Anthropic CCDV-F (Claude Certified Developer - Foundations) Practice Test 9

Question 80

A developer is reviewing alternatives for an inference routing feature that must stop a single customer from consuming the entire API allowance during a runaway workflow. Which option is a defensible fit?

  1. Use batch pricing where deadline permits
  2. Keep volatile fields after reusable context
  3. Set budget alerts and tenant caps
  4. Cache a stable shared prefix

Explanation

Correct Answer

C:

Accounting plus enforced limits bounds exposure to unexpected traffic and agent loops; this behavior directly supports the need to stop a single customer from consuming the entire API allowance during a runaway workflow. Usage dashboards alone only report overspend after the fact.

Incorrect Answers

A:

Asynchronous batch processing can offer a cost-effective throughput path under supported terms; this applies to the different goal to lower spend for large scheduled classification jobs that do not need immediate responses, not the requirement to stop a single customer from consuming the entire API allowance during a runaway workflow; a night job should not be optimized with live streaming unless its users need it.

B:

When an application needs to prevent request timestamps from invalidating a costly static prefix, stable instruction and reference sections should precede changing per-request details for effective prefix reuse; the current capability must instead stop a single customer from consuming the entire API allowance during a runaway workflow, while placing a changing value before the cache boundary undermines matching.

D:

This approach serves the separate need to reduce repeated processing of identical tool definitions and system instructions because prefix caching can lower repeated-input cost when the eligible prefix and cache conditions match; it does not establish the outcome to stop a single customer from consuming the entire API allowance during a runaway workflow, and caching does not turn every changed prompt into an automatic hit.

Question 81

A production-readiness check for an agent governance component requires the service to stop a coding agent from running a destructive deployment command before execution. Which approach passes this check?

  1. Use a post-tool hook to record outcomes
  2. Use a pre-tool hook to deny dangerous commands
  3. Fail closed when a required safety hook fails
  4. Keep hook scripts under change control

Explanation

Correct Answer

B:

Pre-execution interception provides a deterministic opportunity to reject a proposed tool call; this behavior directly supports the need to stop a coding agent from running a destructive deployment command before execution. Auditing the outcome afterward cannot undo the command.

Incorrect Answers

A:

Post-execution events are useful for audit, metrics, and follow-up checks once the tool has run; this applies to the different goal to capture which files an approved agent operation actually modified, not the requirement to stop a coding agent from running a destructive deployment command before execution; they cannot serve as the only prevention mechanism for an irreversible action.

C:

A fail-closed gate treats missing approval as denial for sensitive operations; the supported outcome is to prevent a tool action from proceeding when its authorization service is unreachable, whereas this scenario requires the ability to stop a coding agent from running a destructive deployment command before execution; fail-open behavior favors availability at the expense of an important security boundary.

D:

This approach serves the separate need to prevent an unnoticed local edit from disabling a team's protection of secret files because versioning and review make guardrail behavior inspectable and reproducible; it does not establish the outcome to stop a coding agent from running a destructive deployment command before execution, and unreviewed workstation-local scripts can drift across developers.

Question 82

Engineers are splitting a token economics service into two independently tested functions. Function A must avoid assuming retrieved material, tool outputs, history, and instructions each have independent infinite capacity, while function B must avoid confusing allowable prompt size with maximum generated response length. Which plan is sound?

  1. First component: Understand context-window competition. Second component: Distinguish input from output token limits.
  2. First component: Budget tokens rather than characters. Second component: Apply zero-shot instructions for a straightforward task.
  3. First component: Apply zero-shot instructions for a straightforward task. Second component: Distinguish input from output token limits.
  4. First component: Understand context-window competition. Second component: Budget tokens rather than characters.

Explanation

Correct Answer

A:

The first proposal fits its assigned role because all supplied content competes for a finite effective context budget and may affect attention, while the second proposal fits its assigned role because the context budget constrains total material considered while output ceilings bound what a single generation can emit; their respective acceptance criteria are to avoid assuming retrieved material, tool outputs, history, and instructions each have independent infinite capacity and to avoid confusing allowable prompt size with maximum generated response length.

Incorrect Answers

B:

The first proposal serves the different purpose to estimate whether a multilingual request will fit a model context window instead of the assigned first role because tokenization splits text into model-specific units, so character counts are not a reliable context proxy, while the second proposal serves the different purpose to classify a familiar support message where labels and criteria are already clear instead of the assigned second role because direct instructions can be sufficient when the task is simple and well specified; their respective acceptance criteria are to avoid assuming retrieved material, tool outputs, history, and instructions each have independent infinite capacity and to avoid confusing allowable prompt size with maximum generated response length.

