Guides How an agent works, from request to answer: Follow one request from the first message to the final answer: what goes into the context, how tools are called and why the loop stops. Context engineering: deciding what the model sees: Why more context is not better, what fills the window, and how caching, compaction, retrieval and sub-agents keep it useful. Building an agent, step by step: A ladder from one prompt to a team of agents, with twelve steps, patterns, example routes, open protocols and a launch checklist. Securing an agent: permissions, sandboxes and prompt injection: Permissions, sandboxes, hooks and prompt injection: how to let an agent act without handing over the keys. Testing and running agents in production: Evals, traces, cost and governance: how to know that an agent works, and how to keep it working once it is live. Skill, tool, MCP server, plugin, hook: what is the difference?: Skill, tool, MCP server, plugin or hook? One table shows who starts each, when it enters the context and what it costs. Where to go deeper: Talks, courses, books and papers from the people who build and study these systems, easiest first.