CCAR-F Study Resources
Study resources
The 62 documentation pages the CCAR-F lessons are built on, grouped by exam domain, with the lessons that cite each one. Then runnable notebooks from Anthropic's Claude Cookbook for hands-on practice.
Domain 1Agentic Architecture & Orchestration
- Tool use: how it worksLesson 1.1
- Agent SDK: the loop at a glanceLesson 1.1
- Agent SDK: turns and budgetLesson 1.1
- How we built our multi-agent research system: architectureLesson 1.2
- Multi-agent research system: prompting the coordinatorLessons 1.2, 1.3
- Building effective agents: routingLessons 1.2, 1.6
- Agent SDK: create subagentsLesson 1.3
- Agent SDK: AgentDefinition configurationLesson 1.3
- Agent SDK: what subagents inheritLesson 1.3
- Claude Code hooks guide: hooks versus instructionsLessons 1.4, 1.5
- Customer support agent guide: success criteria and escalationLesson 1.4
- Building effective agents: workflows and agentsLessons 1.4, 1.6
- Claude Code hooks guide: what you can automateLesson 1.5
- Hooks reference: PostToolUseLesson 1.5
- Agent SDK sessions: choose an approachLesson 1.7
- Agent SDK sessions: resume by IDLesson 1.7
- Agent SDK sessions: fork to explore alternativesLesson 1.7
Domain 2Tool Design & MCP Integration
- Define tools: best practices for tool definitionsLesson 2.1
- Writing tools for agents: prompt-engineering tool descriptionsLessons 2.1, 2.4
- Writing tools for agents: choosing the right toolsLessons 2.1, 2.3
- MCP specification: tool error handlingLesson 2.2
- Handle tool calls: errors with is_errorLesson 2.2
- Writing tools for agents: returning meaningful contextLesson 2.2
- Define tools: forcing tool useLesson 2.3
- Agent SDK: restricting a subagent's toolsLesson 2.3
- Claude Code MCP: installation scopesLesson 2.4
- MCP specification: resourcesLesson 2.4
- Claude Code tools referenceLesson 2.5
- Tools reference: Edit tool behaviourLesson 2.5
- Common workflows: understand new codebasesLesson 2.5
Domain 3Claude Code Configuration & Workflows
- Claude Code memory: how CLAUDE.md files loadLesson 3.1
- Claude Code memory: importing filesLesson 3.1
- Claude Code memory: organising rulesLessons 3.1, 3.3
- Skills: where skills liveLesson 3.2
- Skills: run skills in a subagentLesson 3.2
- Skills: allowed-toolsLesson 3.2
- Skills: passing argumentsLesson 3.2
- Claude Code memory: path-specific rulesLesson 3.3
- Permission modes: plan modeLesson 3.4
- Subagents: the built-in Explore subagentLesson 3.4
- Common workflows: understand new codebasesLesson 3.4
- Best practices: give Claude a way to verify its workLesson 3.5
- Best practices: let Claude interview youLesson 3.5
- Best practices: provide specific contextLesson 3.5
- Best practices: course-correct earlyLesson 3.5
- CLI reference: flagsLesson 3.6
- GitHub Actions: project standards in CLAUDE.mdLesson 3.6
- Best practices: an independent review stepLesson 3.6
Domain 4Prompt Engineering & Structured Output
- Prompting best practices: be clear and directLessons 4.1, 4.2
- GitHub Actions: project standards and review criteriaLesson 4.1
- Prompting best practices: use examples effectivelyLesson 4.2
- Structured outputs: strict tool useLesson 4.3
- Define tools: forcing tool useLesson 4.3
- Define tools: schema designLesson 4.3
- Structured outputs: what schemas do and do not guaranteeLesson 4.4
- Handle tool calls: returning errors to ClaudeLesson 4.4
- Writing tools for agents: meaningful contextLesson 4.4
- Batch processing: how the Message Batches API worksLesson 4.5
- Batch processing: retrieving resultsLesson 4.5
- Best practices: an independent review stepLesson 4.6
- Best practices: run multiple Claude sessionsLesson 4.6
Domain 5Context Management & Reliability
- Effective context engineering: long-horizon tasksLesson 5.1
- Prompting best practices: long contextLesson 5.1
- Customer support agent guide: success criteria and escalationLesson 5.2
- Agent SDK: handle clarifying questionsLesson 5.2
- MCP specification: error handlingLesson 5.3
- Multi-agent research system: coverage and recoveryLesson 5.3
- Handle tool calls: errorsLesson 5.3
- Claude Code context window: when context fills upLesson 5.4
- Claude Code context window: what survives compactionLesson 5.4
- Harnesses for long-running agents: incremental progressLesson 5.4
- Develop tests: build evaluationsLesson 5.5
- Develop tests: evaluation design principlesLesson 5.5
- Multi-agent research system: citations and synthesisLesson 5.6
- Multi-agent research system: evaluating outputsLesson 5.6
Go hands-onClaude Cookbook notebooks
Optional and needs a Claude API key. Each notebook builds a pattern from the lessons listed beside it.
- Customer service agent with client-side toolsLesson 1.1
- Orchestrator-workers patternLesson 1.2
- Basic workflows: chaining, routing, parallelizationLesson 1.6
- Tool choiceLesson 2.1
- Extracting structured JSON with tool useLesson 4.3
- Evaluator-optimizer loopLesson 4.4
- Batch processing with the Message Batches APILesson 4.5
- Automatic context compactionLesson 5.1
- Memory and context managementLesson 5.4
- Building evalsLesson 5.5
Learning the topics? Start with the CCAR-F learning guide. Ready to test yourself? Take the practice tests.