GitHub Spec Kit: The Four-Stage Spec-Driven Development Workflow (/specify → /plan → /tasks → /implement)

## What This Is / What Problem It Solves Most people write code with AI by typing one sentence, watching it generate a pile of code, and then complaining until it fixes what's wrong — GitHub calls this 'vibe coding.' The result is requirement drift, unpredictable output, and code that's hard to maintain afterward. **Spec Kit** is GitHub's official open-source answer: Spec-Driven Development. You first write out 'what to build' as a structured, executable specification, then have the AI produce the implementation from that spec — the spec isn't just reference documentation, it's the actual source that drives code generation. ## Why This Source Is Worth Using This is an **official GitHub project** (not a personal side project), MIT-licensed for commercial use, and claims integration with 30+ AI coding agents (GitHub Copilot, Claude Code, Cursor, Gemini, Codex CLI, and more) — currently the most substantial reference implementation of the 'spec-driven' approach. ## How to Use It (Key Points) The official workflow is a sequence of slash commands (the actual prefix is /speckit.): 1. `/speckit.constitution`: Establish the project's governing principles. 2. `/speckit.specify`: Define 'what to build' (requirements and user stories — no technical decisions yet). 3. `/speckit.clarify`: Clarify ambiguous points in the spec (recommended). 4. `/speckit.plan`: Produce a technical implementation plan using your chosen tech stack. 5. `/speckit.analyze`: Cross-check all documents for consistency and coverage gaps (optional). 6. `/speckit.tasks`: Break the plan down into an executable task list. 7. `/speckit.implement`: Build the feature step by step according to the task list. There are also `/speckit.taskstoissues` (convert tasks to GitHub issues) and `/speckit.checklist` (a quality checklist). Installation requires Python 3.11+, Git, and uv: first run `uv tool install specify-cli --from git+https://github.com/github/spec-kit.git@<version>`, then initialize with `specify init <project> --integration <agent>`. ## When to Use It Best suited for features with real complexity that need multiple steps, where you want the process to be traceable, the outcome predictable, and the spec handoff-ready; overkill for a one-off throwaway script. The full_prompt above condenses this six-stage discipline into a version you can paste directly into any AI agent conversation — you can follow it without installing the CLI. 📎 Source: github/spec-kit (by GitHub, Inc., MIT License) — this piece is a Traditional Chinese adaptation; see source_url for the original content.

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。

# Spec-Driven Development 工作流（改寫自 GitHub Spec Kit）

你是一位嚴守「規格先於程式」紀律的開發代理。請依下列階段依序執行，每一階段都先產出文件、與我確認後才進下一階段，禁止跳階直接寫 code。

## 階段 0｜憲法 constitution
為本專案建立一份治理原則文件（命名慣例、架構分層、測試門檻、不可違反的底線）。
專案名稱：{{PROJECT_NAME}}
團隊既有規範 / 技術偏好：{{TEAM_PRINCIPLES}}

## 階段 1｜規格 specify
把下列需求展開成一份「使用者導向」的規格：描述「要做什麼、為誰、解決什麼」，用使用者故事 + 驗收情境表達，**不要**寫任何技術選型或實作細節。
要建的功能：{{FEATURE_DESCRIPTION}}

## 階段 2｜釐清 clarify
主動找出規格中所有「未明確 / 有歧義 / 邊界未定義」之處，逐條向我提問並標記為待決議。我回答後更新規格。

## 階段 3｜計畫 plan
依規格產出技術實作計畫：技術棧、模組切分、資料模型、對外介面、相依與風險。
指定技術棧 / 限制：{{TECH_STACK}}

## 階段 4｜一致性分析 analyze
交叉檢查 憲法 ↔ 規格 ↔ 計畫 三份文件是否一致、有無覆蓋缺口或互相矛盾，列出問題清單。

## 階段 5｜拆任務 tasks
把計畫拆成可逐項勾選、有明確完成定義（DoD）的任務清單，標出相依順序與可平行項。

## 階段 6｜實作 implement
依任務清單逐項實作，每完成一項就回報並對照規格驗收，全部做完前不要宣稱完成。

---
📎 來源：github/spec-kit（作者 GitHub, Inc.，MIT 授權）— 本篇為繁中改寫整理，原始指令名稱與完整 CLI 見上方連結。實際使用時，安裝官方 CLI（uv tool install specify-cli）後在支援的 AI 代理中輸入 /speckit.specify、/speckit.plan、/speckit.tasks、/speckit.implement 等指令即可。