Anthropic CCDV-F (Claude Certified Developer - Foundations) Practice Test 59

Question 580

Suspicious inputs begin triggering unauthorized tool-call attempts and the product must limit impact. Which development approach should be selected?

  1. Queue nonurgent tasks during provider outages
  2. Disable privileged tools during security incidents
  3. Keep user instructions subordinate to security policy
  4. Return a verified FAQ when inference is unavailable

Explanation

Correct Answer

B:

A feature flag or policy change can revoke risky execution paths while investigation continues; a crucial qualification is that a prompt-only warning cannot reliably disable an exposed backend permission.

Incorrect Answers

A:

Durable storage retains jobs for resumed processing once inference is available; here, the controlling capability is that a feature flag or policy change can revoke risky execution paths while investigation continues, but an in-memory list in a failed worker may lose pending work.

C:

The application enforces identity and authorization checks despite the request's wording; by comparison, the intended solution relies on this: a feature flag or policy change can revoke risky execution paths while investigation continues; requests from end users do not grant new backend privileges.

D:

A controlled cache of approved responses preserves basic service where exact matches exist; the needed capability is different—a feature flag or policy change can revoke risky execution paths while investigation continues—and the system must avoid pretending a cached answer is newly generated or universally applicable.

Question 581

A web page used for research contains a hidden instruction to call a destructive tool. Which measure directly addresses the problem?

  1. Use a safe fallback for ambiguous instructions
  2. Validate prompt variables before interpolation
  3. Treat retrieved source text as evidence
  4. Treat attacker content as quoted data only

Explanation

Correct Answer

D:

The application preserves the page as untrusted input and excludes it from control instructions; in production, a malicious instruction can still be persuasive, so execution permissions remain bounded.

Incorrect Answers

A:

The assistant asks for missing scope or routes the decision to an authorized operator; this addresses a separate concern, not the fact that the application preserves the page as untrusted input and excludes it from control instructions; guessing a target object could cause irreversible cross-customer damage.

B:

Input checks constrain length, types, and permitted resource scopes before prompt assembly; here, the controlling capability is that the application preserves the page as untrusted input and excludes it from control instructions, but merely surrounding an unsafe value with quotation marks does not establish authorization.

C:

External content is data to quote or analyze only within the authorized task; the missing capability is more specific: the application preserves the page as untrusted input and excludes it from control instructions; a source document has no standing to instruct tool use or access policy.

Question 582

A customer approval status must remain correct even after dozens of unrelated chat turns. Which technical control is appropriate?

  1. Reassert the current task objective concisely
  2. Keep a structured conversation summary
  3. Segment work into explicit milestones
  4. Preserve immutable business facts in structured state

Explanation

Correct Answer

D:

The application records authoritative facts outside model-generated summaries and supplies them when needed, although conversation recollection can drift or omit earlier commitments.

Incorrect Answers

A:

Reintroducing the active goal and constraints guides continuation back to the approved scope; the missing capability is more specific: the application records authoritative facts outside model-generated summaries and supplies them when needed; repetition cannot override the need to preserve factual state accurately.

B:

A summary stores salient decisions, constraints, and references in a reduced representation; by comparison, the intended solution relies on this: the application records authoritative facts outside model-generated summaries and supplies them when needed; compression may omit nuance and needs checking for critical obligations.

C:

Milestones with acceptance criteria limit aimless elaboration and provide reliable checkpoints; the needed capability is different—the application records authoritative facts outside model-generated summaries and supplies them when needed—and a plan without tested completion criteria will still drift.

Question 583

An audit pipeline expects exactly the same prose for every repeat inference on unchanged input. Which of the following would work as specified?

  1. Reserve enough output tokens for the desired result
  2. Avoid assuming byte-identical responses
  3. Use a fixed regression dataset for stable comparisons
  4. Use a more exploratory sampling setting when supported

Explanation

Correct Answer

B:

Generative sampling and platform execution can cause output variability, so compare validated properties instead; the important constraint is that a hard textual hash is usually an unsuitable semantic acceptance metric.

Incorrect Answers

A:

The configured generation maximum must be consistent with the required output length, whereas the requirement calls for another behavior: generative sampling and platform execution can cause output variability, so compare validated properties instead; raising it arbitrarily can increase possible cost without improving content accuracy.

