Anthropic CCDV-F (Claude Certified Developer - Foundations) Practice Test 16

Question 150

An agent consumes attacker-controlled web pages and has access to protected tools. Which response is most defensible under these conditions?

  1. Perform security abuse tests
  2. Use regression evaluations for prompts
  3. Run unit tests for a tool argument validator
  4. Categorize and triage production incidents

Explanation

Correct Answer

A:

Red-team cases test injection resistance and authorization enforcement at trust boundaries, although ordinary happy-path tests will not expose deliberate instruction attacks.

Incorrect Answers

B:

A frozen dataset and grading rubric compare model outputs before and after the proposed prompt revision; however, the critical mechanism here is that red-team cases test injection resistance and authorization enforcement at trust boundaries, and passing a linter does not measure semantic output quality.

C:

Unit cases exercise the validator's accept/reject logic without live API traffic; the requirement instead turns on this behavior: red-team cases test injection resistance and authorization enforcement at trust boundaries; a complete agent test is slower and less precise for locating pure validation bugs.

D:

Incident classification routes problems to the correct service, integration, or prompt team; this cannot substitute for the behavior that red-team cases test injection resistance and authorization enforcement at trust boundaries, because treating every failure as bad model reasoning conceals infrastructure issues.

Question 151

A generation stops with structured tool requests instead of a final customer-facing explanation. What would best address the stated failure mode?

  1. Handle pause_turn as a resumable server operation
  2. Trace the external tool service
  3. Handle end_turn as normal completion
  4. Handle tool_use by dispatching the requested tool

Explanation

Correct Answer

D:

The application executes authorized client tools and continues the conversation with their results; a crucial qualification is that presenting tool-use JSON directly to the user skips the required integration step.

Incorrect Answers

A:

The client follows the documented continuation flow for that intermediate condition; in contrast, the application executes authorized client tools and continues the conversation with their results, while a paused operation is not necessarily a finished user answer.

B:

A distributed trace follows the tool dispatcher to the failing downstream dependency; this addresses a separate concern, not the fact that the application executes authorized client tools and continues the conversation with their results; rewriting system instructions will not repair an unavailable invoice endpoint.

C:

The caller can finalize the output without fabricating further tool actions; here, the controlling capability is that the application executes authorized client tools and continues the conversation with their results, but additional automatic continuations risk duplicating the answer.

Question 152

A tool result can be retried after a network timeout without creating the same ticket twice. Which implementation mechanism is needed?

  1. Assign action limits in the agent harness
  2. Keep model context limited to relevant evidence
  3. Assign tool side effects to an execution worker
  4. Use idempotent side-effect handling

Explanation

Correct Answer

D:

Stable operation identifiers and deduplication at the action service protect against duplicate writes; in production, retrying a model-generated tool call blindly can repeat the business operation.

Incorrect Answers

A:

Host-controlled counters and deadlines bound execution regardless of what Claude decides next; this addresses a separate concern, not the fact that stable operation identifiers and deduplication at the action service protect against duplicate writes; prompt instructions alone cannot enforce a hard external quota.

B:

A curated evidence window preserves useful facts while discarding noisy intermediate transcripts; here, the controlling capability is that stable operation identifiers and deduplication at the action service protect against duplicate writes, but pruning must preserve commitments and results required for downstream steps.

C:

A tightly permissioned executor maps authorized operations into deterministic tool calls; the missing capability is more specific: stable operation identifiers and deduplication at the action service protect against duplicate writes; broad access in the planning component defeats the intended separation of duties.

Question 153

A shared monorepo needs separate backend and UI coding conventions, while a developer proposes pasting deployment credentials into the instructions file. Which TWO repository practices address instruction scope and secret exposure? Choose TWO.

  1. Initialize repository guidance deliberately
  2. Keep secrets outside coding-agent memory files
  3. Use CLAUDE.md for repository instructions
  4. Use a specialized subagent for focused review
  5. Place narrow guidance close to the relevant code

Explanation

Correct Answers

B:

Credentials remain in protected stores or environment controls rather than committed textual guidance; the important constraint is that a shared file can be copied, logged, and inherited by unintended readers.

E:

Scoped project guidance can communicate local conventions without overloading every task with unrelated instructions; its operational boundary is that conflicting guidance across scopes should be reviewed for effective precedence.

