Anthropic CCA-F (Claude Certified Architect - Foundations (CCAR-F)) Practice Test 4

Question 30

A parent agent asks a child to draft an executive brief from verified research notes. Which data should accompany the delegation?

  1. Every source document in full regardless of relevance
  2. The parent company's secret credentials with the research notes
  3. Only an instruction to be persuasive with no source material
  4. The validated findings, target audience and citation references

Explanation

Correct Answer

D:

The worker can write within scope using evidence and output expectations. It need not receive the entire underlying orchestration transcript.

Incorrect Answers

A:

Indiscriminate context transfer can crowd out useful instructions. Supplying curated evidence is more reliable and efficient.

B:

Credentials do not help draft an authorized brief. Passing secrets needlessly enlarges the exposure surface.

C:

Style guidance cannot substitute for factual inputs. An unsupported brief may invent conclusions and lose auditability.

Question 31

Two subagents are assigned to modify the same configuration file in parallel. What change best prevents lost edits?

  1. Ask the agents to choose more descriptive filenames for logs
  2. Partition ownership or serialize writes with an explicit integration step
  3. Increase the context size of each agent while keeping concurrent writes
  4. Make each worker unaware of the other and accept whichever writes last

Explanation

Correct Answer

B:

Concurrent changes to one file can conflict even if each agent's local edit is valid. Clear file ownership and review make the merge predictable.

Incorrect Answers

A:

Log naming may improve diagnostics but does not prevent two workers from overwriting the same target.

C:

Additional memory does not serialize modifications or resolve race conditions. The conflict is about shared mutable state.

D:

Last-writer-wins can silently discard valid work. Independence is appropriate for research, not unsynchronized shared mutations.

Question 32

A company is creating a read-only research subagent to analyze confidential project files. Which TWO design decisions best enforce a useful, narrow delegation? Choose TWO.

  1. Pass a scoped question and only the authorized supporting context
  2. Let the worker publish an externally binding decision without parent review
  3. Share every unrelated parent conversation to maximize token usage
  4. Provide only the repository inspection tools needed for the research task
  5. Give the research worker unrestricted deployment and billing mutation tools

Explanation

Correct Answers

A:

The worker can investigate relevant sources without inheriting unnecessary customer records. The parent retains integration responsibility.

D:

A restricted tool set prevents the worker from accidentally modifying files or calling unrelated administrative services. Least privilege is implemented at the capability boundary.

Incorrect Answers

B:

The delegated role is investigative. Final synthesis and authorization remain with the responsible coordinator.

C:

Excessive context can expose data and distract from the research objective. More history is not necessarily more useful.

E:

Broad powers are unnecessary for inspection. More privileged options enlarge operational risk without improving research quality.

Question 33

A specialist discovers an unresolved dependency outside its assigned scope. How should it conclude the delegated task?

  1. Return a partial status, supporting evidence and the specific blocker to its parent
  2. Abandon all collected evidence as soon as a blocker appears
  3. Silently create new high-privilege subagents until the dependency disappears
  4. Report success and omit the missing dependency

Explanation

Correct Answer

A:

A truthful structured handoff lets the coordinator reassign or escalate work. The child should not fabricate completion outside its authority.

Incorrect Answers

B:

Verified partial findings remain useful to recovery or handoff. Losing them forces unnecessary repeated work.

C:

Unapproved delegation can overrun cost and authority boundaries. The parent should decide whether to expand scope.

D:

False completion masks the reason the wider workflow cannot finish. The parent needs explicit uncertainty to decide next steps.

Question 34

A coordinator has a defined AgentDefinition for a reviewer, but its allowedTools contains only Read, Grep and Glob. The older exam guide calls the delegation tool Task; the current SDK calls it Agent. Which configuration change enables real subagent invocation?

  1. Give the reviewer write access without changing coordinator permissions
  2. Ask the coordinator to imitate a reviewer in its own response text
  3. Increase the token limit while leaving allowedTools unchanged
  4. Permit the subagent delegation tool for the SDK version

Explanation

Correct Answer

D:

The coordinator must have the delegation capability in its allowed tools. Current SDK configurations expose Agent; older versions called the same invocation path Task. A defined agent alone is not permission to invoke it.

Incorrect Answers

A:

Child tool permissions do not grant the parent the ability to spawn that child. The parent still lacks the delegation tool, and unnecessary writes weaken the reviewer boundary.

B:

A simulated second opinion in the same context is not subagent execution. It lacks isolated context, separate tool boundaries and an identifiable delegation trace.

C:

More context or output tokens cannot authorize a missing tool. The invocation will remain unavailable until the allowed-tools configuration includes the delegation capability.

Question 35

A parent must ask a subagent to verify a reported security issue without exposing unrelated customer tickets. Which handoff is appropriate?

