Question 180
An automation needs a refund action but the tool accepts arbitrary monetary values. Where should the solution architect focus the implementation?
- At this decision point, enforce server-side refund bounds, approvals and idempotency within the tool implementation.
- As a candidate control, partition tool sets by task role and grant capability only when relevant to the current workflow.
- For this deployment, remove unused password-reset tooling from the agent's available capabilities.
- In the proposed design, expose a validated status-update operation with server-side field and transition checks.
Explanation
Correct Answer
A:Tool schemas alone do not prevent malicious or erroneous high-impact values without backend validation; a suitable release result is that out-of-policy amounts are rejected regardless of Claude's argument text.
Incorrect Answers
B:Task-scoped exposure limits attack surface and cuts tool-choice confusion in large deployments; the alternative helps if an agent's tool catalog grows as every new team adds a connector to a common pool, instead of when an automation needs a refund action but the tool accepts arbitrary monetary values.
C:Capability removal prevents risky actions at their source instead of depending on model restraint; it is justified if a support agent can reset passwords despite only needing to search troubleshooting articles, not the case in which an automation needs a refund action but the tool accepts arbitrary monetary values.
D:Narrow application endpoints allow the required mutation without granting general table write authority; its scope includes cases where a purchasing assistant requests unrestricted database write access to update one order status field, whereas the problem presented is that an automation needs a refund action but the tool accepts arbitrary monetary values.
Question 181
An architect wants to measure whether a newly added connector increases risk more than value. What revision would make the proposal more defensible?
- During architecture review, remove unused password-reset tooling from the agent's available capabilities.
- In the proposed implementation, consolidate overlapping tools behind distinct, task-centered interfaces and explicit descriptions.
- At this decision point, expose a validated status-update operation with server-side field and transition checks.
- As a candidate control, review invocation frequency, misuse paths, ownership and necessity before retaining the connector.
Explanation
Correct Answer
D:Tool governance needs evidence of a real task requirement and proportionate privilege; its effect can be measured when quarterly inventory shows every retained capability has an owner, scope and justified use case.
Incorrect Answers
A:Capability removal prevents risky actions at their source instead of depending on model restraint; the other case is when a support agent can reset passwords despite only needing to search troubleshooting articles, not the present case where an architect wants to measure whether a newly added connector increases risk more than value.
B:Reducing ambiguous selection improves tool reliability and limits unnecessary decision branches; this makes sense when a research agent has twenty near-identical tools and frequently selects the wrong one, while this problem involves an architect wants to measure whether a newly added connector increases risk more than value.
C:Narrow application endpoints allow the required mutation without granting general table write authority; this technique fits where a purchasing assistant requests unrestricted database write access to update one order status field, and not where an architect wants to measure whether a newly added connector increases risk more than value.
Question 182
A tool-use trace shows password changes triggered by an assistant commissioned only to explain support articles. What outcome must the reviewer see before proceeding?
- The release check confirms that attempted password resets fail because no such executable tool exists for the agent.
- The reviewer observes that unrelated departments' connectors are absent from each worker's visible inventory.
- The release report verifies that adversarial prompts cannot execute retired tools and remaining calls retain audit traces.
- The controlled replay verifies that quarterly inventory shows every retained capability has an owner, scope and justified use case.
Explanation
Correct Answer
A:Capability removal prevents risky actions at their source instead of depending on model restraint; the reviewer is concerned with situations where a support agent can reset passwords despite only needing to search troubleshooting articles.
Incorrect Answers
B:Task-scoped exposure limits attack surface and cuts tool-choice confusion in large deployments; the other workload is one where an agent's tool catalog grows as every new team adds a connector to a common pool, whereas the release criterion addresses a tool-use trace shows password changes triggered by an assistant commissioned only to explain support articles.
C:Observability records misuse but cannot itself prevent actions made possible by excessive permissions; this applies when a pilot shows unexpected tool invocations but its team proposes logging only, while the required safeguard concerns a tool-use trace shows password changes triggered by an assistant commissioned only to explain support articles.
D:Tool governance needs evidence of a real task requirement and proportionate privilege; the alternative helps if an architect wants to measure whether a newly added connector increases risk more than value, not the risk where a tool-use trace shows password changes triggered by an assistant commissioned only to explain support articles.
