Claude Certified Architect study checklist (CCAR-F)
One self-check question for each of the 30 task statements in the Claude Certified Architect Foundations (CCAR-F) exam guide, grouped by domain. If you cannot answer a question with confidence, open the linked study page.
When you can answer most of them, test yourself under time with the free Claude Certified Architect practice exam: 20 questions with an explanation for every answer option, no sign-up needed. The CCAR-F exam guide covers format, scoring and registration.
Agentic Architecture & Orchestration (27% of the exam)
- 1.1 Agentic loops
Can you say which stop reason ends an agentic loop, and what to do with the others? - 1.2 Coordinator and subagent patterns
Can you decide when a coordinator should hand work to subagents, and what each one receives? - 1.3 Subagent invocation and context passing
Can you explain what context a subagent starts with and how to pass it what it needs? - 1.4 Workflow enforcement and handoff
Can you choose between enforcing a workflow step in code and asking the model to follow it? - 1.5 Agent SDK hooks
Can you place a check in PreToolUse or PostToolUse and explain why? - 1.6 Task decomposition
Can you break a large task into steps an agent can finish and verify? - 1.7 Session state, resumption and forking
Can you choose between resuming a session, forking it or starting fresh?
Tool Design & MCP Integration (18% of the exam)
- 2.1 Tool interface design
Can you write a tool name, description and input schema the model will choose correctly? - 2.2 Structured errors for MCP tools
Can you return a tool error the agent can act on, instead of a free-text failure? - 2.3 Tool distribution and tool choice
Can you decide which tools an agent gets, and when to force a tool choice? - 2.4 MCP server integration
Can you pick an MCP configuration scope and keep approval, permission and data access separate? - 2.5 Built-in tools
Can you tell when a built-in tool fits better than a custom one?
Claude Code Configuration & Workflows (20% of the exam)
- 3.1 CLAUDE.md hierarchy
Can you predict which CLAUDE.md files load for a given working directory? - 3.2 Slash commands and skills
Can you choose between a slash command and a skill for a repeated task? - 3.3 Path-specific rules
Can you scope a rule to only the files it should apply to? - 3.4 Plan mode vs direct execution
Can you decide when to plan first and when to execute directly? - 3.5 Iterative refinement
Can you improve a result through iteration without losing earlier constraints? - 3.6 Claude Code in CI/CD
Can you run Claude Code safely inside a CI/CD pipeline?
Prompt Engineering & Structured Output (20% of the exam)
- 4.1 Prompts with explicit criteria
Can you rewrite a vague prompt with explicit, checkable criteria? - 4.2 Few-shot prompting
Can you choose few-shot examples that show the pattern without the model copying them? - 4.3 Structured output with JSON schemas
Can you get structured output that validates against a JSON schema? - 4.4 Validation and retry loops
Can you design a validation and retry loop with a clear limit? - 4.5 Message Batches API
Can you decide when to use the Message Batches API and handle each result state? - 4.6 Multi-pass review
Can you split a review into passes that each check one thing?
Context Management & Reliability (15% of the exam)
- 5.1 Context across long conversations
Can you keep the right facts in context across a long conversation? - 5.2 Escalation and ambiguity
Can you decide when an agent should escalate to a person instead of guessing? - 5.3 Error propagation
Can you stop an error in one step from spreading through the rest of the workflow? - 5.4 Large codebase context
Can you give Claude enough of a large codebase without flooding its context? - 5.5 Human review and confidence
Can you route low-confidence outputs to human review? - 5.6 Information provenance
Can you trace where each piece of information in an answer came from?