English ｜ [简体中文]
An AI-powered Product Requirements Document writing assistant — from raw ideas to structured, ready-to-build PRDs in minutes.
Writing a good PRD is one of the most important — and most tedious — tasks in product development. PRD-Writer automates the tedious parts: the structure, the prioritization, and the acceptance criteria. You bring the product thinking; PRD-Writer brings the scaffolding.
Whether you have a rough idea, an existing draft that needs polishing, or a Feishu document you want to turn into a full PRD — PRD-Writer adapts to your workflow and delivers consistent, professional results in English or Simplified Chinese.
Disclaimer: This project is 99% AI-generated. The AI's owner has no programming background. Please evaluate the project's feasibility before use.
| 🔄 Anthropic 3-Phase Workflow | Context Gathering → Refinement & Structure → Reader Testing |
| 📊 MoSCoW Prioritization | P0 / P1 / P2 / Won't — every feature has a clear priority label |
| ✅ Given/When/Then Criteria | BDD-style acceptance criteria your engineering team can execute directly |
| 🌐 Bilingual Templates | Full English and Simplified Chinese support with consistent structure |
| 🔗 Feishu Doc Integration | Read existing Feishu documents and generate or refine PRDs from them |
| 📐 7-Chapter PRD Structure | Complete, production-ready output covering every PRD essential |
PRD-Writer adopts Anthropic's proven collaborative writing methodology, ensuring AI and user form a clear, efficient collaboration loop:
- Phase 1 — Context Gathering: Active questioning to clarify product background, target users, core scenarios, and constraints
- Phase 2 — Refinement & Structure: Organizes all information into a complete 7-chapter PRD with MoSCoW-prioritized features
- Phase 3 — Reader Testing: Validates the document from a reader's perspective, confirms key decisions, and iterates on feedback
No more vague "high/medium/low" labels. PRD-Writer applies the industry-standard MoSCoW framework to every feature:
| Priority | Label | Meaning |
|---|---|---|
| P0 | Must Have | Core functionality; a launch blocker if missing |
| P1 | Should Have | High-impact features that significantly improve the product |
| P2 | Could Have | Nice-to-have enhancements for the current release |
| Won't | Won't This Iteration | Explicitly deprioritized; future candidates |
Every P0 and P1 feature includes acceptance criteria in the BDD-inspired Given/When/Then format:
Given [precondition — system state before the action]
When [trigger — the user's specific action]
Then [expected outcome — observable, verifiable result]
Your engineering team can translate these directly into automated test cases. No more ambiguous acceptance criteria.
PRD-Writer ships with full English and Simplified Chinese PRD templates. Input your requirements in any language; output a structurally identical, professionally written PRD in your target language. Ideal for:
- Cross-border product teams collaborating across language barriers
- Startups building for both domestic and international markets
- Freelancers serving clients globally
Whether you're at the very beginning of a product idea or working from existing materials, PRD-Writer meets you where you are:
- Mode A — Raw idea → Complete PRD from scratch
- Mode B — Existing draft → Polished, restructured PRD
- Mode C — Feishu doc link → Iterative, multi-round PRD refinement
Every output follows a proven, production-ready structure:
- Product Overview — Vision, goals, scope, and value proposition
- User Analysis — Personas and user journeys
- Feature Specifications — MoSCoW-prioritized feature list and detailed descriptions
- Success Metrics — North Star metric, leading and lagging indicators
- Acceptance Criteria — Given/When/Then criteria for all P0/P1 features
- Technical Constraints — Tech stack, integrations, security, and compliance
- Appendix — Glossary, version history, and contacts
When you provide a Feishu document link (Feishu/Lark), PRD-Writer automatically reads its content, identifies gaps, and generates or refines a full PRD through multi-round collaboration.
git clone https://github.com/your-org/prd-writer.git
cd prd-writer
# Follow the setup instructions in the repository# Use the ClawHub CLI to install PRD-Writer as an OpenClaw skill
clawhub install prd-writerPrerequisites: Node.js 18+, npm 9+, and an active OpenClaw installation. For detailed installation instructions, see the User Guide 📖.
