agentic-patterns

Multi-agent architecture patterns include sequential, hierarchical, parallel, and event-driven workflows, organized across a five-layer model separating Cognition, Context, Interaction, Runtime, and Governance concerns for production-grade systems.

Design multi-agent workflows by mapping tasks to ReAct or plan-execute styles using a pattern catalog, which provides implementation notes for structuring collaborative, autonomous, and hierarchical agent interactions.

Common multi-agent anti-patterns are architectural flaws that compromise governance and system stability; this pattern catalog identifies them and provides reference designs to ensure safety considerations and proper concern separation.

Use hierarchical agentic patterns for structured task delegation and collaborative patterns when agents require peer-to-peer interaction, mapping tasks to the appropriate architecture based on your workflow's autonomy requirements.

The five-layer model for agentic architectures separates concerns across Cognition, Context, Interaction, Runtime, and Governance layers, providing a structured framework for designing and evaluating complex multi-agent systems.

This focuses on designing and applying agentic architecture patterns, not reviewing existing agentic code or selecting frameworks; use dedicated review-methodology or frameworks skills for those specific tasks.