Anthropic CCAR-P (Claude Certified Architect - Professional) Practice Test 13

Question 120

A multi-tenant application uses one shared prompt that mixes each customer's secret taxonomy. Which change most directly satisfies this requirement?

  1. Keep tenant-specific confidential material isolated and cache only appropriately scoped content.
  2. Version template contracts and test consuming services before releasing the shared update.
  3. Place stable material before volatile request-specific content and observe cache hit rates.
  4. Keep the prompt simple unless instrumentation shows meaningful cache reuse savings.

Explanation

Correct Answer

A:

Prompt reuse must respect tenant data boundaries; efficiency does not authorize cross-customer disclosure. Version the reusable artifact and instrument tenant-safe prompt reuse; reusing content does not automatically grant the right to tenant-safe prompt reuse.

Incorrect Answers

B:

That is primarily a technique for safe changes to shared prompt modules. It lacks the explicit measure for tenant-safe prompt reuse that is necessary here, namely a system that will keep tenant-specific confidential material isolated and cache only appropriately scoped content.

C:

Improving stable prefix ordering for cacheability does not imply tenant-safe prompt reuse. The selected design must keep tenant-specific confidential material isolated and cache only appropriately scoped content, whereas this choice omits that direct protection or verification step.

D:

This recommendation targets cost-effective reuse choice at a different point in the workflow. It leaves tenant-safe prompt reuse outside the design contract and does not require the team to keep tenant-specific confidential material isolated and cache only appropriately scoped content.

Question 121

A procurement assistant needs to read purchase orders but its shared tool token also authorizes refunds. Which solution best resolves this operational gap?

  1. Issue a read-scoped credential and remove refund operations from that agent's callable tool set.
  2. Use tenant-scoped short-lived credentials and enforce per-request tenant identity.
  3. Apply destination allowlists, recipient verification and outbound-send authorization.
  4. Expose parameterized, read-only domain queries through a narrow service interface.

Explanation

Correct Answer

A:

Service-side authorization, not prompt discipline, limits what compromised or mistaken tool calls can do. Verify read-only tool authority under the actual connector identity; the backend must actually enforce read-only tool authority, regardless of instructions.

Incorrect Answers

B:

The control would be justified if per-tenant credential isolation were the limiting factor. In this scenario the decisive target is read-only tool authority; success requires an architecture that will issue a read-scoped credential and remove refund operations from that agent's callable tool set.

C:

The option belongs to the broader work on restricted outbound destinations, but it cannot by itself deliver read-only tool authority. A decision on this case needs the team to issue a read-scoped credential and remove refund operations from that agent's callable tool set and verify that outcome.

D:

This shifts attention toward bounded database query capabilities when the relevant decision is about read-only tool authority. Unless the architecture can issue a read-scoped credential and remove refund operations from that agent's callable tool set, the failure described in the stem remains unresolved.

Question 122

An agent has access to 250 APIs and frequently calls irrelevant administrative endpoints. What design decision would best fit these facts?

  1. Issue a read-scoped credential and remove refund operations from that agent's callable tool set.
  2. Require version pinning, reviewed capability diffs and explicit authorization on tool upgrades.
  3. Apply destination allowlists, recipient verification and outbound-send authorization.
  4. Curate task-specific tools and load infrequent capabilities through controlled discovery.

Explanation

Correct Answer

D:

A smaller capability surface improves tool selection and limits the consequences of prompt injection. Verify smaller authorized tool surface under the actual connector identity; the backend must actually enforce smaller authorized tool surface, regardless of instructions.

Incorrect Answers

A:

The design concern is real but different: read-only tool authority. It does not impose the specific requirement for smaller authorized tool surface; a suitable solution must curate task-specific tools and load infrequent capabilities through controlled discovery and demonstrate the result.

B:

This addresses governed tool-capability changes; however, the problem asks for smaller authorized tool surface. Without the safeguard to curate task-specific tools and load infrequent capabilities through controlled discovery, the original failure mode is still possible.

C:

That choice would be responsive to restricted outbound destinations, but the reported defect is different. It leaves smaller authorized tool surface unproved because there is no commitment to curate task-specific tools and load infrequent capabilities through controlled discovery.

Question 123

A travel agent can book flights for customers without confirming a final fare or itinerary. Which intervention most directly addresses the failure?

  1. Issue a read-scoped credential and remove refund operations from that agent's callable tool set.
  2. Place an explicit user confirmation and server-side policy check before purchase execution.
  3. Apply destination allowlists, recipient verification and outbound-send authorization.
  4. Require version pinning, reviewed capability diffs and explicit authorization on tool upgrades.

Explanation

Correct Answer

B:

Human intent should be bound to the exact amount and itinerary before a costly external state change. Verify transaction-specific confirmation under the actual connector identity; the backend must actually enforce transaction-specific confirmation, regardless of instructions.

Incorrect Answers

A:

This recommendation targets read-only tool authority at a different point in the workflow. It leaves transaction-specific confirmation outside the design contract and does not require the team to place an explicit user confirmation and server-side policy check before purchase execution.