  1. Give the child all current and historical customer records for completeness
  2. Share production access tokens so the child can explore freely
  3. Tell the child to guess the vulnerability from its training alone
  4. Pass the minimal sanitized code excerpts and testable security claim

Explanation

Correct Answer

D:

A narrow evidence package can support verification while respecting data minimization. The child needs relevant artifacts, not the parent's full workload.

Incorrect Answers

A:

Unrelated customer data does not improve validation of the code issue. It unnecessarily broadens confidentiality exposure.

B:

High-privilege secrets create avoidable operational risk. Inspection should use controlled read interfaces and task-specific evidence.

C:

Without the actual evidence, the worker cannot establish whether a claimed issue exists in the repository.

Question 36

A returns agent is permitted to approve refunds under a documented threshold, but higher amounts require staff approval. What should enforce that rule?

  1. A system prompt saying refunds above the threshold are discouraged
  2. A reviewer who only checks refund logs once each quarter
  3. Application-side policy checking before the refund tool runs
  4. A longer customer response explaining refund ethics

Explanation

Correct Answer

C:

A deterministic gate can reject or pause out-of-policy requests regardless of prompt wording. The agent may suggest a refund but cannot authorize its own exception.

Incorrect Answers

A:

Instructions influence behavior but are not a hard transaction constraint. Backend checks remain necessary for financially sensitive side effects.

B:

Post hoc audit may detect abuse but does not block the unapproved transaction at the time it occurs.

D:

Customer-facing prose cannot prevent the backend from processing an unauthorized refund. The control must sit on the execution path.

Question 37

A support chatbot cannot verify whether an account holder owns the billing address being changed. Which step should occur next?

  1. Ask a search subagent to guess the owner's name from public sources
  2. Update the billing address and ask the customer to confirm later
  3. Have the assistant generate a more convincing security answer
  4. Pause the change for approved identity verification

Explanation

Correct Answer

D:

Ownership ambiguity is an authorization blocker. The system must not infer control of an account from a conversational claim.

Incorrect Answers

A:

External guesses are not accepted proof of account control. The operation needs trusted identity verification.

B:

Changing account data before authorization reverses the required trust sequence. The wrong owner could modify another customer's records.

C:

Fluent wording does not establish identity. A separate verification control is needed before the tool can write data.

Question 38

A support workflow drafts a refund response but a high-value payout requires approval. Which TWO independent safeguards should the architecture implement? Choose TWO.

  1. Validate amount and policy authorization in the backend before any refund write
  2. Treat a confident model summary as proof the refund completed
  3. Grant all support specialists unrestricted refund permissions
  4. Rely solely on a prompt that says high-value refunds are unusual
  5. Require a recorded human approval for cases above the policy threshold

Explanation

Correct Answers

A:

A deterministic server-side check prevents the model from bypassing financial limits. Tool-use intent is not proof of authority.

E:

The approval checkpoint establishes accountability for exceptions. The system can pause the action while preserving the case evidence.

Incorrect Answers

B:

A generated description does not establish that the backend accepted a transaction. Actual execution status is needed.

C:

Broad privileges violate least privilege and do not create the required approval trail. Sensitive writes should remain scoped.

D:

Prompt guidance cannot guarantee authorization checks at execution time. A strong policy must be enforced outside the model.

Question 39

A customer support agent receives an ambiguous cancellation request that could refer to two subscriptions. What should it do before issuing a cancellation?

  1. Cancel both subscriptions to ensure one of them matches
  2. Cancel the most expensive subscription as a best guess
  3. Treat the request as proof that all authorization checks passed
  4. Clarify which subscription the customer wants to cancel

Explanation

Correct Answer

D:

The ambiguity affects a destructive side effect. Clarification prevents canceling the wrong service and preserves the customer's intent.

Incorrect Answers

A:

Executing multiple destructive actions compounds the ambiguity. A safe workflow needs a unique target.

B:

An inferred preference is not consent for a specific irreversible action. Cost does not establish which subscription the user intended.

C:

User text alone does not resolve the target or verify permission. Both prerequisites still matter.

Question 40

A policy team wants every support answer mentioning regulated advice to show an escalation notice, even if the model forgets. Which solution provides a stronger guarantee?

  1. Enforce the condition in application logic before releasing the response
  2. Train customer agents to check only a random monthly sample
  3. Add the notice as an optional style tip in the prompt
  4. Increase the model's creative sampling to make notices more prominent

Explanation

Correct Answer

A:

Deterministic checks can block or augment responses that match the defined escalation criteria. Instructions alone cannot ensure every generated answer complies.

Incorrect Answers

B:

Sampling is useful for audits, not per-answer enforcement. The requirement applies to every qualifying response.

C:

A language-model instruction can be overlooked. Mandatory behavior needs a coded guard or approval gate.

D:

Greater output variability does not improve guaranteed compliance. It can make consistency worse.

Claude Certified Architect - Foundations (CCAR-F) (CCA-F) Practice Tests