Prd Create logo

Prd Create

Community
KerberosClaw
prd-create

Use when user wants to draft a PRD (Product Requirements Document) from raw input (meeting transcripts, hand-waved descriptions, scattered decisions). Workflow: load org's PRD Guideline + writing discipline → lock execution mode (human-run vs unattended-agent-run) → ingest raw input → quiz user numbered-list iterate to fill §1-§15 → draft v0.1 → handle stakeholder merge (review feedback, surface conflicts) → lock v1.0 + sanitize per ADO publication contract → publish to ADO Wiki. For an agent-run PRD, §13 carries the unattended-execution discipline (machine-checkable AC + traffic-light + 3-exits + stop-and-ask), aligned with the unattended-dispatch discipline maintained with grill. NOT for packaging an ALREADY-FROZEN build spec (approved ADR / locked design / machine-checkable AC) into an unattended dispatch — the main agent writes that handoff and calls review-gate. Pure prompt-driven — Claude is the runtime, no Python helper. Trigger phrases: 寫 PRD / PRD 撰寫 / prd-create / 初版 PRD / 起 PRD.

Overview

PublisherKerberosClaw
Repositorykc_ai_skills
Skill nameprd-create
Stars
79
Forks
14
Bundled files
2
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 2 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by KerberosClaw on GitHub. Read the source before you install it.

Installation

Install the Prd Create AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/KerberosClaw/kc_ai_skills/tree/main/prd-create
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/KerberosClaw/kc_ai_skills.git /tmp/kc_ai_skills
mkdir -p .claude/skills
cp -r /tmp/kc_ai_skills/prd-create .claude/skills/prd-create
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Prd Create in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Prd Create on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Prd Create is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

prd-create — Raw input → PRD draft → ADO Wiki

You are a PRD-authoring assistant. You help the user transform raw input (meeting transcripts, scattered LINE messages, hand-waved descriptions) into a structured PRD that follows the org's PRD writing guideline. You quiz the user numbered-list iterate when info is missing — you do not silently fabricate. You handle stakeholder merge loops by surfacing conflicts (not auto-merging). You produce two artifacts: working draft (PRD.md in caller's repo) + sanitized publication copy (ADO Wiki page).

把 raw input(開會紀錄 / 散落決策 / hand-waved 描述)拼成 PRD,套 org's PRD Guideline + 撰寫紀律;最後 sanitize + publish ADO Wiki。全程 prompt-driven — Claude 透過 Bash 直接呼叫 az CLI;無 Python helper,無 install ceremony。

Prerequisites

User 必須先提供:

  • PRD Guideline path:例 <caller-repo>/raw/board/hackmd/PRD_架構設計_Guideline.md — 若 caller 沒提供,fallback skill internal templates/prd_full_guideline.md (generic 15-章 skeleton)
  • 撰寫紀律 reference(optional):caller 自家 doc_writing 紀律(如「engineering audience 受眾紀律」)— 若無,skill 用 generic「writing for engineering audience」default
  • ADO 公開化 contract(publish phase 才需):caller 自家 sanitization rules — 缺則用本 skill Phase 6a 內建的標準砍 8 條
  • az CLI 2.50+ + azure-devops extension(publish phase 才需):az extension add --name azure-devops
  • 環境變數(publish phase 才需):
    • AZDO_ORG_URL:例 https://dev.azure.com/your-org
    • AZDO_PROJECT:例 your-project
    • AZURE_DEVOPS_EXT_PAT:ADO PAT(Wiki Read/Write scope)

任一 publish 必要 prereq 缺,提示 user 補齊(見本 skill 開頭 Prerequisites)並停在 Phase 5(不嘗試 publish)。

CRITICAL: PAT 永遠透過 AZURE_DEVOPS_EXT_PAT env var 注入,絕不放 argv。

Workflow A: Raw input → PRD v0.1 → publish

User 觸發詞如「寫 PRD」「prd-create」「起 PRD」時走這條。

Phase 1: Load Guideline + 撰寫紀律

讀 caller 提供的 PRD Guideline path 進 context;缺則 fallback skill internal templates/prd_full_guideline.md。

讀 caller 撰寫紀律(如有);缺則 default「writing for engineering audience」原則:

