Timed Mode Set 2 – Claude Certified Developer Foundations CCDV-F

1. Question

You are designing a generative AI service that prepares draft responses for newly submitted customer-support cases. The business expects agents to receive each draft within two minutes, even when ticket volume reaches its weekday peak.

Which infrastructure specification is most appropriate for meeting this target?

The service should cache frequently used support guidance to reduce repeated retrieval operations.
The service should use a faster Claude model during periods of elevated ticket volume.
The application should create a draft response for each newly submitted case.
The service must automatically scale to process 500 simultaneous requests while maintaining the required response latency.

2. Question

You are designing two MCP integrations: one server will execute development utilities directly on an engineer’s workstation, while another will provide a centralized company service to users connecting from different locations. Each server requires a transport that matches how clients will access it. Which communication approach best meets these requirements?

Configure both servers to use stdio because MCP supports only communication between local processes.
Use stdio for the local server and launch a separate stdio process on the centralized host for every remote client connection.
Use stdio for the server operating on the local workstation, and use a network transport such as Streamable HTTP for the centrally hosted server.
Use a network transport such as Streamable HTTP for both servers to provide a consistent connection model, including the server running on the local workstation.

3. Question

You are configuring an AI support agent with two tools: find_shipments and find_payments. Testing shows that the agent often confuses the tools because each description contains only a brief summary of its name. Which solution will most effectively improve the agent’s tool selection?

Add several shipment and payment request examples to the agent’s system prompt.
Expand each tool description to explain its output, intended use cases, exclusions, and differences from the other tool.
Create a separate tool that selects which search tool the agent should call.
Rename the tools to lookup_customer_shipments and lookup_customer_payments.

4. Question

You are configuring an autonomous coding assistant that can run terminal commands and modify production resources. The security team requires controls that prevent destructive commands from running automatically while still allowing approved high-risk operations when necessary. Which controls best meet these requirements? (Select TWO.)

Record all executed commands in an immutable audit log and review flagged high-risk operations after the session.
Configure a deterministic pre-execution hook that evaluates every command and rejects commands matching destructive patterns.
Restrict the agent to a least-privilege service account that still has the repository and database write permissions required for routine tasks.
Require a human reviewer to approve commands classified as high risk before they are executed.
Increase the model’s temperature so it evaluates risky commands more carefully.

5. Question

You are configuring Claude to follow an organization’s writing standards, reusable report layouts, and content-quality checklist. These materials contain procedural guidance rather than external data, and should be added to Claude’s context only when a related writing task is performed. Which approach best meets these requirements?

Create reusable prompt templates that insert the appropriate guidelines when a team member selects a content workflow
Create a Skill that packages the instructions, templates, and checklist for Claude to load when relevant
Define a custom tool and place all writing guidance inside its tool description
Store the materials in a dedicated Claude Project so all conversations within the marketing workspace can access them

6. Question

You are part of a platform engineering team supporting several business units that independently connect Claude to the same internal analytics system. Because each unit maintains separate integration code, the supported operations and response schemas have become inconsistent, so the company needs one centrally managed implementation that works across its Claude environments. Which solution will best meet these requirements?

Deploy a shared remote MCP server that exposes the analytics system’s operations as tools and can be connected to multiple Claude environments.
Require every business unit to maintain its own integration while following a common API versioning and release-notification policy.
Package an MCP server for each business unit to run locally and allow teams to update their instances independently.
Store standardized prompt templates in a shared repository and instruct each Claude environment to use them when constructing analytics requests.

7. Question

You are evaluating an AI support agent that can call one tool for asset specifications and another for service history. During testing, the agent repeatedly chooses the service-history tool for technical-detail requests and the specifications tool for questions about previous repairs. Which change will most effectively improve the agent’s tool-selection accuracy?

Add a validation step that checks whether the returned fields match the request and retries with the other tool when they do not.
Revise both tool descriptions to clearly define their returned data, appropriate use cases, exclusions, and differences.
Replace both tools with a single retrieval tool that accepts a broad information-category parameter.
Add several example technician requests to the agent’s system prompt.

8. Question

You are reviewing access credentials after a developer’s departure and discover that an API key created by that developer is still used by a critical application. The developer’s console and directory access have already been removed, but the production integration continues to operate normally. What action should the team take to properly secure the production environment?

