Skip to content

Module 03 — Prompt Engineering

Level: Beginner → Intermediate · Time: ~6 hours (pages) + ~2 hours (labs) · Exam relevance: CCAO-F D1, D7 · CCAR-P D2, D4

Prompt engineering is the craft of writing the instructions, context and examples that steer Claude toward a useful, reliable answer. This module turns the basics from Module 02 — Prompt fundamentals into a repeatable method you can use in the Claude apps and in production code.

All guidance here follows Anthropic's living reference, Prompting best practices (checked 2026-10-24). Where an API detail matters (thinking, effort, structured outputs), the examples follow the current rules in docs/STYLE.md section 1.5.

Learning objectives

By the end of this module you can:

  1. Write clear, direct instructions that a capable newcomer could follow without guessing.
  2. Separate instructions, context, examples and inputs with XML tags.
  3. Choose between zero-shot and few-shot prompting, and build a small, diverse example set.
  4. Decide when Claude should reason before answering, and tune that with adaptive thinking and effort.
  5. Break a complex request into sub-tasks and chain them, with checks between steps.
  6. Get machine-readable output with JSON Schema structured outputs, messages.parse() and strict tools, and explain why assistant prefill is no longer used.
  7. Design system prompts, roles, reusable templates and guardrails.
  8. Improve a prompt systematically with test cases and small, measured changes.
  9. Adapt your strategy to the task type: analysis, research, drafting or brainstorming.

Page order

# Page What you learn
1 Clear and direct instructions The "new colleague" test, context and motivation, task-type strategies
2 Structure with XML tags Separating instructions from data, nesting documents, tagged output
3 Examples and few-shot prompting Zero-shot vs few-shot, choosing and formatting examples
4 Reasoning and thinking Prompted chain of thought, adaptive thinking, the effort parameter
5 Task decomposition and chaining Splitting work, sequential and parallel chains, self-correction
6 Structured output JSON Schema outputs, Pydantic parsing, strict tools, life after prefill
7 System prompts and roles Roles, templates, variables, guardrails, prompt injection basics
8 Iterating on prompts Test sets, failure analysis, versioning, knowing when to stop

Prerequisites

Exam relevance

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

Certification Objectives covered Where
CCAO-F CCAO-F-1.1 Write effective prompts Pages 1, 2, 3, 7
CCAO-F CCAO-F-1.2 Decompose complex requests Page 5
CCAO-F CCAO-F-1.3 Iterate on prompts Page 8
CCAO-F CCAO-F-1.4 Adapt strategy to task type Pages 1, 4, 5
CCAO-F CCAO-F-2.6 Choose output formats Page 6
CCAO-F CCAO-F-5.3 Write system-level instructions Page 7
CCAO-F CCAO-F-7.1, 7.2 Diagnose and fix poor outputs Page 8
CCAR-P CCAR-P-2.2 System prompts, templates, guardrails Page 7
CCAR-P CCAR-P-2.3 Zero-shot, few-shot, chain of thought Pages 3, 4
CCAR-P CCAR-P-1.5 Decomposition techniques Page 5
CCAR-P CCAR-P-4.4 Diagnose prompt failure Page 8
CCAR-P CCAR-P-5.1 Guardrails and safety controls Page 7

Foundations candidates should be able to apply every technique in the Claude apps. Professional candidates should also be able to justify the technique in an architecture review and know the API mechanics (structured outputs, effort, strict tools).

Hands-on labs

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

Quick reference

Next module

Continue to Module 04 — Context Engineering, which widens the lens from the words in one prompt to everything that fills Claude's context window: documents, memory, caching and Skills.