Question 70
A shared wrapper renames a required field used by several downstream services. How should the implementation handle this need?
- Use a compatibility test matrix for upgrades
- Document breaking API changes for callers
- Pin a tested exact model version when necessary
- Return actionable validation errors to callers
Explanation
Correct Answer
B:A versioned change notice and migration path help clients adopt the new contract safely—with this limitation: a silent rollout may turn valid old requests into failures.
Incorrect Answers
A:Matrix tests exercise supported combinations against contract and quality checks; however, the critical mechanism here is that a versioned change notice and migration path help clients adopt the new contract safely, and assuming backward compatibility without testing risks hidden behavioral changes.
C:Version pinning keeps the model selection stable until the team deliberately upgrades and retests; this cannot substitute for the behavior that a versioned change notice and migration path help clients adopt the new contract safely, because the pinned model may later require migration if the provider retires it.
D:A structured 4xx response names the offending field without leaking internal implementation details; in contrast, a versioned change notice and migration path help clients adopt the new contract safely, while a generic 500 would misleadingly suggest a server outage.
Question 71
A legal-comparison assistant delegates source gathering to one helper and independently grades citation completeness before publication. Which TWO worker roles make that separation explicit? Choose TWO.
- Assign tool side effects to an execution worker
- Use a summarizing worker for context compression
- Delegate quality checking to a critic worker
- Delegate retrieval to a research worker
- Use a specialist subagent
Explanation
Correct Answers
C:An evaluator examines the artifact against explicit tests instead of composing the primary answer; in production, critique must have actionable criteria and should not be equated with guaranteed correctness.
D:A research worker gathers references and returns a compact evidence set to the lead; it should not be allowed to publish the final legal conclusion as an independent authority.
Incorrect Answers
A:A tightly permissioned executor maps authorized operations into deterministic tool calls; here, the controlling capability is that an evaluator examines the artifact against explicit tests instead of composing the primary answer, but broad access in the planning component defeats the intended separation of duties.
B:A dedicated summarizer condenses retained facts with provenance and open issues for the parent context; the missing capability is more specific: an evaluator examines the artifact against explicit tests instead of composing the primary answer; compression may omit nuance unless the system checks critical commitments and references.
E:An isolated worker handles a bounded analysis and returns a concise finding to the parent, whereas the requirement calls for another behavior: an evaluator examines the artifact against explicit tests instead of composing the primary answer; the subagent still needs a scoped tool set and cannot silently inherit unlimited production privileges.
Question 72
A subscription-support assistant has drifted toward unrelated advice and still relies on a policy copy that predates the latest product update. Which TWO context practices reestablish the authorized goal and up-to-date evidence? Choose TWO.
- Segment work into explicit milestones
- Use an external source refresh for changing facts
- Preserve immutable business facts in structured state
- Reassert the current task objective concisely
- Retrieve relevant evidence per user question
Explanation
Correct Answers
B:Refetching authoritative information avoids treating outdated context as eternally current; its operational boundary is that the model should not invent a change simply because its history is old.
D:Reintroducing the active goal and constraints guides continuation back to the approved scope; the implementation must still recognize that repetition cannot override the need to preserve factual state accurately.
Incorrect Answers
A:Milestones with acceptance criteria limit aimless elaboration and provide reliable checkpoints; this cannot substitute for the behavior that refetching authoritative information avoids treating outdated context as eternally current, because a plan without tested completion criteria will still drift.
C:The application records authoritative facts outside model-generated summaries and supplies them when needed; this addresses a separate concern, not the fact that refetching authoritative information avoids treating outdated context as eternally current; conversation recollection can drift or omit earlier commitments.
E:Retrieval selects bounded passages pertinent to the current question and supplies their source context; the missing capability is more specific: refetching authoritative information avoids treating outdated context as eternally current; selecting on superficial keywords alone may miss nuanced exceptions.
Question 73
A team wants to move across SDK and model versions while retaining tool and output schema behavior. What should the team do at this decision point?
- Use a compatibility test matrix for upgrades
- Run contract tests against a versioned API schema
- Keep rollback artifacts and data migration notes
- Pin a tested exact model version when necessary
Explanation
Correct Answer
A:Matrix tests exercise supported combinations against contract and quality checks—with this limitation: assuming backward compatibility without testing risks hidden behavioral changes.