Move the existing key into the team’s secrets manager, update its label and access controls, and continue using the same credential value.
Create a team-managed secondary key, configure it as a standby credential, and retain the original key as the application’s primary credential until its scheduled expiration.
Create a team-managed replacement key, update the application, and explicitly deactivate the existing key because workspace API keys are not revoked when their creator loses access.
Change the existing key’s display name to the production service’s name because renaming removes its association with the former developer.

9. Question

You are designing a prompt for an automated system that must return the same fields in the same order for every request. Although the prompt clearly describes the required format, the model continues to produce inconsistent layouts. Which prompt design approach is most likely to produce consistently structured outputs?

Present the required fields and their sequence in a clearly labeled schema table.
Include several sample inputs paired with outputs that follow the exact required structure.
Increase the response token limit to prevent incomplete formatting.
Lower the model’s temperature to reduce variation across repeated generations.

10. Question

You are integrating Claude into a reporting workflow that sends generated JSON to downstream systems and incorporates factual summaries into documents delivered to clients. The workflow must prevent malformed data and unsupported claims from reaching production. Which practices should be implemented? (Select TWO.)

Configure the downstream system to populate missing fields with default values and continue processing the response.
Use a parser, validate the output against the required schema, and define fallback behavior for invalid responses.
Ask Claude to review its initial response for accuracy and use the revised response as the final publication-ready version.
Verify significant factual claims against reliable sources before including them in customer-facing content.
Increase randomness to improve the reliability of structured responses.

11. Question

A compliance department is testing two methods for reviewing several lengthy regulatory reports. One method produces stronger comparisons across the reports but occasionally links conclusions to the wrong source, while the other maintains dependable source attribution but identifies fewer relationships between documents. The project requirements do not specify which quality measure is more important. What should the department do before selecting a method?

Assign equal weighting to cross-document reasoning and source attribution, then select the method with the highest combined evaluation score.
Define and prioritize the quality criteria required for the compliance review.
Select the method with stronger cross-document analysis because identifying relationships is the primary purpose of using an AI model.
Run both methods on a representative document set and choose the approach with the strongest average performance across the two quality measures.

12. Question

An incident-response assistant has been running for several hours and has accumulated raw monitoring data, command outputs, and lengthy diagnostic reports in its conversation history. Although the required evidence was already collected, the assistant now overlooks earlier conclusions and repeatedly proposes completed troubleshooting steps. The team must restore the assistant’s focus without losing important investigative progress. Which technique most effectively addresses the underlying issue?

Increase the assistant’s maximum response length so it can provide more detailed troubleshooting guidance.
Compact the conversation by removing or summarizing obsolete tool outputs while preserving critical findings, decisions, and unresolved issues.
Divide the investigation among specialized diagnostic agents and combine their complete reports in a coordinator agent’s conversation.
Start a new conversation and discard all information gathered during the investigation.

13. Question

A generative AI service sends a large block of identical operating instructions with every API request. However, a newly generated session identifier and the current date are inserted before the reusable instructions, causing prompt-caching metrics to show almost no cache reads. What is the most likely reason for the low cache hit rate?

The changing metadata appears at the beginning of the prompt, preventing the reusable content from matching a previously cached prefix.
The cache breakpoint is positioned after the user message, causing the entire request to be evaluated as one cacheable segment.
Prompt caching supports only conversation messages and excludes system-level instructions.
Each API request uses a different model configuration, resulting in separate prompt-cache namespaces.

14. Question

A legal analytics team plans to submit several lengthy case files to Claude and request a detailed, multi-section analysis in the same API call. The team must ensure that the request remains within the model’s available context capacity. Which factor must the team account for when determining the request size?

Estimate document usage from file size and page count, then configure the response length according to the model’s maximum output-token limit.
The case files, instructions, and generated response share the same context window, so the combined token usage must remain within the model’s capacity.
Only the generated analysis consumes the context window, while uploaded case files have no token impact.
Large case files must always be processed with the Batch API before Claude can analyze them.

15. Question

An AI-powered learning platform handles both basic terminology questions and complex analytical exercises. The platform must respond quickly to straightforward prompts while automatically dedicating more computation to difficult tasks, without relying on an external request router. Which model capability best meets these requirements?

