⚠️ The tool interface and all AI-generated outputs are in Traditional Chinese (繁體中文).
The path from "I have an idea" to "engineers can start coding" is often the messiest part of software development:
- Requirements are too vague for engineers to act on
- Technical designs skip security and observability until it's too late
- Tasks are either too coarse to estimate or too granular to track
- AI Coding Tools go off-track because the prompt lacked structure
Feature Planner gives this path structure, repeatability, and consistent quality.
Describe a feature in plain language. The skill guides you through four stages and produces a complete planning document:
Your feature description (plain language)
↓
User Stories (BDD Given / When / Then)
↓
Technical Design (architecture, security, QA, ops — 7 sections)
↓
Task Breakdown (independently verifiable development tasks)
↓
Full Markdown document (hand directly to engineers or an AI Coding Tool)
Note: All output — User Stories, technical design, task list, and the final document — is written in Traditional Chinese (Mandarin). This is intentional: the tool is designed for Mandarin-speaking development teams.
skill/
├── README.md ← This file
├── SKILL.md ← Entry point: workflow and usage rules
│
├── prompts/ ← AI system prompts for each stage
│ ├── 01-user-stories.md ← BDD story generation
│ ├── 02-tech-design.md ← Technical design generation
│ └── 03-task-planning.md ← Task breakdown generation
│
├── templates/ ← Output document formats
│ ├── full-document.md ← Complete planning document
│ └── task-list-minimal.md ← Minimal task list for AI Coding Tools
│
├── examples/ ← End-to-end worked examples
│ └── login-module.md ← Example: login module planning
│
└── references/ ← Reference material for better outputs
├── bdd-patterns.md ← BDD best practices and anti-patterns
└── tech-design-checklist.md ← Technical design quality checklist
No setup required. Works in Claude.ai, the Claude app, or any Claude interface.
- Open a new Claude conversation
- Paste the contents of
SKILL.mdat the start, or upload the file - Describe your feature — Claude will guide you through each stage
Starter message:
[Paste the contents of SKILL.md here]
I want to plan the following feature:
Users can log in with Email + Password. After 5 consecutive failures,
the account is locked for 15 minutes.
Use each prompt file directly as the system parameter.
const fs = require('fs');
const Anthropic = require('@anthropic-ai/sdk');
const client = new Anthropic({ apiKey: process.env.ANTHROPIC_API_KEY });
// Stage 1: Generate User Stories
async function generateStories(featureDescription) {
const system = fs.readFileSync('./prompts/01-user-stories.md', 'utf-8');
const response = await client.messages.create({
model: 'claude-sonnet-4-20250514',
max_tokens: 8192,
system,
messages: [{ role: 'user', content: `功能需求：\n${featureDescription}` }],
});
return response.content[0].text;
}
// Stage 2: Generate Technical Design (include preferences)
async function generateTechDesign(stories, preferences) {
const system = fs.readFileSync('./prompts/02-tech-design.md', 'utf-8');
const response = await client.messages.create({
model: 'claude-sonnet-4-20250514',
max_tokens: 8192,
system,
messages: [{
role: 'user',
content: `User Stories：\n${stories}\n\n使用者偏好：\n${preferences}`
}],
});
return response.content[0].text;
}
// Stage 3: Generate Task Breakdown
async function generateTasks(stories, techDesign) {
const system = fs.readFileSync('./prompts/03-task-planning.md', 'utf-8');
const response = await client.messages.create({
model: 'claude-sonnet-4-20250514',
max_tokens: 8192,
system,
messages: [{
role: 'user',
content: `User Stories：\n${stories}\n\n技術設計：\n${techDesign}`
}],
});
return response.content[0].text;
}Preferences format (sent as plain text in zh-TW):
成本預算：完全免費（只用開源或免費方案）
部署環境：個人本機、對外公開服務
開發重點：最簡架構、長期維護
cp -r /path/to/skill ./planning-skill
claude "請閱讀 planning-skill/SKILL.md，我需要規劃一個新功能"Plain language is fine. Be specific about: who uses it, what the key flow is, and any constraints or edge cases.
Users can check out their shopping cart using a credit card.
A confirmation email is sent on success and inventory is updated.
Unpaid orders are automatically cancelled after 30 minutes.
Claude generates BDD-formatted stories. You can confirm, request changes, or add new scenarios before proceeding.