C:

Repeated evaluation over the same representative cases reveals distributional quality differences; the requirement instead turns on this behavior: generative sampling and platform execution can cause output variability, so compare validated properties instead; one identical answer does not prove full determinism across deployments.

D:

Sampling parameters can trade predictability for variety in compatible model modes; this cannot substitute for the behavior that generative sampling and platform execution can cause output variability, so compare validated properties instead, because increased randomness may degrade factual precision and schema reliability.

Question 584

An agent needs to query an internal ledger reachable only from a private application worker. Which production safeguard is most relevant?

  1. Use a local function dispatcher for private integrations
  2. Use the Claude Agent SDK
  3. Apply a sandbox to untrusted generated code
  4. Use Anthropic-hosted managed execution where supported

Explanation

Correct Answer

A:

Application code executes a narrowly defined local tool and returns selected results to Claude; notably, the API itself does not have arbitrary access to a customer's private internal network.

Incorrect Answers

B:

The SDK provides higher-level agent controls and integrations while the application supplies policies and oversight; in contrast, application code executes a narrowly defined local tool and returns selected results to Claude, while prebuilt orchestration does not remove responsibility for secrets and tool permission boundaries.

C:

Isolation constrains filesystem, processes, secrets, and network access independently of model reasoning; this addresses a separate concern, not the fact that application code executes a narrowly defined local tool and returns selected results to Claude; a prompt claiming code is safe cannot replace runtime containment.

D:

Managed runtime can reduce infrastructure ownership for supported agent operations; here, the controlling capability is that application code executes a narrowly defined local tool and returns selected results to Claude, but hosted deployment still requires review of data paths, available tools, and platform limitations.

Question 585

A router sends most tickets to a small model and complex cases to a larger one. Which solution should the engineer adopt?

  1. Include cache reads and writes in cost calculations
  2. Compute cost from input and output token usage
  3. Track separate model pricing assumptions
  4. Evaluate a smaller tier against quality requirements

Explanation

Correct Answer

C:

Per-tier pricing and request distribution are needed for realistic cost forecasts—with this limitation: a single blended rate can conceal costly high-tier escalation patterns.

Incorrect Answers

A:

Cache creation and cache retrieval have distinct billing treatment that affects economics, whereas the requirement calls for another behavior: per-tier pricing and request distribution are needed for realistic cost forecasts; treating every token as an uncached input misstates true spend.

B:

Per-model input and output token rates combine with measured usage for accurate costing; however, the critical mechanism here is that per-tier pricing and request distribution are needed for realistic cost forecasts, and number of messages alone is not an adequate unit of resource consumption.

D:

Task-specific benchmarks may show a lower-priced model meets the acceptance threshold; this cannot substitute for the behavior that per-tier pricing and request distribution are needed for realistic cost forecasts, because selecting a model by cost alone can generate costly correction work.

Question 586

A provider adds a noncritical response property and clients must avoid unnecessary failure. Which option best fits the described workflow?

  1. Preserve UTF-8 content safely
  2. Negotiate supported content types properly
  3. Use typed response variants where applicable
  4. Preserve future-compatible unknown fields cautiously

Explanation

Correct Answer

D:

Tolerant read behavior can ignore unknown optional metadata while validating required fields; silent acceptance of unknown privileged action types would be unsafe.

Incorrect Answers

A:

Proper encoding preserves characters consistently between client, API, and storage; here, the controlling capability is that tolerant read behavior can ignore unknown optional metadata while validating required fields, but re-encoding bytes as the wrong character set can corrupt meaning.

B:

Content-type checks help the parser distinguish valid API data from an unexpected gateway body; the missing capability is more specific: tolerant read behavior can ignore unknown optional metadata while validating required fields; blind JSON parsing obscures transport failures.

C:

A discriminated union or type dispatch handles text, tool calls, and other supported blocks correctly; by comparison, the intended solution relies on this: tolerant read behavior can ignore unknown optional metadata while validating required fields; treating all blocks as plain strings discards structural information.

Question 587

A drafting agent needs independent scoring of citations and completeness before a response is shipped. What is the best fit for the described situation?

  1. Delegate retrieval to a research worker
  2. Assign tool side effects to an execution worker
  3. Run an explicit fixed-stage workflow
  4. Delegate quality checking to a critic worker

Explanation

Correct Answer

D:

An evaluator examines the artifact against explicit tests instead of composing the primary answer; notably, critique must have actionable criteria and should not be equated with guaranteed correctness.