Extended thinking, which enables the model to perform a dedicated reasoning process before generating its final response.
Extended context, which provides a larger context window for processing lengthy conversations and extensive source material.
Adaptive reasoning that dynamically adjusts the depth of reasoning according to task complexity.
Ultrathink, which temporarily applies the model’s highest reasoning effort to tasks requiring extensive analysis.

16. Question

A media analytics platform sends the same 500-token writing guideline with every request, followed by a unique research summary. After increasing each generated report from approximately 250 words to 1,000 words, the organization observes a significant rise in API charges. A developer recommends prompt caching as the main cost-reduction strategy. Which assessment most accurately addresses the increased cost?

Prompt caching is unlikely to address the primary increase; input and output token usage should be analyzed separately to identify the main source of the higher charges.
Prompt caching will substantially reduce the total cost because the repeated writing guideline can be reused across requests with different research summaries.
Prompt caching will provide the greatest savings when the repeated editorial guideline is placed in the system prompt rather than included with each source brief.
Prompt caching will deliver the strongest cost reduction after the changing source briefs are shortened to increase the proportion of repeated input tokens.

17. Question

A product team is preparing to introduce a Claude-powered document analysis feature, and the finance department requires a projected monthly API budget before approving the launch. The application will submit a combination of repeated instructions and request-specific content to the selected Claude model. Which information is required to create the most reliable monthly cost estimate?

A blended token rate, forecasted user growth, average requests per session, and estimated peak concurrency.
The model's context-window size, projected monthly users, average requests per user, and expected response length.
The selected model's token rates, average input and output tokens per request, monthly request volume, and expected prompt-cache usage.
The application’s expected number of users and average session duration.

18. Question

A customer-service platform uses an AI model for ticket routing across ten support queues and processes hundreds of thousands of requests each day. Several candidate models meet the required accuracy threshold, but differ in latency and cost. Which approach should the company use to select the most suitable model?

Select the model with the strongest latency results and use autoscaling to manage overall operating cost.
Choose the model with the lowest per-request cost and apply batching to improve ticket-routing throughput.
Select the lowest-capacity model that achieves the required accuracy, then continuously evaluate its performance in production.
Distribute requests randomly across all models to balance their strengths and weaknesses.

19. Question

A RAG application is producing inaccurate responses, but the engineering team does not know whether the issue is caused by poor document retrieval, prompt construction, or model interpretation. Which action should the team prioritize first?

Increase the number of retrieved document chunks so the model receives more context for every request.
Review execution traces showing the retrieved documents, constructed prompt, and generated response for each failed request.
Replace the current model with a more capable reasoning model before conducting further testing.
Display a warning that generated answers may contain errors until the system is updated.

20. Question

An enterprise operates a high-volume AI agent service and needs to diagnose failed tool executions without retaining every prompt, response, and tool result in full. Which observability approach best balances troubleshooting capability, storage cost, and data exposure?

Record complete session payloads only after users submit a support ticket.
Store final agent outputs and aggregate latency statistics while excluding tool activity.
Generate structured traces and metrics for all sessions using correlation identifiers, while selectively capturing detailed payloads through sampling and failure-based triggers.
Disable production telemetry and use synthetic test sessions to investigate unexpected behavior.

21. Question

A software organization needs Claude Code to automatically apply consistent repository-specific build procedures, architectural guidance, and development standards for all contributors. Each contributor must also be able to add private preferences without changing the shared instructions. Which configuration meets these requirements?

Commit a project-level CLAUDE.md file with the shared instructions, and store individual preferences in a user-level or local context file.
<span>Require every contributor to copy the shared instructions into <code>~/.claude/CLAUDE.md</code> on each workstation.</span>
Add the development standards to the repository's Claude Code settings file together with permissions and environment configuration.
Publish the instructions in the contributor guide and require developers to provide them manually at the beginning of every Claude Code session.

22. Question

You are migrating an application from an older Claude model to a newer model and need to adjust how much reasoning is applied for complex tasks. The team wants to use the extended thinking token budget effectively without forcing a fixed reasoning amount.

Which approach should be implemented?