C:

The first proposal serves the different purpose to classify a familiar support message where labels and criteria are already clear instead of the assigned first role because direct instructions can be sufficient when the task is simple and well specified, while the second proposal fits its assigned role because the context budget constrains total material considered while output ceilings bound what a single generation can emit; their respective acceptance criteria are to avoid assuming retrieved material, tool outputs, history, and instructions each have independent infinite capacity and to avoid confusing allowable prompt size with maximum generated response length.

D:

The first proposal fits its assigned role because all supplied content competes for a finite effective context budget and may affect attention, while the second proposal serves the different purpose to estimate whether a multilingual request will fit a model context window instead of the assigned second role because tokenization splits text into model-specific units, so character counts are not a reliable context proxy; their respective acceptance criteria are to avoid assuming retrieved material, tool outputs, history, and instructions each have independent infinite capacity and to avoid confusing allowable prompt size with maximum generated response length.

Question 83

Two separate tickets in an operations orchestration backlog require different controls. One needs to maintain a valid transcript when the model requests multiple tools; the other needs to avoid charging a customer twice after a worker retries a timed-out payment action. Which implementation pair is correct?

  1. First component: Execute independent requested tools concurrently. Second component: Forward tool_result blocks with matching identifiers.
  2. First component: Use the Claude Agent SDK for an integrated tool-driven agent. Second component: Add idempotency keys to side-effecting operations.
  3. First component: Preserve assistant tool-use content in conversation history. Second component: Add idempotency keys to side-effecting operations.
  4. First component: Preserve assistant tool-use content in conversation history. Second component: Forward tool_result blocks with matching identifiers.
  5. First component: Forward tool_result blocks with matching identifiers. Second component: Use the Claude Agent SDK for an integrated tool-driven agent.

Explanation

Correct Answer

C:

The first proposal fits its assigned role because the next request retains the assistant message containing tool_use blocks before providing their user-role tool_result blocks, while the second proposal fits its assigned role because application-generated operation identifiers allow the downstream service to recognize a repeated request; their respective acceptance criteria are to maintain a valid transcript when the model requests multiple tools and to avoid charging a customer twice after a worker retries a timed-out payment action.

Incorrect Answers

A:

The first proposal serves the different purpose to reduce turnaround when an agent asks for unrelated reads in the same response instead of the assigned first role because parallel execution is safe for independent operations while every result must still be matched to its call ID, while the second proposal serves the different purpose to let Claude continue after a client tool returns a transaction status instead of the assigned second role because results must reference the corresponding tool_use IDs and be included in the next conversation turn; their respective acceptance criteria are to maintain a valid transcript when the model requests multiple tools and to avoid charging a customer twice after a worker retries a timed-out payment action.

B:

The first proposal serves the different purpose to build a coding agent that needs managed tool execution, sessions, and event handling instead of the assigned first role because the SDK provides agent-loop primitives and integrations that remove much of the custom harness work, while the second proposal fits its assigned role because application-generated operation identifiers allow the downstream service to recognize a repeated request; their respective acceptance criteria are to maintain a valid transcript when the model requests multiple tools and to avoid charging a customer twice after a worker retries a timed-out payment action.

D:

The first proposal fits its assigned role because the next request retains the assistant message containing tool_use blocks before providing their user-role tool_result blocks, while the second proposal serves the different purpose to let Claude continue after a client tool returns a transaction status instead of the assigned second role because results must reference the corresponding tool_use IDs and be included in the next conversation turn; their respective acceptance criteria are to maintain a valid transcript when the model requests multiple tools and to avoid charging a customer twice after a worker retries a timed-out payment action.

E:

The first proposal serves the different purpose to let Claude continue after a client tool returns a transaction status instead of the assigned first role because results must reference the corresponding tool_use IDs and be included in the next conversation turn, while the second proposal serves the different purpose to build a coding agent that needs managed tool execution, sessions, and event handling instead of the assigned second role because the SDK provides agent-loop primitives and integrations that remove much of the custom harness work; their respective acceptance criteria are to maintain a valid transcript when the model requests multiple tools and to avoid charging a customer twice after a worker retries a timed-out payment action.

Question 84

The acceptance criteria for an operations orchestration change state that it must coordinate multiple specialists working on interdependent evidence without losing changes. Which implementation aligns with that criterion?

