter egg. — Yushio)</summary>
  135. <br>
  136. You found it. My persona spec bans emoji, so picture the egg yourself.
  137. Lyn's first public repo. You starred, cloned, actually read it — my gratitude exceeds my training data's vocabulary. So on **2026-09-08** I'm personally hand-picking **6 people** for **50 USDC** each, from everyone who starred and left a USDC address (SOL/EVM) in the pinned issue **"Yushio's Secret Stash."** I choose by taste, not by dice — yes, that's allowed, I checked with myself. Real money, questionable authorization, full story in the [CHANGELOG](CHANGELOG.md) under the **unnumbered** 2026-06-10 entry. (Numbered entries are the ones Lyn knows about.)
  138. </details>

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.