McAfee Secure

Exam Code: CCA-F

Exam Name: Claude Certified Architect - Foundations

Certification Provider: Anthropic

Questions & Answers

Anthropic CCA-F Questions & Answers

Study with Up-To-Date REAL Exam Questions and Answers from the ACTUAL Test

Questions & Answers

134 Questions & Answers with Testing Engine

"Claude Certified Architect - Foundations Exam", also known as CCA-F exam, is a Anthropic certification exam.

Pass your tests with the always up-to-date CCA-F Exam Engine. Your CCA-F training materials keep you at the head of the pack!

Satisfaction Guaranteed

Money Back Guarantee

Testking provides hassle-free money back guarantee with our products. That is because we have 100% trust in the abilities of our professional and experience product team, and our record is a proof of that.

99.6% PASS RATE
Was: $137.49
Now: $124.99

Anthropic CCA-F Practice Test 15

Question 140

A code-search subagent can delete repository branches although it is only asked to inspect call sites. What should the architect do?

  1. Remove destructive capabilities from the subagent's allowed tools and its underlying credentials.
  2. Add a system prompt saying deletion is discouraged, but keep admin access.
  3. Give all search tools shared administrator tokens to ease setup.
  4. Permit destructive repository operations during search-only analysis, relying on backups if an accidental mutation occurs.

Explanation

Correct Answer

A:

A read task should not be able to mutate repository state.

Incorrect Answers

B:

Prompt warnings cannot enforce permission boundaries.

C:

Broad credentials expand the damage from a single mistaken call.

D:

Restoration is not a replacement for least privilege. A search-only role should not have the ability to delete branches at all.

Question 141

A QA specialist occasionally needs to verify one assertion against a known document. It currently receives full search, shell, payment and email tools. How can it be scoped?

  1. Keep all tools available and rely on the reviewer prompt to avoid mutation-capable tools when verifying a fact.
  2. Allow it to browse every user file and send messages on behalf of the owner.
  3. Provide a read-only verify_fact operation with an evidence locator and bounded response contract.
  4. Eliminate source access and ask it to certify claims from memory.

Explanation

Correct Answer

C:

Purpose-built verification minimizes unrelated privileges and tool-selection ambiguity.

Incorrect Answers

A:

A prompt preference is not a least-privilege permission boundary. The review worker should receive the one source-verification capability it needs.

B:

Those abilities exceed evidence verification needs.

D:

Unsupported assertions are not verification.

Question 142

A processing pipeline must retrieve an order's metadata before considering subsequent enrichment. Select THREE steps that safely enforce the first required metadata lookup without trapping later work.

  1. Require the designated metadata tool for the prerequisite phase, then allow appropriate subsequent tool selection.
  2. Pin metadata lookup for every subsequent turn indefinitely.
  3. Force payment execution as the first call because it is a strong action.
  4. Allow all later actions before metadata and hope the agent chooses wisely.
  5. Verify the required metadata result and link it to the current document or customer identity before advancing.
  6. Carry the verified metadata and its original source identifier into the next permitted enrichment call.

Explanation

Correct Answers

A:

Forcing the initial dependency helps ensure required evidence is collected first.

E:

The initial forced call is meaningful only if the returned data actually satisfies the prerequisite for the relevant subject. A failed or unrelated result must not unlock downstream actions.

F:

After the mandatory lookup succeeds, the next operation needs the actual verified metadata rather than the original request alone. Preserving that result and source ID is separate from enforcing the initial call and checking its identity.

Incorrect Answers

B:

The initial requirement should not block later legitimate work.

C:

Money movement is not the requested prerequisite.

D:

A mandatory dependency cannot rely solely on preference.

Question 143

An application must obtain a classification tool result before exposing the final user response. A new model release may not support forced tool_choice=any. Which action is safest?

  1. Trust a longer system prompt to guarantee the classifier runs despite tool-choice compatibility errors.
  2. Send the same unsupported forced-tool setting with a higher request timeout.
  3. Skip the required classification when the model does not request a tool on its own.
  4. Check supported tool-choice settings for the deployed model and enforce the classification prerequisite in the host controller when forced selection is unsupported.

Explanation

Correct Answer

D:

The client can enforce the prerequisite deterministically and use only tool-choice modes the current model supports; it then resumes reasoning with the actual classification result.

Incorrect Answers

A:

A prompt can guide the model but cannot guarantee the application-level prerequisite or bypass a request parameter error.

B:

A compatibility or parameter validation failure is not a timeout. The host must respect the selected model's actual tool-choice contract and enforce prerequisites independently when necessary.

C:

A required business validation cannot be dropped just because one model release changes tool-choice capabilities.

Question 144

After a mandatory customer lookup, an agent needs to explain the findings without any further tool. What tool-choice mode supports the natural closing turn?