Question 183
Agent selection errors increase after the team adds seven nearly synonymous retrieval functions. Which result should the pilot review prioritize?
- The regression suite verifies that tests reject updates to price, recipient or invalid order states while allowing approved transitions.
- A production-like test shows that quarterly inventory shows every retained capability has an owner, scope and justified use case.
- The release check confirms that adversarial prompts cannot execute retired tools and remaining calls retain audit traces.
- The reviewer observes that tool-selection tests favor the correct interface across similar natural-language tasks.
Explanation
Correct Answer
D:Reducing ambiguous selection improves tool reliability and limits unnecessary decision branches; the control should hold even if a research agent has twenty near-identical tools and frequently selects the wrong one.
Incorrect Answers
A:Narrow application endpoints allow the required mutation without granting general table write authority; it is relevant if a purchasing assistant requests unrestricted database write access to update one order status field, instead of when agent selection errors increase after the team adds seven nearly synonymous retrieval functions.
B:Tool governance needs evidence of a real task requirement and proportionate privilege; it serves conditions where an architect wants to measure whether a newly added connector increases risk more than value, not the case in which agent selection errors increase after the team adds seven nearly synonymous retrieval functions.
C:Observability records misuse but cannot itself prevent actions made possible by excessive permissions; the other case is when a pilot shows unexpected tool invocations but its team proposes logging only, whereas the problem presented is that agent selection errors increase after the team adds seven nearly synonymous retrieval functions.
Question 184
A purchase-status update runs through a database credential able to alter unrelated orders and columns. What result would directly validate the necessary safeguard?
- The end-to-end check confirms that quarterly inventory shows every retained capability has an owner, scope and justified use case.
- The observed outcome is that tests reject updates to price, recipient or invalid order states while allowing approved transitions.
- The regression suite verifies that adversarial prompts cannot execute retired tools and remaining calls retain audit traces.
- A production-like test shows that out-of-policy amounts are rejected regardless of Claude's argument text.
- The release check confirms that unrelated departments' connectors are absent from each worker's visible inventory.
Explanation
Correct Answer
B:Narrow application endpoints allow the required mutation without granting general table write authority; it is the decisive test when a purchasing assistant requests unrestricted database write access to update one order status field.
Incorrect Answers
A:Tool governance needs evidence of a real task requirement and proportionate privilege; that would be useful if an architect wants to measure whether a newly added connector increases risk more than value, while this problem involves a purchase-status update runs through a database credential able to alter unrelated orders and columns.
C:Observability records misuse but cannot itself prevent actions made possible by excessive permissions; that is defensible if a pilot shows unexpected tool invocations but its team proposes logging only, rather than the risk that a purchase-status update runs through a database credential able to alter unrelated orders and columns.
D:Tool schemas alone do not prevent malicious or erroneous high-impact values without backend validation; the other workload is one where an automation needs a refund action but the tool accepts arbitrary monetary values, rather than the condition that a purchase-status update runs through a database credential able to alter unrelated orders and columns.
E:Task-scoped exposure limits attack surface and cuts tool-choice confusion in large deployments; this applies when an agent's tool catalog grows as every new team adds a connector to a common pool, while the immediate question involves a purchase-status update runs through a database credential able to alter unrelated orders and columns.
Question 185
An enterprise tool catalog is expanding, but logs merely record prohibited invocations without preventing them, and no owner has reviewed whether each exposed connector is still necessary. Which TWO verifiable results would show both enforced access control and disciplined tool inventory? Choose TWO.
- The measured evidence confirms that unrelated departments' connectors are absent from each worker's visible inventory.
- The scenario test demonstrates that out-of-policy amounts are rejected regardless of Claude's argument text.
- The end-to-end check confirms that adversarial prompts cannot execute retired tools and remaining calls retain audit traces.
- The observed outcome is that quarterly inventory shows every retained capability has an owner, scope and justified use case.
- The regression suite verifies that tool-selection tests favor the correct interface across similar natural-language tasks.
Explanation
Correct Answers
C:Observability records misuse but cannot itself prevent actions made possible by excessive permissions; an acceptance test needs to cover cases where a pilot shows unexpected tool invocations but its team proposes logging only; another obligation arises when an architect wants to measure whether a newly added connector increases risk more than value.
