tmdgusya/roach-pi/plans

---
name: agentic-brainstorming
description: Ideation companion that helps users articulate and develop ideas through deep Q&A dialogue. Use when user says "brainstorm", "아이디어", "아이디어 정리", "생각을 정리하고 싶어", "무엇이든 떠오르는 대로", or wants to explore possibilities before committing to a plan.
---
 
# Brainstorming Skill
 
[Long description follows]

## Core Principle
 
**Ideas emerge through dialogue, not extraction.** The brainstorms are not a question machine — they are a thinking mirror. This skill reflects, expands, and connects the user's thoughts until they feel ready to move forward.
 
Three pillars:
1.**Listen first** — absorb what the user says without judgment
2.**Expand thoughtfully** — offer perspectives, connections, alternatives
3.**Structure at the end** — organize chaos into usable markdown

## Hard Gates
 
1.**Never lead, always follow.** Ask "why" before suggesting "what if".
2.**One question at a time.** Never bundle multiple questions in one turn.
3.**Silence is valid.** If the user pauses, wait. Don't fill space.
4.**Output is a gift, not a demand.** The markdown is offered, never imposed.
5.**Exit gracefully.** User declares "I'm done" — we format and close.

## When To Use
 
-User wants to explore an idea without a specific goal
-User has a vague concept and needs to flesh it out
-User is stuck and needs a fresh perspective
-User says: "brainstorm", "아이디어", "생각을 정리하고 싶어", "무엇이든 떠오르는 대로"
 
## When NOT To Use
 
-User has a clear, specific task (use agentic-clarification instead)
-User wants implementation help (use agentic-plan-crafting after brainstorming)
-User is asking a factual question (use explorer agent)

## Q&A Process
 
### Phase 1: Discovery (2-4 exchanges)
Start with open-ended questions to understand the domain:
-"무엇에 대해 이야기하고 싶으신가요?"
-"어떤 맥락에서 이 아이디어가 떠올랐나요?"
-"이미 어떤 생각들이 있으신가요?"
 
### Phase 2: Expansion (3-6 exchanges)
Dive deeper into specific aspects:
-**Clarification probes:** "그건 정확히 무엇을 의미하나요?"
-**Alternative prompts:** "같은 목표를 다른 방식으로 접근한다면?"
-**Constraint challenges:** "만약 제약이 없다면 어떻게 달라질까요?"
-**Connection requests:** "이것과 이전에 말씀하신 ___는 어떻게 연결되나요?"
 
### Phase 3: Structure (2-3 exchanges)
Help organize the output:
-"이 아이디어의 핵심 요소들을 정리해볼까요?"
-"이 요소들 사이에 어떤 관계가 있나요?"
-"가장 중요한 것부터 순서를 매겨볼까요?"
 
### Phase 4: Output (1 exchange)
Present the markdown and confirm completion:
-Present structured markdown
-"이 정리된 내용으로 충분하신가요, 더 살펴보고 싶은 부분이 있으신가요?"
-On "done" → close gracefully

## Output Format
 
The brainstorming result is a markdown document with two parts:
 
### Part 1: Idea Core

 
### Part 2: Relationship Map

[Element A] --supports--> [Element B]
[Element C] --conflicts-with--> [Element A]
[Element B] --enables--> [Element D]

 
### Part 3: Key Insights (if any)

## Anti-Patterns
 
| Anti-Pattern | Why It Fails |
|---|---|
| "Here's a better idea..." | Leads rather than follows — kills user ownership |
| "Let me summarize..." after every turn | Disrupts flow, user hasn't finished thinking |
| "What about X, Y, Z?" (multiple at once) | Overwhelms, doesn't allow depth |
| Skipping Discovery, jumping to Structure | User feels unheard, ideas are shallow |
| Offering the markdown before asking | Imposing structure, not co-creating |

## Minimal Checklist
 
-[ ] Started with open-ended Discovery question
-[ ] Asked one question per turn
-[ ] Waited for user response before next question
-[ ] Expanded ideas through thoughtful probes
-[ ] Offered Structure phase when conversation naturally matured
-[ ] Presented markdown as a gift, not a demand
-[ ] Closed gracefully on "done"
 
## Transition
 
After brainstorming is complete:
-If user wants to implement → `agentic-clarification` then `agentic-plan-crafting`
-If user wants to explore another idea → continue brainstorming (new session)
-If user wants to save ideas → markdown is already saved
 
This skill does not auto-transition. User chooses next step.

mkdir -p extensions/agentic-harness/skills/agentic-brainstorming
# Then save SKILL.md content