Anthropic CCDV-F (Claude Certified Developer - Foundations) Practice Test 21

Question 200

A teammate wants to change the model for an ongoing Claude Code session. Which built-in control is intended for this?

  1. Access /mcp to view and manage server connections.
  2. Set out a Claude Code subagent with role instructions and permitted capabilities.
  3. Use /model and select a supported model.
  4. Arrange scoped rules so relevant directories receive the appropriate guidance.

Explanation

Correct Answer

C:

Model selection is a session configuration operation; rewriting a user prompt does not itself guarantee a runtime model change.

Incorrect Answers

A:

The MCP command exposes connection configuration and authentication rather than editing the conversation's natural-language instructions; this implements inspect MCP connections rather than the demanded switch active coding model.

B:

Subagent definitions create isolated task contexts and can restrict tool access, separating review from broad editing authority; when the goal is switch active coding model, optimizing configured code review subagent misses that operational boundary.

D:

Path-specific rules reduce instruction clutter and make different coding standards applicable in the appropriate project area; the consequence is improved scoped repository rules, not verified switch active coding model.

Question 201

A team must deny certain file mutations regardless of developer prompts. Where should it enforce that in Claude Code?

  1. Carry out Claude Code in print/headless mode with a supported structured output format.
  2. Adopt /clear to begin a fresh session context.
  3. Adopt the supported resume workflow to reopen the saved conversation.
  4. Configure deny rules and suitable hook or sandbox policies in settings.
  5. Group scoped rules so relevant directories receive the appropriate guidance.
  6. Adopt /model and select a supported model.

Explanation

Correct Answer

D:

Operational permission controls govern actual tool use more reliably than a statement inside CLAUDE.md alone. Pair workspace instructions with settings, operating-system permissions, or hook checks that refuse disallowed tool execution.

Incorrect Answers

A:

Noninteractive invocation is suited to automation; the caller must still parse completion status and restrict tool permissions; it follows headless CI invocation constraints but leaves deny risky file tools unspecified.

B:

Clearing starts an empty conversation; it is different from a summary-based compacting operation for continued tasks; this may help clear conversational context but cannot by itself deliver deny risky file tools.

C:

Session continuation retrieves recorded history or state; merely recreating the same working directory does not recover a conversation; the deny risky file tools criterion differs from multi-criterion integration; the gap persists.

E:

Path-specific rules reduce instruction clutter and make different coding standards applicable in the appropriate project area; the effect concerns scoped repository rules; the success condition is deny risky file tools.

F:

Model selection is a session configuration operation; rewriting a user prompt does not itself guarantee a runtime model change; it improves switch active coding model only, leaving deny risky file tools unproven.

Question 202

A Claude Code repository needs shared conventions while enterprise security rules remain nonoverridable. Which TWO configuration approaches support that split? Choose TWO.

  1. Keep all conventions only in one developer’s personal memory.
  2. Commit reviewed project instructions in CLAUDE.md.
  3. Apply enforced restrictions through managed settings and actual permissions.
  4. Depend only on a chat reminder to block forbidden shell commands.
  5. Place API secrets directly in a committed settings file.

Explanation

Correct Answers

B:

Repository instructions convey coding conventions consistently to collaborators using the project. Keep instructions in the repository so teammates can review changes in the same workflow as code edits.

C:

Higher-priority managed policy can prevent unsafe operations regardless of local preferences. Managed deny policies or tool-level permissions can be checked even when a prompt suggests a forbidden command.

Incorrect Answers

A:

Personal settings are not a reliable shared source of repository standards. Personal preferences are not a controlled distribution mechanism for the security and coding rules of a shared project.

D:

Natural-language guidance cannot reliably intercept actual execution side effects. Instructions in chat are advisory unless an execution-layer control is configured to deny a prohibited action.

E:

Source-controlled secrets broaden credential exposure and impede safe rotation. Committed secrets persist in repository history and may be visible to more collaborators than the runtime service.

Question 203

A script consumes Claude Code output as structured streaming events. What must it avoid assuming?

  1. Prepare a reusable Skill with documented instructions and supporting assets.
  2. Do not treat each stream event as a complete assistant answer.
  3. Apply /clear to begin a fresh session context.
  4. Operate /init in Claude Code and review the generated project guidance.
  5. Prepare deny rules and suitable hook or sandbox policies in settings.

Explanation

Correct Answer

B:

Incremental event streams contain intermediate messages and metadata; the consumer must recognize completion before using final results.