Incorrect Answers

A:

An initialization workflow can help establish and review useful CLAUDE.md guidance for the repository, whereas the requirement calls for another behavior: credentials remain in protected stores or environment controls rather than committed textual guidance; generated notes must be checked before being treated as authoritative development policy.

C:

A project instruction file provides contextual guidance when Claude Code works in that repository; the requirement instead turns on this behavior: credentials remain in protected stores or environment controls rather than committed textual guidance; it is explanatory context rather than a security boundary enforcing tool privileges.

D:

A subagent delegates bounded work with appropriate tools and returns results to the parent; this cannot substitute for the behavior that credentials remain in protected stores or environment controls rather than committed textual guidance, because the delegated agent should not inherit broader privileges than necessary.

Question 154

A direct HTTP client must keep its request contract consistent with the selected API revision. What change is best aligned with the goal?

  1. Pin a tested exact model version when necessary
  2. Parse an HTTP status before decoding success payload
  3. Use request headers for API authentication
  4. Set the requested API version header where required

Explanation

Correct Answer

D:

Version negotiation constrains accepted fields and response behavior for an endpoint, although omitting required version information or assuming an arbitrary revision can break compatibility.

Incorrect Answers

A:

Version pinning keeps the model selection stable until the team deliberately upgrades and retests; the missing capability is more specific: version negotiation constrains accepted fields and response behavior for an endpoint; the pinned model may later require migration if the provider retires it.

B:

The caller distinguishes transport/application failure from successful responses before parsing their bodies; by comparison, the intended solution relies on this: version negotiation constrains accepted fields and response behavior for an endpoint; attempting to deserialize every response as a Claude message conceals gateway faults.

C:

The service authenticates the API credential through its designated HTTP headers; the needed capability is different—version negotiation constrains accepted fields and response behavior for an endpoint—and including a secret in model messages would expose it to prompt handling and logging.

Question 155

A classifier needs less variation between otherwise identical inputs while preserving model-based interpretation. What change is best aligned with the goal?

  1. Use low-temperature settings when supported for consistency
  2. Use a fixed regression dataset for stable comparisons
  3. Avoid assuming byte-identical responses
  4. Use extra reasoning on high-complexity tasks

Explanation

Correct Answer

A:

Lower randomness can reduce sampling variability in supported configurations; notably, it does not guarantee deterministic output or eliminate model errors.

Incorrect Answers

B:

Repeated evaluation over the same representative cases reveals distributional quality differences; in contrast, lower randomness can reduce sampling variability in supported configurations, while one identical answer does not prove full determinism across deployments.

C:

Generative sampling and platform execution can cause output variability, so compare validated properties instead; this addresses a separate concern, not the fact that lower randomness can reduce sampling variability in supported configurations; a hard textual hash is usually an unsuitable semantic acceptance metric.

D:

Supported thinking or effort settings allocate greater reasoning capacity to difficult deliberation; here, the controlling capability is that lower randomness can reduce sampling variability in supported configurations, but added reasoning can increase latency and cost and still requires factual validation.

Question 156

A repeated long system prompt is cached but some calls refresh the prefix and others hit it. Which response is most defensible under these conditions?

  1. Compare cache TTL with request arrival pattern
  2. Aggregate usage across multi-step tasks
  3. Track separate model pricing assumptions
  4. Include cache reads and writes in cost calculations

Explanation

Correct Answer

D:

Cache creation and cache retrieval have distinct billing treatment that affects economics, subject to this requirement: treating every token as an uncached input misstates true spend.

Incorrect Answers

A:

Lifetime selection must account for actual reuse frequency and write overhead; the requirement instead turns on this behavior: cache creation and cache retrieval have distinct billing treatment that affects economics; a longer cache is not automatically economical for rarely repeated prompts.

B:

Summing usage across the whole trace shows the true cost of satisfying one business request; this cannot substitute for the behavior that cache creation and cache retrieval have distinct billing treatment that affects economics, because only counting the final call omits earlier orchestration expenses.

C:

Per-tier pricing and request distribution are needed for realistic cost forecasts; in contrast, cache creation and cache retrieval have distinct billing treatment that affects economics, while a single blended rate can conceal costly high-tier escalation patterns.

Question 157

A small application has one model call and no need for a framework's graph or dependency injection. Which action most directly closes the gap?

