Full curriculum

Every area and chapter in CCAF Prep, in the recommended order. On the guided path each area ends in a final quiz that requires a perfect 100% score — with unlimited fresh-draw retries — before the next area opens up. Prefer to explore? Free navigation lets you open any area right away.

Start free

1. Orientation

How the exam works: format, scoring, scenarios, and how to use this course.

Final quiz: 8 questions — 100% to advance on the guided path.

  • The Exam: Format, Scoring, Scenariosorientation.exam-format
  • How to Study with This Courseorientation.study-strategy

2. Claude API Foundations

The request/response contract, tool use, stop_reason handling, the agentic loop, and the Message Batches API.

Final quiz: 12 questions — 100% to advance on the guided path.

  • Messages, Roles, and the Request Loopapi.messages
  • Tool Use: Definitions, tool_use Blocks, Resultsapi.tool-use
  • stop_reason: tool_use vs end_turnapi.stop-reasons
  • The Agentic Loop, End to Endapi.agentic-loop
  • Message Batches API Fundamentalsapi.batch

3. Prompt Engineering & Structured Output

Explicit criteria and false-positive control, few-shot prompting, tool_use + JSON schemas for guaranteed structure, validation/retry loops, batch processing strategy, and multi-instance review architectures.

Final quiz: 12 questions — 100% to advance on the guided path.

  • Explicit Criteria over Vague InstructionsD4.1
  • Few-Shot Prompting for ConsistencyD4.2
  • Enforcing Structure: tool_use + JSON SchemasD4.3
  • Validation, Retry, and Feedback LoopsD4.4
  • Batch Processing StrategiesD4.5
  • Multi-Instance and Multi-Pass ReviewD4.6

4. Tool Design & MCP Integration

Writing tool descriptions that get selected correctly, structured MCP error responses, scoping tools across agents with tool_choice, MCP server configuration and resources, and choosing the right built-in tool (Read, Write, Edit, Bash, Grep, Glob).

Final quiz: 12 questions — 100% to advance on the guided path.

  • Designing Tool Interfaces That Get Selected CorrectlyD2.1
  • Structured Error Responses for MCP ToolsD2.2
  • Distributing Tools Across Agents & tool_choiceD2.3
  • MCP Servers, Scoping, and ResourcesD2.4
  • Built-in Tools: Read, Write, Edit, Bash, Grep, GlobD2.5

5. Agentic Architecture & Orchestration

Building reliable agentic loops in production, orchestrating coordinator-subagent systems, configuring subagent invocation and context passing, enforcing multi-step workflows with programmatic gates and handoffs, using Agent SDK hooks for interception and normalization, choosing task decomposition strategies, and managing session resumption and forking.

Final quiz: 12 questions — 100% to advance on the guided path.

  • Agentic Loops for Autonomous ExecutionD1.1
  • Coordinator–Subagent OrchestrationD1.2
  • Subagent Invocation, Context Passing, SpawningD1.3
  • Multi-Step Workflows: Enforcement & HandoffsD1.4
  • Agent SDK Hooks: Interception & NormalizationD1.5
  • Task Decomposition StrategiesD1.6
  • Sessions: Resumption and ForkingD1.7

6. Claude Code Configuration & Workflows

Configuring CLAUDE.md hierarchy, @import, and modular .claude/rules/ for team-wide conventions; building custom slash commands and Agent Skills with frontmatter controls; applying path-specific glob rules for conventions spread across a codebase; choosing plan mode versus direct execution; using iterative-refinement techniques (input/output examples, test-driven iteration, the interview pattern); and integrating Claude Code into CI/CD pipelines with non-interactive flags and structured output.

Final quiz: 12 questions — 100% to advance on the guided path.

  • CLAUDE.md Hierarchy, @import, and Modular RulesD3.1
  • Custom Slash Commands and SkillsD3.2
  • Path-Specific Rules with Glob FrontmatterD3.3
  • Plan Mode vs Direct ExecutionD3.4
  • Iterative Refinement TechniquesD3.5
  • Claude Code in CI/CD PipelinesD3.6

7. Context Management & Reliability

Preserving critical facts and figures across long interactions instead of losing them to progressive summarization or the lost-in-the-middle effect; designing escalation and ambiguity-resolution patterns that trigger on explicit requests and policy gaps rather than unreliable sentiment or self-reported confidence; propagating structured error context across multi-agent systems instead of generic failure statuses; managing context in large codebase exploration with scratchpads, subagent delegation, and crash-recovery manifests; calibrating field-level confidence against labeled validation sets to route human review; and preserving provenance, conflicting-source attribution, and temporal metadata through multi-source synthesis.

Final quiz: 12 questions — 100% to advance on the guided path.

  • Preserving Critical Information Across Long InteractionsD5.1
  • Escalation and Ambiguity ResolutionD5.2
  • Error Propagation in Multi-Agent SystemsD5.3
  • Context Management in Large Codebase ExplorationD5.4
  • Human Review Workflows and Confidence CalibrationD5.5
  • Provenance and Uncertainty in Multi-Source SynthesisD5.6

8. Scenario Practice

Applying every domain to the six production contexts the real exam draws its scenarios from: a customer support resolution agent, Claude Code for team code generation, a multi-agent research system, developer-productivity tooling over unfamiliar codebases, Claude Code wired into CI/CD, and a structured data extraction pipeline. Each chapter is a case-study walkthrough of one scenario followed by cross-domain judgment questions in that scenario's voice — the closest rehearsal available for how the real exam actually presents questions.

Final quiz: 12 questions — 100% to advance on the guided path.

  • Scenario: Customer Support Resolution Agentscenario.support-agent
  • Scenario: Code Generation with Claude Codescenario.code-gen
  • Scenario: Multi-Agent Research Systemscenario.research
  • Scenario: Developer Productivityscenario.dev-productivity
  • Scenario: Claude Code for CIscenario.ci
  • Scenario: Structured Data Extractionscenario.extraction

We use a session cookie to keep you signed in, and would like to record your cookie preference for possible future features. See our Privacy Policy.