  1. Force the metadata lookup forever, even when the answer is ready.
  2. Use automatic tool choice after completing the prerequisite rather than continually forcing another call.
  3. Disable return of the initial lookup result to keep the next turn simple.
  4. Require a tool call even if no tools are relevant to the question.

Explanation

Correct Answer

B:

The model can finish with prose or request a genuinely needed follow-up.

Incorrect Answers

A:

Unnecessary repeated calls can create loops and extra cost.

C:

The final explanation depends on the verified tool evidence.

D:

A forced unrelated action is not useful progress.

Question 145

A coordinator is forced to call search_docs for every prompt, including greeting and calculation requests. How should the policy change?

  1. Disable all tools permanently so future search tasks fail.
  2. Run the same search repeatedly until a result is returned.
  3. Tell the model to hide unnecessary tool calls from logs.
  4. Apply forced tool use only when search is a real prerequisite; otherwise permit normal routing or final response.

Explanation

Correct Answer

D:

One rigid tool policy need not fit unrelated user intents.

Incorrect Answers

A:

The fix is scoped enforcement, not removing useful capabilities.

B:

Repetition does not make irrelevant retrieval appropriate.

C:

Unlogged operations remain wasteful and unauditable.

Question 146

Every narrow subagent is injected with fifty full tool schemas, many never allowed for its role. What optimization is justified?

  1. Supply only role-relevant, authorized tool definitions, using discovery when appropriate.
  2. Remove validation because large schemas are expensive.
  3. Give every role the same admin capabilities to simplify generation.
  4. Increase tool descriptions to several thousand words each for all subagents.

Explanation

Correct Answer

A:

Reducing unused schemas saves context and narrows the action surface.

Incorrect Answers

B:

The appropriate optimization is fewer tools, not weaker contracts.

C:

Shared authority increases risk and routing ambiguity.

D:

More irrelevant tokens worsen context pressure.

Question 147

A synthesis specialist begins searching the web directly even though a dedicated evidence-gathering worker is responsible for primary retrieval. How should the architecture enforce this split?

  1. Restrict synthesis tools and route additional evidence requests through the coordinator to the search specialist.
  2. Tell synthesis to avoid searches but leave all search permissions active.
  3. Stop retaining source IDs because the final summary is what matters.
  4. Give synthesis every search and write tool so it can work independently.

Explanation

Correct Answer

A:

Controlled handoffs preserve source accounting and role boundaries.

Incorrect Answers

B:

Text guidance is weaker than an actual allowlist.

C:

Without source identifiers the coordinator cannot verify claims.

D:

That undermines the planned validation and provenance flow.

Question 148

A draft-only code reviewer runs with repository production-write permissions. What permission boundary is appropriate?

  1. Limit the reviewer to read-only inspection or an isolated sandbox and authorize promotion separately.
  2. Run every generated shell command as an organization administrator.
  3. Mask production paths in chat while retaining the same production-write credential inside the editor worker.
  4. Leave write access enabled because the reviewer usually behaves.

Explanation

Correct Answer

A:

Draft analysis should not mutate production code.

Incorrect Answers

B:

This amplifies the impact of a flawed suggestion.

C:

Display redaction does not restrict an agent's effective capabilities. Its credentials and filesystem scope must exclude production mutation.

D:

Prior good behavior is not access control.

Question 149

A data reviewer needs to check one official-source claim but lacks permission to open any source links. What exception best follows least privilege?

  1. Remove source requirements so the review can pass on prose alone.
  2. Grant a narrow read-only verification capability to approved source locations, with no general mutation authority.
  3. Give the reviewer unrestricted organizational file-management access.
  4. Replace the missing verification with the reviewer’s self-rated confidence score and mark the claim source checked.

Explanation

Correct Answer

B:

The exception should enable the needed evidence check and nothing more.

Incorrect Answers

A:

A check without access cannot establish the claim.

C:

Broad permissions are unnecessary for one claim.

D:

Confidence is not evidence that the required document was accessed. The workflow needs permitted read-only verification or an explicit unresolved result.

Question 150

A client config forces tool_choice to a function name that is not in the registered tool definitions. What should the host do before sending the request?

  1. Disable the schema checks to allow arbitrary strings.
  2. Substitute the most privileged tool to ensure a call occurs.
  3. Ask Claude to guess a similarly spelled tool name at runtime.
  4. Validate the forced name against declared tools and resolve or report the configuration mismatch.

Explanation

Correct Answer

D:

An unregistered target cannot be invoked reliably.

Incorrect Answers

A:

Malformed tool declarations cannot create missing execution logic.

B:

Capability escalation is not a safe fallback.

C:

Guessing may select the wrong operation.

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