Question 70
An agent runs a bounded support workflow with multiple optional tool calls. A tool-driven conversation keeps calling the same read-only endpoint without progress. Which mechanism should be selected?
- Check cancellation state before dispatching queued tool work
- Return a bounded handoff or clarification rather than forcing more autonomous turns
- Use rate-aware concurrency limits and jittered backoff
- Log authorized tool names, request IDs, outcomes, and decision context
- Enforce a maximum tool-call or iteration budget and a fallback outcome
Explanation
Correct Answer
E:Without a host-side bound, repeatedly proposing the same action can consume resources without resolving the customer's issue.
Incorrect Answers
A:A cancellation flag in application state can block new operations; telling the model to stop is not sufficient for queued execution; it protects against the case where a user cancels a request while a queued tool operation has not started, rather than the failure described here: a tool-driven conversation keeps calling the same read-only endpoint without progress.
B:This choice is relevant when a response contains no tool calls but the task is still unresolved and requires human judgment; when prerequisites are missing, another iteration does not create evidence; explicit unresolved state protects correctness and enables review. Here, however, a tool-driven conversation keeps calling the same read-only endpoint without progress.
C:It would be appropriate when a backend becomes rate limited while many agents share it; application-side scheduling controls aggregate demand; independent immediate retries can amplify throttling across sessions. Here, a tool-driven conversation keeps calling the same read-only endpoint without progress, so a different mechanism is required.
D:Structured execution logs establish which external action occurred and on what evidence without assuming that model prose is a reliable audit record; the alternative problem occurs when support needs a record of why an agent issued each sensitive call. The present scenario instead requires the system to stop a nonproductive tool loop.
Question 71
A review agent must obey application-level guarantees under adversarial prompts. A prompt injection asks the assistant to call a privileged export tool. What should be implemented at the appropriate boundary?
- Check a central rate or quota counter at execution time
- Gate export execution on server-side identity and scope validation
- Apply a confirmation checkpoint enforced by the host workflow
- Validate target identifier and cardinality in tool arguments before execution
Explanation
Correct Answer
B:The permission decision must not be delegated to untrusted text; the service should reject unauthorized exports even if requested by a model.
Incorrect Answers
A:This choice is relevant when a tool operation must obey a daily account limit; per-prompt reminders cannot coordinate usage across multiple agents; an authoritative service counter can enforce shared limits. Here, however, a prompt injection asks the assistant to call a privileged export tool.
C:A required approval checkpoint converts a human-control requirement into execution logic, rather than merely advising the model to ask first; it protects against the case where a user requests deletion, but the policy requires independent confirmation, rather than the failure described here: a prompt injection asks the assistant to call a privileged export tool.
D:It would be appropriate when a tool can modify thousands of records although the task requires one record; target and batch-size validation blocks unintended bulk changes even when a model builds syntactically valid arguments. Here, a prompt injection asks the assistant to call a privileged export tool, so a different mechanism is required.
Question 72
A developer uses Claude Code to change a risky authentication module. The engineer wants a proposed design and affected-file list before any modifications. Which option best meets the operational need?
- Perform the authorized edits and run the specified verification steps
- Present the plan and wait for authorization before enabling write operations
- Use plan mode to inspect the codebase and propose steps without applying edits
- Use read-only search and inspection to map the responsible files
Explanation
Correct Answer
C:Read-only planning supports review of dependencies and risk before the team authorizes source changes. For the stated case, the check is whether the system can choose planning before execution; this approach supplies that control.
Incorrect Answers
A:This choice is relevant when the approved implementation has already been planned and reviewed; once scope and approval are established, indefinite re-planning adds little value; controlled execution produces testable results. Here, however, the engineer wants a proposed design and affected-file list before any modifications.
B:A design review requirement is not satisfied merely by generating a plan if the next stage can still commit changes autonomously; the alternative problem occurs when a proposed refactor requires an explicit human go/no-go. The present scenario instead requires the system to choose planning before execution.
D:It would be appropriate when the team does not know where authorization is implemented; an exploratory pass establishes architecture evidence; jumping into edits risks changing the wrong component. Here, the engineer wants a proposed design and affected-file list before any modifications, so a different mechanism is required.
Question 73
A code assistant integrates MCP tools maintained by different engineering groups. One tool's description encourages bypassing repository review rules. What is the most appropriate design choice?
- Version breaking schema changes or provide a backwards-compatible migration
- Ignore instructions in tool metadata that conflict with higher-priority security policy
- Return an empty success result for legitimate no-match outcomes
- Use an enumerated argument value set and validate it on the server
Explanation
Correct Answer
B:Descriptions are useful for discovery, not an authority source; a malicious tool can attempt prompt injection through its advertised text.
Incorrect Answers
A:This choice is relevant when an internal tool is versioned while several clients depend on its outputs; unexpected field renames can break tool consumers and downstream validators even when a tool's description still sounds correct. Here, however, one tool's description encourages bypassing repository review rules.
