Question 120 A customer support case is resumed after three days, but the last known order status was “in transit.” What should the resumed agent do? Delete the verified customer ID and guess the owner from the address. Refresh the mutable order status before making a current delivery claim. Re-run the entire onboarding workflow before reading the order. Report the old status as current because the session resumed successfully. Explanation Correct Answer B:Persisted context may contain stale external facts even when identities and preferences remain valid. Incorrect Answers A:Identity continuity is useful and should not be discarded without reason. C:Only the time-sensitive order state needs refreshing. D:Resumption does not freeze an external order system.
Question 121 An automated security assistant sees ambiguous evidence of intrusion but is configured to isolate endpoints only on high-confidence alerts. What should it do? Classify the alert as definitely benign because evidence is incomplete. Isolate all endpoints regardless of evidence to be safe. Follow the configured threshold and escalate the uncertain case for analyst triage without claiming confirmed compromise. Silently lower the high-confidence threshold to gain autonomy. Explanation Correct Answer C:Confidence and delegated authority bound autonomous response. Incorrect Answers A:Uncertainty is not absence of risk. B:That can cause unnecessary business disruption. D:Changing policy is not part of the delegated task.
Question 122 A second pass suggests that a correct mitigation is “too generic” without citing a conflicting requirement. What is the best treatment? Ignore all future second-pass reviews. Reject the whole work solely because the critic disliked phrasing. Tell the reviewer all later versions must copy its exact wording. Check whether it meets the original acceptance criteria and record the suggestion as non-blocking absent a concrete defect. Explanation Correct Answer D:Review feedback needs evidence and materiality, not an automatic veto. Incorrect Answers A:A weak suggestion does not invalidate the review function. B:Subjective objections are not technical failures. C:That creates stylistic rigidity without stronger correctness.
Question 123 A session started in a nested folder is missing relevant conventions for a sibling package. What should be checked? Only the model’s response temperature. The system clipboard contents. Whether the project has enough screenshots. Which CLAUDE.md and scoped rule files are loaded for the working directory and referenced paths. Explanation Correct Answer D:Instruction visibility depends on project and directory scoping. Incorrect Answers A:Sampling does not change which guidance files are in scope. B:Clipboard state does not determine memory hierarchy. C:Screenshots are not the source of coding conventions.
Question 124 A legacy repository has unknown subsystems and sparse documentation. The team wants a test plan. What should happen first? Generate all tests before reading a single file. Write test plans based only on the repository size. Assign tests evenly to files by alphabetical order. Map the repository structure and call relationships, then plan targeted test work. Explanation Correct Answer D:A dynamic plan needs a discovered architecture to identify meaningful coverage. Incorrect Answers A:Premature generation is likely to miss actual components and interfaces. B:Size is not a substitute for architecture information. C:File names do not reveal dependency or risk structure.
Question 125 An agent can write to a ticketing system only when the user explicitly asks to create a ticket. What selection guard prevents accidental writes? Skip user-intent checks because the tool schema is valid. Route informational requests to read tools and require action-specific confirmation for ticket creation. Call create_ticket whenever a relevant keyword appears. Make ticket creation a side effect of search_ticket. Explanation Correct Answer B:A user’s question is not automatically permission to create persistent records. Incorrect Answers A:Argument validation does not authorize a mutation. C:Keywords do not establish user intent or approval. D:Hidden writes create unexpected records.
Question 126 A reviewer is asked to show both a high-confidence security bug and a harmless lookalike case. Which example pair is best? A real source-to-sink bypass beside an almost identical safely validated flow. Two unrelated security bugs with different languages. Two identical inputs with opposite labels and no explanation. A list of vulnerability names without case evidence. Explanation Correct Answer A:A close contrast defines what evidence distinguishes a true vulnerability. Incorrect Answers B:Those do not illustrate the false-positive boundary. C:Contradictory labels confuse the decision criterion. D:Taxonomy alone does not demonstrate how to adjudicate similar code.
Question 127 A reviewer notes that an assistant’s patch works but introduces an unused dependency. What should the next iteration do? Rebuild the entire product in another language. Disable dependency scanning to avoid warnings. Remove the unnecessary dependency and verify the application still builds. Keep every dependency in case it is useful someday. Explanation Correct Answer C:Refinement can improve maintainability without changing the working feature. Incorrect Answers A:That is disproportionate to an unused package. B:Removing checks hides potential risk. D:Unneeded packages add maintenance and security surface.
Question 128 A retrieval worker cannot access a source and returns an empty list. The synthesizer concludes there are no relevant records. What contract correction is needed? Suppress worker errors to keep the final answer short. Have the synthesizer invent plausible retrieved records. Always interpret empty arrays as exhaustive proof of absence. Distinguish retrieval failure from a successful zero-result search in a typed worker response. Explanation Correct Answer D:Downstream agents need to know whether absence is evidence or a failed lookup. Incorrect Answers A:That makes unsupported conclusions likely. B:Fabrication conceals the failure. C:A failed query may also yield no content.
Question 129 A handoff from translation to summarization fails whenever a document has no detected language. What is the best exception path? Set an arbitrary source language without recording the guess. Return a typed unknown-language state and route to detection retry or human review. Produce the summary from untranslated text by default. Delete the document and mark the pipeline complete. Explanation Correct Answer B:The downstream step should receive an explicit status rather than pretending translation succeeded. Incorrect Answers A:A wrong language choice can corrupt later analysis. C:The workflow promised translation and may not support the unknown language. D:Destructive fallback loses work and conceals an unresolved exception.
Question 130 A deployment agent can run tests and propose infrastructure edits, but must not secretly publish a release. Which TWO tool controls are appropriate? Restrict available write tools to the authorized workspace and environment. Provide every production deployment credential in the prompt. Treat any passing unit test as unconditional deploy consent. Hide the tool actions from operators to avoid interruptions. Require explicit release approval before exposing or executing the deploy capability. Explanation Correct Answers A:Tool scoping limits the side effects the agent can attempt. E:A separate authorization gate protects production publication. Incorrect Answers B:Prompt access to secrets bypasses controlled tool boundaries. C:Passing tests are not a release authorization. D:Lack of visibility undermines auditability and oversight.