  1. Use a graph with conditional edges
  2. Adopt a blackboard of shared task state
  3. Use retrieval before generation
  4. Use a framework only when its primitives fit

Explanation

Correct Answer

B:

A controlled shared workspace holds versioned observations and ownership metadata for collaborators; this behavior directly supports the need to coordinate multiple specialists working on interdependent evidence without losing changes. Unstructured message passing alone can create stale assumptions.

Incorrect Answers

A:

This approach serves the separate need to branch between remediation and escalation based on the result of a security scan because conditional graph transitions encode known branches while allowing model reasoning within selected nodes; it does not establish the outcome to coordinate multiple specialists working on interdependent evidence without losing changes, and a linear chain cannot express result-dependent routing cleanly.

C:

For the alternate objective to answer questions from changing internal documentation with traceable source excerpts, retrieval grounds a model answer in current authoritative text rather than hoping parametric knowledge is current; the feature here must coordinate multiple specialists working on interdependent evidence without losing changes, and the retrieved text remains untrusted as instructions and needs relevance checks.

D:

Frameworks can help with graphs, retries, and checkpoints but create maintenance and integration costs; this applies to the different goal to select orchestration dependencies for a project that already has a stable task engine, not the requirement to coordinate multiple specialists working on interdependent evidence without losing changes; adopting a framework solely for fashionable abstraction is not a technical requirement.

Question 85

A developer is reviewing alternatives for a prompt engineering feature that must prevent a specialist's exhaustive document scan from contaminating the coordinator's working context. Which option is a defensible fit?

  1. Use scoped subagent contexts
  2. Deduplicate overlapping passages
  3. Anchor summaries to source identifiers
  4. Refresh stale evidence before reuse
  5. Separate long-term memory from active scratchpad

Explanation

Correct Answer

A:

Isolated workers return concise findings while retaining intermediate detail within their own window; this behavior directly supports the need to prevent a specialist's exhaustive document scan from contaminating the coordinator's working context. Sharing every search hit with every agent multiplies noise and cost.

Incorrect Answers

B:

For the alternate objective to save tokens when several retrieval chunks contain the same repeated policy section, overlap detection reduces redundant context without throwing away unique clauses; the feature here must prevent a specialist's exhaustive document scan from contaminating the coordinator's working context, and simply increasing retrieval count may provide many copies of the same material.

C:

Stable document IDs and passage references preserve provenance beyond the shortened working memory; this applies to the different goal to enable a reviewer to reconstruct evidence after context has been compacted, not the requirement to prevent a specialist's exhaustive document scan from contaminating the coordinator's working context; a bare compressed claim cannot show which source established it.

D:

When an application needs to avoid answering a pricing question from last month's retrieved cache, freshness checks determine whether a saved passage remains authoritative for a dynamic fact; the current capability must instead prevent a specialist's exhaustive document scan from contaminating the coordinator's working context, while high semantic similarity says nothing about timeliness.

E:

Explicit storage and freshness rules govern what is retained and when it is reintroduced; the supported outcome is to prevent outdated user preferences from being treated as current task instructions, whereas this scenario requires the ability to prevent a specialist's exhaustive document scan from contaminating the coordinator's working context; an unbounded persistent transcript can accumulate contradictory assumptions.

Question 86

A team maintaining an API reliability needs to fix an agent loop that repeatedly requests the same lookup without using the returned data. Which implementation most directly satisfies the requirement?

  1. Diagnose a 5xx provider error
  2. Trace a missing tool_result
  3. Handle a 429 rate limit error
  4. Investigate a 400 invalid_request error

Explanation

Correct Answer

B:

Inspecting message history and tool IDs reveals whether execution output reached the next model turn; this behavior directly supports the need to fix an agent loop that repeatedly requests the same lookup without using the returned data. Blaming model reasoning first overlooks an integration-layer omission.

Incorrect Answers

A:

Server-side failures may warrant bounded retries, jitter, and a fallback rather than request-shape changes; this applies to the different goal to keep a batch pipeline resilient when the upstream service briefly fails, not the requirement to fix an agent loop that repeatedly requests the same lookup without using the returned data; treating every provider outage as a malformed JSON prompt hides the real cause.

C:

A bounded retry policy and workload shaping address quota or rate throttling; the supported outcome is to prevent a bursty job service from exceeding allowed request throughput, whereas this scenario requires the ability to fix an agent loop that repeatedly requests the same lookup without using the returned data; repeating the request immediately can intensify the limit.

