· Joseph · AI & Machine Learning  · 6 min read

BMAD-Method intro

這是自發性的連續寫30篇教學文章,不是很想把文章發在ithelp,來這邊挑戰一下自己寫30天BMAD-Method相關的技術文章,預計會用BMAD-Method做各種不擅長的專案。期間可能會視情況購置需要的AI agent plans,可能是Claude Code, OpenAI, 或Gemini都說不定,看token燃燒速度而定。

第一篇先來介紹介紹BMAD-Method這個 AI Agent Framework吧。

TOC

這是自發性的連續寫30篇教學文章,不是很想把文章發在ithelp,來這邊挑戰一下自己寫30天BMAD-Method相關的技術文章,預計會用BMAD-Method做各種不擅長的專案。期間可能會視情況購置需要的AI agent plans,可能是Claude Code, OpenAI, 或Gemini都說不定,看token燃燒速度而定。

第一篇先來介紹介紹BMAD-Method這個 AI Agent Framework吧。

TOC

基本介紹

常用ChatGPT的都知道,我們常常需要假定AI成為某領域專家、或者喂給他事先準備好的海量資料給他讀(RAG),這樣AI agent才不會在他廣泛的知識庫裡迷航。這種方法像是現在1對1的家教一樣,一次只能管理一個AI agent。接著,就有專門的cursorrule去定義各式各樣的roles,可以根據不同語言不同技術產生cursorrule template。然後就出現了Subagentsagents.md,每個agent都有他自己的專業領域、能力及工具,他們也可以互相傳遞工作。

From: https://docs.anthropic.com/en/docs/claude-code/sub-agents

Subagents are pre-configured AI personalities that Claude Code can delegate tasks to. Each subagent:

  • Has a specific purpose and expertise area
  • Uses its own context window separate from the main conversation
  • Can be configured with specific tools it’s allowed to use
  • Includes a custom system prompt that guides its behavior

BMAD-Method恰恰就是這個概念,他把上面的事情整合在一起:

  1. 很多定義好的team
  2. 很多定義好且不同領域專家的角色
  3. 我需要做什麼的時候,引入整個team或單一角色。

工作流workflow

除此之外,他還把Agile敏捷開發的概念也引入進來,讓agent與agent之間的交付更有規章也更讓大家熟悉,我們來看看下面這張流程圖:

詳細可以看這裡:https://github.com/bmad-code-org/BMAD-METHOD/blob/main/docs/user-guide.md#the-planning-workflow-web-ui-or-powerful-ide-agents

Start: TheNewProject

Optional: 分析研究

專案概要?

UX/UI?

早期測試? (Optional)

開發週期

每個跑過Agile / Scrum的團隊應該都差不多,從 Analyst ResearchProject Brief,然後交付給 UI/UX,最後Documents Aligned之後進入開發Dev Cycle。 接著就是一個sprint一個sprint的循環。

角色

最後我們來看看各個定義好的角色:

下表為ChatGPT生成,prompt:

https://github.com/bmad-code-org/BMAD-METHOD/tree/main/bmad-core/agents 幫我介紹BMAD-Method裡面這些角色的功用

角色名稱英文名稱核心責任 / 功用典型使用時機
AnalystBusiness Analyst做市場調研、競品分析、需求蒐集與創意發掘,產出 project brief 來幫助定義專案方向專案剛開始,需要收集與整理需求、探索方向時
ArchitectArchitect設計技術/系統架構,從 PRD 中規劃模組、API、資料庫與整體系統架構需求明確後,需要落實到技術架構階段
PMProduct Manager管理產品策略與需求文件 (PRD),拆解 epics / user stories,設定優先順序決定要做哪些功能、排定優先順序時
POProduct Owner維護並排序 backlog,驗證 user stories 是否符合方向與品質標準開發過程中,確保開發團隊執行正確的功能
SMScrum Master (Story Preparation Specialist)將大型需求或 epics 拆成可開發的 user stories,補充驗收標準與背景規劃轉入開發階段前,準備故事與任務時
Developer (Dev)Developer撰寫程式碼、實作 user story、修 bug,完成具體功能有明確 user story 可以著手開發時
QATest Architect & Quality Advisor設計測試案例、評估非功能性需求,審查成果是否符合驗收標準功能完成後進行驗收,或開發中確保品質
UX ExpertUX-Expert提供使用者體驗與前端設計建議,產出 UX spec 或 UI 規範專案需要 UI/UX 規劃與互動設計時
BMAD OrchestratorOrchestrator核心協調者,監督 agent 流程,分派任務並維持從需求到測試的順暢交接整體流程中保持一致性與協調
BMAD MasterMaster agent全域總控者,能生成文件、檢查 checklist 或發起任務需要跨角色的總體檢查或指令下達時

到這邊介紹完基本的BMAD-Method了,期待明天開始安裝跟上機演練。

References:

Back to Blog

Related Posts

View All Posts »
Use Grafana MCP with Gemini and n8n

Use Grafana MCP with Gemini and n8n

The Model Context Protocol (MCP) is extremely useful. An AI assistant helps you decide when and how to use connected tools, so you only need to configure them. After integrating MCP logging management systems into several of my projects, it has saved me a significant amount of time. In this article, I'm going to integrate Grafana with the Gemini CLI and n8n. I will chat with the Gemini CLI and n8n and have them invoke the Grafana MCP server. structure TOC

Use Figma MCP server with Gemini CLI

Use Figma MCP server with Gemini CLI

In this article, I won't introduce what MCP is. Instead, I will explain how to set up the Figma MCP server and use Gemini as an MCP client to work with Figma. I will also show you how to run a prompt to get a Figma design with Gemini. TOC

Install Gemini CLI

Install Gemini CLI

Introduction Gemini CLI has been one of the most popular AI agents in the first half of 2025. It's similar to Claude Code, bringing its power directly into your terminal. Although other terminal AI agents exist, their pricing plans are quite different. Gemini CLI provides a free tier with 100 requests per day using Gemini 2.5 Pro, and you can unlock Tier 1 by upgrading to a paid plan. Prerequisites I'm going to use npm to install Gemini. My Node.js version is v24.4.1, and my npm version is 11.4.2. Gemini needs Node.js version 20 or higher installed. If you're using macOS, you can also choose Homebrew to install the Gemini CLI. Installation Now, let's install it using npm. After installation, you can run gemini directly in your terminal. npm install -g @google/gemini-cli installation I'm using the Use Gemini API key authentication method, so I need to generate a key from Google AI Studio and set it in .zshrc (or .bashrc) by adding this line: And then you can try Gemini now! Run some examples example Prompt: give me suggestions for the socket functionality of this project? Response: Conclusion: The Gemini installation is very simple. Although I am using Neovim with Avante, Gemini gives me more power to use the terminal. Next, I will explore how to use Gemini with an MCP server and integrate the workflow into my daily tasks.

AI code review with n8n

AI code review with n8n

Previously I read a post "Automate and Accelerate GitLab Code Reviews with OpenAI and n8n.io". This made me wonder: If I don’t choose GitHub Copilot for code reviews, can I still integrate AI and n8n with GitHub PR reviews? I haven’t written a blog in a long time—it’s time to start again!