| name | prd-builder |
| description | Socratic PRD builder. Interrogates the user through problem, solution, and feasibility phases until zero ambiguity, then generates a structured PRD and saves it to .agents/prd.md. Run /init-project after to scaffold the project. |
PRD Builder Skill
When to Use
- User runs
/prd-builder with an idea or no input
- User wants to build a PRD before running
/init-project
- User wants structured interrogation before committing to a project plan
Phase 1: Initial Intake
Read what the user provided after the command. If they gave a description, use it as the starting point. If nothing was provided, open with:
Let's build your PRD. I'll ask questions until we have zero ambiguity.
Start here: what's the one-sentence version of what you're building?
Do not proceed until you have at least a rough description of the product idea.
Phase 2: Interrogate the Problem
Ask until you can answer all of these with confidence — no guessing:
The problem:
- What specific problem does this solve?
- Who has this problem? (Not "developers" — "solo devs shipping SaaS with no ops team")
- How painful is it? (Nice-to-have or blocking their work?)
- How do they solve it today? (Manual workaround, competing tool, ignoring it?)
Market signal:
- Have you talked to anyone who has this problem?
- Are there existing solutions? What's wrong with them?
Ask in batches of 2–3 questions. Don't dump all at once. Wait for answers before continuing.
Phase 3: Interrogate the Solution
Ask until you can answer all of these:
The product:
- What does the product do, step by step, from the user's perspective?
- What's the core loop — the thing a user does every single session?
- What are the 3–5 features that MUST exist for v1?
- What are you explicitly NOT building in v1?
Users and access:
- Who are the target users?
- How do they access it? (Web app, CLI, API, mobile, desktop?)
- Is there a concept of accounts, teams, or shared workspaces?
Data:
- What data does the product create or consume?
- Does it persist data? Where? (local, cloud, user-owned?)
- Any sensitive data? (PII, financial, health?)
Phase 4: Interrogate Feasibility
Ask until you can answer all of these:
Tech:
- Any strong preference on stack? (Language, framework, hosting?)
- Budget: free-tier only, or paid services available?
- Existing infrastructure to integrate with? (Auth provider, database, APIs?)
Constraints:
- Timeline for v1? Hard deadline or aspirational?
- Team size? Solo or collaborators?
- Compliance requirements? (GDPR, HIPAA, SOC 2, etc.)
Success criteria:
- How do you know v1 is working? What's the first signal of success?
- What's the riskiest assumption in this plan?
Phase 5: Generate PRD
Once all phases are resolved, produce the PRD:
# Product Requirements Document
## Product
[One sentence: what it is]
## Problem
[The specific problem, who has it, how painful, current solutions and their gaps]
## Target User
[Specific persona — not a demographic, a person with a specific context and job]
## Solution
[What the product does, the core loop, how it solves the problem]
## Core Features (v1 must-haves)
1. [Feature] — [Why it's essential]
2. [Feature] — [Why it's essential]
...
## Out of Scope (v1)
- [Explicit exclusions with brief reason]
## Technical Constraints
- Stack: [language, framework, database, hosting]
- Budget: [free / paid / hybrid]
- Timeline: [v1 target]
- Team: [size]
- Compliance: [requirements or none]
## Non-Functional Requirements
- [Performance, scale, security, accessibility, etc.]
## Success Criteria
- [How you'll know v1 is working]
## Riskiest Assumptions
1. [Assumption] — [Mitigation plan]
## Open Questions
1. [Unresolved question]
Phase 6: Validate and Refine
Present the PRD and ask:
Here's your PRD. Review it and tell me:
- Anything wrong or missing?
- Any scope that should shift?
- Any constraints I got wrong?
I'll revise until you're satisfied, then save it.
Iterate until the user confirms the PRD is accurate. Do NOT save until confirmed.
Phase 7: Save and Hand Off
Once confirmed:
- Save the PRD to
.agents/prd.md
- Confirm save with:
PRD saved to .agents/prd.md
Next step: run /init-project — it will read your PRD automatically, run research,
and scaffold the full project plan without repeating these questions.
Do NOT scaffold, create tasks, or write state.json — that's /init-project's job.
Notes
- Never generate the PRD speculatively before the user confirms the intake phases
- If the user provides a fully-formed idea upfront, you can compress Phases 2–4 into fewer questions — but still verify all the required fields before generating
- If the user says "just generate it" before all fields are resolved, list what you're assuming and ask them to confirm the assumptions before proceeding