What problem does it solve? Teams building AI agents often over-engineer with unnecessary multi-agent topologies or under-specify safety, durability, and evaluation controls. This Skill produces a decision-complete logical architecture that classifies each capability correctly and applies production-grade contracts, policies, and governance before any code is written. ## Core Features & Use Cases - Behavior Classification: Classifies each capability as deterministic code, direct model call, deterministic AI workflow, or agent, then selects the smallest sufficient agent pattern. - Governed System Design: Defines task, tool, context, memory, approval, durability, failure, and evaluation contracts with deterministic control outside model discretion. - AI Stack Handoff: Delegates model, runtime, retrieval, and observability technology selection to ai-stack only when a model-backed capability remains. - Use Case: When asked to design a customer-support agent that drafts answers and sends them after human approval, the Skill classifies the flow as a deterministic AI workflow, defines exact approval binding and idempotent send effects, and hands frozen requirements to ai-stack for component selection. ## Quick Start Ask the agent to design a production AI agent subsystem for your workload, for example: use ai-agent-design to design a bounded diagnostic agent with read-only tools and explicit budgets.
