Question 360 Two sources report different versions of an API. One is a current official API reference and the other an undated blog post. Which approach is best? Average the two version numbers. Trust the blog because it has more comments. Present both as simultaneous exact requirements without qualification. Use the versioned official reference for implementation while documenting the conflicting secondary account. Explanation Correct Answer D:Source authority and currency matter when technical contracts change. Incorrect Answers A:Version identifiers are categorical, not averaged. B:Popularity does not establish current API behavior. C:That may produce an invalid request.
Question 361 A codebase has common testing rules but special instructions for database migrations. Which TWO organizational decisions are sound? Scope migration-only conventions to the migration directory or matching files. Store common test and safety conventions in a concise root CLAUDE.md. Duplicate all instructions into every file in the repository. Place migration database passwords directly in the shared guidance. Remove test commands so Claude chooses any command it likes. Explanation Correct Answers A:Local rules avoid burdening unrelated frontend or service work. B:Common guidance should be discoverable to all project sessions. Incorrect Answers C:Excess repetition makes maintenance and context usage worse. D:Secrets must be provisioned securely outside committed instructions. E:Deterministic project-specific commands reduce errors.
Question 362 A coordinator sends raw web pages directly to a final-writing agent, which begins quoting irrelevant pages. Which division of responsibilities is preferable? Make the coordinator perform every low-level page parse itself. Use a retrieval specialist to gather sources, an analysis specialist to extract claims, and synthesis to write from vetted claims. Suppress source metadata from all workers to save tokens. Let synthesis invent source titles instead of examining retrieval outputs. Explanation Correct Answer B:Separating retrieval from evidence selection reduces noise and improves provenance. Incorrect Answers A:That removes useful specialization and can overwhelm the coordinator context. C:The writer then cannot link conclusions to evidence. D:Fabricated references undermine the deliverable.
Question 363 An assistant creates database migration code that may drop a table. What is an appropriate independent gate? Check only whether SQL syntax parses. Allow any second model to execute the migration automatically. Require a reviewer to assess destructive operations, rollback and approved change scope before execution. Treat two matching model outputs as approval. Explanation Correct Answer C:A high-impact side effect warrants explicit authorization and recovery review. Incorrect Answers A:Valid SQL can still destroy data. B:A second model is not an authorized change approver. D:Agreement cannot replace policy authorization.
Question 364 A build agent is permitted to run tests but not deploy to production. Which pre-tool condition should be checked? Wait for the deploy to complete, then ask for permission. Approve any deployment if the command parses as shell syntax. Remove deployment traces so the change is not visible. Block deploy commands outside the explicitly authorized environment and action set. Explanation Correct Answer D:A pre-tool hook can implement capability restrictions even when the model proposes the command. Incorrect Answers A:A post-deployment approval cannot prevent the unauthorized change. B:Syntactic validity is not change authorization. C:Hiding activity worsens the security violation.
Question 365 A review job must run on merge requests but not block a release solely for speculative low-confidence findings. What policy is appropriate? Define severity and evidence thresholds, preserving human escalation for uncertain high-impact cases. Deploy regardless of deterministic failing security tests. Block every PR whenever the model mentions any possible issue. Suppress all review findings to guarantee throughput. Explanation Correct Answer A:A calibrated gate avoids noisy false blocks while treating genuine risks seriously. Incorrect Answers B:Objective test failures still require resolution. C:Uncalibrated alerts can halt healthy work. D:That discards the intended quality control.
Question 366 A security team wants to log attempted shell commands but redact tokens before storing them. What hook behavior is appropriate? Store raw environment variables and secrets in a shared analytics table. Log only the final assistant summary and never the command. Inspect command arguments in a pre-execution hook and log sanitized metadata under policy. Rewrite every shell command into an unrelated command for privacy. Explanation Correct Answer C:The hook can record the request while preventing sensitive values from entering telemetry. Incorrect Answers A:Overcollection creates a new credential exposure. B:The investigator would lack the actual action evidence. D:Uncontrolled rewriting changes execution semantics.
Question 367 A document extractor needs to return null when the field is absent, but examples always show populated values. What few-shot addition is best? Make up plausible field values for absent examples. Include genuine absent-field cases with null outputs and no invented placeholders. Delete the field from the schema without downstream agreement. Use only more fully populated documents. Explanation Correct Answer B:Negative examples teach that missing data are not grounds for guessing. Incorrect Answers A:Fabrication would teach the wrong behavior. C:The consumer may require a stable nullable field. D:The relevant edge case would remain unaddressed.
Question 368 A coordinator invokes the same data-extraction worker for two independent files in parallel. How should returned results be associated with the requests? Use the individual invocation or tool-use identifiers and document IDs. Associate by whichever result arrives first. Discard the worker input references after dispatch. Combine both extractions into one undifferentiated text block. Explanation Correct Answer A:Identity fields prevent one file’s extraction from being mistaken for the other’s. Incorrect Answers B:Network timing does not identify which input a result belongs to. C:Without request identifiers, outputs become ambiguous. D:Loss of document provenance makes downstream interpretation unreliable.
Question 369 A monorepo has organization-wide naming and release conventions that every Claude Code session must follow. Where should shared project guidance be kept? Only in one engineer’s private terminal history. In an ephemeral tool response that will disappear. In a version-controlled project CLAUDE.md with stable cross-repository conventions. In a production database password variable. Explanation Correct Answer C:Project memory makes the team rules available consistently to work in that codebase. Incorrect Answers A:Other users and fresh sessions cannot reliably access that instruction. B:Transient output is not a durable team policy. D:Secrets are not a replacement for documented development instructions.
Question 370 An insurer needs documents translated, extracted, and checked for policy coverage. Some files are already translated. Which approach avoids waste? Check artifact state and execute only missing, dependency-ordered steps for each document. Skip extraction when translation exists. Run every stage concurrently even when one requires the previous output. Retranslate every file whenever any stage changes. Explanation Correct Answer A:Workflow routing can reuse completed translations while preserving required extraction and review. Incorrect Answers B:Translation does not create the coverage facts the next stage needs. C:Dependency violations can produce invalid or empty results. D:This repeats completed work and may introduce inconsistencies.
