Brainstorm Refine

Overview

Engage in iterative idea refinement using the brainstorm-refiner agent. Each cycle:

• Agent explores, challenges, and presents options
• You choose direction
• Agent refines based on your choice
• Repeat until ready for implementation

When to Use

• Developing vague ideas into concrete concepts
• Evaluating multiple approaches to a problem
• Challenging assumptions systematically
• Sharpening requirements before implementation

Workflow

digraph brainstorm_loop {
    "Start" -> "Get initial idea";
    "Get initial idea" -> "Launch brainstorm-refiner agent";
    "Launch brainstorm-refiner agent" -> "Present options/questions to user";
    "Present options/questions to user" -> "User chooses direction";
    "User chooses direction" -> "Exit?" [label="check"];
    "Exit?" -> "Write design doc" [label="yes"];
    "Exit?" -> "Launch brainstorm-refiner agent" [label="no, with context"];
    "Write design doc" -> "Offer implementation path";
}

Implementation

Phase 1: Initial Exploration

1. If user provides an idea, proceed directly
2. If no idea provided, ask: "What idea would you like to explore?"
3. Launch brainstorm-refiner agent with:
   • User's idea/concept
   • Any context from the codebase if relevant
   • Request for structured exploration

Phase 2: Iteration Loop

For each iteration:

1. Launch Agent: Use Task tool with subagent_type=brainstorm-refiner

   • Provide current idea state
   • Include previous refinements and decisions
   • Request structured output (options, questions, validation)

2. Present Results: Share agent's output with user

3. Get Direction: Use AskUserQuestion with options:

   • Go deeper: Explore one of the presented options in detail
   • New angle: Pivot to explore a different aspect
   • Refine: Sharpen the current direction
   • Ready: Move to design documentation
   • Exit: End brainstorming session

4. Update Context: Based on user choice:

   • If "Go deeper": Note which option, launch agent focused on that option
   • If "New angle": Ask what angle, launch agent with new focus
   • If "Refine": Launch agent with refinement request
   • If "Ready" or "Exit": Exit loop

Phase 3: Completion

When user is ready:

1. Summarize the refined idea
2. Write design document to docs/plans/YYYY-MM-DD-<topic>-design.md
3. Offer next steps:
   • Use superpowers:writing-plans for implementation planning
   • Use superpowers:using-git-worktrees for isolated development

Exit Conditions

Exit the loop when user:

• Selects "Ready" or "Exit" option
• Says: "done", "that's enough", "let's implement", "ready to build"
• Explicitly asks to stop brainstorming

Context Accumulation

Track across iterations:

• Original idea
• Key decisions made
• Options explored and rejected (with reasons)
• Constraints identified
• Success criteria defined

Pass this context to each agent invocation to maintain coherence.

Example Flow

1. User: /brainstorm-refine
2. You: "What idea would you like to explore?"
3. User: "A notification system for our app"
4. Launch agent -> Returns options (push notifications, in-app, email, etc.)
5. Present options, ask direction
6. User chooses: "Go deeper on in-app notifications"
7. Launch agent focused on in-app notifications -> Returns refined options
8. Present options, ask direction
9. User chooses: "Ready"
10. Write design doc, offer implementation path

Critical Rules

1. Use the agent - Always launch brainstorm-refiner for exploration, don't do it yourself
2. Accumulate context - Pass full history to each agent invocation
3. Respect exit signals - Watch for completion phrases
4. One direction at a time - Don't overwhelm with multiple explorations
5. Document decisions - Track what was explored and why choices were made

Integration Points

• After completion: Suggest superpowers:writing-plans for implementation
• For code context: Read relevant files before launching agent if idea relates to existing code
• For challenges: Can pivot to brainstorm-challenger agent for more critical perspective