Anthropic CCDV-F (Claude Certified Developer - Foundations) Practice Test 4

Question 30

A research agent must make a final answer only after three data sources have been queried successfully. How should completion be determined?

  1. Track successful source results in harness state and guard finalization on that state.
  2. Leave the tool loop and deliver the completed assistant response.
  3. Provide a structured tool_result describing the failure without fabricating inventory.
  4. Save the session identifiers and supported state needed for resumption.

Explanation

Correct Answer

A:

Model confidence is not evidence of completed external calls; explicit state permits an auditable completion predicate.

Incorrect Answers

B:

End_turn indicates the current assistant turn has finished, so blindly issuing another tool call would add unnecessary steps; the effect concerns recognizing completion; the success condition is verifiable completion criteria.

C:

Reporting the real error lets the model revise its plan or explain the limitation; silent invention hides a failed integration; the consequence is improved surfacing client-tool errors, not verified verifiable completion criteria.

D:

Long-running interaction cannot rely on in-memory local variables across process lifetimes; session continuity needs an explicit persistence design; even if session resumption state succeeds, verifiable completion criteria still needs separate verification.

Question 31

A deployed agent accumulates expensive calls when a tool returns the same transient error. What resilience mechanism belongs outside the model?

  1. Establish and validate a typed input schema at the tool boundary.
  2. Deliver a structured tool_result describing the failure without fabricating inventory.
  3. Run both calls and return a result associated with each distinct tool_use ID.
  4. Cap retries and apply backoff with a terminal failure path.

Explanation

Correct Answer

D:

Retry policy is operational control of the executor, not a reasoning preference; limits prevent a looping agent from exhausting cost and capacity.

Incorrect Answers

A:

Explicit fields and runtime validation reject malformed inputs before touching external systems; model-generated arguments are not trusted by default; when the goal is bounded transient error retries, optimizing tool argument validation misses that operational boundary.

B:

Reporting the real error lets the model revise its plan or explain the limitation; silent invention hides a failed integration; the effect concerns surfacing client-tool errors; the success condition is bounded transient error retries.

C:

Multiple tool-use blocks may occur together; correlation is required so Claude can attribute observations to the correct requests; for this request, bounded transient error retries is the missing piece, not parallel tool result correlation.

Question 32

A team weighs a hosted agent service against running an agent SDK in its own containers. The key requirement is a proprietary in-network tool that cannot be exposed externally. What should lead the choice?

  1. Document and validate a typed input schema at the tool boundary.
  2. Carry out both calls and return a result associated with each distinct tool_use ID.
  3. Decode content blocks by type rather than treating the response as one string.
  4. Apply a pre-execution hook or permission gate that can deny disallowed commands.
  5. Favor an execution environment with authorized network access to the private tool.

Explanation

Correct Answer

E:

Deployment placement determines reachability and data boundaries; neither agent quality nor an SDK name can bypass those infrastructure constraints.

Incorrect Answers

A:

Explicit fields and runtime validation reject malformed inputs before touching external systems; model-generated arguments are not trusted by default; the scenario's controlling requirement is private-network deployment fit, not the related tool argument validation concern.

B:

Multiple tool-use blocks may occur together; correlation is required so Claude can attribute observations to the correct requests; the consequence is improved parallel tool result correlation, not verified private-network deployment fit.

C:

Tool_use blocks carry structured inputs and IDs, not user-facing prose; the harness must branch on content type and stop reason; a remedy for content-block-aware parsing is not a substitute for private-network deployment fit.

D:

A deterministic inspection point can block a call before side effects occur; a prompt-based reminder cannot replace the gate; for this request, private-network deployment fit is the missing piece, not command pre-execution denial.

Question 33

A tool call would mutate an order, but the same request might be retried after a network timeout. What should the executor add?

  1. Perform the requested client tools and return linked tool_result blocks in a new user message.
  2. Perform both calls and return a result associated with each distinct tool_use ID.
  3. Deploy deterministic hooks around the tool lifecycle.
  4. Use an idempotency key or deduplication record for the mutation.

Explanation

Correct Answer

D:

A repeated call after an uncertain outcome must not create duplicate side effects; idempotent execution protects state across retries.

