Feature Brainstorm
A structured three-phase approach to brainstorming complex feature implementations using parallel Sonnet sub-tasks for exploration, validation, and feasibility assessment.
Overview
Phase 1: EXPLORE → Launch parallel sub-tasks to analyze codebase
Phase 2: VALIDATE → Launch parallel sub-tasks to research and validate ideas
Phase 3: ASSESS → Launch parallel sub-tasks to evaluate feasibility and create proposals
Before Starting
- Confirm the feature scope with the user
- Identify the codebase location (ask if not provided)
- Create a working directory:
mkdir -p /home/claude/brainstorm-sessions
Phase 1: Exploration
Launch 3-5 parallel sub-tasks using Sonnet to explore the codebase from different angles.
Sub-task prompt template:
You are exploring a codebase to understand how to implement: [FEATURE]
Your specific focus: [EXPLORATION_ANGLE]
Codebase location: [PATH]
Instructions:
1. Explore relevant files using view and bash tools
2. Document key findings in /home/claude/brainstorm-sessions/explore-[ANGLE].md
3. Note: existing patterns, relevant abstractions, potential integration points, dependencies
Keep findings concise and actionable.
Exploration angles to assign:
- Architecture: Overall structure, module boundaries, data flow patterns
- Similar features: Existing features with comparable patterns
- Data layer: Models, schemas, database interactions, state management
- API surface: Endpoints, interfaces, contracts that would be affected
- Testing patterns: How similar features are tested, test infrastructure
Example sub-task launch:
claude --model claude-sonnet-4-20250514 --print \
"You are exploring a codebase to understand how to implement: user authentication with OAuth.
Your specific focus: Architecture - overall structure and module boundaries.
Codebase location: /path/to/repo
Explore relevant files and document findings in /home/claude/brainstorm-sessions/explore-architecture.md"
After all exploration sub-tasks complete, synthesize findings:
cat /home/claude/brainstorm-sessions/explore-*.md > /home/claude/brainstorm-sessions/exploration-summary.md
Phase 2: Validation
Based on exploration findings, launch 3-4 parallel sub-tasks to validate ideas and perform research.
Sub-task prompt template:
You are validating an implementation approach for: [FEATURE]
Exploration context:
[PASTE KEY FINDINGS FROM PHASE 1]
Your validation focus: [VALIDATION_ANGLE]
Instructions:
1. Research best practices using web search if needed
2. Validate against codebase patterns in [PATH]
3. Document findings in /home/claude/brainstorm-sessions/validate-[ANGLE].md
4. Flag any concerns, risks, or open questions
Be critical - identify potential problems early.
Validation angles to assign:
- Pattern alignment: Does the approach fit existing codebase patterns?
- Technical research: Best practices, library options, security considerations
- Edge cases: Error handling, failure modes, boundary conditions
- Performance: Scalability concerns, potential bottlenecks
- Dependencies: Impact on existing code, breaking changes
After validation sub-tasks complete:
cat /home/claude/brainstorm-sessions/validate-*.md > /home/claude/brainstorm-sessions/validation-summary.md
Phase 3: Assessment
Launch 2-3 parallel sub-tasks to assess feasibility and create concrete proposals.
Sub-task prompt template:
You are creating an implementation proposal for: [FEATURE]
Context:
- Exploration findings: [SUMMARY]
- Validation results: [SUMMARY]
Your assessment focus: [ASSESSMENT_ANGLE]
Instructions:
1. Create a concrete implementation proposal
2. Estimate effort and complexity
3. Identify prerequisites and blockers
4. Document in /home/claude/brainstorm-sessions/assess-[ANGLE].md
Output format:
## Approach Summary
## Implementation Steps
## Effort Estimate (T-shirt size + reasoning)
## Risks and Mitigations
## Prerequisites
Assessment angles to assign:
- MVP approach: Minimum viable implementation, fastest path
- Robust approach: Production-ready, handles edge cases
- Alternative approach: Different architecture or pattern (if applicable)
After assessment sub-tasks complete:
cat /home/claude/brainstorm-sessions/assess-*.md > /home/claude/brainstorm-sessions/assessment-summary.md
Synthesis
After all phases, create a final brainstorm report:
- Read all summary files
- Create
/home/claude/brainstorm-sessions/[FEATURE]-brainstorm-report.mdwith:- Executive summary
- Recommended approach with justification
- Alternative approaches considered
- Key risks and mitigations
- Suggested next steps
- Open questions for user
Present the report to the user and discuss.
Sub-task Management
Launching sub-tasks:
# Run in background, capture output
claude --model claude-sonnet-4-20250514 --print "[PROMPT]" > output.md 2>&1 &
Waiting for completion:
wait # Wait for all background jobs
Checking results:
ls -la /home/claude/brainstorm-sessions/
Customization
Adjust the number and focus of sub-tasks based on:
- Small feature: 2-3 sub-tasks per phase
- Medium feature: 3-4 sub-tasks per phase
- Large/complex feature: 4-5 sub-tasks per phase
Skip phases if appropriate:
- Skip Phase 2 if the implementation is straightforward
- Combine Phase 2 and 3 for well-understood patterns