Incorrect Answers
B:Contract tests enforce agreed field names, types, and error shapes across producer and client; this cannot substitute for the behavior that matrix tests exercise supported combinations against contract and quality checks, because casual manual testing may miss a breaking field change.
C:Release planning preserves prior binaries and any required data conversion strategy; in contrast, matrix tests exercise supported combinations against contract and quality checks, while reverting only source files may not undo incompatible database changes.
D:Version pinning keeps the model selection stable until the team deliberately upgrades and retests; this addresses a separate concern, not the fact that matrix tests exercise supported combinations against contract and quality checks; the pinned model may later require migration if the provider retires it.
Question 74
A desktop coding workflow must limit which local actions the coding agent may perform. Which technical control is appropriate?
- Invalidate session privileges after role changes
- Inject task-specific retrieval into an API call
- Read project guidance from Claude Code files
- Configure permissions within the relevant client
Explanation
Correct Answer
D:Client-specific settings define the environment and its permitted tool behavior; the implementation must still recognize that configuring a distant backend does not necessarily constrain a local coding session.
Incorrect Answers
A:Authorization checks must reflect current permissions rather than those cached at session creation; this cannot substitute for the behavior that client-specific settings define the environment and its permitted tool behavior, because prior approval of one request does not authorize later actions forever.
B:Retrieval supplies authorized source excerpts as contextual content in the request; in contrast, client-specific settings define the environment and its permitted tool behavior, while model training data is not a reliable substitute for current private documents.
C:Claude Code can load contextual project instructions such as CLAUDE.md during its sessions; this addresses a separate concern, not the fact that client-specific settings define the environment and its permitted tool behavior; those instructions do not automatically govern an isolated custom API service.
Question 75
An agent integration attempts restricted tool operations more often after a new prompt rollout. What is the best fit for the described situation?
- Use a golden evaluation set for regressions
- Apply authorization before retrieval
- Instrument output schema failure rate
- Track side-effect authorization denials
Explanation
Correct Answer
D:Auditing denied calls helps identify policy confusion or prompt injection pressure, subject to this requirement: denials should remain enforced even if error metrics worsen.
Incorrect Answers
A:Fixed labeled cases detect performance changes across relevant user segments; the requirement instead turns on this behavior: auditing denied calls helps identify policy confusion or prompt injection pressure; one successful demo query does not establish production quality.
B:The retrieval layer filters sources by verified caller access before any text reaches Claude; this cannot substitute for the behavior that auditing denied calls helps identify policy confusion or prompt injection pressure, because filtering only the final answer cannot undo earlier disclosure to the model.
C:Tracking validation failures isolates content contract problems from transport availability; in contrast, auditing denied calls helps identify policy confusion or prompt injection pressure, while HTTP success is not equivalent to a valid business payload.
Question 76
A report generator needs several paragraphs but the response cap only permits a short fragment. Which production safeguard is most relevant?
- Reserve enough output tokens for the desired result
- Estimate token usage with supported token counting
- Use zero-shot instructions for a straightforward task
- Count the complete request against context limits
Explanation
Correct Answer
A:The configured generation maximum must be consistent with the required output length, subject to this requirement: raising it arbitrarily can increase possible cost without improving content accuracy.
Incorrect Answers
B:Token counting estimates model tokens more accurately than character length or word count; here, the controlling capability is that the configured generation maximum must be consistent with the required output length, but different formats and models may tokenize the same text differently.
C:Direct instructions define the output without consuming tokens on demonstrations; the missing capability is more specific: the configured generation maximum must be consistent with the required output length; complex edge cases may still need examples and evaluations.
D:The context budget encompasses the relevant input contents and model-specific output allowances; by comparison, the intended solution relies on this: the configured generation maximum must be consistent with the required output length; counting only the latest user text understates true token pressure.
Question 77
A long discussion has wandered away from the user's original authorized deliverable. Which of the following would work as specified?
