Agent-Native Architecture
A 15-area architecture checklist for systems where AI agents are primary actors, spanning tools, context, approvals, permissions, and auditability.
What is Agent-Native Architecture?
Agent-Native Architecture is an opinionated skill from iliaal's AI Skills collection. It guides system design across 15 areas that matter when agents perform application work, including tool design, execution patterns, secure context injection, approval gates, permission boundaries, and audit trails. The skill can be installed on its own or used through the broader Whetstone workflow plugin.
Best for
Architects and engineers designing agent systems, MCP tools, approval workflows, or products that give AI controlled access to real actions.
Not for
Simple prompt-only features that do not grant an agent tools, durable context, permissions, or meaningful execution authority.