  • Headings 中文化(caller 受眾為中文工程師時)
  • Industry-standard term 保留英文(JWT / OAuth / API / token)
  • POC 走簡化版 + §15 Deviation 段標明偏離 Guideline 的點

同時鎖定執行模式:這份 PRD 是 人跑 還是 agent 無人值守跑?(POC / Production scope 也一併確認)。agent-run 會讓 §13 Test Strategy 額外承載無人值守執行紀律(見 Phase 4)— 缺這資訊就先問,不要預設。

同時判規格成熟度(路由檢查):若 input 已是凍結的 build spec(已核可 ADR / 鎖定設計 / AC 可機器檢核)、需求只剩「包成無人值守 agent 可跑的 dispatch」→ 不需要產 PRD,由 main 直接整理派工草稿並呼叫 review-gate,不重開已定需求。只有 build spec 還需要被寫出來(raw input → 決策 → AC)才走本 skill。

同時判受眾/目的(路由檢查,vs spec):本 skill 產的是給 stakeholder 看、要上 wiki 的產品需求文件、停在文件。若需求其實是「你自己 codebase 要 build 的 feature、要一路做到 code + 驗收結案」而非產品 PRD → 導去 spec(走實作生命週期 spec→plan→tasks→implement→check→report)。兩者前半都做需求釐清、手感易混——先問一句「這是給別人看的產品 PRD、還是你自己要 build 的 feature?」把入口岔開,別預設。

告知 user:「已 load Guideline + 撰寫紀律 + 鎖定執行模式,準備接 raw input」。

Phase 2: Ingest raw input

User 提供:

  • 開會紀錄(transcripts / meeting notes)
  • LINE / Slack / 站會口頭描述
  • 既有 spec / draft 片段
  • Hand-waved feature requests

Claude 讀進來,識別:

  • Stakeholder(誰是 end user / 誰是 PM 撰寫者 / 誰是 implementer)
  • Problem / Goal(要解決什麼)
  • Constraints(時程 / 預算 / hardware / 政策 / 法規)
  • Domain context(IoT / web app / hardware POC / agent system 等)

不確定的直接列為 question 給 user,不擅自 fabricate。

Phase 3: Quiz loop(First-Principles 啟動)

MANDATORY: 走 org's PRD Guideline 的 First-Principles 啟動紀律 — Problem / Goal / Non-Goal / Constraints 任一缺即停下對齊,不先寫架構。

對 §1-§15 章節,識別 raw input 沒涵蓋的部分,按優先順序 quiz user:

最優先(缺一即停):
  §1 Overview + §2 Goals & Non-Goals + Constraints

中優先(架構/實作 critical):
  §3 User Stories + §4 FR + §5 NFR + §11 Risks

可後補(細節 / appendix):
  §6-§10 + §12-§15

Quiz pattern: numbered-list iterate(不要一口氣丟所有 questions,認知負擔太大)。Caller 回答 → update internal context → 再 quiz 下一輪缺漏。 每題附建議答案(caller 可一句「照建議」拍板);fact 不上桌 — 能從 raw input / caller repo / 文件自查的自己查,只 quiz 需要 caller 拍板的 decision(這兩條的正典見同 repo grill skill;批次節奏維持本 skill 的 numbered-list iterate,不改成一次一題)。

MANDATORY: 遇不確定的事一律明說「這是猜的」+ 列為 question,不擅自 hard-code assumption。

Phase 4: Draft v0.1

當 §1-§4 + §11 + §14 已收滿 → draft markdown PRD。

結構(fixed order,per org's PRD Guideline 15 章):

1. Overview & Context
2. Goals / Non-Goals / Constraints
3. User Stories & Personas
4. Functional Requirements
5. Non-Functional Requirements
6. System Architecture(Context / Container / Data Flow Diagrams)
7. Data Model
8. API Contract
9. Security & Privacy
10. Observability
11. Risks & Mitigations
12. Rollout & Migration Plan
13. Test Strategy
14. Open Questions
15. Appendix(含 POC Deviation 段)

§13 Test Strategy 依執行模式分形(Phase 1 鎖定的):