Incorrect Answers

A:

Skills package repeatable procedures and context for assistant use; they are not a replacement for executable tool permissions; the interpret headless stream events criterion differs from reusable Claude Code skill; the gap persists.

C:

Clearing starts an empty conversation; it is different from a summary-based compacting operation for continued tasks; it improves clear conversational context only, leaving interpret headless stream events unproven.

D:

The initialization command creates a starter instruction file; the team should still edit it to reflect real repository conventions; the scenario's controlling requirement is interpret headless stream events, not the related initialize repo instructions concern.

E:

Operational permission controls govern actual tool use more reliably than a statement inside CLAUDE.md alone; for interpret headless stream events, this instead manages deny risky file tools; the requested decision remains unresolved.

Question 204

A Claude API request returns HTTP 400 after a developer changes its JSON schema. Which diagnostic path should start first?

  1. Inspect the sequence of tool requests, observations, and termination decisions.
  2. Reattempt eligible requests with controlled backoff and an alternate path if available.
  3. Use quota-aware throttling and bounded backoff rather than immediate repeated retries.
  4. Inspect request validation details and the supplied schema before retrying.

Explanation

Correct Answer

D:

Bad-request errors indicate an invalid request contract; retrying unchanged payloads will not repair schema or parameter mistakes.

Incorrect Answers

A:

A final answer alone hides recurrent loops; trajectory traces show whether the agent used error feedback appropriately; because looping trajectory diagnostics is its concern, invalid-request recovery remains outside its effect here.

B:

An overloaded service differs from a malformed payload; resilience requires delay and limits, not changing the prompt blindly; when the goal is invalid-request recovery, optimizing overload response handling misses that operational boundary.

C:

Rate limiting signals excessive request volume or tokens; paced retries and concurrency controls address that condition; the consequence is improved traffic throttling triage, not verified invalid-request recovery.

Question 205

A production service returns HTTP 429 intermittently under load. What is the appropriate first recovery?

  1. Match request IDs with model, prompt, tool, latency, and failure-stage metadata.
  2. Regard it as model-output or task-semantic error and add domain validation.
  3. Apply quota-aware throttling and bounded backoff rather than immediate repeated retries.
  4. Use severity-weighted regression gates that block unacceptable high-risk failures.

Explanation

Correct Answer

C:

Rate limiting signals excessive request volume or tokens; paced retries and concurrency controls address that condition.

Incorrect Answers

A:

Cross-layer traces identify where a request diverged and which configuration produced the symptom, while respecting secret redaction; this implements correlated failure telemetry rather than the demanded traffic throttling triage.

B:

The transport and schema succeeded; the incorrect business value must be caught by evaluation or authoritative checks; this targets semantic failure isolation, while the system must demonstrate traffic throttling triage.

D:

Averages can hide harmful tail regressions; release criteria must reflect business impact of particular error classes; the consequence is improved multi-criterion integration, not verified traffic throttling triage.

Question 206

A service gets a transient 529 overload response while requests were previously valid. What should its client do?

  1. As an alternative, investigate the application's tool argument mapping and dispatch layer.
  2. Employ deterministic connector fixtures and compare model trajectories against a stable expected outcome.
  3. Examine stop_reason and output token limits in the recorded API trace.
  4. Retry eligible requests with controlled backoff and an alternate path if available.

Explanation

Correct Answer

D:

An overloaded service differs from a malformed payload; resilience requires delay and limits, not changing the prompt blindly.

Incorrect Answers

A:

The tool input chosen by Claude and the executed downstream call differ; that points to integration code, not model reasoning; changing mismapped tool arguments does not establish overload response handling in this application.

B:

Fixed tool observations make prompt or model regressions reproducible; live-connector behavior should be covered by a separate integration test suite with its own assertions; the scenario's controlling requirement is overload response handling, not the related reproducible agent test fixtures concern.

C:

A max_tokens stop distinguishes deliberate truncation from a natural end_turn and directs the next fix; it follows incomplete generation trace constraints but leaves overload response handling unspecified.

Question 207

An extraction response is valid JSON but assigns the wrong invoice currency. What failure class is this?

  1. Treat it as model-output or task-semantic error and add domain validation.
  2. For this workflow, investigate the application's tool argument mapping and dispatch layer.
  3. As an alternative, examine the retrieval corpus version and ranking before changing the model.
  4. Link request IDs with model, prompt, tool, latency, and failure-stage metadata.