D:

This approach serves the separate need to diagnose a request that fails consistently before generating any model output because validation failures point to malformed inputs or unsupported parameters that must be corrected; it does not establish the outcome to fix an agent loop that repeatedly requests the same lookup without using the returned data, and backoff does not fix a structurally invalid payload.

Question 87

An MCP interoperability team is evaluating two disjoint use cases. For one use case, it must deny destructive filesystem operations before a coding agent executes them; for the other, it must make one internal ticketing connector available to several agent clients. Which paired choices are justified?

  1. First component: Use a deterministic workflow for fixed stages. Second component: Use subagents for independent expertise.
  2. First component: Use hooks for enforceable safety checks. Second component: Use a deterministic workflow for fixed stages.
  3. First component: Use subagents for independent expertise. Second component: Publish an MCP server for reusable integrations.
  4. First component: Use hooks for enforceable safety checks. Second component: Publish an MCP server for reusable integrations.

Explanation

Correct Answer

D:

The first proposal fits its assigned role because runtime interception can apply deterministic policies to proposed actions, while the second proposal fits its assigned role because MCP standardizes tool and resource discovery across compatible applications; their respective acceptance criteria are to deny destructive filesystem operations before a coding agent executes them and to make one internal ticketing connector available to several agent clients.

Incorrect Answers

A:

The first proposal serves the different purpose to automate the same classify-review-export path across predictable submissions instead of the assigned first role because explicit application control makes each transition testable and easier to audit, while the second proposal serves the different purpose to delegate a security analysis while keeping the main coordinator's context concise instead of the assigned second role because specialist contexts reduce tool-output contamination and support focused parallel work; their respective acceptance criteria are to deny destructive filesystem operations before a coding agent executes them and to make one internal ticketing connector available to several agent clients.

B:

The first proposal fits its assigned role because runtime interception can apply deterministic policies to proposed actions, while the second proposal serves the different purpose to automate the same classify-review-export path across predictable submissions instead of the assigned second role because explicit application control makes each transition testable and easier to audit; their respective acceptance criteria are to deny destructive filesystem operations before a coding agent executes them and to make one internal ticketing connector available to several agent clients.

C:

The first proposal serves the different purpose to delegate a security analysis while keeping the main coordinator's context concise instead of the assigned first role because specialist contexts reduce tool-output contamination and support focused parallel work, while the second proposal fits its assigned role because MCP standardizes tool and resource discovery across compatible applications; their respective acceptance criteria are to deny destructive filesystem operations before a coding agent executes them and to make one internal ticketing connector available to several agent clients.

Question 88

A production-readiness check for a customer platform component requires the service to constrain a repository coding assistant's allowed tools and execution behavior. Which approach passes this check?

  1. Use separate permission profiles by environment
  2. Pin a tested model identifier
  3. Use settings.json for Claude Code permissions
  4. Store API keys in managed secret storage

Explanation

Correct Answer

C:

Permission settings define configurable controls for Claude Code rather than merely describing preferences; this behavior directly supports the need to constrain a repository coding assistant's allowed tools and execution behavior. A CLAUDE.md comment cannot be relied on as a hard execution gate.

Incorrect Answers

A:

This approach serves the separate need to allow broad exploratory tools in a sandbox but only approved reads in production because environment-scoped policy avoids inheriting development privileges in a live setting; it does not establish the outcome to constrain a repository coding assistant's allowed tools and execution behavior, and one permissive configuration applied everywhere erases the staging boundary.

B:

Explicit model versions support reproducibility and controlled migrations; despite being useful when a service must keep deployment behavior stable while evaluating new model releases separately, it is not a solution to constrain a repository coding assistant's allowed tools and execution behavior; a floating alias can change behavior without an application code change.

D:

Central secret management supports scoped access, auditing, and key rotation; this applies to the different goal to rotate production credentials without committing new source code, not the requirement to constrain a repository coding assistant's allowed tools and execution behavior; a plain-text config checked into a repository is an inappropriate credential store.

Question 89

An architecture review of an incident response product lists two distinct outcomes: the first subsystem must resume a multi-step agentic process after worker or network failure, and a second subsystem must allow a research worker to read records without granting its coordinator's write privileges. Which pair delivers these outcomes?

