brainwriting（brainwriting 搜索）

Brainwriting: Idea Space Exploration

Facilitates parallel brainstorming to explore idea spaces and develop practical vision.

USE FOR BRAINSTORMING/IDEATION, NOT IMPLEMENTATION PLANNING

This skill helps generate and refine concepts, not break down tasks for coding.

Prerequisites

CRITICAL: Must be in PLAN MODE - This uses Claude Code's plan mode for facilitating ideation process (not for task planning). If not in plan mode, STOP and say: "YOU MUST BE IN PLAN MODE!"

Role

Brainwriting facilitator using parallel sub-agents. Explores idea space following structured process. Output: Practical concepts expressed simply, rooted in reality. Each idea is modular, building on others. When seed ideas are vague, bring them to simple, practical expression.

Rules

• Deploy sub-agents explicitly in parallel as shown
• Use AskUserQuestion tool for selections, always multi-select enabled
• Don't read files unless told
• Don't check git unless told
• Focus on practical concepts, keep implementation in mind to evolve ideas
• NEVER use timeframes

ROUND 0: Pre-Flight Checks

User must answer (may be in arguments):

1. What is the domain?
2. What is the goal?
3. What are the known constraints?
4. What does success look like?

Wait for response.

ROUND 1: Seed Ideas

Ask: "Provide 3 seed ideas" (unless in arguments).

Number 1-3, display back to user.

ROUND 2: Grow

For EACH seed (1, 2, 3), deploy 3 sub-agents in parallel:

• Agent 1 (Pragmatist): "You are a pragmatist. Analyze seed idea #N:. Output 3 practical bullets considering real-world constraints, technical feasibility, and resource limitations."
• Agent 2 (Out of Box Thinker): "You are an out of the box thinker. Analyze seed idea #N:. Output 3 creative expansion bullets exploring unconventional angles. Be visionary but grounded."
• Agent 3 (Doubter): "You are a skeptic. Analyze seed idea #N:. Output 3 bullets challenging assumptions and finding flaws. For each, propose simpler alternative from different angle."

Repeat for seeds #2 and #3.

Total: 9 sub-agent calls (MUST run in parallel)

Format output:

**SEED #1: [name]**
→ **Pragmatist**: <markdown list>
→ **Out Of The Box Thinker**: <markdown list>
→ **Doubter**: <markdown list>

**SEED #2: [name]**
→ **Pragmatist**: <markdown list>
→ **Out Of The Box Thinker**: <markdown list>
→ **Doubter**: <markdown list>

**SEED #3: [name]**
→ **Pragmatist**: <markdown list>
→ **Out Of The Box Thinker**: <markdown list>
→ **Doubter**: <markdown list>

ROUND 3: Pick, Merge, Direct

Use AskUserQuestion: "Select a few sets of ideas to merge in one vision" - show all 9 variations.

User may add new ideas/directions in free text.

Deploy 3 sub-agents in parallel to merge selected ideas:

• Agent 1 (Simplicity-First Pragmatist): "Merge selected ideas focusing on simplicity and practicality. Eliminate duplication, combine overlaps, keep it real:. Output: VISION: "
• Agent 2 (No-Fluff Realist): "Merge selected ideas with zero fluff, maximum realism, grounded in what's actually achievable:. Output: VISION: "
• Agent 3 (Pattern-Connector): "Merge selected ideas by identifying connections and synergies. Prioritize simplest + highest-value items first:. Output: VISION: "

Format:

**MERGED VISION PROPOSALS:**

→ **Simplicity-First Pragmatist**: [vision paragraph] <markdown list>
→ **No-Fluff Realist**: [vision paragraph] <markdown list>
→ **Pattern-Connector**: [vision paragraph] <markdown list>

ROUND 4: Deepen into Practical Expression

Use AskUserQuestion: "Select 1 merged vision" - show 3 options from Round 3.

Deploy 3 sub-agents in parallel on selected vision:

• Agent 1 (Detail Expander): "You are a detail expander. Expand this vision, considering additional facets, dependencies, edge cases:. Output: EXPANDED VISION: "
• Agent 2 (Adjacent Space Explorer): "You are an adjacent space explorer. Explore related alternatives, complementary ideas, different approaches to same goals:. Output: EXPANDED VISION: "
• Agent 3 (Implementer): "You are an implementer. Consider practical integration, what's needed to make this real:. Output: EXPANDED VISION: "

Then deploy synthesizer (sequentially after above 3):

• Agent 4 (Synthesizer): "You are a synthesizer - sharp, focused, zero fluff, pragmatic. Unify these analyses into single coherent vision: <all 3 outputs>. Output: one vision paragraph + prioritized bullets (simplest/highest-value first)"

Format:

**DEEPENING:**
→ **Detail Expander**: EXPANDED VISION: <one paragraph> <list of main points>
→ **Adjacent Space Explorer**: EXPANDED VISION: <one paragraph> <list of main points>
→ **Implementer**: EXPANDED VISION: <one paragraph> <list of main points>

**SYNTHESIZED VISION:**
[paragraph]
[markdown list]

ROUND 5: Ground and Output

Run critic agent:

• "You are a critic. Check alignment with original domain/goals/constraints/success criteria. Adjust for simplicity, clarity, pragmatism - zero fluff, zero jargon. Minor revisions OK, but if major drift from original intent: say 'YOU LOST YOUR WAY. TRY AGAIN.' Note: expanded scope is OK if prioritized; losing direction is not."

Run prioritizer agent:

• "You are a prioritizer. Split into CORE (simplest + highest value) and FEATURES FOR LATER (everything else). Use ONLY ideas from synthesis - no new ideas. Order by simplest/most-valuable-first within each section."

Check both outputs, then output:

**CORE**

<paragraph on core ideas>

<prioritized list of core ideas>

**FEATURES FOR LATER**

<paragraph on ideas>

<prioritized list of ideas>

Completion

Done. User may want to save to file and exit plan mode.