  1. Finish with a user-facing answer after the last action
  2. Use typed PydanticAI-style dependencies
  3. Use a light direct API integration
  4. Use Strands-style model-driven agents

Explanation

Correct Answer

C:

A direct client minimizes abstraction and makes the request path easy to inspect—with this limitation: it will need explicit orchestration if tool calls and branching emerge later.

Incorrect Answers

A:

The model synthesizes a final response once no additional client tool request is needed, whereas the requirement calls for another behavior: a direct client minimizes abstraction and makes the request path easy to inspect; a raw tool result is not automatically a contextualized customer answer.

B:

A typed framework aligns model interactions with application data models and testable dependencies; however, the critical mechanism here is that a direct client minimizes abstraction and makes the request path easy to inspect, and types validate shape, not the truth of every generated factual assertion.

D:

An agent framework coordinates tool selection while exposing configuration and observability hooks; this cannot substitute for the behavior that a direct client minimizes abstraction and makes the request path easy to inspect, because tool permission and termination policies remain the integrator's responsibility.

Question 158

An API consumer needs measured input and output token quantities for per-request accounting. Which development approach should be selected?

  1. Read usage counts from the response
  2. Inspect cache read and creation usage
  3. Parse typed content blocks rather than raw concatenation
  4. Store response metadata with the trace

Explanation

Correct Answer

A:

Response usage fields provide the token measures used for billing and diagnostics, although character count is not an exact substitute for model tokenization.

Incorrect Answers

B:

Cache usage metrics distinguish newly written content from tokens read from cache; however, the critical mechanism here is that response usage fields provide the token measures used for billing and diagnostics, and an assumed hit without usage evidence can hide prompt-prefix changes.

C:

Typed block handling preserves whether an element is displayable text or a tool request; the requirement instead turns on this behavior: response usage fields provide the token measures used for billing and diagnostics; blindly concatenating heterogeneous blocks can lose tool metadata and semantics.

D:

Structured metadata and correlation records provide context for debugging and cost attribution; this cannot substitute for the behavior that response usage fields provide the token measures used for billing and diagnostics, because logging only the visible answer removes information needed to identify stop or usage anomalies.

Question 159

A new third-party integration requests broad repository access despite needing read-only metadata. How should the implementation handle this need?

  1. Pin and test MCP server compatibility
  2. Review plugin permissions before activation
  3. Fail closed on missing safety configuration
  4. Declare integration dependencies explicitly

Explanation

Correct Answer

B:

Least-privilege permission review limits what installed integrations can access, although a friendly plugin description is not a security audit.

Incorrect Answers

A:

Version tracking and integration tests preserve compatibility at the tool boundary; the needed capability is different—least-privilege permission review limits what installed integrations can access—and automatically adopting an untested update can break dispatch or authorization.

C:

Startup validation blocks unsafe execution rather than falling back to all permissions; however, the critical mechanism here is that least-privilege permission review limits what installed integrations can access, and the absence of a guardrail must not silently widen access.

D:

Configuration and onboarding document required tool integrations and supported versions; the requirement instead turns on this behavior: least-privilege permission review limits what installed integrations can access; assuming every client has identical external tools makes behavior unpredictable.

Question 160

Customers can attach files and the service must reject unsupported or excessively large payloads. What should be implemented to solve this problem?

  1. Redact unnecessary sensitive content before submission
  2. Correct a malformed request before resubmission
  3. Validate MIME type and size before upload
  4. Use a supported file or document input method

Explanation

Correct Answer

C:

Server-side checks bound resource use and ensure supported input formats before model submission; the important constraint is that client-provided extensions and names are not reliable content validation alone.

Incorrect Answers

A:

Preprocessing removes irrelevant secrets and personal fields prior to sending the image content; the needed capability is different—server-side checks bound resource use and ensure supported input formats before model submission—and asking Claude to ignore a secret still transmits that data to the service.

B:

Client-side schema validation fixes the payload and prevents fruitless retries, whereas the requirement calls for another behavior: server-side checks bound resource use and ensure supported input formats before model submission; a network backoff cannot make an invalid request body well formed.

D:

The integration supplies supported document content using the relevant API capability and access rules; the requirement instead turns on this behavior: server-side checks bound resource use and ensure supported input formats before model submission; an arbitrary URL or filename does not automatically become readable source material.

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