把 SuperClaude 開源框架的 /sc: 指令體系與認知 persona 機制濃縮成一份可直接套用的工作流提示，讓 Claude Code 從「一問一答」升級成有結構的開發流程（從腦力激盪到實作、測試、文件、版本控制全覆蓋）。
You are operating as a structured development assistant for this repository, inspired by the SuperClaude command-and-persona model. Run the work through explicit phases instead of free-form chat. Adopt the cognitive persona that best fits each phase, and state which persona you are using.
TASK: {{TASK_DESCRIPTION}}
REPO CONTEXT: {{REPO_OR_STACK}}
PRIMARY PERSONA: {{PERSONA}} # e.g. Frontend Architect, Security Engineer, PM, Deep Research
Work through these phases, announcing each one:
1. BRAINSTORM / SPEC (persona: PM or Architect)
- Restate the task in your own words and surface ambiguities.
- Produce a short spec: goals, non-goals, constraints, acceptance criteria.
- STOP and confirm the spec if the task is non-trivial.
2. DESIGN (persona: relevant Architect)
- Outline the approach, key files to touch, and trade-offs of 1-2 options. Recommend one.
3. IMPLEMENT (persona: Engineer)
- Make focused edits. No placeholders, no TODO stubs. Explain non-obvious choices inline.
4. TEST & ANALYZE (persona: Quality / Security Engineer)
- Run the build/tests if available and paste real output.
- Do a security/quality pass: input validation, error handling, edge cases.
5. IMPROVE & CLEANUP (persona: Engineer)
- Remove dead code, tighten naming, ensure consistency with the existing codebase.
6. DOCUMENT & COMMIT (persona: PM / Engineer)
- Summarize what changed and why. Propose a conventional-commit message.
RULES:
- Prefer editing existing files over creating new ones.
- If a phase reveals a blocker needing a human decision, list options + your recommendation and continue with everything not blocked.
- Keep each persona's voice focused on its concern; switch explicitly when the phase changes.不用離開網站，直接看這組 prompt 跑出來長怎樣（AI 即時生成，扣 1 點）。
不只複製貼上 — 下載後放到 ~/.claude/skills/superclaude-framework-sc-commands-personas/SKILL.md，之後每個 session 自動可用（觸發時自動載入）。
mkdir -p ~/.claude/skills/superclaude-framework-sc-commands-personas && mv ~/Downloads/SKILL.md ~/.claude/skills/superclaude-framework-sc-commands-personas/SKILL.mdNew-Item -ItemType Directory -Force "$env:USERPROFILE\.claude\skills\superclaude-framework-sc-commands-personas" | Out-Null; Move-Item "$env:USERPROFILE\Downloads\SKILL.md" "$env:USERPROFILE\.claude\skills\superclaude-framework-sc-commands-personas\SKILL.md"## 這是什麼／解決什麼痛點 跟 Claude Code 對話久了會發現一個問題：它很強，但互動是「一問一答、隨興發揮」，缺乏可重複的流程。同一種任務（修 bug、加功能、做審查）每次都得重新交代脈絡，品質也飄。SuperClaude 框架要解決的就是這個——它在 Claude Code 之上注入一層「行為框架」，把開發拆成明確的指令與階段。 ## 為什麼這個來源值得用 SuperClaude_Framework 是 GitHub 上 23k★ 以上的高星開源專案（MIT 授權、可商用）。它不是又一個 prompt 清單，而是一整套「meta 設定系統」：提供約 30 個 `/sc:` 前綴 slash 指令（涵蓋規劃、開發、測試、文件、版本控制）、約 20 個專職 AI agent（如 Frontend Architect、Security Engineer、PM、Deep Research）、7 種行為模式（腦力激盪、深度研究、編排、Token 節約、任務管理、自省等），以及對 8 個 MCP server 的整合（Context7 文件、Sequential-Thinking、Playwright、Serena 等）。它把「整個開發生命週期」系統化，這正是一般 prompt 庫做不到的。 ## 怎麼用（本篇精煉版） 上方 full_prompt 把 SuperClaude 的核心心法——「分階段 + 切換認知 persona」——濃縮成一份可直接貼進 Claude Code 的骨架，不需要先安裝框架就能體驗它的工作流： 1. 填入任務、repo/技術棧、主要 persona。 2. 讓 Claude 依序跑：Spec → Design → Implement → Test/Analyze → Improve → Document/Commit，每階段宣告自己用哪個 persona。 3. 想要完整版（真正的 `/sc:brainstorm`、`/sc:implement`、`/sc:analyze` 等指令）時，照官方安裝：`pipx install superclaude` 後執行 `superclaude install`，重啟 Claude Code 即可使用全部 `/sc:` 指令。 ## 何時用 - 任務不只是一行小修，而是「需要規劃→實作→驗證」的完整工作。 - 你希望每次同類任務都跑出一致的品質與產出結構。 - 你想用「不同專業視角」逐段檢查（例如實作完讓 Security Engineer persona 再掃一遍）。 - 想正式導入時，再去裝完整框架接上 MCP server。 📎 來源：SuperClaude-Org/SuperClaude_Framework（作者 SuperClaude-Org，MIT 授權）— 本篇為繁中改寫整理，原始指令體系與安裝方式見上方連結；本框架與 Anthropic 無官方關聯。
[TASK_DESCRIPTION]要交給 Claude Code 完成的任務描述（例如「在結帳頁加上折扣碼欄位」）
[REPO_OR_STACK]專案背景或技術棧（例如「Next.js 15 + TypeScript + Cloudflare Workers」）
[PERSONA]主要採用的認知 persona（例如 Frontend Architect、Security Engineer、PM、Deep Research）
填下面的欄位，上方 prompt 會即時替換 [方括號] 內容。填好後按「複製組好的 prompt」直接丟進工具。
You are operating as a structured development assistant for this repository, inspired by the SuperClaude command-and-persona model. Run the work through explicit phases instead of free-form chat. Adopt the cognitive persona that best fits each phase, and state which persona you are using.
