Question 50
For a dependency upgrade initiative, acceptance depends on demonstrating that a straightforward lookup avoids unnecessary agent coordination. Which approach is the strongest fit?
- limit concurrency and schedule independent jobs against execution budgets
- add a synthesis stage with explicit conflict-resolution criteria
- use one focused tool-assisted agent for the bounded task
- spawn a coordinator plus multiple specialists for every trivial query
Explanation
Correct Answer
C:For a dependency upgrade initiative, orchestration overhead is not justified when there is no useful specialization or concurrency; this directly establishes that a straightforward lookup avoids unnecessary agent coordination instead of relying on an unrelated workflow control.
Incorrect Answers
A:Although bounded orchestration balances throughput with tool quotas and downstream service capacity, that only supports the distinct goal that workload growth does not cause uncontrolled parallel fan-out, not the requirement that a straightforward lookup avoids unnecessary agent coordination in a dependency upgrade initiative.
B:Although merging divergent work requires an authoritative step rather than choosing by arrival time, that only supports the distinct goal that two teams' analyses are reconciled before publication, not the requirement that a straightforward lookup avoids unnecessary agent coordination in a dependency upgrade initiative.
D:In a dependency upgrade initiative, extra agents add latency and handoff risk without new capabilities; as a result, the stated requirement that a straightforward lookup avoids unnecessary agent coordination would still not be met.
Question 51
Production readiness for a source-grounded policy synthesis service depends on ensuring that a selector accepts only supported operation names and that a delegated agent receives the background it needs to act. Select the most appropriate combined approach.
- ask the model to invent a suitable action string; and pass the necessary task context and references explicitly at invocation
- declare an operation enum in the schema and reject values outside it in the tool handler; and assume the child agent sees all previous coordinator turns
- ask the model to invent a suitable action string; and assume the child agent sees all previous coordinator turns
- declare an operation enum in the schema and reject values outside it in the tool handler; and pass the necessary task context and references explicitly at invocation
Explanation
Correct Answer
D:In a source-grounded policy synthesis service, schema constraints guide valid selections and runtime validation blocks unsupported operation values; alongside this, subagents do not automatically inherit the coordinator's entire conversational history, satisfying the distinct requirements that a selector accepts only supported operation names and that a delegated agent receives the background it needs to act.
Incorrect Answers
A:In a source-grounded policy synthesis service, subagents do not automatically inherit the coordinator's entire conversational history, but free-form labels can name unimplemented operations, and a prompt is not an enforced allowlist; therefore the separate requirement that a selector accepts only supported operation names would remain unsatisfied.
B:In a source-grounded policy synthesis service, schema constraints guide valid selections and runtime validation blocks unsupported operation values, but implicit context inheritance makes delegation fragile and incomplete; therefore the separate requirement that a delegated agent receives the background it needs to act would remain unsatisfied.
C:In a source-grounded policy synthesis service, free-form labels can name unimplemented operations, and a prompt is not an enforced allowlist and implicit context inheritance makes delegation fragile and incomplete; the design would fail both conditions: a selector accepts only supported operation names and a delegated agent receives the background it needs to act.
Question 52
An audit of a backend API maintenance team asks how to make sure that project guidance fits within a limited context budget. Which implementation should be selected?
- keep the top-level instruction set concise and move specialized details to scoped files
- use CLAUDE.md imports to organize referenced instructions
- paste complete design manuals into the top-level CLAUDE.md
- put concise shared guidance in the repository root CLAUDE.md
Explanation
Correct Answer
A:For a backend API maintenance team, focused guidance preserves useful context and reduces instruction competition; this directly establishes that project guidance fits within a limited context budget instead of relying on an unrelated workflow control.
Incorrect Answers
B:Although modular imports keep a short entry point while preserving reviewed details in separate files, that only supports the distinct goal that long project guidance is divided into maintainable source files, not the requirement that project guidance fits within a limited context budget in a backend API maintenance team.
C:In a backend API maintenance team, indiscriminate loading spends tokens on facts unrelated to the active task; as a result, the stated requirement that project guidance fits within a limited context budget would still not be met.
D:Although project instructions travel with the codebase and establish common architecture and build constraints, that only supports the distinct goal that all contributors receive the repository's stable engineering conventions, not the requirement that project guidance fits within a limited context budget in a backend API maintenance team.
