| name | brainstorm |
| description | Socratic clarification skill that surfaces assumptions before `/prd`. Triggers: brainstorm, surface-assumptions, clarify, ambiguity-check, pre-prd, what-am-i-missing. Read-only; output to `output/brainstorm/` only. |
/brainstorm — Pre-/prd Socratic clarification
Prerequisites
- A requirement note exists in
input/requirements/<file>.md, or a one-line topic can be passed in interactively
- Write permission for
output/brainstorm/ is granted (in settings.local.json)
/prd has not yet been started (this skill is upstream of /prd)- Applicability: any of — the note is under half a page / the "what we won't do" is unclear /
stakeholders or success metrics are undefined
(reference:
@.claude/learnings/L0001-brainstorm-before-prd.md)
Principles
- Question-driven: never inflate features by guessing. Surface ambiguity via
AskUserQuestion
- No code or production-doc changes: this skill's charter is structuring assumptions only
- Socratic pacing: max 4 questions per round, max 3 rounds (avoid user fatigue)
Usage
Examples
/brainstorm input/requirements/REQ_subscription.md
/brainstorm Add a subscription feature (one-liner)
Output destination
output/brainstorm/BRAINSTORM_<topic>_<YYYY-MM-DD>.md only- No writes to other directories
Integration with Other Skills
- Downstream (required):
/prd <output-file> — build the formal PRD off the seed
- Parallel (optional):
/adr <title> — record significant design judgments as they emerge
- Upstream: none (this is the top of the PRD pipeline)
Workflow
- Read input: read the requirement note. Or ask for a 1-2 sentence seed interactively
- Socratic question salvo (
AskUserQuestion, max 4 per round, max 3 rounds):
- Motivation (Why now? Would the same value hold one month from now?)
- Users (Who is the primary persona? Niche cases?)
- Success (What does post-launch success look like? Which metrics?)
- Out-of-scope (What are we deliberately not doing? Why?)
- Exit (How do we roll back if it fails? Data implications?)
- Constraints (Which legal / security / SLA constraints can't be relaxed?)
- Alternatives (Could the same value be delivered differently?)
- Surface implicit assumptions: mark with
[Assumption]
- Propose 3 scope variants: Minimum / Standard / Ambitious
- List open questions: star (★) the ones that must resolve before
/prd
- Save: write to
output/brainstorm/ and present to the user
- Gate decision: ask whether to proceed to
/prd (see "Gates" below)
Output Contract
output/brainstorm/BRAINSTORM_<topic>_<YYYY-MM-DD>.md must contain these sections:
## 1. Core request (one sentence)
## 2. Assumption inventory (explicit + implicit)
## 3. Stakeholders and stakes (who expects what, who fears what)
## 4. In-scope / out-of-scope candidates (3+ proposals)
## 5. Open-question list (★-prioritized)
## 6. Next step (Are we ready to run `/prd`?)
Prohibitions
- No code, test, or production-doc changes (
docs/, src/, tests/ off-limits)
- No requirement inflation: never add features the input didn't request
- No write outside
output/brainstorm/
- No question fatigue: max 4 questions per round; max 3 rounds total
- No secret transcription: redact discovered API keys, etc., as
[REDACTED]
Gates
- Gate 1: At end of brainstorm, ask "Ready to proceed to
/prd?"
- Gate 2: If 3+ ★-marked open questions remain, hold off on transitioning to
/prd
(only pass through if the user explicitly says "proceed")
Socratic question templates
| Category | Example |
|---|
| Motivation | Why now? Would the same value hold one month from now? |
| Users | Who is the primary persona? Which niche cases exist? |
| Success | What does post-launch success look like? What metrics? |
| Out-of-scope | What are we deliberately not doing? Why? |
| Exit | If this fails, how do we roll back? Data implications? |
| Constraints | Which legal / security / SLA constraints can't be relaxed? |
| Alternatives | Could the same value be delivered differently? |
