This repository provides a production-oriented manual for building, evaluating, and shipping LLM agent systems. It offers over 60 runnable skeletons, adapters for 24+ LLM providers, MCP integration guides, skill design patterns, and coding-agent workflows for tools like Claude Code and Cursor. The handbook covers the full agent stack including memory architecture, observability, prompt injection defense, and a curated catalog of existing agent frameworks and evaluation tools.