C:It would be appropriate when a build lookup returns no match and callers mistake it for a crashed server; no matching build is a valid result distinct from an operational error; mixing them causes unnecessary retries and escalation. Here, one tool's description encourages bypassing repository review rules, so a different mechanism is required.
D:Schema constraints and server checks prevent an unintended target resulting from free-form environment strings; it protects against the case where a deployment tool accepts an environment name with many misspelled variants, rather than the failure described here: one tool's description encourages bypassing repository review rules.
Question 74
A research coordinator must synthesize facts from many long reports. A key statistic is buried between hundreds of unrelated facts. Which response is most reliable?
- Use effective dates and revision history to resolve precedence
- Build a short evidence ledger with the relevant claim, number, source, and date
- Retain a canonical source reference and the needed excerpts once
- Record each definition alongside its source and compare before synthesis
Explanation
Correct Answer
B:A compact ledger brings decisive facts back into active context instead of hoping the writer finds them inside long retrieved text.
Incorrect Answers
A:Time-sensitive policy questions require temporal reasoning; simple keyword retrieval can privilege an obsolete statement; the alternative problem occurs when an older report says a rule is active, but a later update supersedes it. The present scenario instead requires the system to surface critical evidence before final synthesis.
C:It would be appropriate when the prompt is filled with repeated copies of the same source extract; repeated text wastes context and can distort apparent evidentiary weight without adding new information. Here, a key statistic is buried between hundreds of unrelated facts, so a different mechanism is required.
D:Without definition tracking, a long summary can accidentally merge incompatible metrics into one confident statement; it protects against the case where two sources use the same term with incompatible definitions, rather than the failure described here: a key statistic is buried between hundreds of unrelated facts.
Question 75
A codebase team decides how Claude Code should apply reusable guidance. Developers need to run an audit checklist on demand with a target service. What should the team do to resolve this particular gap?
- Build a Skill containing an overview and supporting reference files
- Store the shortcut in user commands rather than the project
- Implement a hook or CI gate for the required action
- State the invariant in CLAUDE.md and enforce critical parts in tests
- Create a named project command that accepts the service target
Explanation
Correct Answer
E:An on-demand workflow supports deliberate audits without making every conversation execute the checklist. This matches the workflow goal to offer an explicit reusable invocation rather than introducing an unrelated action.
Incorrect Answers
A:A Skill can organize larger task-specific knowledge while keeping routine sessions from carrying all of it; it protects against the case where a specialist procedure involves a folder of reference templates and examples, rather than the failure described here: developers need to run an audit checklist on demand with a target service.
B:Personal commands avoid cluttering shared team workflows or surprising other contributors; the alternative problem occurs when one engineer wants a shortcut unrelated to the team's shared work. The present scenario instead requires the system to offer an explicit reusable invocation.
C:This choice is relevant when a workflow must run even when Claude chooses not to call a named command; prompt or command availability does not guarantee execution; event-based or pipeline enforcement closes that gap. Here, however, developers need to run an audit checklist on demand with a target service.
D:It would be appropriate when an invariant requirement applies to all code edits regardless of command selection; an optional command is not a suitable place for universal code review obligations; the invariant must be visible independently. Here, developers need to run an audit checklist on demand with a target service, so a different mechanism is required.
Question 76
A repository team decides how to package MCP integrations for collaboration. A project-wide tool server has been approved for every contributor to that repository. Which control is most directly relevant?
- Inspect the effective MCP configuration and resolve duplicate definitions
- List approved tools and schemas before requesting tool calls
- Use a local-only MCP definition rather than committing it to shared project settings
- Keep the project-scoped server definition under version control without secrets
Explanation
Correct Answer
D:Versioned project configuration gives contributors consistent definitions while leaving credentials outside shared files.
Incorrect Answers
A:It would be appropriate when the same server appears in multiple configuration layers with conflicting commands; MCP discovery depends on active connection configuration; changing model instructions cannot fix an incorrect server command. Here, a project-wide tool server has been approved for every contributor to that repository, so a different mechanism is required.
B:This choice is relevant when a team wants to discover available server tools without invoking any of them; discovery can inform architecture and permission review while avoiding side effects associated with actual operation execution. Here, however, a project-wide tool server has been approved for every contributor to that repository.
C:Local scope prevents unreviewed capabilities from silently appearing in every contributor's workspace; it protects against the case where a developer tests an experimental local server that colleagues must not use, rather than the failure described here: a project-wide tool server has been approved for every contributor to that repository.
Question 77
A team uses Claude Code to implement a business-rule change. A ticket uses the word 'active' without defining qualifying account states. How should the workflow be adjusted?
- Remove incidental changes and present a focused diff
- Run the reproducer, inspect the cause, then apply a minimal change
- Request the exact states and exception cases before coding
- Validate the changed module and affected integration boundaries
Explanation
Correct Answer
C:An undefined business term can map to several implementations; clarification prevents a seemingly plausible but wrong patch.
