Skip to content

Module 04 — Context Engineering

Level: Intermediate → Advanced · Time: ~6 hours (pages) + ~2 hours (labs) · Exam relevance: CCAO-F D3, D5 · CCAR-P D2, D3, D4

Prompt engineering asks "what should I write?". Context engineering asks a larger question: "of everything that could be in Claude's context window (instructions, documents, tool definitions, tool results, conversation history, memories), what is the smallest, most useful set for this step, and how do I keep it that way as the work goes on?"

This module covers the techniques Anthropic documents for that job, as of 2026-10-24: careful placement of long documents, prompt caching, memory, compaction, context editing and progressive disclosure with Agent Skills. It is the conceptual foundation for Professional objectives CCAR-P-2.4 (optimise context windows and token usage), CCAR-P-2.5 (reuse prompts via caching, modular prompts and Skills) and CCAR-P-3.8 (progressive discovery versus monolithic context).

Learning objectives

By the end of this module you can:

  1. Explain context engineering, the idea of a finite "attention budget", and how it differs from prompt engineering.
  2. Structure prompts with long documents for accuracy: placement, document tags and quote-first grounding.
  3. Design prompts for caching: stable prefixes, breakpoints, TTL choice, verification and cost estimation.
  4. Choose between memory options: Projects memory in the Claude apps and the memory tool in the API.
  5. Keep long conversations and agent loops healthy with compaction and context editing.
  6. Compare progressive disclosure (Skills, just-in-time retrieval) with monolithic context and justify a choice.

Page order

# Page What you learn
1 What is context engineering? The context window as a budget; the components of context; the core strategies
2 Managing long context Document placement, multi-document structure, quotes-first grounding, when to restart
3 Prompt caching Prefix matching, breakpoints, TTLs, usage fields, silent invalidators, cost maths
4 Memory and persistence Apps memory and Projects; the API memory tool; designing what to remember
5 Compaction and context editing Server-side summarisation, clearing tool results and thinking, when to use which
6 Progressive disclosure and Skills Agent Skills, just-in-time loading, tool search, monolithic versus progressive context

Prerequisites

Exam relevance

Objective IDs come from 16-certification-tracks/registry.yaml.

Certification Objectives covered Where
CCAR-P CCAR-P-2.4 Optimise context windows and token usage Pages 1, 2, 5
CCAR-P CCAR-P-2.5 Reuse prompts via caching, modular prompts and Skills Pages 3, 6
CCAR-P CCAR-P-3.8 Progressive discovery versus monolithic context Pages 1, 6
CCAR-P CCAR-P-4.5 Optimise token usage, latency and cost-performance Pages 3, 5
CCAR-P CCAR-P-1.4 Multi-agent systems (context isolation) Page 1
CCAO-F CCAO-F-3.4 Manage context limits and memory Pages 2, 4, 5
CCAO-F CCAO-F-5.1, 5.4 Configure and maintain Projects Page 4

Foundations candidates should focus on pages 1, 2 and 4 (the apps view: when to restart, summarise or persist). Professional candidates need all six, including the API mechanics and cost reasoning.

Hands-on labs

Cost: requires API credits (live mode). All labs run offline in mock mode by default.

Next module

Continue to Module 05 — Claude Products and Models, which covers the product surfaces (apps, Projects, artifacts, connectors) and how to choose a model.