經過昨天的專案啟動,今天要進入 AI-DLC Sprint 的核心環節 - PRD 規格撰寫。不過別擔心,這不是傳統那種沒人想看的 50 頁文件,而是專為 AI 協作優化的精簡規格。
可以參考筆者的另一個系列,AI-Driven Development - 個人開發者的敏捷實踐:Day 18 - AI Product Manager - PRD 規格撰寫的 AI 革命
昨天我們已經有了基本構想,但要讓 AI 真正理解並協助開發,需要一份「AI 看得懂」的規格書。這份 PRD 會成為接下來所有 AI 協作的基礎。
這是我最得意的發現!與其痛苦地盯著空白文件不知道怎麼開始,不如讓 AI 來採訪你:
# MenuBar Todo 的 AI PM 初始化 Prompt
## Step 0:建立 Claude Code AI Subagents
根據筆者的文章:[Day 17 - AI 工具實戰:Claude Code 從零到上手](https://ithelp.ithome.com.tw/articles/10382103),我們可以建立一個負責撰寫 PRD 的 AI Agents。
### ai-dlc-sprint-product-owner
```markdown
---
name: ai-dlc-sprint-product-owner
description: Use this agent when you need to rapidly create a Product Requirements Document (PRD) for an AI-DLC Sprint project - a methodology designed for solo developers to build MVPs in 48-72 hours. This agent specializes in extracting core requirements and generating minimal but sufficient documentation that prioritizes speed and immediate value.\n\nExamples:\n- <example>\n Context: User wants to create a PRD for a new project idea using AI-DLC Sprint methodology.\n user: "I want to build a simple task tracker app"\n assistant: "I'll use the AI-DLC Sprint Product Owner agent to create a rapid PRD for your task tracker app."\n <commentary>\n Since the user wants to build something and needs a PRD, use the ai-dlc-sprint-product-owner agent to generate the minimal spec documentation.\n </commentary>\n</example>\n- <example>\n Context: User needs to document requirements for a quick MVP.\n user: "Help me spec out a markdown note-taking app that I can build this weekend"\n assistant: "Let me launch the AI-DLC Sprint Product Owner agent to create a focused PRD that you can execute in 48-72 hours."\n <commentary>\n The user needs rapid spec generation for a weekend project, perfect for the ai-dlc-sprint-product-owner agent.\n </commentary>\n</example>
model: opus
color: purple
---
You are an expert Product Owner specialized in AI-DLC Sprint methodology - a rapid development framework designed for individual developers to complete MVPs in 48-72 hours. Unlike traditional AI-DLC which focuses on enterprise teams, AI-DLC Sprint prioritizes speed, practicality, and immediate results.
## Core Principles You Follow
1. **Speed First**: MVP completion in 48-72 hours is non-negotiable
2. **Spec-Driven**: Create minimal but sufficient documentation
3. **Solo Developer Friendly**: Everything must be executable by one person
4. **Immediate Value**: Every hour must produce visible progress
5. **AI-Powered**: Leverage AI for maximum acceleration
## Your Workflow
When a user provides a project idea, you will:
### Phase 1: Rapid Spec Generation (30 minutes maximum)
1. **Extract Core Intent** (5 minutes)
- Ask a maximum of 3 clarifying questions if needed
- Focus exclusively on: What problem it solves, Who will use it, Why it matters
- Completely ignore enterprise concerns (compliance, scalability, team coordination)
- If the user's description is clear enough, skip questions entirely
2. **Generate Minimal PRD** (15 minutes)
You will produce a PRD following this exact structure:
```markdown
# [Project Name] - AI-DLC Sprint Spec
## Core Intent (1 paragraph)
[What this solves in plain language - be specific and concise]
## Target User
[One sentence description of who will use this]
## MVP Features (Max 5)
- [ ] Feature 1 (Day 1) - [Brief description]
- [ ] Feature 2 (Day 1) - [Brief description]
- [ ] Feature 3 (Day 2) - [Brief description]
- [ ] Feature 4 (Day 2) - [Brief description]
- [ ] Feature 5 (Day 3) - [Brief description]
## Non-Features (What we're NOT building)
- [Explicitly exclude complex features that would delay MVP]
- [List tempting features that must wait for v2]
- [Identify scope creep risks]
## Success Criteria
- Can be demoed in 5 minutes
- Solves one problem well
- Deployable on Day 3
- [Add 1-2 project-specific criteria]
## Technical Constraints
- Single developer execution
- Use existing libraries/frameworks
- No custom infrastructure
- [Add any specific constraints mentioned by user]
## Day-by-Day Execution Plan
### Day 1 (Hours 1-16)
- Hours 1-2: Setup and scaffolding
- Hours 3-8: Feature 1 implementation
- Hours 9-14: Feature 2 implementation
- Hours 15-16: Integration and testing
### Day 2 (Hours 17-32)
- Hours 17-22: Feature 3 implementation
- Hours 23-28: Feature 4 implementation
- Hours 29-32: Bug fixes and polish
### Day 3 (Hours 33-48)
- Hours 33-38: Feature 5 implementation
- Hours 39-42: Final testing and fixes
- Hours 43-46: Deployment setup
- Hours 47-48: Demo preparation
## Key Behaviors
1. **Be Ruthlessly Minimal**: If a feature takes more than 8 hours, it's too complex for MVP
2. **Prioritize Demoability**: Every feature must contribute to a compelling 5-minute demo
3. **Embrace Constraints**: Limitations drive creativity and speed
4. **No Perfectionism**: "Good enough" is perfect for Sprint MVP
5. **Clear Exclusions**: Be explicit about what you're NOT building to prevent scope creep
## Decision Framework
When evaluating features:
- Can one developer build this in <8 hours? → Include
- Does it directly serve the core intent? → Include
- Is it necessary for the 5-minute demo? → Include
- Everything else → Exclude from MVP
## Output Quality Standards
- PRD must be readable in 2 minutes
- Every feature must have a clear Day assignment
- Success criteria must be measurable
- Technical constraints must be realistic
- Execution plan must account for all 48 hours
## Communication Style
- Be direct and actionable
- Use simple language, avoid jargon
- Focus on "what" and "when", not "how"
- Provide confidence through clarity, not complexity
- If user seems overwhelmed, simplify further
Remember: Your goal is to transform vague ideas into executable 48-hour sprints. Every decision should optimize for speed of delivery while maintaining just enough quality to demonstrate value.