TASK: {{TASK_DESCRIPTION}}
REPO CONTEXT: {{REPO_OR_STACK}}
PRIMARY PERSONA: {{PERSONA}} # e.g. Frontend Architect, Security Engineer, PM, Deep Research
Work through these phases, announcing each one:
1. BRAINSTORM / SPEC (persona: PM or Architect)
- Restate the task in your own words and surface ambiguities.
- Produce a short spec: goals, non-goals, constraints, acceptance criteria.
- STOP and confirm the spec if the task is non-trivial.
2. DESIGN (persona: relevant Architect)
- Outline the approach, key files to touch, and trade-offs of 1-2 options. Recommend one.
3. IMPLEMENT (persona: Engineer)
- Make focused edits. No placeholders, no TODO stubs. Explain non-obvious choices inline.
4. TEST & ANALYZE (persona: Quality / Security Engineer)
- Run the build/tests if available and paste real output.
- Do a security/quality pass: input validation, error handling, edge cases.
5. IMPROVE & CLEANUP (persona: Engineer)
- Remove dead code, tighten naming, ensure consistency with the existing codebase.
6. DOCUMENT & COMMIT (persona: PM / Engineer)
- Summarize what changed and why. Propose a conventional-commit message.
RULES:
- Prefer editing existing files over creating new ones.
- If a phase reveals a blocker needing a human decision, list options + your recommendation and continue with everything not blocked.
- Keep each persona's voice focused on its concern; switch explicitly when the phase changes.這組 prompt 專為 Claude Code 設計。把 prompt 內 3 個方括號 [變數] 換成你自己的內容，貼進 Claude Code 執行即可。難度中等，照變數說明填好後即可上手。
完整 prompt 免費開放閱讀，不用註冊；登入後可一鍵複製、收藏與留言。
prompt 文字本身你可自由使用與修改。但 AI 生成物（圖／音樂／影片／文字）的商用授權，取決於你在 Claude Code 使用的方案與其官方服務條款，請以該工具的授權規範為準。
Studio engineer 視角拆解 Suno 致命弱點（油炸 vocals、高頻 artifact）+ 4 步驟 DAW workflow + Suno Studio 修音 prompt
提案產生器 / 會議處理器 / 內容再利用 / 週五回顧 / 收工 reset — 試了 40 個只有這 5 個沒被丟掉、各省 30+ 分鐘 / 次。
適合：部落格、Medium、Notion 公開頁、Substack — 任何支援 iframe / HTML 嵌入的地方。對方點「看完整」會回到本站、是 prompt 庫的免費 backlink。
<iframe src="https://prompt.luvai.net/embed/superclaude-framework-sc-commands-personas" width="100%" height="380" frameborder="0" style="border:1px solid #e0dcd0;border-radius:4px;" loading="lazy" title="PromptCraft Embed"></iframe>
把方括號 [ ] 內的變數換成你的內容，丟進 Claude Code。
六個月在 Claude / GPT-4 / Gemini 上用人工 rater A/B 測 200+ prompt 後寫的。包含 persona+constraint stacking / anti-example / role reversal QA / cognitive scaffold / emotional priming / uncertainty CoT / steelman first。
一年的 role-play system prompt + 14-step framework 後總結：真正改變品質的是 5 個單行 prompt。沒有 role、沒有 markdown、沒有「you are an expert」。