Incorrect Answers

A:

Client tools do not run merely because the model emitted a tool_use block; the caller must dispatch them and preserve the tool_use identifiers; it follows handling a tool request constraints but leaves retry-safe external writes unspecified.

B:

Multiple tool-use blocks may occur together; correlation is required so Claude can attribute observations to the correct requests; the effect concerns parallel tool result correlation; the success condition is retry-safe external writes.

C:

Hook handlers are executed by the framework at defined events and can record metadata independently of model-generated prose; it improves deterministic tool event instrumentation only, leaving retry-safe external writes unproven.

Question 34

An agent must coordinate 15-minute investigations even when an HTTP worker is recycled. What hosting property is needed?

  1. Perform file tools under filesystem permissions that deny that directory.
  2. Reduce or filter tool observations before reinserting them into context.
  3. Implement a custom Messages API harness that owns dispatch and state.
  4. Persist job state outside ephemeral request workers and resume from checkpoints.

Explanation

Correct Answer

D:

Long-running tasks must survive worker termination; relying on a single request's memory risks losing progress and repeating side effects.

Incorrect Answers

A:

Operating-system boundaries block writes even if the model chooses an unsafe path; a natural-language prohibition alone is not sufficient; this implements filesystem scope restriction rather than the demanded durable long-running agent execution.

B:

Returning only the relevant log evidence preserves the token budget and reduces distraction while retaining pointers to original records; when the goal is durable long-running agent execution, optimizing controlled observation size misses that operational boundary.

C:

Direct orchestration exposes scheduling, persistence, error handling, and tool-execution boundaries; it trades framework convenience for precise integration control; this may help custom orchestrator control but cannot by itself deliver durable long-running agent execution.

Question 35

The model offers a free-text command to a tool that expects a structured issue identifier. What should the implementer do?

  1. For this workflow, remove comment-posting authority from the agent's available tools.
  2. Introduce deterministic hooks around the tool lifecycle.
  3. Define and validate a typed input schema at the tool boundary.
  4. Apply an idempotency key or deduplication record for the mutation.

Explanation

Correct Answer

C:

Explicit fields and runtime validation reject malformed inputs before touching external systems; model-generated arguments are not trusted by default.

Incorrect Answers

A:

The harness can deny the capability entirely; prompt instructions or post-hoc logs cannot enforce a missing permission boundary; when the goal is tool argument validation, optimizing limiting agent write authority misses that operational boundary.

B:

Hook handlers are executed by the framework at defined events and can record metadata independently of model-generated prose; this implements deterministic tool event instrumentation rather than the demanded tool argument validation.

D:

A repeated call after an uncertain outcome must not create duplicate side effects; idempotent execution protects state across retries; it follows retry-safe external writes constraints but leaves tool argument validation unspecified.

Question 36

An agent keeps consuming its context with full log dumps from each diagnostic call. What harness improvement matters most?

  1. Summarize or filter tool observations before reinserting them into context.
  2. Record successful source results in harness state and guard finalization on that state.
  3. At the application layer, cap retries and apply backoff with a terminal failure path.
  4. Process content blocks by type rather than treating the response as one string.

Explanation

Correct Answer

A:

Returning only the relevant log evidence preserves the token budget and reduces distraction while retaining pointers to original records.

Incorrect Answers

B:

Model confidence is not evidence of completed external calls; explicit state permits an auditable completion predicate; when the goal is controlled observation size, optimizing verifiable completion criteria misses that operational boundary.

C:

Retry policy is operational control of the executor, not a reasoning preference; limits prevent a looping agent from exhausting cost and capacity; this implements bounded transient error retries rather than the demanded controlled observation size.

D:

Tool_use blocks carry structured inputs and IDs, not user-facing prose; the harness must branch on content type and stop reason; the consequence is improved content-block-aware parsing, not verified controlled observation size.

Question 37

A coding agent has permission to change files but must never touch the deployment directory. Which enforcement is strongest?

  1. Run file tools under filesystem permissions that deny that directory.
  2. Maintain the session identifiers and supported state needed for resumption.
  3. Aggregate or filter tool observations before reinserting them into context.
  4. Invoke the requested client tools and return linked tool_result blocks in a new user message.