Question 53
For an internal knowledge research system, a reviewer identifies two independent obligations: a tool receives a typed and bounded request and a large research job combines specialized findings into one answer. Which proposal should be approved?
- define required properties and types in the input schema and validate arguments at the execution boundary; and use a coordinator that delegates focused tasks and owns final synthesis
- accept one unstructured free-text argument for every operation; and have all subagents independently publish the final answer
- accept one unstructured free-text argument for every operation; and use a coordinator that delegates focused tasks and owns final synthesis
- define required properties and types in the input schema and validate arguments at the execution boundary; and have all subagents independently publish the final answer
Explanation
Correct Answer
A:In an internal knowledge research system, input schemas describe legal arguments, while explicit validation prevents unsupported values from reaching tool logic; alongside this, the hub can integrate distinct subagent outputs and resolve overlap without peer-to-peer chatter, satisfying the distinct requirements that a tool receives a typed and bounded request and that a large research job combines specialized findings into one answer.
Incorrect Answers
B:In an internal knowledge research system, unconstrained prose leaves parsing and validation errors to ad hoc handler inference and independent final outputs offer no single point to reconcile contradictions; the design would fail both conditions: a tool receives a typed and bounded request and a large research job combines specialized findings into one answer.
C:In an internal knowledge research system, the hub can integrate distinct subagent outputs and resolve overlap without peer-to-peer chatter, but unconstrained prose leaves parsing and validation errors to ad hoc handler inference; therefore the separate requirement that a tool receives a typed and bounded request would remain unsatisfied.
D:In an internal knowledge research system, input schemas describe legal arguments, while explicit validation prevents unsupported values from reaching tool logic, but independent final outputs offer no single point to reconcile contradictions; therefore the separate requirement that a large research job combines specialized findings into one answer would remain unsatisfied.
Question 54
In a technical review of a nightly code quality job, the architect must address the following constraints together: a risky refactor is evaluated before any files change; test-specific naming conventions run only when touching test files. Which change set is most defensible?
- let the agent immediately edit every matching file; and use a test-file glob in the rules frontmatter
- use plan mode to inspect dependencies and propose steps before execution; and use a test-file glob in the rules frontmatter
- let the agent immediately edit every matching file; and apply the same test naming rule to every source file
- use plan mode to inspect dependencies and propose steps before execution; and apply the same test naming rule to every source file
Explanation
Correct Answer
B:In a nightly code quality job, planning separates investigation and approval from code mutation; alongside this, glob matching limits enforcement guidance to the intended file families, satisfying the distinct requirements that a risky refactor is evaluated before any files change and that test-specific naming conventions run only when touching test files.
Incorrect Answers
A:In a nightly code quality job, glob matching limits enforcement guidance to the intended file families, but early writes can spread an unapproved assumption through the repository; therefore the separate requirement that a risky refactor is evaluated before any files change would remain unsatisfied.
C:In a nightly code quality job, early writes can spread an unapproved assumption through the repository and production files and configuration do not share the test naming context; the design would fail both conditions: a risky refactor is evaluated before any files change and test-specific naming conventions run only when touching test files.
D:In a nightly code quality job, planning separates investigation and approval from code mutation, but production files and configuration do not share the test naming context; therefore the separate requirement that test-specific naming conventions run only when touching test files would remain unsatisfied.
Question 55
The release checklist for a release-triage developer agent explicitly checks that sensitive tool calls are denied before they reach an executor; it also checks that the agent executes tests and examines their exit status. What coordinated solution meets the checklist?
- register a PreToolUse hook that checks the operation against policy; and inspect the test configuration only with Read and claim tests passed
- register a PreToolUse hook that checks the operation against policy; and use Bash with an approved test command
- write a stronger instruction in the system prompt; and inspect the test configuration only with Read and claim tests passed
- write a stronger instruction in the system prompt; and use Bash with an approved test command
Explanation
Correct Answer
B:In a release-triage developer agent, pre-execution interception can reject a disallowed call rather than relying on post-hoc advice; alongside this, the shell can run build tools and return execution feedback that Read cannot supply, satisfying the distinct requirements that sensitive tool calls are denied before they reach an executor and that the agent executes tests and examines their exit status.