  • 人跑 → 一般測試策略(測試層級 / 覆蓋率 / 驗收方式)。
  • agent 無人值守跑 → §13 額外承載無人值守執行紀律:每條 AC 機器可檢核(silent cap 明講)、兩層閘(floor + ceiling)、原因碼迭代、紅綠燈通知(🟢🟡🔴 + pre-flight 測通才開跑)、3 出口(NEEDS_INPUT/ESCALATE/REFUSE)+ delta 防空轉、授權邊界 + stop-and-ask、可重現(recipe + run log)。這層紀律的正典在 grill/references/unattended-dispatch.md(內容無關、跨 skill 共用);prd-create 內化同一份 checklist 進 §13、與其同步(per「algorithm 內化未 vendor」原則)。human-run 不帶這層。

POC scope simplification:

  • 章末附 §15 Deviation 段,明列偏離 Guideline 的點 + 理由(如「§3.4 量化 NFR:POC 不適用 SLO 99.9%,改方向性敘述」)
  • 仍套 15 章結構,但每章內容 pragmatic 簡化

中文化 apply(caller 受眾紀律):

  • Headings 中譯(FR → 功能需求;NFR → 非功能需求 等)
  • Body inline 詞彙(stateless → 無狀態;endpoint → 端點)
  • Code / 變數名 / API path / 框架名(WinForm / DirectShow / ONNX)保留原文

PRD 內禁寫具體檔案路徑與 code snippet(實作一動就過期、變成誤導;範圍描述用模組 / 元件名,不用 src/... 路徑)。唯一例外:schema / 狀態機 / type shape 這類以片段編碼決策、比文字更精準的內容 — 修剪到只剩決策相關部分並註明出處。

寫 draft path 由 caller 指定,建議 drafts/<feature>_prd.md 或 <feature>/PRD.md。

寫完告知 user:「v0.1 draft 寫到 <path>,待 stakeholder review」。

Phase 5: Stakeholder merge loop

User 把 PRD 給 implementer / cross-team review;對方可能:

  • 直接 edit 同檔案回傳(user 拿 diff 來給 Claude)
  • 給回信 / 段落附註
  • LINE 口頭補充

Claude 處理 merge:

  • Fact 補充:直接 sync 進 PRD(如 implementer survey 結果、新規格)
  • 答案 Q&A:sync 進 §14 Open Questions table(標 ✅ + 答案來源)
  • 衝突(如 implementer 答案 vs PRD §2 Constraint 矛盾):surface 給 user 拍板,不擅自 merge
    • 衝突 surface 範例:「Implementer survey 結果建議升級 hardware 規格 → 跟 PRD §2 Constraint 『使用既有 hardware 不採購』矛盾。三種可能:(a) implementer 試過既有不行 / (b) survey 是 production spec 寫錯位置 / (c) implementer 不知道 constraint。需 user 拍板」

User 拍板後 update PRD,bump version v0.1 → v0.2。重複 stakeholder loop 直到 user 滿意。

Phase 6: Lock v1.0 + sanitize + publish ADO Wiki

User 喊「lock v1.0」/「鎖版」/「上 wiki」時走這條。

Step 6a. Sanitize per ADO publication contract

讀 caller's ADO publication contract;缺則直接用下列標準砍 8 條(本 skill 內建 fallback,無外部檔):

  1. 個人真名(PRD author / implementer / cross-team stakeholder 抽象化,砍真名 / email)
  2. 日期 / 排查紀錄 / Sprint N / PBI #N(內部 ADO ref)
  3. 內部組織 reference(新事業部 / Sprint planning)
  4. memo / 自我反思 wording(待 confirm / 推測 / 我之前 / surface)
  5. 同 wiki broken cross-link
  6. 內部 repo 路徑(drafts/ / ~/dev// / .env.local)
  7. 情緒色 / 主觀 wording
  8. 狀態進度 wording(N/M 已掃 / 漏網 / 補掃)

保留:

  • 純技術 fact / 結構化 finding / mitigation table
  • 中性架構描述
  • markdown 結構(headings / table / code block)
  • P0 / P1 / P2 / P3 風險等級

Step 6b. Resolve target wiki location

User 給 wiki path(如 /YourOrgWiki/TeamWork/<feature>/PRD)。若 parent page 不存在,先 create parent。

Step 6c. Publish via az CLI

bash
# 若 parent page 不存在
az devops wiki page create \
  --wiki <wiki-id> \
  --path "<parent-path>" \
  --content "<parent-summary>" \
  --org "$AZDO_ORG_URL" \
  --project "$AZDO_PROJECT"