Explanation

Correct Answer

A:

Operating-system boundaries block writes even if the model chooses an unsafe path; a natural-language prohibition alone is not sufficient.

Incorrect Answers

B:

Long-running interaction cannot rely on in-memory local variables across process lifetimes; session continuity needs an explicit persistence design; this may help session resumption state but cannot by itself deliver filesystem scope restriction.

C:

Returning only the relevant log evidence preserves the token budget and reduces distraction while retaining pointers to original records; the consequence is improved controlled observation size, not verified filesystem scope restriction.

D:

Client tools do not run merely because the model emitted a tool_use block; the caller must dispatch them and preserve the tool_use identifiers; for filesystem scope restriction, this instead manages handling a tool request; the requested decision remains unresolved.

Question 38

An agent encounters a user cancellation while waiting for a long external tool. What should the harness provide?

  1. Select an execution environment with authorized network access to the private tool.
  2. Invoke both calls and return a result associated with each distinct tool_use ID.
  3. Propagate cancellation and cleanly stop outstanding work where possible.
  4. Isolate external tools so deterministic test doubles can replace real connectors.
  5. Choose a pre-execution hook or permission gate that can deny disallowed commands.

Explanation

Correct Answer

C:

The execution layer owns task lifecycle and resource cleanup; another model message cannot reliably cancel a pending network operation.

Incorrect Answers

A:

Deployment placement determines reachability and data boundaries; neither agent quality nor an SDK name can bypass those infrastructure constraints; changing private-network deployment fit does not establish cancellation lifecycle in this application.

B:

Multiple tool-use blocks may occur together; correlation is required so Claude can attribute observations to the correct requests; the cancellation lifecycle criterion differs from parallel tool result correlation; the gap persists.

D:

Dependency boundaries allow simulated responses and failure injection, making tool-loop behavior reproducible without real side effects; it improves testing agent tool adapters only, leaving cancellation lifecycle unproven.

E:

A deterministic inspection point can block a call before side effects occur; a prompt-based reminder cannot replace the gate; testing cancellation lifecycle would still fail after improving command pre-execution denial.

Question 39

A team wishes to test agent behavior without contacting billing systems. How should the harness be structured?

  1. Brief or filter tool observations before reinserting them into context.
  2. Record the session identifiers and supported state needed for resumption.
  3. Supply a structured tool_result describing the failure without fabricating inventory.
  4. Abstract external tools so deterministic test doubles can replace real connectors.

Explanation

Correct Answer

D:

Dependency boundaries allow simulated responses and failure injection, making tool-loop behavior reproducible without real side effects.

Incorrect Answers

A:

Returning only the relevant log evidence preserves the token budget and reduces distraction while retaining pointers to original records; because controlled observation size is its concern, testing agent tool adapters remains outside its effect here.

B:

Long-running interaction cannot rely on in-memory local variables across process lifetimes; session continuity needs an explicit persistence design; it improves session resumption state only, leaving testing agent tool adapters unproven.

C:

Reporting the real error lets the model revise its plan or explain the limitation; silent invention hides a failed integration; for testing agent tool adapters, a control centered on surfacing client-tool errors is not the appropriate substitute.

Question 40

An assistant spends half its context replaying a month of old chat when only three decisions still matter. What is the best memory pattern?

  1. Keep a compact, explicit summary of durable decisions and retrieve details when needed.
  2. Keep only the most recent sentence and discard the three durable decisions.
  3. As an alternative, compute aggregates with a tool and pass only validated results back to Claude.
  4. Provide the web-reading worker limited tools and treat its findings as data, not instructions.

Explanation

Correct Answer

A:

Separating long-term task facts from raw transcript prevents context waste while preserving the information needed for new decisions.

Incorrect Answers

B:

A last-sentence-only summary loses the decisions the scenario explicitly still needs. The model may save tokens initially, but the next turn would lack essential task constraints and commit history.

C:

Moving mechanical data processing into code avoids token-heavy raw inputs while keeping Claude focused on interpretation; even if offloading deterministic computation succeeds, compact durable memory still needs separate verification.

D:

Isolating untrusted retrieval prevents document text from inheriting authority over high-impact actions in the parent process; for compact durable memory, a control centered on untrusted research containment is not the appropriate substitute.

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