Incorrect Answers
A:In a release-triage developer agent, pre-execution interception can reject a disallowed call rather than relying on post-hoc advice, but reading configuration does not execute a test suite; therefore the separate requirement that the agent executes tests and examines their exit status would remain unsatisfied.
C:In a release-triage developer agent, an instruction alone does not provide deterministic execution-time authorization and reading configuration does not execute a test suite; the design would fail both conditions: sensitive tool calls are denied before they reach an executor and the agent executes tests and examines their exit status.
D:In a release-triage developer agent, the shell can run build tools and return execution feedback that Read cannot supply, but an instruction alone does not provide deterministic execution-time authorization; therefore the separate requirement that sensitive tool calls are denied before they reach an executor would remain unsatisfied.
Question 56
Before expanding an automated CI review service, the owner must demonstrate that the team obtains a read-only survey of an unfamiliar module, while also demonstrating that CI failures provide actionable evidence to maintainers. Which design provides both controls?
- investigate in plan-oriented or read-only mode first; and discard the runner logs and keep only the assistant's final sentence
- investigate in plan-oriented or read-only mode first; and persist structured errors, relevant logs, and a bounded diagnostic summary
- run write tools to discover how the module behaves; and discard the runner logs and keep only the assistant's final sentence
- run write tools to discover how the module behaves; and persist structured errors, relevant logs, and a bounded diagnostic summary
Explanation
Correct Answer
B:In an automated CI review service, inspection can establish architecture and change scope without side effects; alongside this, stable artifacts help maintainers reproduce the problem without rerunning a chat, satisfying the distinct requirements that the team obtains a read-only survey of an unfamiliar module and that CI failures provide actionable evidence to maintainers.
Incorrect Answers
A:In an automated CI review service, inspection can establish architecture and change scope without side effects, but the missing trace impedes diagnosis of command or infrastructure failures; therefore the separate requirement that CI failures provide actionable evidence to maintainers would remain unsatisfied.
C:In an automated CI review service, mutation is unnecessary for gaining initial understanding and the missing trace impedes diagnosis of command or infrastructure failures; the design would fail both conditions: the team obtains a read-only survey of an unfamiliar module and CI failures provide actionable evidence to maintainers.
D:In an automated CI review service, stable artifacts help maintainers reproduce the problem without rerunning a chat, but mutation is unnecessary for gaining initial understanding; therefore the separate requirement that the team obtains a read-only survey of an unfamiliar module would remain unsatisfied.
Question 57
An incident postmortem for a dependency upgrade initiative requires controls so that a repeatable code-review procedure is available as a named workflow and that the team obtains a read-only survey of an unfamiliar module. Which paired remedy covers both failure modes?
- retype the entire procedure in every conversation; and investigate in plan-oriented or read-only mode first
- retype the entire procedure in every conversation; and run write tools to discover how the module behaves
- create a documented reusable skill in SKILL.md; and run write tools to discover how the module behaves
- create a documented reusable skill in SKILL.md; and investigate in plan-oriented or read-only mode first
Explanation
Correct Answer
D:In a dependency upgrade initiative, skills package instructions and supporting resources for invocation across similar tasks; alongside this, inspection can establish architecture and change scope without side effects, satisfying the distinct requirements that a repeatable code-review procedure is available as a named workflow and that the team obtains a read-only survey of an unfamiliar module.
Incorrect Answers
A:In a dependency upgrade initiative, inspection can establish architecture and change scope without side effects, but manual repetition produces drift and loses shared revisions; therefore the separate requirement that a repeatable code-review procedure is available as a named workflow would remain unsatisfied.
B:In a dependency upgrade initiative, manual repetition produces drift and loses shared revisions and mutation is unnecessary for gaining initial understanding; the design would fail both conditions: a repeatable code-review procedure is available as a named workflow and the team obtains a read-only survey of an unfamiliar module.
C:In a dependency upgrade initiative, skills package instructions and supporting resources for invocation across similar tasks, but mutation is unnecessary for gaining initial understanding; therefore the separate requirement that the team obtains a read-only survey of an unfamiliar module would remain unsatisfied.
Question 58
An audit of a test failure triage bot identifies two unmet requirements: an unusual null-value case is handled without inventing data and a consequential claim receives independent human sign-off. Select TWO actions that remedy those particular deficiencies.
