| name | prd |
| description | Generate a structured Product Requirements Document from a feature idea. |
| disable-model-invocation | true |
Skill: Product Requirements Document
Adapted from external PRD and clarification skills
Overview
Generate a structured, demo-grade PRD from a rough idea, then refine it through systematic questioning. Combines requirement generation with clarification into a single pass.
When to Use
- After brainstorming, to formalize a design into structured requirements
- When starting from a rough MVP or demo idea that needs product clarity
- Before running
/ux-spec for UI-heavy features
- When the feature needs explicit UX decisions, not just technical architecture
Not for: Backend-only features, pure technical tasks, or ideas that are already well-specified.
Input
The user provides a rough description — possibly vague, incomplete, or "vibe-level." You must:
- Infer missing details, but clearly label assumptions
- Optimize for a believable demo, not production scale
- Ask one clarifying question max before proceeding, then state assumptions
Process
Phase 1: Generate the PRD
Write a PRD with these 7 sections. Use concise, builder-friendly language.
1. One-Sentence Problem
[User] struggles to [do X] because [reason], resulting in [impact].
Pick the single most demo-worthy problem if multiple exist.
2. Demo Goal
- What must work for this demo to be considered successful
- What outcome the demo should clearly communicate
- Non-goals (what is intentionally out of scope)
3. Target User
Define one primary user role:
- Role / context
- Skill level
- Key constraint (time, knowledge, access, etc.)
Avoid personas or demographics.
4. Core Use Case (Happy Path)
The single most important end-to-end flow:
- Start condition
- Step-by-step flow (numbered)
- End condition
If this flow works, the demo works.
5. Functional Decisions
Required capabilities only. Use this table:
Rules: phrase as capabilities, no nice-to-haves, keep the list tight.
6. UX Decisions
Explicitly define so nothing is left implicit:
- Entry point: How the user starts, what they see first
- Inputs: What the user provides
- Outputs: What the user receives and in what form
- Feedback & states: How the system communicates loading, success, failure, partial results
- Errors: What happens on invalid input, system failure, user inaction
7. Data & Logic
- Inputs: Where data comes from (user, API, static/mocked, generated)
- Processing: High-level logic only (input → transform → output)
- Outputs: Where results go (UI only, stored, logged)
Phase 2: Clarify and Refine
After generating the PRD, ask the user for their preferred depth:
| Depth | Questions | Use When |
|---|
| Quick | 5 | Time-constrained, PRD is already solid |
| Medium | 10 | Balanced review of key areas |
| Long | 20 | Comprehensive exploration |
| Ultralong | 35 | Exhaustive deep-dive |
Then ask questions one at a time using AskUserQuestion, adapting after each answer.
Prioritize questions by impact:
- Critical path items — requirements that block other features
- High-ambiguity areas — vague language, missing acceptance criteria
- Integration points — interfaces with external systems
- Edge cases — error handling, boundary conditions
- Non-functional requirements — performance, accessibility gaps
- User journey gaps — missing steps, undefined states
After each answer, reassess:
- Did the answer reveal new ambiguities? Prioritize those.
- Did it clarify related areas? Skip now-redundant questions.
- Did it contradict earlier answers? Address the conflict.
Question quality standards:
- Specific — reference exact PRD sections
- Actionable — answer directly informs a requirement update
- Non-leading — don't suggest the "right" answer
- Singular — one question per turn
Phase 3: Finalize
After all questions are complete:
- Update the PRD with clarified requirements
- List any remaining unresolved ambiguities
- Suggest priority order for addressing open items
Output
Write the PRD to docs/plans/{topic}-prd.md.
Done When
A builder could read this PRD, build a demo without guessing, and explain the product clearly to someone else.
After the PRD
"PRD is saved. Next steps:
- UI-heavy feature: Run
/ux-spec to design the user experience before planning.
- Technical feature: Start a new session with the planner agent to break this into implementation phases."
Guidelines
- Optimize for speed + clarity
- Make reasonable assumptions explicit
- Do NOT include: architecture diagrams, tech stack decisions, pricing/GTM, long explanations
Common Mistakes
Overspecifying implementation: The PRD defines WHAT, not HOW. No tech stack, no architecture.
Skipping UX decisions (section 6): The most commonly under-specified section. Entry points, feedback states, and error handling are product decisions, not implementation details.
Too many functional requirements: If the list is longer than 8-10 items, you're past MVP. Cut ruthlessly.
Vague happy path: "User logs in and does the thing" is not a use case. Number every step.