C:

This would make sense as an additional project activity for restricted outbound destinations. The priority required here is transaction-specific confirmation, so the workflow still needs a control that will place an explicit user confirmation and server-side policy check before purchase execution.

D:

By itself, this improves governed tool-capability changes, but the case is not asking the architect to optimize that dimension. The requirement is transaction-specific confirmation, supported by an approach to place an explicit user confirmation and server-side policy check before purchase execution.

Question 124

An MCP integration sends broad employee attributes to every tool even when only a customer ID is needed. What should the solution architect prioritize?

  1. Require version pinning, reviewed capability diffs and explicit authorization on tool upgrades.
  2. Issue a read-scoped credential and remove refund operations from that agent's callable tool set.
  3. Minimize arguments and redact unnecessary sensitive fields before tool invocation.
  4. Use tenant-scoped short-lived credentials and enforce per-request tenant identity.

Explanation

Correct Answer

C:

Data minimization reduces exposure at integration boundaries without blocking the requested lookup. Verify least-data tool invocation under the actual connector identity; the backend must actually enforce least-data tool invocation, regardless of instructions.

Incorrect Answers

A:

This shifts attention toward governed tool-capability changes when the relevant decision is about least-data tool invocation. Unless the architecture can minimize arguments and redact unnecessary sensitive fields before tool invocation, the failure described in the stem remains unresolved.

B:

For a workflow whose primary weakness was read-only tool authority, that proposal could be reasonable. Here the unmet obligation is least-data tool invocation, and that requires the system to minimize arguments and redact unnecessary sensitive fields before tool invocation.

D:

It is aimed at a different outcome, per-tenant credential isolation. No evidence is given that it provides least-data tool invocation; the needed mechanism is to minimize arguments and redact unnecessary sensitive fields before tool invocation at the appropriate boundary.

Question 125

The service stores one long-lived API token for every tenant and cannot revoke a single customer's access. Which design choice best matches the stated constraint?

  1. Use tenant-scoped short-lived credentials and enforce per-request tenant identity.
  2. Remove destructive privileges at the API policy and gateway layer.
  3. Issue a read-scoped credential and remove refund operations from that agent's callable tool set.
  4. Minimize arguments and redact unnecessary sensitive fields before tool invocation.

Explanation

Correct Answer

A:

Scoped identity reduces blast radius and makes revocation meaningful for a specific customer. Verify per-tenant credential isolation under the actual connector identity; the backend must actually enforce per-tenant credential isolation, regardless of instructions.

Incorrect Answers

B:

Consider its actual operating scope: backend-enforced mutation restrictions. The requested assurance is per-tenant credential isolation, which calls for an implementation that can use tenant-scoped short-lived credentials and enforce per-request tenant identity, not merely for a related improvement.

C:

The alternative could reduce risk associated with read-only tool authority, yet the described behavior requires per-tenant credential isolation. That outcome depends on the capability to use tenant-scoped short-lived credentials and enforce per-request tenant identity.

D:

The proposed measure is aimed at least-data tool invocation, not per-tenant credential isolation. The architecture must instead use tenant-scoped short-lived credentials and enforce per-request tenant identity; a separate change is still needed for per-tenant credential isolation.

Question 126

An assistant can run SQL queries across an unrestricted production database to answer inventory questions. Which architecture adjustment is most appropriate?

  1. Build a permission matrix from required workflows and test with the actual production identity.
  2. Issue a read-scoped credential and remove refund operations from that agent's callable tool set.
  3. Expose parameterized, read-only domain queries through a narrow service interface.
  4. Curate task-specific tools and load infrequent capabilities through controlled discovery.
  5. Require version pinning, reviewed capability diffs and explicit authorization on tool upgrades.

Explanation

Correct Answer

C:

A constrained gateway validates inputs and prevents arbitrary database operations from model-generated text. Verify bounded database query capabilities under the actual connector identity; the backend must actually enforce bounded database query capabilities, regardless of instructions.

Incorrect Answers

A:

By itself, this improves production-role permission testing, but the case is not asking the architect to optimize that dimension. The requirement is bounded database query capabilities, supported by an approach to expose parameterized, read-only domain queries through a narrow service interface.

B:

Its main effect concerns read-only tool authority. The case explicitly hinges on bounded database query capabilities; to resolve it reliably, the implementation has to expose parameterized, read-only domain queries through a narrow service interface, which this choice does not establish.

D:

The proposed action prioritizes smaller authorized tool surface. That can coexist with the right solution, but it cannot replace bounded database query capabilities: the application must still expose parameterized, read-only domain queries through a narrow service interface.

E:

The design concern is real but different: governed tool-capability changes. It does not impose the specific requirement for bounded database query capabilities; a suitable solution must expose parameterized, read-only domain queries through a narrow service interface and demonstrate the result.

Question 127

A tool description says to avoid deletions, yet its endpoint accepts DELETE with the agent's bearer token. Which approach should be recommended?

