Question 430
A Claude-based coding assistant generates patch-review records. The model returns a plausible patch summary even though required static checks were skipped. What should the architect specify in the design?
- Mark missing checks as blocked and do not assign a passing verdict
- Run permitted commands and report any unavailable or failing test conditions
- Request targeted correction while retaining validated findings
- Check line ranges against the exact repository revision before accepting findings
- Cross-check referenced symbols with source or build metadata
Explanation
Correct Answer
A:A fluent summary cannot turn skipped checks into evidence; state should reflect actual execution results. For the stated case, the check is whether the system can separate generated narrative from validation status; this approach supplies that control.
Incorrect Answers
B:Describing test commands does not prove they are executable or passing; CI evidence is required; it is a safeguard for the assistant proposes tests that do not run in the current build, not the specific situation described: the model returns a plausible patch summary even though required static checks were skipped.
C:For the separate situation where a correction retry rewrites already verified findings unnecessarily, narrow repair reduces regression risk and prevents fresh hallucinations in sections that already passed checking; the current task is to separate generated narrative from validation status.
D:This choice is relevant when a line number in a valid review object exceeds the length of the referenced file; schema integers can still point outside actual files; source-aware validation prevents misleading review annotations. Here, however, the model returns a plausible patch summary even though required static checks were skipped.
E:Factual checking must follow formatting validation because a correctly typed nonexistent method is still an invalid finding; the benefit is tied to a finding names an API method that has been removed. It is not the immediate control for the need to separate generated narrative from validation status.
Question 431
A document-extraction pipeline exposes conversion and validation tools to Claude. The team continually adds loosely related parameters to one giant function. Which change most directly addresses this condition?
- Constrain the input schema to supported currency codes and document normalization
- Split distinct operations into purpose-built tools with focused schemas
- Include lookup time and source provenance in the result contract
- Return page identifiers, text, status, and errors as separate structured records
Explanation
Correct Answer
B:Narrow functions are easier for Claude to select and for the host to validate than a parameter-heavy command that mixes unrelated side effects.
Incorrect Answers
A:This choice is relevant when a validator requires normalized currency codes but callers send symbols; a typed canonical contract prevents ambiguous symbols from becoming silently incorrect financial records. Here, however, the team continually adds loosely related parameters to one giant function.
C:For the separate situation where an address verifier uses an external data source that may be outdated, a timestamped external check lets downstream logic decide whether the verification is fresh enough for the task; the current task is to reduce ambiguous multi-purpose tool surfaces.
D:Per-page status distinguishes a partly readable document from a fully successful extraction and supports targeted retries; the benefit is tied to an OCR tool can return multiple pages with partial failures. It is not the immediate control for the need to reduce ambiguous multi-purpose tool surfaces.
Question 432
A customer-service agent uses hooks around tool operations. A dangerous path appears in file-write arguments before a tool runs. What is the most defensible next step?
- Inspect and reject paths outside the approved workspace in pre-tool logic
- Use a pre-execution hook or service policy that rejects unauthorized refund calls
- Transform dates at a controlled post-tool boundary into a canonical format
- Remove or deny shell execution in the actual tool permissions or hook
Explanation
Correct Answer
A:Argument-level checks prevent an otherwise permitted write tool from escaping the authorized directory. Here it serves the explicit need to validate target paths before writes without adding a separate unverified inference.
Incorrect Answers
B:This choice is relevant when a refund tool must never execute for amounts above the permitted threshold without approval; a deterministic tool boundary can block a high-value mutation even if a model instruction fails to deter it. Here, however, a dangerous path appears in file-write arguments before a tool runs.
C:A post-tool normalization step makes structured evidence consistent without hoping a later prompt corrects every variant; the benefit is tied to a source-search tool returns inconsistent date formats that confuse downstream parsing. It is not the immediate control for the need to validate target paths before writes.
D:For the separate situation where a prompt says 'never run shell commands,' but the tool remains available and callable, text guidance is probabilistic, whereas execution controls can reject a forbidden operation regardless of model output; the current task is to validate target paths before writes.
Question 433
A Claude Code team audits the quality of its persistent project instructions. One developer asks Claude Code to ignore a required security gate for speed. Which approach satisfies the stated constraint?
- Summarize service boundaries and point to authoritative files for details
- Document the actual lint, test, and review procedures with relevant paths
- Resolve the conflict and express one authoritative formatting rule
- Enforce the gate in CI or permission policy, while keeping guidance aligned
Explanation
Correct Answer
D:Prompt instructions are not an authoritative security boundary; a required check should fail independently of the assistant's compliance.
Incorrect Answers
A:For the separate situation where the team wants Claude to know repository architecture without pasting full source files, a navigational map improves codebase discovery while leaving large or frequently changing implementation text in its source files; the current task is to protect mandatory controls outside optional prose.
B:This choice is relevant when CLAUDE.md contains vague advice to 'write good code' but omits repository commands; specific commands and expected outcomes give Claude verifiable steps; broad style slogans offer little operational guidance. Here, however, one developer asks Claude Code to ignore a required security gate for speed.