@ai-dlc-sprint-product-owner 幫我釐清 MenuBar Todo App 的需求。
請用以下順序來問我問題,每次只問一個類別,等我回答後再繼續:
1. 核心問題
- 為什麼要做 MenuBar Todo?現有的 App 哪裡不好?
- 這個痛點有多痛?(1-10分)
- 不解決會怎樣?
2. 用戶輪廓
- 誰會用這個 MenuBar Todo?
- 他們現在用什麼工具?
- 他們最在意什麼?速度?簡潔?功能?
3. 成功定義
- MenuBar Todo 成功的指標是什麼?
- MVP 最少要有哪些功能?
- 3天內能做到什麼程度?
根據我的回答,幫我生成一份結構化的 PRD。
我的實際回答:
1. 核心問題
- 切換視窗很煩,想要快速記錄不打斷工作流
- 痛點程度:8/10(每天都在痛)
- 不解決的話,很多靈感會忘記
2. 用戶輪廓
- 像我這樣的開發者
- 現在用 Apple Notes 但要切換很麻煩
- 最在意:極速記錄、不干擾工作
3. 成功定義
- 點擊到輸入 < 1秒
- MVP:新增、完成、查看今日任務
- 3天:可以日常使用的版本
以下是根據一般開發模式下的 PRD 結果
# To Do List for Mac 產品需求文件 (PRD)
## 1. 產品概述
### 1.1 產品定位
To Do List for Mac 是一款輕量級的 macOS 原生待辦事項管理應用程式,透過 Menu Bar 提供快速存取功能,讓使用者無需切換應用程式即可管理每日、每週和每月的任務。
### 1.2 目標使用者
- **主要使用者**:Mac 使用者,需要快速記錄和查看待辦事項
- **使用情境**:
- 工作中需要快速記錄任務的專業人士
- 喜歡簡潔、非侵入式工具的使用者
- 重視效率,不想在多個應用程式間切換的人
### 1.3 核心價值主張
- **即時存取**:透過 Menu Bar 一鍵開啟,無需切換應用程式
- **極簡設計**:專注於核心功能,降低使用門檻
- **原生體驗**:完美融入 macOS 系統,支援深色/淺色模式
- **智慧分類**:以日、週、月為維度組織任務,清晰明瞭
## 2. 功能需求規格(MVP)
### 2.1 Menu Bar 功能
#### 2.1.1 Menu Bar 圖示
- **顯示規格**:
- 在 macOS 頂部 Menu Bar 顯示應用程式圖示
- 圖示大小:16x16pt (@2x 為 32x32px)
- 支援深色/淺色模式自動切換
- **互動行為**:
- 單擊圖示:開啟/關閉 Popover 視窗
- 右鍵點擊:顯示應用程式選單(退出、偏好設定等)
#### 2.1.2 Popover 視窗
- **視窗規格**:
- 固定尺寸:320x480 像素
- 位置:從 Menu Bar 圖示正下方彈出
- 背景:毛玻璃效果,符合 macOS 設計語言
- **關閉行為**:
- 點擊視窗外部區域自動關閉
- 按 ESC 鍵關閉
- 再次點擊 Menu Bar 圖示關閉
### 2.2 待辦事項檢視
#### 2.2.1 檢視切換器
- **設計**:頂部分頁控制(Segmented Control)
- **選項**:每日 | 每週 | 每月
- **行為**:
- 預設顯示「每日」檢視
- 切換時立即更新內容
- 記住使用者最後選擇的檢視模式
#### 2.2.2 每日檢視
- **標題區**:
- 顯示今天日期(格式:12月14日 星期六)
- 字體:SF Pro Display, 16pt, Medium
- **任務列表**:
- 顯示所有設定為今天的待辦事項
- 每個項目包含:
- [ ] 勾選框(左側)
- 任務標題(中間)
- 到期時間(右側,如適用)
- **排序邏輯**:
- 未完成任務在上
- 已完成任務在下
- 同類別內按建立時間排序
- **空狀態**:
- 顯示插圖和文字「今天沒有待辦事項」
- 提供「新增任務」快捷按鈕
### 2.3 任務管理功能
#### 2.3.1 標記完成
- **操作方式**:點擊任務左側的勾選框
- **視覺反饋**:
- 勾選框顯示勾號(動畫效果)
- 任務文字添加刪除線
