PRD Workshop — Product Requirement Workshop
Guide product managers from their initial draft to a confirmed, structured PRD through modular interactive confirmation.
<HARD-GATE>
Do NOT generate any PRD output until all 5 modules have been confirmed one by one. Each module must be individually confirmed before proceeding to the next. This applies to EVERY session.
</HARD-GATE>
Overview
The product manager has already written an initial draft (on Feishu, Yuque, Axhub, or local files). This skill:
1. Reads and parses the input
2. Generates an information matrix (what's there vs what's missing)
3. Walks through 5 modules one by one for confirmation
4. Generates two PRD documents (product-readable + AI-engineering)
5. Pushes documents back to the remote repository
Checklist
You MUST complete these steps in order:
- Receive input — accept link, file, or natural language description
- Parse & analyze — read content, extract information, generate information matrix
- Module A: Product Overview — confirm/supplement background, goals, users, competitors
- Module B: User Stories — confirm/supplement user stories with priority
- Module C: Acceptance Criteria — confirm/supplement Given/When/Then criteria
- Module D: Non-Functional Requirements — confirm/supplement performance, security, compatibility
- Module E: Prioritization & Planning — confirm/supplement priority order, iteration plan, risks
- Generate dual PRD documents — product-readable + AI-engineering versions
- Push to remote — via MCP (Feishu) or save locally with naming convention
Step 1: Receive Input
Ask the user for the input source. Accept:
- Feishu document link — use Feishu MCP to read remotely
- Web URL (Yuque, Axhub, etc.) — use WebFetch or browser tools to scrape
- Local file (.md, .txt, .doc, .docx, .wps) — use Read tool or Python to parse
- Natural language — the user describes directly
Present the information matrix:
✅ Identified:
- Business goal: ...
- Features: ...
- Target users: ...
- Constraints mentioned: ...
❓ Need confirmation/supplement:
- Module A: [ ] Background [ ] Metrics [ ] User personas
- Module B: [ ] User stories [ ] Priority
- Module C: [ ] Acceptance criteria [ ] Edge cases
- Module D: [ ] Performance [ ] Security
- Module E: [ ] Iteration plan [ ] Risks
Ask user to confirm the matrix before proceeding.
Step 2-6: Modular Workshop
Rule: Complete one module, get explicit confirmation, then move to the next.
For each module, Read the corresponding guide:
- Module A: modules/product-overview.md
- Module B: modules/user-stories.md
- Module C: modules/acceptance-criteria.md
- Module D: modules/non-functional.md
- Module E: modules/prioritization.md
Follow the guide's questions and checklist. Record confirmed content.
Step 7: Generate Dual PRD
After all 5 modules are confirmed, generate two documents:
Document A: Product-Readable PRD
- Use templates/prd-product.md as the template
- Fill with confirmed content
- Save to: {original-name}-产品版.md (current directory)
Document B: AI-Engineering PRD
- Use templates/prd-ai-engineering.md as the template
- Fill with confirmed content
- Save to: {original-name}(AI版本).md (current directory)
Step 8: Push to Remote
If the original document came from Feishu MCP:
- Push both documents back via Feishu MCP
- Naming: original name + suffix (AI版本) for the engineering version
- Naming: original name + suffix -产品版 for the product version
If the original document came from a local file or natural language:
- Save both documents locally in the specified paths
- Suggest commit and push
Error Handling
| Scenario | Response |
|---|---|
| Feishu link inaccessible | Ask user to check permissions, offer manual paste fallback |
| Web page unreachable | Retry with WebFetch, ask user to download locally |
| Unsupported file format | Ask to convert to .docx or .txt |
| User exits mid-workshop | Save confirmed modules to temp file, offer resume next session |
Key Principles
- One module at a time — never skip ahead
- Confirm explicitly — each module requires user confirmation before proceeding
- Content from confirmed workshop only — do not invent requirements
- Dual output — always generate both versions, never just one
- Preserve original naming — AI version suffix matches original document name