Question 100 An agent displays service status from two tools that can disagree. What response logic is safest? Show source-specific statuses and flag unresolved contradictions for review Combine both statuses into 'all systems normal' Choose whichever status sounds most positive Delete the disagreeing tool result Explanation Correct Answer A:Contradictory systems should not be silently collapsed into false certainty. Incorrect Answers B:An invented consensus hides possible outage information. C:Optimism is not evidence of a correct service state. D:Discarding the conflicting service response erases evidence needed to determine which system holds the authoritative state for that specific user request.
Question 101 An agent topic has to provide an FAQ answer only when retrieval finds adequate evidence. Which fallback is appropriate? Disable all search logs to conceal gaps Make up a citation resembling the company site Return an unrelated general answer confidently Explain missing evidence and route to support rather than fabricate an answer Explanation Correct Answer D:Grounding failure should yield uncertainty or escalation. Incorrect Answers A:Missing logs makes content issues harder to correct. B:Invented provenance does not support an answer. C:Irrelevant content can mislead the user.
Question 102 A topic needs to ask a user whether to proceed with a nonrefundable purchase. Which UI construct fits? A confirmation question or card that collects explicit approval An unvalidated free-form conversation guess Use the agent's high confidence in the itinerary as authorization to purchase A passive message saying the purchase is complete Explanation Correct Answer A:The action should depend on a recorded affirmative response. Incorrect Answers B:Ambiguous language should not authorize spending. C:Confidence cannot substitute for explicit informed user confirmation. D:A message cannot supply consent.
Question 103 A topic receives a string from one tool and needs to compare a numeric threshold. What should it do? Convert and validate the typed value before comparing Skip validation for malformed values Compare text lexicographically as if it were numeric Parse the first digit of the string as the full numeric value Explanation Correct Answer A:Numeric conditions require a well-defined number. Incorrect Answers B:Invalid inputs can produce erroneous decisions. C:String order gives wrong results for values like 9 and 100. D:Truncating a multi-digit amount changes the threshold comparison.
Question 104 A product troubleshooting topic needs to branch on an error code and preserve the code for later logging. Which variable use fits? Discard the code after the first node Ask the model to recall the code without storing it Store the error code in a topic variable and reference it in the condition and log action Hardcode the last test error code in every message Explanation Correct Answer C:A reusable variable links diagnosis and telemetry. Incorrect Answers A:Later reporting would lose key evidence. B:Unstructured recall is less reliable than a variable. D:Test data does not represent the current user's device.
Question 105 An employee asks the same information in two unrelated topics. How can the topic avoid redundant questions when the value was already collected? Pass the existing variable into the called topic's input Store all employee identifiers in public messages Require users to repeat every identifier forever Copy the value into an unrelated global variable without indicating what it represents Explanation Correct Answer A:Variable passing can skip duplicate collection with the same validated value. Incorrect Answers B:Public messages are inappropriate for private identifiers. C:Repeated questions harm experience and increase mistakes. D:An untyped or mislabeled global variable is not a reliable parameter contract for the called topic.
Question 106 A service topic must format a response as a validated JSON object for downstream consumption. What design is best? Request that the model print JSON-shaped prose and skip downstream schema validation Use an explicit structured output schema with validated fields Omit required fields when the model is unsure without signaling errors Ask the agent to append arbitrary text after each JSON value Explanation Correct Answer B:A known contract is safer than prose that merely resembles JSON. Incorrect Answers A:JSON-looking text can still have missing fields or incorrect types. C:Consumers need explicit validation failures. D:Extra text can break machine parsing.
Question 107 A topic calls an HTTP endpoint whose URL changes by environment. How should the endpoint be managed? Use environment configuration or a managed connection rather than literal production URLs in every node Hardcode the same development URL in production Let users enter any backend host URL as a topic parameter for the privileged tool Let the user supply any URL for privileged requests Explanation Correct Answer A:Centralized configuration separates dev and production endpoints. Incorrect Answers B:The agent could call a test backend accidentally. C:Unvalidated destinations can redirect sensitive traffic to unauthorized hosts. D:Unvalidated destinations can cause SSRF or data leaks.
Question 108 A conversational tool returns confidential notes that should only be shown to supervisors. Which response rule matters? Check the supervisor's verified authorization before returning fields Rely on the user not clicking a spoiler Check only the user's department name typed into chat Hide the notes behind a collapsed visual section Explanation Correct Answer A:Output must follow the tool's data access policy. Incorrect Answers B:A social convention is not authorization. C:A self-reported role is not verified authorization for confidential notes. D:Hidden UI still contains sensitive information.
Question 109 An agent must show a standard severity label for several possible incident categories. Which topic design supports predictable mapping? Ask the model to improvise labels at each turn Use conditions or an explicit mapping expression based on validated category values Always label every event Critical Choose a label based on message length Explanation Correct Answer B:Deterministic mapping avoids invented severity categories. Incorrect Answers A:Free-form severity names can vary arbitrarily. C:An unconditional severity is inaccurate. D:Length does not determine incident severity.
Question 110 An HR agent needs a separate prompt for a sensitive policy summary, grounded by current knowledge. Which advanced topic element is appropriate? A broad delete-record tool called before summarization A custom prompt using approved knowledge inputs with explicit output constraints A hardcoded answer regardless of source changes An instruction to omit all uncertainty and citations Explanation Correct Answer B:Constrained prompt plus grounded context can produce a reviewable summary. Incorrect Answers A:Deleting records is unrelated and dangerous. C:Static responses can misstate current policy. D:The user loses evidence and important limitations.