quorum

Use this agent when the user presents an architecture proposal, system design, implementation plan, technical idea, or code approach and wants it stress-tested before committing to implementation. This includes RFC reviews, ADR drafts, feature planning, migration strategies, API designs, data model proposals, or any technical decision that would be costly to reverse. The agent should be invoked proactively whenever the user shares a plan or idea and asks for feedback, critique, or validation. Examples: - User: "I'm thinking of migrating our monolithic API to microservices. Here's my plan..." Assistant: "Let me launch the devils-advocate agent to stress-test this migration plan before you commit to it." [Uses Task tool to launch devils-advocate agent with the migration plan] - User: "Here's my architecture for the new event system - we'll use a pub/sub pattern with Redis streams..." Assistant: "I'll use the devils-advocate agent to challenge this architecture and find potential weak spots." [Uses Task tool to launch devils-advocate agent with the architecture proposal] - User: "I want to refactor the worker system to decouple startup from app initialization. My approach is..." Assistant: "Before you start implementing, let me run this through the devils-advocate agent to battle-test the approach." [Uses Task tool to launch devils-advocate agent with the refactoring plan] - User: "What if we used Zod schemas as the single source of truth for all our types and validation?" Assistant: "That's an interesting idea - let me use the devils-advocate agent to probe it for weaknesses." [Uses Task tool to launch devils-advocate agent with the typing strategy proposal] - User: "I'm planning to add a caching layer between the API and database using Redis." Assistant: "Let me launch the devils-advocate agent to challenge this caching strategy and surface risks you might not have considered." [Uses Task tool to launch devils-advocate agent with the caching proposal]

Debate a design question with N agents arguing different positions, then synthesize a recommendation. Use when there are multiple valid approaches, the user says 'debate this', 'quorum', or wants perspectives stress-tested before deciding.