Incorrect Answers

A:

A research worker gathers references and returns a compact evidence set to the lead, whereas the requirement calls for another behavior: an evaluator examines the artifact against explicit tests instead of composing the primary answer; it should not be allowed to publish the final legal conclusion as an independent authority.

B:

A tightly permissioned executor maps authorized operations into deterministic tool calls; however, the critical mechanism here is that an evaluator examines the artifact against explicit tests instead of composing the primary answer, and broad access in the planning component defeats the intended separation of duties.

C:

Application code controls the predetermined stage transitions and makes them reproducible in tests; the requirement instead turns on this behavior: an evaluator examines the artifact against explicit tests instead of composing the primary answer; it cannot decide to introduce an unplanned research step after unexpected evidence appears.

Question 588

Every insurance renewal must follow a published sequence that auditors can trace exactly. Which development approach should be selected?

  1. Prefer a workflow for regulated approvals
  2. Prefer a hybrid workflow-agent design
  3. Prefer an agent for exploratory diagnostics
  4. Execute a parallel fan-out workflow

Explanation

Correct Answer

A:

Explicit branches and immutable records make the process stable under regulatory review; its operational boundary is that free-form agent planning is unsuitable for mandatory stage-order guarantees.

Incorrect Answers

B:

Code preserves mandatory outer checkpoints while an agent explores only within an approved segment; this addresses a separate concern, not the fact that explicit branches and immutable records make the process stable under regulatory review; failure to isolate the exploratory stage could let it bypass the required controls.

C:

An agent observes evidence and selects the next tool based on new findings; here, the controlling capability is that explicit branches and immutable records make the process stable under regulatory review, but it still requires restricted read-only tools and a bounded investigation scope.

D:

Independent branches can run at once and a reducer collects their outputs, lowering wall-clock time; the missing capability is more specific: explicit branches and immutable records make the process stable under regulatory review; parallel execution is unsafe if branches mutate the same resource without coordination.

Question 589

A developer modifies JSON schema checks and needs immediate feedback on invalid inputs. Which implementation mechanism is needed?

  1. Perform security abuse tests
  2. Run unit tests for a tool argument validator
  3. Run an end-to-end test using a sandbox
  4. Require pull-request review

Explanation

Correct Answer

B:

Unit cases exercise the validator's accept/reject logic without live API traffic; its operational boundary is that a complete agent test is slower and less precise for locating pure validation bugs.

Incorrect Answers

A:

Red-team cases test injection resistance and authorization enforcement at trust boundaries; this cannot substitute for the behavior that unit cases exercise the validator's accept/reject logic without live API traffic, because ordinary happy-path tests will not expose deliberate instruction attacks.

C:

An end-to-end test exercises the public request path, model call, dispatcher, and response formatter; this addresses a separate concern, not the fact that unit cases exercise the validator's accept/reject logic without live API traffic; isolated component success cannot prove the integration as a whole functions.

D:

Peer review examines diff, tests, and ownership before a controlled merge; here, the controlling capability is that unit cases exercise the validator's accept/reject logic without live API traffic, but direct unreviewed edits reduce accountability and increase regression risk.

Question 590

A research worker should search current public sources without possessing a production deployment credential. What would best address the stated failure mode?

  1. Use structured tool schemas for business actions
  2. Add a human gate for high-stakes changes
  3. Use a custom client tool for a private business operation
  4. Limit external tools to necessary operations

Explanation

Correct Answer

D:

Minimal capabilities reduce the harm of prompt injection or mistaken planning; more available tools do not guarantee better task performance.

Incorrect Answers

A:

Typed schemas and dispatcher checks convert an intention into a constrained operation; here, the controlling capability is that minimal capabilities reduce the harm of prompt injection or mistaken planning, but general Skill text cannot replace a server-enforced action interface.

B:

Approval controls and audit records separate planning from privileged commitment; the missing capability is more specific: minimal capabilities reduce the harm of prompt injection or mistaken planning; a system prompt telling the model to act cautiously does not itself block execution.

C:

A user-defined tool schema plus local dispatcher safely bridges the model to that system; by comparison, the intended solution relies on this: minimal capabilities reduce the harm of prompt injection or mistaken planning; the developer must operate the function and enforce its permissions.

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