agent-decision-single-vs-multi

What problem does it solve? Teams often jump straight to building multi-agent systems without verifying whether the problem actually requires more than one agent, leading to unnecessary coordination cost, harder debugging, and wasted LLM calls. This Skill forces a disciplined, evidence-based topology decision before any design pattern is chosen. ## Core Features & Use Cases - Structured topology rubric: Asks five batched, multiple-choice questions covering step predictability, impact severity, decomposition needs, parallelism, and isolation requirements. - Deterministic decision logic: Maps answers to a single-agent or multi-agent verdict, names the rejected alternative, and carries forward signals (human-governance needs, sub-branch hints) to later pipeline stages. - Approval-gated artifact: Writes system_design/01_agent_topology.md and stops for explicit user confirmation before stage 2 runs. - Use Case: A user says "build me a multi-agent system for customer support triage." The Skill restates the business problem, walks through the rubric, and determines whether one bounded agent suffices — documenting why multi-agent was rejected if so. ## Quick Start Ask the agent to run stage 1 of the agent-system-design pipeline to decide whether your use case needs a single agent or multiple agents.