CCDV-F Practice Exam 2 — 20 Questions

Question 1

A developer is iterating on a classification prompt for support tickets. Each test run uses a different random sample of 200 tickets, and accuracy swings by 8 points between runs. The prompt itself is unchanged. What is the best first step to get trustworthy signal about whether a prompt edit actually helped?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Freeze a held-out evaluation set and run every prompt version against the same fixed examples.. A fixed held-out evaluation set is the foundation of reliable prompt iteration. By scoring every prompt version against identical examples, the developer isolates the effect of the prompt change from the effect of sampling. Temperature and example count affect generation, but neither controls which inputs are being compared.

Answer

Increase max_tokens so the model has more room to reason before classifying.

Answer

Add more few-shot examples until accuracy stops changing between runs.

Answer

Lower the temperature to 0 and rerun the same random sample twice.

Question 2

A developer is drafting a system prompt for a coding assistant. They want Claude to always answer in the same tone and follow the same safety constraints regardless of what the user types. Where should these persistent rules be placed?

Accepted Answer

In the system prompt, which applies across the whole conversation.. Persistent behavioral rules belong in the system prompt because it frames every turn and is not scoped to a single user or assistant message. This gives consistent tone and safety enforcement regardless of what the user types. User-message rules are turn-scoped, assistant messages represent output, and non-message fields are not read as instructions, so none of those achieve the required consistency.

Answer

In a final assistant message appended after the user's question.

Answer

In the first user message, mixed with the developer's coding question.

Answer

In a separate metadata field outside the messages array.

Question 3

A developer wants Claude Code to run a series of shell commands as part of a build task but is concerned about the assistant executing arbitrary or destructive commands. Which TWO practices reduce this risk while still allowing the needed commands to run? (Choose two.)

Accepted Answer

Add a PreToolUse hook on the Bash tool that inspects the command and blocks anything outside the approved set.. Least-privilege controls are the right answer. An allow list in settings.json restricts execution to the specific build commands, and a PreToolUse hook on the Bash tool can inspect and block disallowed commands before they run. Together they permit the intended build while constraining arbitrary or destructive execution. Skipping permissions removes safeguards, memory-file advice is not enforced, and container hardening changes the environment rather than governing command selection.

Answer

Instruct Claude Code in CLAUDE.md to never run dangerous commands.

Answer

Run the build inside a container with no network access and a read-only filesystem.

Answer

Enable `--dangerously-skip-permissions` to avoid prompts during the build.

Question 4

A developer is building an MCP server that exposes several tools for managing a cloud infrastructure. The server will be used by multiple Claude clients. The developer wants to ensure that tool calls are handled efficiently and that the server can scale. Which two practices should the developer implement? (Choose two.)

Accepted Answer

Design tools to be idempotent where possible, so that repeated calls with the same parameters produce the same result.. The two best practices are statelessness and idempotency. Statelessness allows any server instance to handle any request, enabling horizontal scaling. Idempotency ensures that retries are safe, which is important for reliable infrastructure operations. Together, they improve efficiency and scalability. Other options like caching, global locking, or in-memory session state either introduce complexity or hinder scalability.

Answer

Cache the results of expensive tool calls on the server side and return cached responses for identical requests.

Answer

Use a single global lock to serialize all tool calls, preventing concurrent execution.

Answer

Store session state in memory on each server instance to track client interactions.

Question 5

What is the role of the user when using Claude Code for a complex refactoring task?

Accepted Answer

The user acts as the strategic architect and final validator.. The user remains the architect and final auditor of the code. While Claude Code handles the heavy lifting of writing code and executing tests, the human user must provide strategic direction, approve individual changes, and verify the final outcomes. This synergy is key to effective AI-assisted development, as it allows the developer to leverage the agent's speed and breadth of knowledge while maintaining ultimate control over the project's design and long-term maintainability.

Answer

The user is only responsible for providing the initial prompt.

Answer

The user should avoid interfering to prevent conflicts.

Answer

The user must perform all terminal commands manually.

Question 6

You are designing a multi-agent system where a 'Router' agent directs tasks to a 'Research' agent or a 'Writer' agent. What is the most efficient way to maintain state when the Research agent completes its task and needs to hand back control?

