These 12 questions are drawn from the official exam guide to illustrate the format, difficulty level, and reasoning expected. Each includes a detailed explanation.
Production data shows that in 12% of cases, your agent skips get_customer entirely and calls lookup_order using only the customer's stated name, occasionally leading to misidentified accounts and incorrect refunds. What change would most effectively address this reliability issue?
When a specific tool sequence is required for critical business logic, programmatic enforcement provides deterministic guarantees that prompt-based approaches cannot. B and C rely on probabilistic compliance. D addresses tool availability rather than ordering.
Production logs show the agent frequently calls get_customer when users ask about orders (e.g., "check my order #12345"), instead of calling lookup_order. Both tools have minimal descriptions and accept similar formats. What is the most effective first step?
Tool descriptions are the primary mechanism LLMs use for tool selection. B directly addresses the root cause. A adds token overhead without fixing the underlying issue. C is over-engineered. D requires too much effort for a first step.
Your agent achieves 55% first-contact resolution (target 80%). It escalates straightforward cases while handling complex ones autonomously. What is the most effective fix?
Explicit criteria address the root cause: unclear decision boundaries. B fails because LLM confidence is poorly calibrated. C is over-engineered. D solves the wrong problem — sentiment does not equal complexity.
You want to create a team /review command available to all developers on clone/pull. Where should you create it?
Project-scoped commands in .claude/commands/ are version-controlled and auto-available on clone. B is personal only. C is for instructions. D doesn't exist.
You are restructuring a monolith into microservices — dozens of files, service boundary decisions. Which approach is best?
Plan mode is designed for complex tasks with large-scale changes and architectural decisions. B risks rework. C assumes knowledge without exploration. D ignores stated complexity.
Your codebase has different conventions per area (React, API, DB). Test files are spread throughout. What is the most maintainable way to auto-apply correct conventions?
.claude/rules/ with globs (e.g., **/*.test.tsx) auto-applies based on file paths regardless of directory. B relies on inference. C requires manual invocation. D can't handle cross-directory files.
A report on "impact of AI on creative industries" only covers visual arts. The coordinator decomposed the topic into only visual arts subtasks. All subagents completed successfully. What is the root cause?
The coordinator's logs reveal it only assigned visual arts subtasks. Subagents worked correctly within their scope. The decomposition is the root cause.
The web search subagent times out. What is the best error propagation approach?
Structured context enables intelligent coordinator recovery. B hides context. C suppresses the error. D terminates unnecessarily.
The synthesis agent needs fact verification. 85% are simple fact-checks, 15% require deep investigation. How do you reduce the 40% latency overhead?
A applies least privilege for the 85% case. B creates blocking dependencies. C over-provisions. D relies on speculative caching.
Your CI pipeline hangs — Claude Code is waiting for interactive input. What is the correct approach?
-p (--print) is the documented non-interactive mode. B and D reference non-existent features. C is a workaround, not proper syntax.
Your manager proposes using the Batch API for both pre-merge checks and an overnight technical debt report. How do you evaluate this?
The Batch API has up to 24-hour processing, unsuitable for blocking pre-merge checks. Ideal for overnight jobs. B is wrong because latency is unacceptable for blocking workflows. C misses valid cost optimization. D adds unnecessary complexity.
A 14-file PR review produces inconsistent results, missed bugs, and contradictory feedback. How should you restructure the review?
Multi-pass addresses the attention dilution root cause. B shifts burden without improving the system. C misunderstands that context window size does not equal attention quality. D suppresses intermittent bug detection.
Everything you need for the Claude Certified Architect exam