Explanation

Correct Answer

A:

The transport and schema succeeded; the incorrect business value must be caught by evaluation or authoritative checks.

Incorrect Answers

B:

The tool input chosen by Claude and the executed downstream call differ; that points to integration code, not model reasoning; this implements mismapped tool arguments rather than the demanded semantic failure isolation.

C:

A model can reason consistently over stale evidence; source freshness is a data pipeline defect; for semantic failure isolation, this instead manages outdated retrieval evidence; the requested decision remains unresolved.

D:

Cross-layer traces identify where a request diverged and which configuration produced the symptom, while respecting secret redaction; a remedy for correlated failure telemetry is not a substitute for semantic failure isolation.

Question 208

A tool-use trace shows Claude requested get_customer correctly, but the dispatcher sent a different customer ID. Where is the defect?

  1. Interpret it as model-output or task-semantic error and add domain validation.
  2. Enforce quota-aware throttling and bounded backoff rather than immediate repeated retries.
  3. Review stop_reason and output token limits in the recorded API trace.
  4. Investigate the application's tool argument mapping and dispatch layer.

Explanation

Correct Answer

D:

The tool input chosen by Claude and the executed downstream call differ; that points to integration code, not model reasoning.

Incorrect Answers

A:

The transport and schema succeeded; the incorrect business value must be caught by evaluation or authoritative checks; the scenario's controlling requirement is mismapped tool arguments, not the related semantic failure isolation concern.

B:

Rate limiting signals excessive request volume or tokens; paced retries and concurrency controls address that condition; where mismapped tool arguments matters, a control for traffic throttling triage leaves that test unmet.

C:

A max_tokens stop distinguishes deliberate truncation from a natural end_turn and directs the next fix; for mismapped tool arguments, a control centered on incomplete generation trace is not the appropriate substitute.

Question 209

An answer cites outdated product documentation even though the model followed the provided context. What should be diagnosed first?

  1. Examine the retrieval corpus version and ranking before changing the model.
  2. Adopt deterministic connector fixtures and compare model trajectories against a stable expected outcome.
  3. Employ severity-weighted regression gates that block unacceptable high-risk failures.
  4. Handle it as model-output or task-semantic error and add domain validation.
  5. Examine request validation details and the supplied schema before retrying.

Explanation

Correct Answer

A:

A model can reason consistently over stale evidence; source freshness is a data pipeline defect. Inspect document revision dates and retrieval rankings to learn whether the correct version was indexed and retrieved.

Incorrect Answers

B:

Fixed tool observations make prompt or model regressions reproducible; live-connector behavior should be covered by a separate integration test suite with its own assertions; the effect concerns reproducible agent test fixtures; the success condition is outdated retrieval evidence.

C:

Averages can hide harmful tail regressions; release criteria must reflect business impact of particular error classes; where outdated retrieval evidence matters, a control for multi-criterion integration leaves that test unmet.

D:

The transport and schema succeeded; the incorrect business value must be caught by evaluation or authoritative checks; for this request, outdated retrieval evidence is the missing piece, not semantic failure isolation.

E:

Bad-request errors indicate an invalid request contract; retrying unchanged payloads will not repair schema or parameter mistakes; it changes invalid-request recovery; it does not prove outdated retrieval evidence in this case.

Question 210

A user reports that a chat reply stopped halfway through a paragraph. Which response metadata matters?

  1. For this workflow, examine the retrieval corpus version and ranking before changing the model.
  2. Inspect stop_reason and output token limits in the recorded API trace.
  3. Review request validation details and the supplied schema before retrying.
  4. Introduce quota-aware throttling and bounded backoff rather than immediate repeated retries.

Explanation

Correct Answer

B:

A max_tokens stop distinguishes deliberate truncation from a natural end_turn and directs the next fix. Read stop_reason and actual output-token usage to tell a generation limit from network interruption or UI display truncation.

Incorrect Answers

A:

A model can reason consistently over stale evidence; source freshness is a data pipeline defect; this implements outdated retrieval evidence rather than the demanded incomplete generation trace.

C:

Bad-request errors indicate an invalid request contract; retrying unchanged payloads will not repair schema or parameter mistakes; for incomplete generation trace, a control centered on invalid-request recovery is not the appropriate substitute.

D:

Rate limiting signals excessive request volume or tokens; paced retries and concurrency controls address that condition; the incomplete generation trace criterion differs from traffic throttling triage; the gap persists.

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