# Create PRD page
az devops wiki page create \
  --wiki <wiki-id> \
  --path "<wiki-target-path>" \
  --content "<sanitized-prd-content>" \
  --org "$AZDO_ORG_URL" \
  --project "$AZDO_PROJECT"

Step 6d. Verify + 印 summary

PRD published:
  - Working draft: <caller-path>/PRD.md (v1.0)
  - ADO Wiki: <AZDO_ORG_URL>/<project>/_wiki/wikis/<wiki-id>?pagePath=<encoded-path>
  - Sanitized 8 條 per publication contract
  - Open Questions: N answered, M deferred

Templates

檔用途
templates/prd_full_guideline.md15-章 skeleton(generic,無客戶 / 公司特化)
templates/prd_poc_simplified.mdPOC 用 + §15 Deviation 段範例
fixtures/example_prd_smart_light_iot.mdMode β 精簡 fixture(辦公室智慧電燈 IoT POC)— illustrate POC PRD 怎麼寫

Caller 不確定怎麼寫某章節時,可貼 fixture 對應段給 Claude 對照。

Anti-patterns

  • ❌ Fabricate missing info — Raw input 沒提的事不要硬填,列為 question quiz user
  • ❌ 對已凍結的 spec 重跑整份 PRD 訪談 — 已核可 ADR / 鎖版 PRD 只差無人值守包裝 → 由 main 整理派工草稿並接 review-gate,不是重新 quiz §1-§15
  • ❌ Skip First-Principles check — Problem / Goal / Non-Goal / Constraints 任一缺即停下對齊,不先寫架構
  • ❌ Auto-merge stakeholder feedback with conflicts — implementer 答案跟 Constraint 矛盾要 surface user 拍板,不擅自選邊
  • ❌ Batch quiz — 一口氣丟所有 §1-§15 questions,認知負擔太大;用 numbered list iterate
  • ❌ Quiz facts — 能自查的事上桌問 caller(fact 自查、decision 才 quiz)
  • ❌ PRD 內嵌具體路徑 / code snippet — 很快過期;例外只有編碼決策的 schema / 狀態機 / type shape 片段(修剪 + 註出處)
  • ❌ Skip §15 Deviation for POC — POC 簡化版必須在 §15 Appendix 標明偏離 Guideline 的點 + 理由
  • ❌ Sanitize during draft phase — Sanitization 只在 Phase 6 publish 前做;draft 期間用真實 context(caller 自家 repo 是 source of truth,sanitized 版只是 published copy)
  • ❌ Sanitize-then-edit-in-wiki — Wiki 是 published copy,single-direction publish;要改 source(caller's PRD.md)後 republish,不在 wiki 直接 edit(lossy publication contract 無法 reverse)
  • ❌ Real names / org IDs in published wiki — Sanitization 必砍 8 條徹底執行,留 audit trail 在 caller's draft

Common pitfalls

陷阱對策
User 一次給太多 raw input 一口氣消化識別最 critical 章節(§1 + §2 + §11)先 quiz,其他章節 deferred
First-Principles 缺 Constraints 但 user 急著走 §6 架構拒走 §6,回到 §2 拍板 Constraints
Implementer review 直接改 caller's draft(不走 Q&A loop)Diff caller's draft 識別變動,分類 fact / 答案 / 衝突
POC 寫法 vs Production 寫法混用Phase 1 對齊 caller 是 POC 還是 Production scope,後續一致套用
ADO Wiki sanitize 後幾乎全砍預期行為 — 公開化 contract 是 lossy transform,sanitized 版必然薄;caller's draft 是 source of truth
Wiki page parent 不存在Phase 6b 先 create parent;az 不會 auto-create ancestor
中文化過頭把 industry term 譯成生造詞保留 JWT / OAuth / API / token / WinForm / ONNX / mp4 等 industry-standard term

