Everything administrative is out of the way so that from Chapter 1 every session is pure learning.

Read any Messages request or response and explain every field; build a streaming, multi-turn, multimodal chat client.

Run the tool-use loop from memory, cut cost with caching and batches, and make a client that survives 429s and 529s.

Pick the right model and the cheapest configuration for a workload, and write prompts the Anthropic way. This chapter is ~28% of the exam.

Choose between a single call, a workflow and an agent for any scenario; name the five patterns; build a guarded agent with the Agent SDK.

Explain MCP's architecture and primitives, build and register your own server, and know when plain tool use is the better choice.

Turn daily Claude Code use into exam-grade knowledge of memory, settings, hooks, subagents, headless mode and CI. Only ~3% of the exam — keep it to the schedule and do not over-invest.

Threat-model any Claude application and name the defence; build an eval you trust; debug from stop_reason, usage and traces. First timed practice exam.

Three practice exams at 80%+, every domain above 75%, cheat sheet internalised, logistics done.