- accept model confidence as a substitute for authorization
- route high-impact or low-evidence results to an authorized reviewer
- show only records where every field is present
- include a demonstration showing how an absent field is represented
Explanation
Correct Answers
B:For a test failure triage bot, human approval supplies accountability where a mistaken model result has material cost, which directly establishes that a consequential claim receives independent human sign-off; the separate requirement that an unusual null-value case is handled without inventing data is handled by a different action.
D:For a test failure triage bot, explicit missing-value examples clarify that omissions are not permission to hallucinate, which directly establishes that an unusual null-value case is handled without inventing data; the separate requirement that a consequential claim receives independent human sign-off is handled by a different action.
Incorrect Answers
A:For a test failure triage bot, a model score does not confer business or legal authority, so this option cannot establish that a consequential claim receives independent human sign-off; correctly addressing an unusual null-value case is handled without inventing data elsewhere does not repair that deficiency.
C:For a test failure triage bot, the model lacks a model of the required missing-data behavior, so this option cannot establish that an unusual null-value case is handled without inventing data; correctly addressing a consequential claim receives independent human sign-off elsewhere does not repair that deficiency.
Question 59
For an order-status triage agent, acceptance depends on demonstrating that invalid parameters lead to a clear correction rather than a blind retry. Which approach is the strongest fit?
- return a non-retryable validation error with the field and expected constraint
- return an explicit partial status with missing-item identifiers
- hide the invalid field behind a generic internal-error message
- check operation status or use an idempotency key before retry
Explanation
Correct Answer
A:For an order-status triage agent, field-specific feedback lets the caller fix the input before resubmission; this directly establishes that invalid parameters lead to a clear correction rather than a blind retry instead of relying on an unrelated workflow control.
Incorrect Answers
B:Although failure metadata preserves the difference between a sparse valid response and incomplete execution, that only supports the distinct goal that partial data is not presented as a complete successful result, not the requirement that invalid parameters lead to a clear correction rather than a blind retry in an order-status triage agent.
C:In an order-status triage agent, the client lacks the information needed to repair the request; as a result, the stated requirement that invalid parameters lead to a clear correction rather than a blind retry would still not be met.
D:Although a timeout can occur after the server commits the action, so retry safety needs state checks, that only supports the distinct goal that ambiguous write timeouts do not trigger duplicate side effects, not the requirement that invalid parameters lead to a clear correction rather than a blind retry in an order-status triage agent.
Question 60
For an unfamiliar monorepo investigation, a reviewer identifies two independent obligations: client-side tool requests trigger another model turn after execution and every high-impact tool invocation produces an audit event. Which proposal should be approved?
- branch on stop_reason tool_use and execute returned tool_use blocks; and record the tool name, permitted inputs, outcome, and task identifier in a hook
- branch on stop_reason tool_use and execute returned tool_use blocks; and infer tool activity from the final natural-language answer
- end the loop whenever assistant text appears; and record the tool name, permitted inputs, outcome, and task identifier in a hook
- end the loop whenever assistant text appears; and infer tool activity from the final natural-language answer
Explanation
Correct Answer
A:In an unfamiliar monorepo investigation, the response explicitly signals pending client execution and the loop must provide results before another model decision; alongside this, runtime instrumentation observes actual proposed or completed tool activity, satisfying the distinct requirements that client-side tool requests trigger another model turn after execution and that every high-impact tool invocation produces an audit event.
Incorrect Answers
B:In an unfamiliar monorepo investigation, the response explicitly signals pending client execution and the loop must provide results before another model decision, but a summary can omit failed calls or delegated actions; therefore the separate requirement that every high-impact tool invocation produces an audit event would remain unsatisfied.
C:In an unfamiliar monorepo investigation, runtime instrumentation observes actual proposed or completed tool activity, but text may accompany a tool request and is not the completion signal; therefore the separate requirement that client-side tool requests trigger another model turn after execution would remain unsatisfied.
D:In an unfamiliar monorepo investigation, text may accompany a tool request and is not the completion signal and a summary can omit failed calls or delegated actions; the design would fail both conditions: client-side tool requests trigger another model turn after execution and every high-impact tool invocation produces an audit event.