- Start a new clean window after task boundary
- Preserve immutable business facts in structured state
- Reassert the current task objective concisely
- Use an external source refresh for changing facts
Explanation
Correct Answer
C:Reintroducing the active goal and constraints guides continuation back to the approved scope, although repetition cannot override the need to preserve factual state accurately.
Incorrect Answers
A:Separating independent sessions avoids accidental carryover of assumptions and private context; by comparison, the intended solution relies on this: reintroducing the active goal and constraints guides continuation back to the approved scope; the new session must still receive any relevant persistent project policy.
B:The application records authoritative facts outside model-generated summaries and supplies them when needed; the needed capability is different—reintroducing the active goal and constraints guides continuation back to the approved scope—and conversation recollection can drift or omit earlier commitments.
D:Refetching authoritative information avoids treating outdated context as eternally current; however, the critical mechanism here is that reintroducing the active goal and constraints guides continuation back to the approved scope, and the model should not invent a change simply because its history is old.
Question 78
A build pipeline needs a Claude Code result without an interactive terminal conversation. Which of the following would work as specified?
- Use an interactive session for exploratory editing
- Review effective settings before broad automation
- Use structured or streaming output for pipeline consumers
- Use non-interactive print mode for CI automation
Explanation
Correct Answer
D:Headless command-line invocation can run a prompt and return consumable output under controlled settings, subject to this requirement: CI still needs noninteractive credentials and appropriately limited permissions.
Incorrect Answers
A:An interactive coding session supports iterative commands and review as context evolves; the requirement instead turns on this behavior: headless command-line invocation can run a prompt and return consumable output under controlled settings; automating every command without supervision is unnecessary for a human-led investigation.
B:Examining effective permission and hook configuration catches dangerous inherited allowances; this cannot substitute for the behavior that headless command-line invocation can run a prompt and return consumable output under controlled settings, because a printed prompt cannot override an enforced policy hierarchy.
C:Machine-friendly output formats allow programmatic consumers to handle individual events; in contrast, headless command-line invocation can run a prompt and return consumable output under controlled settings, while human-oriented terminal decoration is not a stable protocol contract.
Question 79
An invoice must be created before its identifier can be used to attach a payment. Which approach targets the deciding condition?
- Run independent read-only tool calls concurrently
- Serialize dependent tool calls
- Separate planning from execution
- Set a maximum tool iteration count
Explanation
Correct Answer
B:Sequential execution preserves the data dependency and passes the created ID to the following action, subject to this requirement: running the payment call early risks missing or stale identifiers.
Incorrect Answers
A:Concurrent dispatch shortens latency when tool requests are independent and safely isolated; in contrast, sequential execution preserves the data dependency and passes the created ID to the following action, while shared write operations require separate coordination and consistency protections.
C:The application can review and authorize the plan before handing it to a restricted executor; here, the controlling capability is that sequential execution preserves the data dependency and passes the created ID to the following action, but treating a plan as permission to execute is a serious privilege error.
D:A harness enforces a fixed ceiling on action cycles and reports incomplete execution safely; the missing capability is more specific: sequential execution preserves the data dependency and passes the created ID to the following action; model self-assessment is too probabilistic to serve as the only loop brake.
Question 80
A support bot must quote the latest policy instead of relying on older snippets in a conversation. How should the implementation handle this need?
- Identify external integration contracts
- Specify human-in-the-loop responsibilities
- Document data-handling and retention constraints
- Identify the upstream source of truth
Explanation
Correct Answer
D:Architecture requirements name the maintained policy repository and its freshness constraints; in production, instructing Claude to remember policy text does not keep that source current.
Incorrect Answers
A:Interface requirements document endpoints, authentication, payloads, limits, and failure handling; this addresses a separate concern, not the fact that architecture requirements name the maintained policy repository and its freshness constraints; a prose-level prompt is not a substitute for a validated service contract.
B:Requirements distinguish permitted recommendations from authority to execute a binding business commitment; here, the controlling capability is that architecture requirements name the maintained policy repository and its freshness constraints, but a general claim that the assistant is safe does not enforce this distinction.
C:Privacy requirements identify permitted collection, transmission, storage, retention, and deletion controls; the missing capability is more specific: architecture requirements name the maintained policy repository and its freshness constraints; developer convenience cannot override legal or organizational handling obligations.