What problem does it solve? Choosing the right agent architecture pattern (RAG, ReAct, supervisor, planner-executor, etc.) is error-prone: teams often over-engineer with complex multi-agent setups or skip simpler deterministic solutions. This Skill provides a structured, question-driven decision process that picks the simplest justified pattern and documents the reasoning. ## Core Features & Use Cases - Branch-aware pattern selection: Takes the single-agent vs multi-agent decision from stage 1 as fixed input and asks targeted questions to select the specific pattern within that branch. - Evidence-grounded recommendations: Covers deterministic code, fixed RAG, Agentic RAG, bounded ReAct, planner-executor, parallel fan-out, critic-actor, supervisor, hierarchical, peer-to-peer, and sequential pipeline patterns, always naming rejected alternatives. - Approval-gated output: Writes system_design/02_design_pattern.md with a decision walkthrough, ASCII topology diagram, and model/provider notes, then stops for explicit user approval. - Use Case: As stage 2 of an agent-system-design pipeline, after topology is approved, run this to decide whether a customer-support bot needs a fixed RAG workflow or a dynamic Agentic RAG loop, with the reasoning recorded for later build stages. ## Quick Start Run stage 2 of the agent-system-design pipeline to choose the design pattern for my usecase based on the approved topology in system_design/01_agent_topology.md.