In any conversation, trigger PRD-Writer using any of the 11 trigger words, for example:
write a PRD
or in Chinese:
写PRD
PRD-Writer enters Phase 1 (Context Gathering) and asks 2–3 targeted questions about your product idea:
- Who is the target user?
- What are the top 3 core features?
- What does success look like (quantifiable)?
Answer as many or as few as you can. PRD-Writer will use
[待确认]placeholders for anything unspecified.
PRD-Writer enters Phase 2 and generates the full 7-chapter PRD with:
- ✅ MoSCoW-prioritized feature list
- ✅ Given/When/Then acceptance criteria for every P0/P1 feature
- ✅ Completion indicators on each chapter (✅ = complete,
⚠️ = needs confirmation)
Then Phase 3 asks you to review and confirm key decisions before finalizing.
Goal: Collect enough background to write the PRD without guessing.
PRD-Writer collects information across 8 dimensions, one at a time:
| Dimension | PRD Chapter | Example Question |
|---|---|---|
| Project background & goals | Product Overview | What problem does this product solve? |
| Target users | User Analysis | Who are the primary users and what are their pain points? |
| Core features | Feature Specifications | What are the top 3 must-have features? |
| Priority decisions | Feature Specifications | What gets cut if you only have 2 weeks? |
| Success metrics | Success Metrics | What quantifiable outcome do you expect? |
| Acceptance criteria | Acceptance Criteria | What does "done" look like for the core feature? |
| Technical constraints | Technical Constraints | Any platform, tech stack, or integration limits? |
| Output format | Delivery | Should I write this to a Feishu doc or reply here? |
PRD-Writer asks 2–3 questions per round to avoid overwhelming you. At the end of Phase 1, it outputs a Context Checklist for your confirmation before generating the document.
Goal: Generate a complete, well-structured PRD document.
Based on confirmed context, PRD-Writer produces all 7 chapters in order of priority:
| Order | Chapter | Priority |
|---|---|---|
| 1 | Product Overview | P0 (mandatory) |
| 2 | User Analysis | P0 (mandatory) |
| 3 | Feature Specifications (MoSCoW) | P0 (mandatory) |
| 4 | Success Metrics | P1 (should have) |
| 5 | Acceptance Criteria (Given/When/Then) | P1 (should have) |
| 6 | Technical Constraints | P2 (could have) |
| 7 | Appendix | P2 (could have) |
Any information not provided by the user is marked with
[待确认]and flagged in the chapter's completion indicator.
Goal: Validate clarity, completeness, and consistency before final delivery.
PRD-Writer reviews the completed PRD from the reader's perspective, covering:
- ✅ Does the Product Overview clearly explain why we are building this?
- ✅ Are target users and their journeys complete?
- ✅ Are P0/P1/P2 priorities aligned with your business goals?
- ✅ Are Given/When/Then criteria specific and verifiable?
- ✅ Are success metric target values realistic?
PRD-Writer also surfaces key decision points for explicit confirmation:
"The target for first response time is 24 hours. Would you prefer a more aggressive target like 4 hours, or a more conservative one like 48 hours?"
You can request updates to any chapter at any time. PRD-Writer marks changes with a version diff (e.g., [v0.9 → v0.91]).
PRD-Writer activates when you use any of the following trigger words or phrases. Use them in either English or Chinese — PRD-Writer auto-detects the language and adapts.
