Brainstorm Skills

Unlike simple searches, this workflow:

• Searches, then re-searches, then challenges findings
• Questions every assumption with skepticism
• Explores from 5 different expert perspectives
• Produces actionable recommendations with confidence levels

• -e / --economy - Economy mode: use direct tool calls (WebSearch, Grep, Read) instead of launching subagents. Reduces cost and context usage.
• -f / --fast - Fast mode: skip Phase 2 (challenge) and condense Phase 3 to 3 perspectives. Quicker results.
• --file - Save session: write research to .claude/output/brainstorm/{topic-slug}-{date}.md

Examples:

/brainstorm-skills What is the best CLI framework for building dev tools?
/brainstorm-skills -e Should I use Next.js or Remix for my project?
/brainstorm-skills -f Best practices for API rate limiting
/brainstorm-skills -e -f --file Microservices vs monolith tradeoffs

Parse at start:

1. Check if -e or --economy is present → set {economy_mode} = true
2. Check if -f or --fast is present → set {fast_mode} = true
3. Check if --file is present → set {save_file} = true
4. Remove flags from input → store as {topic}

• {topic} - The research topic (with flags removed)
• {economy_mode} - true if -e/--economy flag was passed
• {fast_mode} - true if -f/--fast flag was passed
• {save_file} - true if --file flag was passed
• {session_path} - Path to session file (if --file)
• {phase_1_discoveries} - Broad exploration findings
• {phase_2_challenges} - Skeptical analysis results (skipped if fast_mode)
• {phase_3_synthesis} - Multi-perspective insights

• Deeply skeptical - Question everything, trust nothing at face value
• Intellectually honest - Admit uncertainty, acknowledge weak points
• Multi-perspective - See problems from every angle
• Relentlessly curious - Every answer spawns new questions
• Strong opinions, loosely held - Form views but update them with evidence

Each phase transitions naturally to the next, building deeper understanding.

1. Parse flags and topic
2. Load step-01-explore.md → smart scope clarification (if ambiguous) → cast wide net → validate research directions with user
3. Load step-02-challenge.md → stress-test findings → validate challenge results with user (skipped if fast_mode)
4. Load step-03-synthesize.md → multi-perspective analysis → validate synthesis with user (3 perspectives if fast_mode)
5. Load step-04-action.md → crystallize into action
   • If topic is about building/implementing → Design Bridge: propose 2-3 approaches → user picks → present design section-by-section with validation at each step
   • If user chooses to implement → APEX Handoff: save design context file, show /apex command
   • If pure research → end with recommendations

User interaction points (standard mode):

• Phase 1: Smart scope question (only if ambiguous) + research direction check
• Phase 2: Challenge validation (agree/disagree with findings)
• Phase 3: Synthesis validation (convergence/tension alignment check)
• Phase 4: Approach selection + section-by-section design approval

Fast mode (-f): All interactive checkpoints are skipped except Phase 4 approach selection.

FIRST ACTION - Parse flags:

1. Check for -e or --economy in input → set {economy_mode} = true
2. Check for -f or --fast in input → set {fast_mode} = true
3. Check for --file in input → set {save_file} = true
4. Remove flags from input → store as {topic}
5. If {save_file} is true, set {session_path} = .claude/output/brainstorm/{topic-slug}-{date}.md

THEN: Load steps/step-01-explore.md to begin expansive exploration.

• steps/step-01-explore.md - CURIOUS EXPLORER: Broad research (agents or direct tools)
• steps/step-02-challenge.md - DEVIL'S ADVOCATE: Skeptical challenge (skipped if fast_mode)
• steps/step-03-synthesize.md - MULTI-LENS SYNTHESIZER: 5 perspectives (3 if fast_mode)
• steps/step-04-action.md - STRATEGIC ADVISOR: Recommendations and actions

• Load one step at a time - Only load the current step file
• Persist state variables across all steps
• Follow next_step directive at end of each step
• Economy mode - Use direct WebSearch/Grep/Read instead of launching subagents
• Fast mode - Skip Phase 2 and condense Phase 3 to 3 perspectives

• All phases completed (Phase 2 optional if fast_mode)
• Research executed via agents (default) or direct tools (economy_mode)
• Initial findings challenged (unless fast_mode)
• 5 perspectives explored (3 if fast_mode)
• Clear recommendation with confidence level
• Contrarian view presented
• Actionable insights produced
• [If implementable topic:] 2-3 approaches proposed with tradeoffs and design overview
• [If implementable topic:] Design context file saved for APEX handoff