Enable adaptive thinking and use the effort parameter to guide the amount of reasoning Claude applies based on task complexity.
Always allocate the maximum extended thinking token budget so Claude uses the highest reasoning level for every request.
Disable thinking features because newer Claude models automatically reason the same way as older models.
Set a fixed thinking token budget for all requests regardless of task complexity.

23. Question

You are reviewing the security posture of a Claude application where the team relies on one system-prompt to prevent unsafe behavior. The team wants to confirm whether this approach follows safe deployment layers.

Which assessment is most accurate?

It is sufficient because a strong system prompt alone provides complete protection against all application security risks.
It is insufficient because safe deployment requires layered defenses beyond prompt guidance, including input validation, output filtering, least-privilege access, and monitoring.
It is unnecessary because security controls should be avoided to preserve the model’s flexibility.
It is sufficient because Claude automatically blocks all unsafe requests regardless of application controls.

24. Question

A technology company is developing an assistant to coordinate complex cloud migration projects. The required activities cannot be fully defined beforehand because configuration findings may trigger additional compatibility checks, dependency updates, resource changes, or validation tasks. Which implementation pattern best meets these requirements?

A hierarchical planning system that creates a complete migration plan upfront and assigns each activity to specialized executors.
A routing system that assigns each project to one of several predetermined migration pipelines.
An event-driven orchestrator that advances the migration through predefined states as configuration signals are received.
An autonomous agent that selects actions, evaluates intermediate results, and determines the next step iteratively.

25. Question

A software team is developing an autonomous literature-review workflow with the Claude Agent SDK and wants to minimize custom orchestration while retaining control over tool access and agent behavior. Which implementation approach best meets these requirements?

Use the SDK-managed agent loop and context handling while configuring custom tools, permissions, and task instructions.
Build an application-level loop that stores every message, executes all requested tools, and reconstructs the context after each action.
Allow the SDK to invoke tools, but require the application to maintain and resend the complete conversation state between reasoning steps.
Use a separate SDK session for each research phase and provide the next session with a structured summary of the preceding phase.

26. Question

A financial services company deploys a customer-support chatbot that produces accurate, detailed responses, but customers experience a noticeable blank delay before the complete answer appears. The team must improve perceived responsiveness without shortening the responses or reducing the quality. Which solution will most directly meet this requirement?

Configure a smaller maximum token limit to reduce the total response-generation time.
Use a lower-capability model optimized primarily for reduced cost and latency.
Stream generated tokens to the application as soon as they become available.
Enable prompt caching for frequently reused system instructions to reduce input-processing latency.

27. Question

A software engineering team plans to integrate a Claude-based code review agent into the pull request workflow without weakening its existing quality controls. What function should the agent perform throughout each pull request review?

Publish suggested findings as non-blocking review comments while retaining required human approvals and automated test checks.
Run the agent after human review and automated testing are complete, then publish a consolidated quality summary for the maintainers.
Configure the agent as an additional required approval gate that must pass before the pull request becomes eligible for merge.
Review only changes made to authentication, authorization, and other security-related components.

28. Question

A research organization needs to categorize 750,000 historical survey responses before loading the results into an analytics platform. Individual responses are not time-sensitive, processing may take up to 24 hours, and minimizing inference costs is the primary objective. Which approach best meets these requirements?

Place the transcripts in an application-managed queue and send synchronous requests at a steady rate while using prompt caching for the shared classification instructions.
Submit the requests through asynchronous batch processing to receive discounted pricing for a large, latency-tolerant workload.
Use synchronous streaming requests so each classification is returned incrementally and processed immediately.
Use synchronous requests with an advanced reasoning configuration to improve classification accuracy and avoid repeated processing.

29. Question

A company plans to introduce an AI assistant that converts executive meeting transcripts into structured board reports. Before the development team selects a model or designs prompts, the product manager must determine how to translate the proposal into an actionable engineering initiative. What should the product manager do first?

Compare available language models based on long-form writing quality and inference pricing.
Design the end-to-end workflow for transcript ingestion, document generation, human review, and report distribution.
Define the feature’s functional and non-functional requirements, including quality constraints, security needs, integrations, and operational requirements.
Calculate projected token consumption and request approval for the expected operating costs.