brainstorm-synthesis

<objective>
Launch N agents with the same problem from different perspectives. Each returns an
approach with trade-offs. Synthesize the best unified solution from their collective
insights.

This is the F-thread (fusion) pattern: multiple independent analyses converge into one
superior answer. Use it when the decision is hard enough that diverse viewpoints add
value. </objective>

<when-to-use>
Architectural decisions with multiple valid approaches. Complex problems where no single
perspective captures the full picture. High-stakes choices where missing a consideration
is costly. Design decisions in /autotask deep mode.

Skip for straightforward implementations, well-established patterns, or decisions where
one approach is obviously correct. </when-to-use>

<agent-perspectives>
Select 3-5 perspectives that illuminate different facets of the problem:

**Standard perspectives:**

- Pragmatist: Simplest solution that works, minimize complexity
- Architect: Long-term maintainability, extensibility, patterns
- Performance: Efficiency, scalability, resource usage
- Security: Attack surface, data protection, access control
- User-focused: UX impact, developer experience, API ergonomics

**Domain-specific perspectives** (add based on problem):

- Data engineer: For data pipeline decisions
- DevOps: For deployment and infrastructure choices
- Testing: For testability and verification concerns
- Integration: For API and system boundary decisions

Choose perspectives that will genuinely disagree. Similar viewpoints waste the pattern's
value. The goal is productive tension that surfaces trade-offs. </agent-perspectives>

<execution>
Frame a clear problem statement that includes the decision to be made, relevant
constraints, and success criteria.

Launch agents in parallel using the Task tool. Each agent receives:

- The problem statement
- Their assigned perspective
- Instructions to return: recommended approach, key trade-offs, complexity estimate,
  risks

Wait for all agents