| # | English Trigger | 中文触发词 |
|---|---|---|
| 1 | write a PRD |
写PRD |
| 2 | create a PRD |
做PRD |
| 3 | generate PRD |
生成PRD |
| 4 | PRD writing |
PRD写作 |
| 5 | requirements document |
需求文档 |
| 6 | product requirements document |
产品需求文档 |
| 7 | write a requirements document |
编写需求文档 |
| 8 | prd-writer |
prd-writer |
| 9 | product document |
产品文档 |
| 10 | output a PRD |
输出PRD |
| 11 | PRD template |
PRD模板 |
| 12 | draft a PRD |
起草PRD |
| 13 | plan a PRD |
规划PRD |
| 14 | compose a PRD |
撰写PRD |
| 15 | organize a PRD |
整理PRD |
| 16 | feature requirements |
功能需求 |
| 17 | turn requirements into a document |
把需求写成文档 |
| 18 | refine a PRD |
完善PRD |
| 19 | optimize a PRD |
优化PRD |
| 20 | complete a PRD |
补全PRD |
Tip: You don't need an exact match — PRD-Writer understands intent. Just say "I need a PRD for a new feature" and it activates automatically.
Best for: Products from 0 to 1, or new features with no existing documentation.
You describe your product idea in plain language — rough, incomplete, and conversational is fine:
You: I want to build a user feedback system where customers can submit
issues and get notified when they're resolved.
PRD-Writer activates → asks 2–3 clarifying questions → generates the full 7-chapter PRD.
Best for: Unstructured notes, half-written drafts, or outdated PRDs that need a refresh.
You paste your existing content:
You: [paste a rough PRD draft or bullet-point wishlist]
PRD-Writer analyzes it → identifies missing chapters and content gaps → asks targeted follow-up questions → outputs a fully structured, prioritized PRD.
Best for: Teams already working in Feishu who have scattered requirements across multiple documents.
You share a Feishu document URL:
You: https://feishu.cn/doc/xxxxx — Generate a PRD from this.
PRD-Writer reads the document → identifies gaps → confirms missing info with you → generates a polished PRD. Supports multi-round iteration for partial updates to specific chapters.
Most AIs generate a generic, template-filling document. PRD-Writer follows a structured three-phase workflow — it asks targeted questions, applies MoSCoW prioritization, and outputs Given/When/Then acceptance criteria that your engineering team can act on directly. The output is not just text; it's a working document with verifiable completion criteria.
Currently, the full template and workflow are optimized for English and Simplified Chinese. Other languages are partially supported — the workflow functions, but template labels and section headers will fall back to English.
The complete user guide with detailed examples, troubleshooting, and advanced usage tips is available in the 飞书云文档用户指南 📖.
Yes. PRD-Writer supports partial updates. Simply tell it which chapter to revise (e.g., "Please update Chapter 3 — add a P1 feature for email notifications"). Changes are marked with a version diff.
Items marked [待确认] indicate information you did not provide during Context Gathering. PRD-Writer will not guess — it leaves a clear placeholder so you know exactly what needs a business decision. You can fill these in later by telling PRD-Writer to update the specific section.
Yes. When you trigger PRD-Writer, specify your preferred output method:
- Default: Markdown output in the conversation
- Optional: Written directly to a Feishu Cloud Document (provide the folder or wiki link)
PRD-Writer's design philosophy draws direct inspiration from Anthropic's official doc-coauthoring project, which pioneered the application of large language model collaboration techniques to structured document creation.
The three-phase workflow — Context Gathering → Refinement & Structure → Reader Testing — is directly derived from Anthropic's collaborative writing methodology. Anthropic demonstrated how structured AI collaboration can transform open-ended document tasks into efficient, high-quality outputs.
We are grateful to the Anthropic team for advancing the state of the art in AI-assisted writing and documentation, and for providing a framework that PRD-Writer builds upon.
Last updated: 2026-03-30 | PRD-Writer v2.0.0
- [Bug] Fixed Lagging/Leading concept confusion in Success Metrics (Leading = 领先, Lagging = 滞后)
- [Improvement] Template paths changed to relative paths for better portability
- [Improvement] Trigger words expanded from 11 → 20
- [Improvement] Added mode priority rules for A/B/C input mode conflicts
- [Improvement] Added common error examples for Given/When/Then format
- [Improvement] Reader Testing checklist split into 3 rounds (P0/P1/P2)
- [Improvement] Added invalid input rejection rules
- [Improvement] Added Entry/Exit Criteria for each phase
- [Enhancement] Reader Testing now includes specific operational guidance
- Initial release