  1. Issue a read-scoped credential and remove refund operations from that agent's callable tool set.
  2. Apply destination allowlists, recipient verification and outbound-send authorization.
  3. Use tenant-scoped short-lived credentials and enforce per-request tenant identity.
  4. Remove destructive privileges at the API policy and gateway layer.

Explanation

Correct Answer

D:

Textual tool warnings are not access controls; the backend must reject unauthorized mutations. Verify backend-enforced mutation restrictions under the actual connector identity; the backend must actually enforce backend-enforced mutation restrictions, regardless of instructions.

Incorrect Answers

A:

It is aimed at a different outcome, read-only tool authority. No evidence is given that it provides backend-enforced mutation restrictions; the needed mechanism is to remove destructive privileges at the API policy and gateway layer at the appropriate boundary.

B:

It emphasizes restricted outbound destinations without closing the stated gap in backend-enforced mutation restrictions. By contrast, the remedy to remove destructive privileges at the API policy and gateway layer acts on the condition that makes this scenario fail.

C:

That is primarily a technique for per-tenant credential isolation. It lacks the explicit measure for backend-enforced mutation restrictions that is necessary here, namely a system that will remove destructive privileges at the API policy and gateway layer.

Question 128

A team tests an agent using administrator permissions and the production role unexpectedly violates least privilege. What is the best production design response?

  1. Remove destructive privileges at the API policy and gateway layer.
  2. Issue a read-scoped credential and remove refund operations from that agent's callable tool set.
  3. Apply destination allowlists, recipient verification and outbound-send authorization.
  4. Build a permission matrix from required workflows and test with the actual production identity.

Explanation

Correct Answer

D:

Authorization coverage must be validated under real service roles, not inferred from an overpowered test account.

Incorrect Answers

A:

The proposed measure is aimed at backend-enforced mutation restrictions, not production-role permission testing. The architecture must instead build a permission matrix from required workflows and test with the actual production identity; a separate change is still needed for production-role permission testing.

B:

The relevant trade-off is not primarily read-only tool authority. A successful fix needs production-role permission testing; otherwise production-role permission testing remains unverified even after this change.

C:

The control would be justified if restricted outbound destinations were the limiting factor. In this scenario the decisive target is production-role permission testing; success requires an architecture that will build a permission matrix from required workflows and test with the actual production identity.

Question 129

A customer support agent has an email-sending tool but a prompt injection makes it target an unapproved address. Which change most directly satisfies this requirement?

  1. Issue a read-scoped credential and remove refund operations from that agent's callable tool set.
  2. Curate task-specific tools and load infrequent capabilities through controlled discovery.
  3. Use tenant-scoped short-lived credentials and enforce per-request tenant identity.
  4. Apply destination allowlists, recipient verification and outbound-send authorization.

Explanation

Correct Answer

D:

Tool schemas and prompts alone do not provide trusted recipient policy for external communications. Verify restricted outbound destinations under the actual connector identity; the backend must actually enforce restricted outbound destinations, regardless of instructions.

Incorrect Answers

A:

The proposed action prioritizes read-only tool authority. That can coexist with the right solution, but it cannot replace restricted outbound destinations: the application must still apply destination allowlists, recipient verification and outbound-send authorization.

B:

The design concern is real but different: smaller authorized tool surface. It does not impose the specific requirement for restricted outbound destinations; a suitable solution must apply destination allowlists, recipient verification and outbound-send authorization and demonstrate the result.

C:

This addresses per-tenant credential isolation; however, the problem asks for restricted outbound destinations. Without the safeguard to apply destination allowlists, recipient verification and outbound-send authorization, the original failure mode is still possible.

Question 130

A plugin marketplace adds tool permissions automatically after updates, exposing new write capabilities. Which solution best resolves this operational gap?

  1. Curate task-specific tools and load infrequent capabilities through controlled discovery.
  2. Issue a read-scoped credential and remove refund operations from that agent's callable tool set.
  3. Require version pinning, reviewed capability diffs and explicit authorization on tool upgrades.
  4. Place an explicit user confirmation and server-side policy check before purchase execution.

Explanation

Correct Answer

C:

Dependency upgrades can expand attack surface even when prompt templates are unchanged. Verify governed tool-capability changes under the actual connector identity; the backend must actually enforce governed tool-capability changes, regardless of instructions.

Incorrect Answers

A:

Improving smaller authorized tool surface does not imply governed tool-capability changes. The selected design must require version pinning, reviewed capability diffs and explicit authorization on tool upgrades, whereas this choice omits that direct protection or verification step.

B:

This addresses read-only tool authority; however, the problem asks for governed tool-capability changes. Without the safeguard to require version pinning, reviewed capability diffs and explicit authorization on tool upgrades, the original failure mode is still possible.

D:

This would make sense as an additional project activity for transaction-specific confirmation. The priority required here is governed tool-capability changes, so the workflow still needs a control that will require version pinning, reviewed capability diffs and explicit authorization on tool upgrades.

Claude Certified Architect - Professional (CCAR-P) Practice Tests