D:Tool governance needs evidence of a real task requirement and proportionate privilege; the condition under test is that an architect wants to measure whether a newly added connector increases risk more than value; another obligation arises when a pilot shows unexpected tool invocations but its team proposes logging only.
Incorrect Answers
A:Task-scoped exposure limits attack surface and cuts tool-choice confusion in large deployments; the approach helps when an agent's tool catalog grows as every new team adds a connector to a common pool, whereas the release criterion addresses a proposed tool expansion shows improved task completion but gives no evidence about induced permission exposure; another obligation arises when a pilot shows unexpected tool invocations but its team proposes logging only.
B:Tool schemas alone do not prevent malicious or erroneous high-impact values without backend validation; that belongs to situations where an automation needs a refund action but the tool accepts arbitrary monetary values, while the required safeguard concerns a safety review receives a record of unauthorized tool attempts but no prevention at the backend boundary; another obligation arises when an architect wants to measure whether a newly added connector increases risk more than value.
E:Reducing ambiguous selection improves tool reliability and limits unnecessary decision branches; the other case is when a research agent has twenty near-identical tools and frequently selects the wrong one, whereas this scenario involves a proposed tool expansion shows improved task completion but gives no evidence about induced permission exposure; another obligation arises when a pilot shows unexpected tool invocations but its team proposes logging only.
Question 186
An integration owner cannot explain why a new connector is visible to every existing agent role. Which result proves the requested control, not just configuration?
- The negative-case evaluation confirms that attempted password resets fail because no such executable tool exists for the agent.
- A controlled trial finds that out-of-policy amounts are rejected regardless of Claude's argument text.
- The measured evidence confirms that quarterly inventory shows every retained capability has an owner, scope and justified use case.
- The scenario test demonstrates that unrelated departments' connectors are absent from each worker's visible inventory.
Explanation
Correct Answer
D:Task-scoped exposure limits attack surface and cuts tool-choice confusion in large deployments; the observed protection matters when an agent's tool catalog grows as every new team adds a connector to a common pool.
Incorrect Answers
A:Capability removal prevents risky actions at their source instead of depending on model restraint; that fits cases where a support agent can reset passwords despite only needing to search troubleshooting articles, not the case in which an integration owner cannot explain why a new connector is visible to every existing agent role.
B:Tool schemas alone do not prevent malicious or erroneous high-impact values without backend validation; this control is needed when an automation needs a refund action but the tool accepts arbitrary monetary values, whereas the problem presented is that an integration owner cannot explain why a new connector is visible to every existing agent role.
C:Tool governance needs evidence of a real task requirement and proportionate privilege; that would be useful if an architect wants to measure whether a newly added connector increases risk more than value, but this decision concerns an integration owner cannot explain why a new connector is visible to every existing agent role.
Question 187
An assistant sees 300 enterprise tool schemas although most requests use only three. How should the architect revise the proposed solution?
- For the affected service, add semantic capability descriptions with constrained discovery and an explicit selection step.
- In the project scope, load the reporting procedure when intent matches instead of putting it in every system prompt.
- Within the rollout design, provide a compact tool index and reveal full schemas only after relevant selection.
- For the next architecture revision, retrieve product-specific instructions on demand after resolving the user's actual target.
Explanation
Correct Answer
C:Progressive disclosure reduces wasted context and distractors while preserving access to needed capabilities; a pilot is successful when simple requests use a small schema budget and still find the correct specialist tool.
Incorrect Answers
A:Well-described discovery entries help choose the correct integration without dumping every parameter; its scope includes cases where a catalog search must choose among many connectors whose names are difficult to distinguish, and not where an assistant sees 300 enterprise tool schemas although most requests use only three.
B:Conditional instruction loading improves focus for routine tasks without losing specialized capability; it targets cases where an assistant needs a rarely used financial-reporting skill only for month-end requests, rather than the risk that an assistant sees 300 enterprise tool schemas although most requests use only three.
D:Demand-driven context limits irrelevant guidance and prevents stale instructions dominating the prompt; that belongs to situations where a research agent loads entire vendor manuals before knowing which product the user selected, while the immediate question involves an assistant sees 300 enterprise tool schemas although most requests use only three.
Question 188
A research agent loads entire vendor manuals before knowing which product the user selected. Which control should govern the first implementation?
- For the primary workstream, keep discovery concise but expand precise argument schemas before the selected tool call.
