simple-sdd
simple sdd workflow for coding tasks. use when the user says 提案, 實作, or 歸檔 for a development request, or asks to follow a lightweight sdd process before coding. enforce proposal before implementation, wait for confirmation, complete one task at a time, report after each small step, and archive completed sdd work.
- License 许可
- MIT
- 所属仓库
- cdcd72/skills
- 最近更新
- 2026-08-12
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- simple-sdd
- description
- simple sdd workflow for coding tasks. use when the user says 提案, 實作, or 歸檔 for a development request, or asks to follow a lightweight sdd process before coding. enforce proposal before implementation, wait for confirmation, complete one task at a time, report after each small step, and archive completed sdd work.
03 SKILL.md 正文
Simple SDD
核心規矩只有一條:「動手寫程式之前,先把要做什麼寫清楚」。
通用規則
- 寫完要讓使用者確認,使用者同意了才開始寫。
- 每做完一小步就回報。
- 不要一次把所有東西做完。
- 不要一個對話想把所有功能做完;一次做一個就好。
- 不要自己加使用者沒要求的功能,也不要過度設計。
- 沒把握的地方就問使用者,不要用猜的。
- 整個流程分三個階段:提案 → 實作 → 歸檔。使用者會用這三個詞告訴你現在做哪一階段。
階段一:提案
當使用者說「提案」並描述一個需求時:
- 幫這個需求取一個英文短名稱,小寫、用連字號連接,例如
add-todo、fix-login。 - 判斷它是哪一種:新功能 / 修 bug / 重構,三選一。
- 在目前的專案資料夾,也就是現在的工作目錄底下,建立檔案:
sdd/<短名稱>/proposal.md,內容包含:## 為什麼做:這個需求想解決什麼問題## 要改什麼:會新增或改動哪些行為,用條列## 影響範圍:大概會碰到哪些檔案,包含新增的、要改的
- 再建立檔案:
sdd/<短名稱>/tasks.md,把工作拆成一條一條可以打勾的小步驟,每一條前面用- [ ]。- 每一條要小到 1 小時內能做完。
- 整份不要超過 10 條;如果超過,代表這個需求太大,提醒使用者先拆小一點。
- 在
tasks.md最後加一段## 驗收條件,用白話的「情境」描述「做完之後應該長怎樣」,格式如下:- 情境:當使用者點擊「新增」按鈕,就把輸入框的文字加到清單最下面
- 情境:當輸入框是空的就按新增,就不新增,並提示「請先輸入內容」
這就是驗收標準,之後要照這個自我檢查。寫完這兩個檔案後,把重點貼給使用者看,然後停下來等使用者確認。除非使用者說「開始實作」,否則不要寫任何程式。
階段二:實作
當使用者說「實作」時:
- 先讀
sdd/<短名稱>/proposal.md和tasks.md這兩個檔案,搞清楚要做什麼。 - 從上到下,一次只做
- [ ]裡的一條任務。每做一條之前,先看看專案裡有沒有現成的東西可以重複使用,不要重複造輪子。 - 做完一條之後,在打勾之前回頭再讀一次那條任務和「驗收條件」,確認真的做到了、沒有漏。確認沒問題就把該條的
- [ ]改成- [x]。改完後再看一眼tasks.md,確認那一行真的從- [ ]變成- [x]了。 - 每完成一條就簡短回報「第幾條完成」,然後做下一條。
- 如果做到一半發現規格漏了東西或方向不對,停下來告訴使用者,等使用者決定,不要自己亂改規格、也不要硬幹。
- 全部打勾後,告訴使用者「全部完成」,並請使用者驗收。
階段三:歸檔
當使用者驗收沒問題、說「歸檔」時:
- 打開
sdd/<短名稱>/tasks.md,確認每一條都是- [x]。如果還有沒打勾的,先告訴使用者,不要直接歸檔。 - 確認都完成後,把整個
sdd/<短名稱>/資料夾移動到sdd/archive/<今天日期>-<短名稱>/,日期格式YYYY-MM-DD。 - 回報歸檔完成,並用一句話總結這次做了什麼,方便使用者以後回頭查。
04 同仓库更多 Skills
template-skill
Replace with description of the skill and when Agent should use it.
presentation-flow-writer
Guides the user through co-authoring a "簡報流程檔" (presentation flow document) — a structured input document written FOR a slide-generation AI agent, not a script or final slide copy — then turns it into a client-facing HTML review page. Distilled from a workflow proven to let a slide-generation agent run long sessions without stalling or redoing work. Use when the user wants to plan a talk/deck's narrative before generating slides, asks to design a "簡報流程" / "簡報大綱" / "投影片流程檔", wants a template for briefing a presentation-generation AI, or asks to turn an existing 簡報流程檔 into something a client can review and sign off on ("給客戶看的版本" / "簡報流程確認版" / HTML 確認頁).
merge-branch
Merge one local Git branch into another through a guarded workflow: inspect repository state, fetch remote references, analyze divergence, perform the local merge, resolve conflicts according to both branches' intent, run project validation, and report the result. Use when the user asks to merge source branch X into target branch Y, integrate one branch into another, or resolve conflicts caused by that local branch merge. Operate only with local Git commands. Never handle pull requests or merge requests. Never push unless the user gives a separate explicit instruction after reviewing the local result.
implementation-orchestrator
Turn engineering work into a reviewed execution baseline with a context pack, implementation plan, complexity routing, optional workstream breakdowns, execution briefs for coding agents or human executors, QA verification, and optional delivery updates. Use when the user wants to turn a bug, feature, refactor, review request, or technical task into an implementation plan, coordinate work across coding agents or human executors, or verify completed implementation against the approved plan.
gemma4-prompt-engineer
Based on the official Gemma 4 documentation and a curated practical knowledge base, assist in writing, optimizing, and diagnosing prompts tailored for the Google Gemma 4 model. When a user mentions Gemma 4 (including any of its sizes, such as E2B, E4B, 12B, 26B A4B, or 31B), needs to write/modify a Gemma 4 prompt, or asks about Gemma 4 control tokens, thinking mode, multimodal ordering, tool/function calling formats, or why an existing Gemma 4 prompt is underperforming, you must utilize this Skill rather than relying solely on general prompting knowledge.
doc-coauthoring
Guide users through a structured workflow for co-authoring documentation. Use when user wants to write documentation, proposals, technical specs, decision docs, or similar structured content. This workflow helps users efficiently transfer context, refine content through iteration, and verify the doc works for readers. Trigger when user mentions writing docs, creating proposals, drafting specs, or similar documentation tasks.