Anthropic CCDV-F (Claude Certified Developer - Foundations) Practice Test 23

Question 220

How does Claude Code auto memory differ from a deliberately authored project CLAUDE.md?

  1. CLAUDE.md can never be changed by its maintainers.
  2. Auto memory is a guaranteed permission-enforcement engine.
  3. Auto memory is identical to retraining the model’s weights.
  4. Auto memory records learned context, while CLAUDE.md contains maintained instructions.

Explanation

Correct Answer

D:

The distinction concerns authorship and purpose; both need review when stale or inappropriate information affects behavior.

Incorrect Answers

A:

The project can update its instruction file through its normal review process.

B:

Remembered context is not a substitute for execution controls.

C:

Stored contextual notes are not a model-training operation.

Question 221

Two Claude Code sessions need independent branches and file contents while working on the same Git repository, which already has commits. Which launch approach provides separate checkouts?

  1. Start each session with a distinct named --worktree checkout.
  2. Use two terminals in the same directory without changing checkout setup.
  3. Alternate branch switches in the shared checkout while both sessions run.
  4. Give the sessions different conversation names in the same checkout.

Explanation

Correct Answer

A:

The worktree workflow creates a separate checkout and branch for each session. This isolates working files while retaining the repository relationship; eventual integration still requires review.

Incorrect Answers

B:

Separate terminals do not isolate the filesystem or Git working tree, so edits remain visible to both sessions.

C:

Switching one checkout changes the branch and files seen by both processes, defeating the requirement for simultaneous independent contents.

D:

Conversation names identify histories, but both sessions would still operate on the same working files and current branch.

Question 222

A custom Claude Code subagent should review code but never edit it. What should its configuration reflect?

  1. Give every editing tool and call it read-only in prose.
  2. Restrict its available tools to the necessary read and analysis capabilities.
  3. Allow all write tools but reject only output text containing the word "edit."
  4. Allow shell access that bypasses all restrictions.

Explanation

Correct Answer

B:

A review role should have execution permissions matching its task, rather than rely solely on descriptive intent.

Incorrect Answers

A:

A name does not remove available mutation capabilities.

C:

Text filtering cannot reliably prevent mutations performed through available tools, especially through indirect execution paths.

D:

An unrestricted alternate path undermines the intended read-only boundary.

Question 223

A CI wrapper receives some Claude Code output but the process exits unsuccessfully. What should it do?

  1. Mark success whenever any text was printed.
  2. Discard the exit status permanently.
  3. Inspect the process outcome and structured result before reporting success.
  4. Assume failure means every prior file change was rolled back.

Explanation

Correct Answer

C:

Partial output does not prove the automated task completed; the wrapper must preserve failure signals. A failing process may still have printed progress or modified files, so report the outcome honestly and inspect any partial artifacts.

Incorrect Answers

A:

Errors and partial progress can also produce text.

B:

That removes an important execution signal from automation.

D:

Process failure does not automatically reverse completed side effects.

Question 224

The API returns 401 after a credential was revoked. What is the relevant remediation?

  1. Retry the revoked credential indefinitely.
  2. Increase the output-token budget.
  3. Remove schema validation from responses.
  4. Restore valid authorized authentication through the credential-management process.

Explanation

Correct Answer

D:

The failure concerns identity credentials, so prompt or generation changes cannot resolve it. After repairing authentication, retry only through the normal authorized credential path; repeated calls with the revoked key cannot become valid through delay.

Incorrect Answers

A:

A persistent authentication defect is not repaired by repeated identical requests.

B:

Generation capacity does not validate a revoked credential.

C:

Response parsing occurs after authentication and does not address the rejected key.

Question 225

A valid credential receives 403 when accessing a restricted resource. What should be checked?

  1. Only the request's content type, without checking access scope.
  2. The credential’s authorization and resource scope.
  3. Only a new prompt asking the model to assume administrator status.
  4. Only whether the secret has the expected length.

Explanation

Correct Answer

B:

Authentication proves an identity, while authorization determines whether that identity may access this resource or operation.

Incorrect Answers

A:

A valid request representation can still be forbidden because the authenticated identity lacks entitlement to the resource.

C:

A prompted role does not grant backend permissions to the credential making the request.

D:

Credential appearance does not establish the required entitlement.

Question 226

A request consistently returns 400 with a message naming an unsupported field. What should happen first?

  1. Correct the request to the documented supported contract.
  2. Rotate every production secret.
  3. Retry the unchanged payload with increasing delays.
  4. Report the failed request as a completed answer.

Explanation

Correct Answer

A:

The error identifies a deterministic request-shape issue rather than a transient service problem. Use the error details to locate the unsupported field and validate the corrected request before treating the issue as a provider outage.

Incorrect Answers

B:

Credential rotation does not repair request schema.

C:

Waiting does not make an unsupported field valid.

D:

An error response is not a successful generation.

Question 227

A client fails DNS resolution before reaching the API. Which layer should troubleshooting prioritize?

  1. The tool’s business validation rules.
  2. Network name resolution and endpoint configuration.
  3. The model’s classification accuracy.
  4. Only the application's JSON response parser.

Explanation

Correct Answer

B:

Confirm the hostname and DNS resolution first. With no connection or provider response yet, model behavior and response parsing cannot explain the observed failure.

Incorrect Answers

A:

The failure occurs before the remote API call, not within tool business logic.

C:

Inference has not been reached, so accuracy is not the observed failure.

D:

The request has not reached the API, so response parsing cannot explain the observed name-resolution failure.

Question 228

A request takes twelve seconds: model generation takes two and a tool lookup takes ten. Where should optimization start?

  1. Only the final response formatter.
  2. The tool lookup on the measured critical path.
  3. Only the model family.
  4. Reduce log retention while keeping the measured slow tool call unchanged.

Explanation

Correct Answer

B:

The trace identifies the dominant delay; changing the model alone cannot remove the ten-second tool wait. Even eliminating model generation would leave most of the measured wait, so instrument and improve the slow dependency first.

Incorrect Answers

A:

No evidence shows formatting dominates this measured latency.

C:

The model accounts for a smaller portion of the observed total.

D:

Storage retention does not remove the ten-second dependency on the request's critical path.

Question 229

Search finds the correct passage, but the assembled model request omits it. Which component is implicated?

  1. Only the search index contents.
  2. Context assembly between retrieval and inference.
  3. Only the model’s knowledge cutoff.
  4. The document’s original author.

Explanation

Correct Answer

B:

Successful retrieval is insufficient when its evidence never reaches the model; inspect selection and prompt construction.

Incorrect Answers

A:

The correct passage was already retrieved, so absence from the index is not the demonstrated cause.

C:

The needed current evidence could have been supplied directly.

D:

The trace points to application omission rather than missing source content.

Question 230

A change simultaneously modifies model, prompt, and retrieval. Quality declines. What helps identify the cause?

  1. Assume the newest model caused every failure.
  2. Controlled comparisons that vary one component while holding the others fixed.
  3. Compare only deployment timestamps.
  4. Revert random lines until the score improves once.

Explanation

Correct Answer

B:

Ablation isolates contributions that a bundled change makes difficult to attribute. Retain comparable inputs and scoring criteria during the investigation so differences can be attributed to the component being varied.

Incorrect Answers

A:

Several variables changed; the evidence does not isolate that conclusion.

C:

Timing identifies correlation, not which changed component caused the regression.

D:

Uncontrolled experimentation can confuse noise with a causal improvement.

Claude Certified Developer - Foundations (CCDV-F) Practice Tests