After confirming stories, answer three questions:
Cost budget: Completely free / Low cost (<$10/mo) / Medium ($10–$100/mo) / Unlimited
Deployment: Local machine / Internal system / Public-facing (multiple allowed)
Dev priorities: Minimal architecture / Security-first / Fast delivery / Long-term maintenance
Seven sections are generated. Focus on: does the tech stack fit your budget? Are security measures concrete? Is the QA strategy realistic?
You can edit any section or delete sections you don't need.
Each task has a goal, work items, and acceptance criteria. Confirm the scope feels right (1–3 days per task), then export:
For engineers: Paste into GitHub Issues, Linear, or Notion.
For AI Coding Tools:
Please read the following planning document.
Start with Stage 1, Task T1.1. Tell me when each task is done before moving on.
[paste document]
Can I skip a stage?
Yes. If you already have User Stories, start from Stage 2 with prompts/02-tech-design.md.
The output got cut off. What do I do?
The prompts are set to max_tokens: 8192. If it still cuts off, ask Claude to continue from where it stopped, or break the request into smaller parts.
Can I customize the prompts?
Yes — edit any file in prompts/. Common customizations: add your organization's tech stack constraints to 02-tech-design.md, or adjust task size expectations in 03-task-planning.md.
Can I use this for a feature in an existing project? Yes. Include your current stack in the feature description:
Existing stack: React + Node.js/Express + PostgreSQL, deployed on AWS EC2
New feature: ...
| Stage | Check |
|---|---|
| User Stories | Does every Scenario have Given/When/Then? Are edge cases covered? |
| Technical Design | Does the stack fit the budget? Are security measures specific (not generic)? |
| Task Breakdown | Can each task be completed in 1–3 days? Is the acceptance criteria verifiable? |
Full checklists: references/tech-design-checklist.md and references/bdd-patterns.md
在軟體開發中，「從想法到任務」這段路常常是最艱難的——
- 需求描述太模糊，工程師不知道從哪裡開始
- 技術設計缺乏安全與營運考量，上線才發現漏洞
- 任務拆得太粗或太細，估時不準、驗收困難
- 給 AI Coding Tool 的提示詞（prompt）不夠具體，生出來的程式碼走鐘
Feature Planner 的目標是讓這段路有結構、可重複、品質穩定。
輸入一段功能描述，Skill 會引導你完成四個階段，最終輸出一份完整的文件：
功能描述（你的輸入）
↓
User Stories（BDD Given/When/Then 格式）
↓
技術設計文件（架構、資安、QA、維運等 7 個章節）
↓
任務規劃（可獨立驗收的開發任務）
↓
完整 Markdown 文件（可直接給工程師或 AI Coding Tool 使用）
注意： 工具的所有輸出——User Stories、技術設計、任務清單與最終文件——皆以zh-TW撰寫。這是刻意設計的，本工具主要協助以正體中文溝通的開發團隊。
skill/
├── README.md ← 你正在看的這份文件
├── SKILL.md ← Skill 入口：工作流程與使用規則
│
├── prompts/ ← 每個階段的 AI System Prompt
│ ├── 01-user-stories.md ← BDD Story 生成
│ ├── 02-tech-design.md ← 技術設計生成
│ └── 03-task-planning.md ← 任務拆解生成
│
├── templates/ ← 輸出文件的格式模板
│ ├── full-document.md ← 完整規劃文件
│ └── task-list-minimal.md ← 精簡任務清單（直接給 AI Coding Tool）
│
├── examples/ ← 完整走過四個步驟的實際範例
│ └── login-module.md ← 範例：登入模組規劃
│
└── references/ ← 參考資料，讓輸出品質更好
