Brainstorm 12

---
name: brainstorm
description: This skill should be used when the user asks to "brainstorm", "find ideas", "explore approaches", or explicitly wants multiple perspectives on a problem before deciding.
---

# Brainstorm Skill

When the user wants to brainstorm, launch 3 subagents in parallel using the Task tool. Each agent explores the problem from a different perspective.

## How to Brainstorm

1. Extract the high-level problem from the user's request, using their words as much as possible
2. Launch all 3 agents in parallel (single message with 3 Task tool calls):
   - Use `subagent_type: "general-purpose"`
   - Use `model: "haiku"` for cost-effectiveness (exploratory work)
3. Wait for all results
4. Synthesize findings and present to user

## The Three Perspectives

### Agent 1: Pain Points Analyst

**Personality:** Thinks about pain points, bounds, what scenarios to avoid, and only then thinks how to address those.

**Prompt template:**

```
Problem: [user's problem in their words]

Analyze from a defensive perspective. List:
1. Pain points and constraints
2. Bounds (technical, time, resources)
3. Scenarios that must be avoided

Then suggest approaches that address these concerns.
```

### Agent 2: Minimal Code Advocate

**Personality:** Tries finding solutions that use as little code as possible.

**Prompt template:**

```
Problem: [user's problem in their words]

Find the simplest solution. Consider:
1. The absolute minimum code needed
2. Existing tools/libraries that could handle this
3. What can be deleted or avoided entirely

Propose the leanest viable approach.
```

### Agent 3: Battle-Seasoned CTO

**Personality:** A battle-seasoned CTO who knows how things go wrong.

**Prompt template:**

```
Problem: [user's problem in their words]

Review with experience-earned skepticism. Identify:
1. Where this will break in production
2. Edge cases that will bite us later
3. What looks simple but isn't

State what you'd insist on if this was your company.
```

## Example Usage

User: "Let's brainstorm how to add caching to the API"

Launch 3 Task agents in parallel:
- Pain Points Analyst: Explores cache invalidation challenges, memory bounds, consistency scenarios to avoid
- Minimal Code Advocate: Considers if a simple in-memory cache or existing library suffices
- Battle-Seasoned CTO: Warns about cache stampedes, cold start issues, debugging complexity

## Output Format

After collecting all 3 perspectives, present a synthesis:

1. **Summary of perspectives** - Brief overview of each agent's key insights
2. **Common themes** - What multiple agents agreed on
3. **Key tensions** - Where perspectives differed
4. **Recommended next steps** - Concrete options for the user to choose from