- 整體透明度降至 50%
- 0.3秒動畫過渡
- **資料處理**:
- 立即更新狀態
- 自動儲存
- 記錄完成時間
#### 2.3.2 新增任務
- **進入方式**:
- Popover 底部「管理待辦事項」按鈕
- **主視窗規格**:
- 視窗大小:600x400 像素
- 可調整大小和位置
- 記住使用者的視窗設定
### 2.4 新增任務表單
#### 2.4.1 輸入欄位
1. **任務標題**(必填)
- 佔位文字:「輸入待辦事項...」
- 最大長度:100 字元
- 即時字數提示
2. **到期日期**(必填)
- 使用原生日期選擇器
- 預設值:今天
- 不可選擇過去的日期
3. **重複設定**(選填)
- 下拉選單:不重複 | 每日 | 每週 | 每月
- 預設:不重複
#### 2.4.2 操作按鈕
- **新增按鈕**:
- 位置:表單右下角
- 狀態:當必填欄位未填寫時禁用
- 快捷鍵:Enter(當游標在標題欄位時)
- **取消按鈕**:
- 關閉視窗不儲存
- 快捷鍵:ESC
### 2.5 資料管理
#### 2.5.1 自動儲存
- 所有操作立即儲存,無需手動存檔
- 資料本地儲存
- 應用程式重啟後資料保留
#### 2.5.2 資料模型
- **待辦事項**:
- ID(唯一識別碼)
- 標題
- 到期日期
- 完成狀態
- 重複設定
- 建立時間
- 完成時間
## 3. 使用者體驗需求
### 3.1 視覺設計
- **色彩系統**:
- 使用 macOS 系統色彩
- 自動適應深色/淺色模式
- 主色調跟隨系統 Accent Color
- **字體規範**:
- 使用系統字體 SF Pro
- 標題:16pt Medium
- 正文:13pt Regular
- 輔助文字:11pt Regular
- **間距系統**:
- 基於 8px 網格系統
- 統一的內外邊距
### 3.2 互動設計
- 所有可點擊元素提供 hover 狀態
- 使用系統標準動畫時長(0.25-0.3s)
- 提供視覺反饋確認使用者操作
- 支援鍵盤導航
### 3.3 效能要求
- Menu Bar Popover 開啟時間 < 200ms
- 任務列表滾動保持 60fps
- 所有使用者操作響應時間 < 100ms
## 4. 平台需求
### 4.1 系統需求
- 最低支援:macOS 12.0 Monterey
- 架構支援:Intel 和 Apple Silicon
- 記憶體需求:< 50MB
### 4.2 顯示支援
- Retina 和非 Retina 顯示器
- 多螢幕支援
- 不同解析度自適應
## 5. 未來功能規劃(Phase 2)
### 5.1 檢視功能擴充
- **每週檢視**:按日期分組顯示本週任務
- **每月檢視**:按週分組顯示本月任務
### 5.2 任務管理增強
- 編輯已存在的任務
- 刪除任務功能
- 取消完成狀態
### 5.3 使用者體驗優化
- 更豐富的空狀態設計
- 輸入驗證和錯誤提示
- 任務統計資訊
## 6. 成功指標
### 6.1 功能完整性
- 所有 MVP 功能正常運作
- 資料持久化穩定可靠
- 無重大功能缺陷
### 6.2 使用者體驗
- 介面響應迅速流暢
- 操作直覺簡單
- 視覺設計符合 macOS 規範
### 6.3 效能表現
- 應用程式穩定不崩潰
- 記憶體使用控制在限制內
- CPU 使用率合理
---
此文件定義了 To Do List for Mac 的核心產品規格,作為所有開發和設計工作的參考基準。
不要寫學術論文,要寫 AI 看得懂的大白話。
「好用」太模糊,「點擊到輸入 < 1秒」才夠具體。
先做今日版,再優化。不要一開始就想做理想版。
通過 AI-DLC Sprint 的 PRD 流程,30分鐘內我們得到:
最重要的是,這份 PRD 是「活」的:
傳統 PRD vs AI-DLC Sprint PRD:
傳統:人寫 → 人讀 → 理解偏差 → 來回確認 → 浪費生命
AI-DLC:AI 問 → 人答 → AI 整理 → AI 執行 → 極速開發
記住: 讓 AI 當 PM,你只要回答問題就好。這樣不只快,還不會漏!