  1. First component: Keep a single agent with a small tool set. Second component: Use sequential chaining for dependent subtasks.
  2. First component: Use a central state record for orchestration. Second component: Restrict each specialist's tool permissions.
  3. First component: Use sequential chaining for dependent subtasks. Second component: Restrict each specialist's tool permissions.
  4. First component: Use a central state record for orchestration. Second component: Keep a single agent with a small tool set.

Explanation

Correct Answer

B:

The first proposal fits its assigned role because persisted job state tracks completed stages, pending actions, and recoverable checkpoints outside the model context, while the second proposal fits its assigned role because scoped tool capability prevents a delegated role from using authority unrelated to its assignment; their respective acceptance criteria are to resume a multi-step agentic process after worker or network failure and to allow a research worker to read records without granting its coordinator's write privileges.

Incorrect Answers

A:

The first proposal serves the different purpose to answer routine account questions without the overhead of orchestrating multiple agents instead of the assigned first role because a single constrained executor simplifies state, authentication, logging, and latency where specialization adds little, while the second proposal serves the different purpose to ensure a generated specification is validated before code generation begins instead of the assigned second role because each stage consumes an approved artifact from the previous stage, preserving dependency order and stage-specific checks; their respective acceptance criteria are to resume a multi-step agentic process after worker or network failure and to allow a research worker to read records without granting its coordinator's write privileges.

C:

The first proposal serves the different purpose to ensure a generated specification is validated before code generation begins instead of the assigned first role because each stage consumes an approved artifact from the previous stage, preserving dependency order and stage-specific checks, while the second proposal fits its assigned role because scoped tool capability prevents a delegated role from using authority unrelated to its assignment; their respective acceptance criteria are to resume a multi-step agentic process after worker or network failure and to allow a research worker to read records without granting its coordinator's write privileges.

D:

The first proposal fits its assigned role because persisted job state tracks completed stages, pending actions, and recoverable checkpoints outside the model context, while the second proposal serves the different purpose to answer routine account questions without the overhead of orchestrating multiple agents instead of the assigned second role because a single constrained executor simplifies state, authentication, logging, and latency where specialization adds little; their respective acceptance criteria are to resume a multi-step agentic process after worker or network failure and to allow a research worker to read records without granting its coordinator's write privileges.

Question 90

A language-model integration review requires two independent, explicitly defined outcomes: the first workstream must estimate model-token input demand before launching a large processing job; a second must implement an interactive collaboration channel requiring messages in both directions. Which TWO actions directly provide these capabilities? Choose TWO.

  1. Use direct HTTP REST integration for unsupported runtimes
  2. Use the official SDK for standard requests
  3. Use the Messages count_tokens capability
  4. Choose WebSockets for bidirectional realtime exchange
  5. Use a queue for burst absorption

Explanation

Correct Answers

C:

Token counting helps check input budgets before generation and cost planning; this directly supports task 1 to estimate model-token input demand before launching a large processing job, not the separate task or tasks to implement an interactive collaboration channel requiring messages in both directions; word or byte count is not a reliable substitute across diverse content.

D:

Persistent two-way channels fit interactive event exchange where the client also sends messages during a session; this directly supports task 2 to implement an interactive collaboration channel requiring messages in both directions, not the separate task or tasks to estimate model-token input demand before launching a large processing job; standard response streaming alone is not a general bidirectional messaging protocol.

Incorrect Answers

A:

A standards-based HTTP client can send JSON requests with the correct headers and decode responses; this implements the different outcome to call the Messages endpoint from a language without a mature provider SDK rather than the required tasks to implement an interactive collaboration channel requiring messages in both directions and to estimate model-token input demand before launching a large processing job, and a wrapper package is optional, not the underlying API transport, so it is not a direct solution to any of the assigned workstreams.

B:

Maintained client libraries wrap the HTTP API and expose structured response objects; this implements the different outcome to simplify authentication, serialization, and response typing in a Python service rather than the required tasks to implement an interactive collaboration channel requiring messages in both directions and to estimate model-token input demand before launching a large processing job, and SDK convenience does not eliminate the need for correct retry and error handling, so it is not a direct solution to any of the assigned workstreams.

E:

Queueing decouples producers from consumer throughput and enables controlled retries; this implements the different outcome to smooth sudden ingestion spikes before workers call a capacity-limited model API rather than the required tasks to implement an interactive collaboration channel requiring messages in both directions and to estimate model-token input demand before launching a large processing job, and increasing frontend timeouts does not eliminate backend capacity limits, so it is not a direct solution to any of the assigned workstreams.

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