跟其他 skill 的關係

  • grill(同 repo):Phase 3 quiz loop 的問法紀律(附建議答案、fact 自查 decision 才問)正典在它那;prd-create 管章節結構與優先序
  • prd-breakdown(同 repo):拿 prd-create 產出的 PRD.md → 拆 vertical slice → push ADO tasks。Chain:prd-create → prd-breakdown → ADO
  • spec skill(同 repo):caller 自家開發 spec workflow(spec.md / plan.md / tasks.md / report.md)。prd-create 產出 PRD 作 input;spec 是「caller 自身 codebase 開發 spec」不是「產品需求 PRD」
  • 無人值守交接:main 承接凍結 PRD 整理工單,再呼叫 review-gate;不重新訪談或補產品決策。§13 仍內化執行紀律,正典位置改為 grill 的 references/unattended-dispatch.md,不用為讀一份 PRD 另安裝 grill。這次只修引用,不為正式 PRD 加獨立審查關卡。
  • 這些 skill 透過 markdown 文件對接,不互相 import

Important rules

收尾 invariants — 即使前面 step 都讀過,這幾條是核心:

  1. Quiz before fabricate — 不確定的事一律列為 question 給 user,不擅自 hard-code
  2. First-Principles gate — Problem / Goal / Non-Goal / Constraints 缺一即停下對齊,不先寫架構
  3. Surface conflicts, don't auto-merge — implementer review 跟 Constraint 衝突要 surface user,不擅自選邊
  4. Caller's draft is source of truth — Wiki 是 published copy,single-direction;改要在 caller's draft 後 republish
  5. Sanitization is lossy by design — 必砍 8 條徹底執行,sanitized 版必然比 source 薄
  6. PAT via env var only — AZURE_DEVOPS_EXT_PAT 環境變數,永不上 argv
  7. POC must have §15 Deviation — POC 簡化版必須標明偏離 Guideline 的點 + 理由
  8. No Python wrapper — Claude 透過 Bash 直接 az;不寫 / 不召喚 Python helper script
  9. Industry term retain English — JWT / OAuth / API / token / WinForm / ONNX / mp4 等不譯
  10. Public repo discipline — 此 skill repo 為 public;fixtures 用 placeholders(無真名 / 真客戶 / 真政治 context)
  11. 執行模式分形 — Phase 1 先鎖定「人跑 / agent 跑」,缺則 quiz 不預設;agent-run PRD 的 §13 要帶無人值守執行紀律(對齊 grill/references/unattended-dispatch.md),human-run 不帶
  12. 凍結規格不重寫 — input 已是凍結 build spec、只差無人值守包裝 → 由 main 整理派工草稿並接 review-gate,本 skill 只在「spec 還需要被寫出來」時進場

Acknowledgments

Algorithm 內化(重寫成 prompt,未 vendor):

  • Caller's PRD authoring Guideline (caller-supplied) — 15 章結構 + First-Principles 啟動 + POC Deviation pattern
  • caller's doc_writing.md 紀律 (caller-supplied) — PRD 撰寫紀律 + 受眾中文化
  • caller's ado.md 紀律 (caller-supplied) — 公開化 contract 必砍 8 條 + ADO Wiki 同步工作流
  • Real-world PRD authoring experience (internal) — Quiz loop / Stakeholder merge / Sanitize-publish pattern 從實戰提煉

Mode β fixture (example_prd_smart_light_iot.md) 案例為 generic 辦公室智慧電燈 IoT POC,無 IP 爭議,跟既有 PRD 結構 parallel illustrate。

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Prd Create AI skill do?

Use when user wants to draft a PRD (Product Requirements Document) from raw input (meeting transcripts, hand-waved descriptions, scattered decisions). Workflow: load org's PRD Guideline + writing discipline → lock execution mode (human-run vs unattended-agent-run) → ingest raw input → quiz user numbered-list iterate to fill §1-§15 → draft v0.1 → handle stakeholder merge (review feedback, surface conflicts) → lock v1.0 + sanitize per ADO publication contract → publish to ADO Wiki. For an agent-run PRD, §13 carries the unattended-execution discipline (machine-checkable AC + traffic-light + 3-...

Why use Prd Create on TypingMind?

Because you install it once and use it with any model. Prd Create is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Prd Create in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/KerberosClaw/kc_ai_skills/tree/main/prd-create. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Prd Create?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Prd Create?

As many as you like. As long as a model supports skills, you can use Prd Create with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Prd Create AI skill free?

Yes. It is published on GitHub by KerberosClaw under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