C:Conflicting directives can produce nondeterministic edits; a clear standard prevents wasted patch iterations; the benefit is tied to two instruction files issue contradictory formatting requirements. It is not the immediate control for the need to protect mandatory controls outside optional prose.
Question 434
A repository assistant uses several parallel agents to examine unfamiliar services. One specialist's findings are needed by three downstream specialists. Which mechanism should be selected?
- Partition write ownership or serialize changes to the shared file
- State the expected test command and success condition in the delegated task
- Return a concise file-indexed summary and paths to relevant excerpts
- Write the validated summary once and pass its identifier to downstream jobs
Explanation
Correct Answer
D:A versioned artifact prevents three copies of the same evolving findings from diverging across worker contexts. The relevant requirement is to make a durable shared result artifact, and this measure operates at that decision point.
Incorrect Answers
A:This choice is relevant when two agents intend to edit the same config file concurrently; parallel reads may be safe, but simultaneous edits to shared state invite races and difficult-to-merge patches. Here, however, one specialist's findings are needed by three downstream specialists.
B:A specialist with explicit verification criteria can return evidence of completion instead of an untested code suggestion; it is a safeguard for a worker must verify that its proposed fix passes a particular unit test, not the specific situation described: one specialist's findings are needed by three downstream specialists.
C:A narrow handoff saves parent context while retaining references that can be reopened when deeper inspection is needed; the benefit is tied to a search agent finishes with fifty raw file excerpts. It is not the immediate control for the need to make a durable shared result artifact.
Question 435
An agent runs a bounded support workflow with multiple optional tool calls. A response contains no tool calls but the task is still unresolved and requires human judgment. What should be implemented at the appropriate boundary?
- Return a bounded handoff or clarification rather than forcing more autonomous turns
- Log authorized tool names, request IDs, outcomes, and decision context
- Use rate-aware concurrency limits and jittered backoff
- Enforce a maximum tool-call or iteration budget and a fallback outcome
- Check cancellation state before dispatching queued tool work
Explanation
Correct Answer
A:When prerequisites are missing, another iteration does not create evidence; explicit unresolved state protects correctness and enables review.
Incorrect Answers
B:For the separate situation where support needs a record of why an agent issued each sensitive call, structured execution logs establish which external action occurred and on what evidence without assuming that model prose is a reliable audit record; the current task is to terminate safely with an unresolved status.
C:Application-side scheduling controls aggregate demand; independent immediate retries can amplify throttling across sessions; it is a safeguard for a backend becomes rate limited while many agents share it, not the specific situation described: a response contains no tool calls but the task is still unresolved and requires human judgment.
D:This choice is relevant when a tool-driven conversation keeps calling the same read-only endpoint without progress; without a host-side bound, repeatedly proposing the same action can consume resources without resolving the customer's issue. Here, however, a response contains no tool calls but the task is still unresolved and requires human judgment.
E:A cancellation flag in application state can block new operations; telling the model to stop is not sufficient for queued execution; the benefit is tied to a user cancels a request while a queued tool operation has not started. It is not the immediate control for the need to terminate safely with an unresolved status.
Question 436
A development team wants consistent Claude Code behavior across a shared monorepo. The team changes a shared convention but contributors have conflicting personal notes. Which option best meets the operational need?
- Place package-specific guidance near the relevant subproject or use path-scoped rules
- Inspect project and user guidance together and resolve conflicting directions explicitly
- Place concise project instructions in a committed project CLAUDE.md file
- Keep the personal preference in user-level Claude Code guidance
Explanation
Correct Answer
B:Overlapping memory files may supply inconsistent directions; an architect should make shared requirements clear rather than hope the model infers priority.
Incorrect Answers
A:Local rules reduce noise in unrelated tasks and clarify when a special test command actually applies; it is a safeguard for a single package has unusual testing requirements not relevant elsewhere, not the specific situation described: the team changes a shared convention but contributors have conflicting personal notes.
C:This choice is relevant when every contributor must follow the same build and review commands for the repository; project instructions make conventions available when working in the repository; they should describe relevant workflow rules rather than replicate all documentation. Here, however, the team changes a shared convention but contributors have conflicting personal notes.
D:User-level configuration supports individual habits without imposing personal formatting conventions on the entire repository; the benefit is tied to one developer prefers a personal response style that colleagues do not share. It is not the immediate control for the need to clarify the applicable instruction hierarchy.
Question 437
A document classifier uses Claude to tag contracts with legal categories. The team has hundreds of redundant examples in a single prompt. What is the most appropriate design choice?
- Include labeled examples that explain how to break ties
- Select diverse boundary cases based on observed error patterns
- Choose examples from different layouts and writing styles while keeping the rubric stable
- Replace obsolete examples and rerun evaluation on current cases
Explanation
Correct Answer
B:A smaller curated set often gives clearer task guidance and avoids burying the decision rule under duplicated demonstrations.
