跳到主要內容Skip to content
中EN
選單Menu
概念作品Concept

AIAI軟體開發團隊Software engineering teams

工程團隊 AI 程式開發導入手冊AI coding playbook for an engineering team

為虛構的軟體團隊「沐光資訊」設計的 AI 程式開發導入手冊概念:一份網頁版手冊,包含使用政策、AI 協助程式的 PR 審查清單、機密處理規則、AGENTS.md 範本與導入追蹤方式。

An AI coding playbook concept for Muguang, a fictional software team: a web-based handbook with a usage policy, a PR review checklist for AI-assisted code, secrets rules, an AGENTS.md template and a way to track adoption.

此為概念設計,用於展示我們的設計與開發能力,並非真實客戶案例。

This is a concept design that demonstrates our design and development capabilities; it is not a real client project.

Concept: the usage-policy page of a web-based AI coding playbook with core principles and approved tools, beside an AGENTS.md template

團隊裡有人已經天天用 AI 寫程式,有人還不敢碰,大家用的工具和帳號也不一樣。沒有共同規則時,常見的問題是金鑰或客戶資料被貼進對話、測試被悄悄放寬,以及一次送出幾千行沒人看得完的改動。

這個概念把規則寫成一份團隊自己的網頁手冊:核心原則與核准的工具清單、AI 協助程式的 PR 審查清單、機密資料能做與不能做的事、放在 repo 根目錄的 AGENTS.md/CLAUDE.md 範本,以及導入追蹤頁。追蹤頁只記錄描述性的指標,例如標記 AI 協助的 PR 比例與審查退回原因,用來找出要補強的地方,而不是評比個人。

實際執行時,我們會先了解團隊的 repo、CI 與資安要求,再依情況搭配 Claude Code、Codex、Cursor 或 GitHub Copilot 等工具,用團隊自己的程式碼示範從寫測試、實作到審查的完整流程,並把規則落實到 PR 範本與 CI 檢查。手冊中的團隊名稱與數據皆為虛構。

Some engineers already write code with AI every day while others avoid it, and everyone uses different tools and accounts. Without shared rules, the usual problems are keys or customer data pasted into a chat, tests quietly loosened, and changes thousands of lines long that nobody can review.

This concept writes the rules down as the team’s own web handbook: core principles and an approved-tools list, a PR review checklist for AI-assisted code, what is and is not allowed with sensitive data, an AGENTS.md / CLAUDE.md template for the repo root, and an adoption page. That page tracks descriptive measures only, such as the share of PRs marked as AI-assisted and the reasons reviews send changes back, to find what needs more guidance rather than to rate individuals.

In practice we would first look at the team’s repositories, CI and security requirements, then work with tools such as Claude Code, Codex, Cursor or GitHub Copilot as fits, walk through writing tests, implementing and reviewing on the team’s own code, and build the rules into the PR template and CI checks. The team name and figures in the handbook are fictional.

有類似的需求?Planning something similar?

先聊 30 分鐘:你的目標、預算與時程,我們免費提供初步建議。

Start with a free 30-minute call about your goals, budget and timeline.

預約免費 30 分鐘諮詢Book a free 30-minute consultation