Guides

A Claude academy in practitioner guides

This is a Claude academy in the only sense that matters: ten guides, each answering one question a practitioner actually hits, with a position rather than a survey of options.

• Context & Memory

  CLAUDE.md, AGENTS.md or Cursor rules — which does what?

  How CLAUDE.md, AGENTS.md, Cursor rules and Copilot instructions differ in loading and precedence, and keep one source of truth instead of four.

  Published 10 August 2026

• Agents & Orchestration

  Agentic AI: how do you stop an agent loop that won’t finish?

  Every agent framework makes starting easy and stopping your problem. The four non-termination patterns in agentic AI, and six stop conditions.

  Published 1 August 2026

• Reliability, Security & Trust

  AI hallucination: how do you stop an LLM from hallucinating?

  AI hallucination is pattern completion without grounding. Classify the four LLM hallucination types, then apply controls enforced in code.

  Published 1 August 2026

• Context & Memory

  Claude memory: compaction or truncation when context runs out?

  How Claude memory behaves when the context window fills: truncation drops the oldest turns, compaction rewrites them. Only one keeps decisions.

  Published 1 August 2026

• Context & Memory

  Claude Projects, Claude memory — or a written handoff?

  Neither Claude Projects nor Claude memory carries a half-finished task across sessions. A handoff template that lets a fresh session resume.

  Published 1 August 2026

• Agents & Orchestration

  Claude subagents or multi agent systems — when worth it?

  Claude subagents buy parallelism and cost coherence. When multi agent systems win, when they fail at the joins, and the decision test.

  Published 1 August 2026

• Prompting & Output Craft

  Structured output or function calling — how do you stop drift?

  Structured output drifts because you asked in prose. The fix is a schema or function calling plus a parser that rejects rather than repairs.

  Published 1 August 2026

• Context & Memory

  What is CLAUDE.md and what should you put in it?

  CLAUDE.md for Claude Code: where it loads from, what belongs in it, a working example, and how long a CLAUDE.md should actually be.

  Published 1 August 2026

• Tools & MCP

  What is MCP, and what is an MCP server?

  What MCP is, what an MCP server does, and the build order for MCP servers: primitives, transports, auth, tool design and deployment.

  Published 1 August 2026

• Tools & MCP

  Which MCP transport — stdio, SSE or Streamable HTTP?

  Which MCP transport to use: Streamable HTTP is the default, SSE is legacy, stdio is local. Chosen by deployment reality, not preference.

  Published 1 August 2026

How the guides relate to the tracks

A track is a Socratic dialogue: the coach probes your reasoning and surfaces the concepts you fumbled. A guide is the same concept written down, for when you want the answer rather than the questions. Read the guide, then practise the decision with the coach.

Anthropic Academy is Anthropic’s own course platform. These guides are independent and not affiliated with it.