Incorrect Answers
A:This choice is relevant when a particular clause belongs to two related categories but reviewers expect one primary label; examples should show which category wins at overlapping boundaries rather than leaving Claude to infer an unstated hierarchy. Here, however, the team has hundreds of redundant examples in a single prompt.
C:Varied examples prevent the model from associating a label with incidental formatting instead of the clause meaning; the benefit is tied to the training examples use only one document style, but production inputs vary. It is not the immediate control for the need to curate a small representative set.
D:For the separate situation where a prompt example contains outdated policy thresholds, few-shot content can encode policy mistakes as strongly as explicit instructions; stale exemplars should be treated as configuration debt; the current task is to curate a small representative set.
Question 438
A Claude Agent SDK workflow delegates repository investigations to subagents. A subagent returns a broad recommendation without citing inspected files. Which response is most reliable?
- Run the investigation in a subagent and return a concise evidence-based summary
- Have the worker report changed or inspected paths and evidence for its conclusion
- Give the Task-oriented specialist a concrete goal, scope, and expected handoff
- Restrict the worker's allowedTools to safe inspection operations
Explanation
Correct Answer
B:The parent can validate a specialist's claims only when the output identifies what was examined or modified. This matches the workflow goal to require a verifiable handoff from the worker rather than introducing an unrelated action.
Incorrect Answers
A:An isolated worker context avoids flooding the coordinator with irrelevant file reads while preserving the actionable findings; the benefit is tied to a specialist's large file-search transcript is not useful to the parent. It is not the immediate control for the need to require a verifiable handoff from the worker.
C:For the separate situation where a parent agent needs a specialist to take a bounded action rather than just offer advice, structured delegated tasks clarify which worker owns the operation and how its outcome will return to the parent; the current task is to require a verifiable handoff from the worker.
D:This choice is relevant when a worker should be able to inspect code but not run destructive shell commands; tool allowlists define an execution boundary that remains effective even when the worker's text suggests a dangerous action. Here, however, a subagent returns a broad recommendation without citing inspected files.
Question 439
An enterprise design team compares agent organization patterns for different requests. Research outputs frequently conflict and need judgment beyond automated rules. What should the team do to resolve this particular gap?
- Expose a narrowly defined reusable translation worker behind a clear input/output contract
- Use sequential stages with explicit prerequisite checks
- Route disputed findings to a reviewer or human decision before publication
- Use one constrained model call instead of a multi-agent chain
Explanation
Correct Answer
C:Uncertain conflicts should be surfaced and resolved by an accountable stage rather than concealed by the model's confidence.
Incorrect Answers
A:For the separate situation where the same specialized translation step repeats across many workflows, a well-defined specialist reduces duplicated orchestration logic while keeping translation behavior consistent across callers; the current task is to introduce an explicit adjudication stage.
B:The second stage depends on verified clauses, so parallel dispatch would let it act on absent or provisional data; it is a safeguard for a legal review cannot begin until a contract-clause extractor finishes, not the specific situation described: research outputs frequently conflict and need judgment beyond automated rules.
D:This choice is relevant when a small deterministic classification needs no external calls; simple classification has no meaningful subtask parallelism; multiple agents add latency and reconciliation cost without improving the required outcome. Here, however, research outputs frequently conflict and need judgment beyond automated rules.
Question 440
A support assistant assigns risk and confidence to proposed customer actions. Monitoring shows confident unsupported answers increasing after a prompt change. Which control is most directly relevant?
- Require actual policy checks and relevant source fields before approving action
- Route the transaction through the mandatory human approval gate
- Compare confidence buckets against observed correctness and adjust routing thresholds
- Audit a representative sample of routine resolutions while reviewing high-risk exceptions individually
- Request the missing field or hand off instead of guessing
Explanation
Correct Answer
C:Reliability requires empirical error measurement; a fluent explanation is not proof that reported certainty matches accuracy.
Incorrect Answers
A:This choice is relevant when a high-confidence model verdict has no linked eligibility evidence; confidence scores are not evidence; external authorization and validated facts determine whether a sensitive operation may proceed. Here, however, monitoring shows confident unsupported answers increasing after a prompt change.
B:For the separate situation where a high-impact refund crosses a configured approval threshold, the threshold is an execution policy independent of the model's expressed confidence about the refund; the current task is to measure calibration with labeled examples.
D:Risk-stratified sampling gives reviewers evidence of routine decision quality without making every low-risk response wait for approval; mandatory high-risk gates remain in place; it is a safeguard for low-risk policy responses need ongoing quality monitoring, but reviewing every response is overwhelming the support team, not the specific situation described: monitoring shows confident unsupported answers increasing after a prompt change.
E:Uncertainty arising from missing information needs a data remedy, not stronger prompt language or fabricated certainty; the benefit is tied to a decision has low confidence because required account information is missing. It is not the immediate control for the need to measure calibration with labeled examples.