| name | prd-extract |
| description | Extracts a structured PRD.md and tactical implementation plan from the current conversation — captures goals, non-goals, scope, risks, and acceptance criteria. Use when ending a brainstorm before implementation, before invoking the architect agent, or "PRD 뽑아줘", "기획서 정리", "스펙 문서 추출". |
| author | subinium |
| user-invocable | true |
| disable-model-invocation | true |
| allowed-tools | Read, Write |
| argument-hint | [--with-plan] [--output=<path>] |
PRD Extract
Read the conversation context, extract a clean PRD.md (and optionally a tactical plan), save to disk.
Arguments
--with-plan — also generate PLAN.md using the structure from ~/.claude/rules/plan-template.md
--output=<path> — save to this path instead of ./PRD.md (and ./PLAN.md)
Process
1. Re-read conversation
Scan the visible conversation for:
- The user's stated GOAL (what they want to build / fix / explore)
- Constraints and non-goals (what they explicitly excluded)
- Trade-offs discussed (and which side won)
- Risks raised
- Decisions made (and the reasoning)
- Open questions left unanswered
If the conversation is sparse / unclear, ASK 1-2 clarifying questions before writing — do not invent.
2. Write PRD.md
Use this structure:
# PRD: <feature name>
> **Status:** Draft · **Owner:** <user> · **Date:** YYYY-MM-DD
## Problem
What problem are we solving? Who has it? Why now?
## Goals
- Goal 1 (measurable if possible)
- Goal 2
## Non-goals
- Explicitly NOT solving X
- Explicitly NOT supporting Y
## Scope
- In scope: [bullet list]
- Out of scope: [bullet list, with brief reason for each]
## User flow
1. User does X
2. System responds with Y
3. ...
## Acceptance criteria
- [ ] Concrete, testable criterion 1
- [ ] Concrete, testable criterion 2
## Risks
| Risk | Impact | Mitigation |
|---|---|---|
## Open questions
- Question 1 — needs answer from [whom]
- Question 2
## References
- Conversation: <date / session ID if known>
- Related docs / issues
3. (Optional) Write PLAN.md
If --with-plan: use ~/.claude/rules/plan-template.md as the structure. Map PRD goals → tactical steps with:
- File ownership per step
- Dependencies between steps
- Risk-first sequencing (highest uncertainty step first)
- Parallel groups identified
4. Output paths
- Default:
./PRD.md (and ./PLAN.md if --with-plan)
- If file exists: append timestamp suffix (
PRD-YYYYMMDD-HHMMSS.md) — never silently overwrite
- Print the saved paths and a 3-line summary of what was extracted
Constraints (SOFT)
- Don't fabricate goals/criteria the user never mentioned — it's better to leave a section empty than to invent
- Don't include implementation details in the PRD (those belong in PLAN.md or architect output)
- Don't write more than ~600 words in PRD body — it's a contract, not a manifesto
- Korean response per global CLAUDE.md (PRD body itself can be Korean if conversation was Korean)