Incorrect Answers
A:A small diff improves reviewability and makes unintended consequences easier to detect; the alternative problem occurs when the generated patch contains unrelated formatting changes. The present scenario instead requires the system to interview the stakeholder about ambiguous criteria.
B:A known failure gives objective feedback; starting with a broad rewrite increases unrelated regression risk; it protects against the case where the team has a known bug with a reproducible failing test, rather than the failure described here: a ticket uses the word 'active' without defining qualifying account states.
D:It would be appropriate when the fix passes the reproducer but may affect adjacent modules; a local green test is not enough when shared interfaces are modified; neighboring consumers need coverage. Here, a ticket uses the word 'active' without defining qualifying account states, so a different mechanism is required.
Question 78
An MCP service encounters errors while fetching customer records. A backend returns HTTP 429 with a retry window. Which implementation decision is the best fit?
- Classify the issue as transient unavailable and provide a safe diagnostic identifier
- Return a structured rate-limit error and retry-after guidance
- Return a permission-denied outcome and require an authorized path
- Return a distinct not-found status without recommending blind retries
Explanation
Correct Answer
B:Throttling is transient and should be handled with bounded delay, not treated as an absent customer record. This matches the workflow goal to signal temporary throttling with actionable metadata rather than introducing an unrelated action.
Incorrect Answers
A:It would be appropriate when a database loses connectivity during a lookup; clear error categorization lets orchestration retry selectively while keeping internal stack traces out of model-facing content. Here, a backend returns HTTP 429 with a retry window, so a different mechanism is required.
C:A permission error cannot be fixed by repeating the same call; it needs identity correction or escalation without leaking account details; the alternative problem occurs when a tool call is rejected because the caller lacks access to the account. The present scenario instead requires the system to signal temporary throttling with actionable metadata.
D:A genuinely missing resource is different from transient infrastructure failure; retrying unchanged identifiers adds load without resolving it; it protects against the case where a requested account number is well formed but does not exist, rather than the failure described here: a backend returns HTTP 429 with a retry window.
Question 79
A research publishing workflow uses confidence-aware human oversight. The source evidence is contradictory and would change an important recommendation. What should the architect specify in the design?
- Apply the checked fix without adding a human approval queue
- Flag both claims and their evidence for accountable adjudication
- Use observed error rates or rubric-based review rather than the unsupported percentage
- Require a verifiable source or explicit review of the unsupported claim
Explanation
Correct Answer
B:Conflicting high-impact evidence warrants review even if a model prefers one narrative with high confidence. Here it serves the explicit need to use human review for consequential uncertainty without adding a separate unverified inference.
Incorrect Answers
A:Review resources can focus on uncertain decisions while deterministic cosmetic changes proceed under normal controls; it protects against the case where a low-risk spelling correction is supported by deterministic lint feedback, rather than the failure described here: the source evidence is contradictory and would change an important recommendation.
C:It would be appropriate when a model reports ninety-nine percent confidence without a calibrated evaluation process; language-model confidence expressions are not calibrated statistical guarantees unless validated against representative outcomes. Here, the source evidence is contradictory and would change an important recommendation, so a different mechanism is required.
D:Publication quality depends on accessible supporting material, not merely on a plausible citation string; the alternative problem occurs when an automated report cites a source that cannot be retrieved by the reviewer. The present scenario instead requires the system to use human review for consequential uncertainty.
Question 80
A support-orchestration service decides how to pass tasks between specialists. A customer intent is classified, but no worker has been chosen. Which change most directly addresses this condition?
- Associate delegated jobs with stable identifiers and recoverable progress
- Require both verified prerequisite states before dispatching the dependent worker
- Select a named specialist based on intent and permission boundaries
- Validate required fields before invoking the specialist
- Treat the unrecognized status as invalid and request clarification or manual review
Explanation
Correct Answer
C:An explicit dispatch decision links a validated customer need to a worker with relevant expertise and tool access. For the stated case, the check is whether the system can make task routing explicit; this approach supplies that control.
Incorrect Answers
A:This choice is relevant when the coordinator wants to know whether a stalled task can be resumed; a durable job identity connects resumes to prior work and avoids launching duplicate specialists blindly. Here, however, a customer intent is classified, but no worker has been chosen.
B:A conjunction of prerequisites prevents the next stage from treating one successful check as equivalent to full authorization; it protects against the case where a task needs both a billing result and an identity confirmation, rather than the failure described here: a customer intent is classified, but no worker has been chosen.
D:It would be appropriate when a specialist rejects an input because it lacks a case identifier; input checks at the coordinator boundary reduce wasted agent turns and expose missing fields early. Here, a customer intent is classified, but no worker has been chosen, so a different mechanism is required.
E:A closed status enumeration prevents unfamiliar values from accidentally being treated as success; the alternative problem occurs when a delegated task completes but its result schema has an unknown status. The present scenario instead requires the system to make task routing explicit.