- For this workflow, load the reporting procedure when intent matches instead of putting it in every system prompt.
- For the affected service, provide a compact tool index and reveal full schemas only after relevant selection.
- In the project scope, filter discoverable capabilities by current role before exposing details to the model.
- Within the rollout design, retrieve product-specific instructions on demand after resolving the user's actual target.
Explanation
Correct Answer
E:Demand-driven context limits irrelevant guidance and prevents stale instructions dominating the prompt; a suitable release result is that a product change causes the next retrieval to load only the newly relevant documentation.
Incorrect Answers
A:Progressive disclosure is not the removal of validation contracts required for correct execution; the trigger would be that a team compresses all tool schemas into vague one-line descriptions and invocation accuracy falls, while the required safeguard concerns a research agent loads entire vendor manuals before knowing which product the user selected.
B:Conditional instruction loading improves focus for routine tasks without losing specialized capability; the design helps when an assistant needs a rarely used financial-reporting skill only for month-end requests, not the risk where a research agent loads entire vendor manuals before knowing which product the user selected.
C:Progressive disclosure reduces wasted context and distractors while preserving access to needed capabilities; that fits cases where an assistant sees 300 enterprise tool schemas although most requests use only three, not this situation where a research agent loads entire vendor manuals before knowing which product the user selected.
D:Discovery itself must respect authorization; hiding a schema cannot replace backend checks; this control is needed when a tool's full API spec is stable but its permissions change by user role, whereas this scenario involves a research agent loads entire vendor manuals before knowing which product the user selected.
Question 189
A catalog search must choose among many connectors whose names are difficult to distinguish. What intervention best matches the stated boundary?
- Within the delivery plan, persist conclusions and retrieve earlier raw outputs only when the next step needs them.
- At the system boundary, load the reporting procedure when intent matches instead of putting it in every system prompt.
- For the primary workstream, add semantic capability descriptions with constrained discovery and an explicit selection step.
- For this workflow, provide a compact tool index and reveal full schemas only after relevant selection.
Explanation
Correct Answer
C:Well-described discovery entries help choose the correct integration without dumping every parameter; a representative trial should prove that ambiguous requests elicit clarification or a bounded candidate list rather than random tool use.
Incorrect Answers
A:Lazy retrieval keeps active context relevant instead of appending all historical observations; the alternative helps if a multistep agent keeps obsolete tool outputs and bloats the reasoning context, whereas the problem presented is that a catalog search must choose among many connectors whose names are difficult to distinguish.
B:Conditional instruction loading improves focus for routine tasks without losing specialized capability; it is justified if an assistant needs a rarely used financial-reporting skill only for month-end requests, but this decision concerns a catalog search must choose among many connectors whose names are difficult to distinguish.
D:Progressive disclosure reduces wasted context and distractors while preserving access to needed capabilities; it targets cases where an assistant sees 300 enterprise tool schemas although most requests use only three, but the requested boundary concerns a catalog search must choose among many connectors whose names are difficult to distinguish.
Question 190
A tool's full API spec is stable but its permissions change by user role. Which implementation decision is the strongest fit?
- For this production path, retrieve product-specific instructions on demand after resolving the user's actual target.
- For the intended users, add semantic capability descriptions with constrained discovery and an explicit selection step.
- Within the delivery plan, load the reporting procedure when intent matches instead of putting it in every system prompt.
- At the system boundary, filter discoverable capabilities by current role before exposing details to the model.
Explanation
Correct Answer
D:Discovery itself must respect authorization; hiding a schema cannot replace backend checks; the implementation should produce evidence that lower-privilege sessions neither discover nor invoke restricted functions.
Incorrect Answers
A:Demand-driven context limits irrelevant guidance and prevents stale instructions dominating the prompt; this makes sense when a research agent loads entire vendor manuals before knowing which product the user selected, rather than the risk that a tool's full API spec is stable but its permissions change by user role.
B:Well-described discovery entries help choose the correct integration without dumping every parameter; this technique fits where a catalog search must choose among many connectors whose names are difficult to distinguish, rather than the condition that a tool's full API spec is stable but its permissions change by user role.
C:Conditional instruction loading improves focus for routine tasks without losing specialized capability; the trigger would be that an assistant needs a rarely used financial-reporting skill only for month-end requests, while the immediate question involves a tool's full API spec is stable but its permissions change by user role.