agulli/atlas-agents — reverse-engineered prompt

Build me the hands on code repo for an AI agents book in Python. I want it organized as chapter folders, with each chapter showing a practical agent pattern someone can run and learn from.

Start with a simple ReAct agent from scratch, then add examples for prompt routing, context injection, tools and structured outputs, agent handoffs, state graphs with retries and a human review step, multi agent workflows, model fallback routing, local model support, MCP servers, agent discovery, memory, sandboxes, and always on agents.

Please make the examples runnable from the command line, with clear comments and small demos instead of huge production code. Include a shared config area, an env example for API keys, and a requirements file. Use current Python agent libraries where they make sense, like LangGraph, CrewAI, PydanticAI, MCP, and model provider APIs. Look up current docs online if you need to.

Keep it beginner friendly, but make the code real enough that I can study it and adapt each chapter into my own projects.