├── bdd-patterns.md ← BDD 最佳實踐與常見錯誤
└── tech-design-checklist.md ← 技術設計品質檢查清單
不需要任何設定，在 Claude.ai、Claude App 或任何 Claude 介面均可使用。
- 開啟新對話
- 將
SKILL.md的內容貼到對話開頭，或直接上傳檔案 - 描述你的功能需求，Claude 會引導你逐步完成
範例開場白：
[貼上 SKILL.md 的內容]
我想規劃以下功能：
使用者可以用 Email + 密碼登入，連續失敗 5 次後鎖定 15 分鐘。
Admin 可以在後台停用帳號。
把每個 prompt 檔案作為 system 參數，逐步呼叫 API。
const fs = require('fs');
const Anthropic = require('@anthropic-ai/sdk');
const client = new Anthropic({ apiKey: process.env.ANTHROPIC_API_KEY });
// 第一步：生成 User Stories
async function generateStories(featureDescription) {
const system = fs.readFileSync('./prompts/01-user-stories.md', 'utf-8');
const response = await client.messages.create({
model: 'claude-sonnet-4-20250514',
max_tokens: 8192,
system,
messages: [{ role: 'user', content: `功能需求：\n${featureDescription}` }],
});
return response.content[0].text;
}
// 第二步：生成技術設計（需附上偏好資訊）
async function generateTechDesign(stories, preferences) {
const system = fs.readFileSync('./prompts/02-tech-design.md', 'utf-8');
const response = await client.messages.create({
model: 'claude-sonnet-4-20250514',
max_tokens: 8192,
system,
messages: [{
role: 'user',
content: `User Stories：\n${stories}\n\n使用者偏好：\n${preferences}`
}],
});
return response.content[0].text;
}
// 第三步：生成任務規劃
async function generateTasks(stories, techDesign) {
const system = fs.readFileSync('./prompts/03-task-planning.md', 'utf-8');
const response = await client.messages.create({
model: 'claude-sonnet-4-20250514',
max_tokens: 8192,
system,
messages: [{
role: 'user',
content: `User Stories：\n${stories}\n\n技術設計：\n${techDesign}`
}],
});
return response.content[0].text;
}技術偏好格式範例：
成本預算：完全免費（只用開源或免費方案）
部署環境：個人本機、對外公開服務
開發重點：最簡架構、長期維護
cp -r /path/to/skill ./planning-skill
claude "請閱讀 planning-skill/SKILL.md，我需要規劃一個新功能"不需要技術格式，自然語言描述即可。越具體越好，特別是：
- 主要使用者是誰
- 最重要的操作流程
- 有沒有特殊限制或邊界情況
使用者可以在購物車選擇商品後進行結帳，
支援信用卡付款，付款成功後寄送確認 Email 並更新庫存。
訂單在 30 分鐘內未付款自動取消並釋放庫存。
Claude 會生成 BDD 格式的 Stories。你可以確認、要求修改、或新增 Scenario 後再繼續。
確認 Stories 後，回答三個問題：
成本預算：完全免費 / 低成本（<$10/mo）/ 中等（$10~$100/mo）/ 不限
部署環境：個人本機 / 內部系統 / 對外公開（可複選）
開發重點：最簡架構 / 資安優先 / 快速交付 / 長期維護（可複選）
生成 7 個章節。重點確認：技術棧是否符合預算？資安措施是否具體？QA 策略是否務實？
可以編輯任意章節，或刪除不需要的章節。
確認每個任務的範圍合理（1~3 天）後匯出：
給工程師： 貼入 GitHub Issues、Linear 或 Notion
給 AI Coding Tool：
請閱讀以下規劃文件，從第一階段 T1.1 開始實作。
每完成一個任務告訴我，確認後再繼續下一個。
[貼上文件]
可以跳過某個階段嗎？
可以。如果你已有 User Stories，直接使用 prompts/02-tech-design.md 從技術設計開始。
輸出被截斷怎麼辦？
Prompt 預設 max_tokens: 8192。若仍被截斷，請告訴 Claude「請從上次截斷的地方繼續輸出」。
可以客製化 prompt 嗎？
完全可以，直接編輯 prompts/ 下的 .md 檔案。常見做法：在 02-tech-design.md 加入組織的技術棧限制，或在 03-task-planning.md 調整任務粒度定義。
可以用在現有專案的新功能嗎？ 可以。在描述功能時加上現有技術棧資訊：
現有技術棧：React + Node.js/Express + PostgreSQL，部署在 AWS EC2
新功能需求：...
| 階段 | 確認項目 |
|---|---|
| User Stories | 每個 Scenario 都有 Given/When/Then？涵蓋了邊界案例？ |
| 技術設計 | 符合我的成本預算？技術選型有說明理由？資安設計具體？ |
| 任務規劃 | 每個任務 1~3 天可完成？完成條件可以被驗收？ |
詳細清單見 references/tech-design-checklist.md 與 references/bdd-patterns.md。