Accepted Answer

Appending the Research agent's tool results to the main message thread.. Handoffs in multi-agent systems rely on passing the conversation history or a summarized state between specialized instances. This ensures that the Router has the necessary context to decide the next step without losing the progress made by the Research agent. Proper state management prevents redundant work and keeps the agentic flow aligned with the user goal.

Answer

Persisting the Research agent's output to a global database for the Router.

Answer

Restarting the entire session with the Research output as a new prompt.

Answer

Using a separate API key for each agent to isolate their environments.

Question 7

A security auditor is reviewing an application that uses the Model Context Protocol (MCP). Which TWO security considerations are most important for the developer to address when implementing the MCP Server?

Accepted Answer

The server should strictly validate all tool arguments against its internal logic.. Security in MCP is a shared responsibility. While the protocol facilitates communication, the server developer must ensure that the tools and resources exposed do not create vulnerabilities. This includes limiting access to the file system and ensuring that tool inputs are properly validated before being used in potentially dangerous operations like shell commands.

Answer

The server must use a unique API key for every tool call it receives.

Answer

The server must rotate its connection ID every sixty seconds for privacy.

Answer

All MCP tool results must be manually approved by a human operator.

Question 8

When implementing a 'Computer Use' agent, which THREE capabilities are provided by the standardized Anthropic beta tools 'computer', 'text_editor', and 'bash'?

Accepted Answer

Capturing screenshots and performing mouse/keyboard interactions.. Anthropic provides specialized tools for computer interaction to ensure consistent behavior across different environments. These tools allow the agent to perform GUI actions, modify files with precision, and execute shell commands. Together, they form a comprehensive suite for agents that need to operate as a virtual developer or assistant on a computer system.

Answer

Rewriting entire binary files to optimize system performance.

Answer

Automatically upgrading the host operating system to the latest version.

Question 9

Refer to the exhibit. A developer is implementing the provided JSON structure to optimize an application that repeatedly analyzes the same large report. What is the primary financial implication of using the 'cache_control' block in this specific API request?

Accepted Answer

Subsequent requests with the same report will be billed at a reduced rate.. The 'cache_control' block enables Prompt Caching, which is a key tool for cost management. By marking a block as ephemeral, Anthropic caches the preceding content. If the same content is sent again within the cache's lifetime, the developer is charged a significantly lower 'cache hit' rate instead of the full input token price. This is essential for reducing costs in applications with static contexts.

Answer

It eliminates the cost of the first 1024 tokens in every request.

Answer

It triggers a 50% discount on all output tokens for the session.

Answer

The request will be processed using the Batch API pricing model.

Question 10

In the context of Anthropic's pricing model, why is it generally recommended to provide only the necessary context rather than the entire available dataset in a single prompt?

Accepted Answer

To minimize the input token count and keep the per-request cost low.. Token-based pricing means that every piece of information sent to the model has a direct financial cost. Large prompts not only increase the bill but can also lead to 'context stuffing,' which might degrade the model's focus. Efficient prompt engineering—sending only what is needed—is the fundamental practice for both cost management and ensuring high-quality, relevant responses from the AI.

Answer

Because Anthropic charges a 'search fee' for every 1,000 tokens of context.

Answer

Because Claude models cannot process more than 10,000 tokens at a time.

Answer

To prevent the model from reaching its daily 'knowledge limit'.

Question 11

A developer needs to estimate the monthly budget for a new internal knowledge base application powered by Claude. Which TWO factors directly influence the total token consumption and subsequent cost of the API requests?

Accepted Answer

The total number of input tokens in the prompt. Estimating costs requires understanding the components of the billing model used by Anthropic. Total cost is derived from the volume of input tokens sent to the model and the volume of output tokens generated by the model. Developers must account for both segments because they are priced at different rates across the Claude 3 model family to reflect processing requirements.

Answer

The number of concurrent users accessing the application

Answer

The physical geographic location of the application server

Answer

The programming language used to make the API calls

Question 12

Which THREE of these actions are considered safe and standard for Claude Code to perform during an interactive session?

Accepted Answer

Reading and summarizing codebase files.. Claude Code is built with safety as a priority. It is expected to handle file editing, command execution, and code review as part of its normal operation. Understanding these standard actions allows developers to trust the agent's output and integrate it into their workflow efficiently. By focusing on safe, verified, and transparent interactions, the agent maintains its role as a helpful assistant while minimizing risks associated with uncontrolled automation in a local project.

Answer

Deleting arbitrary user files without asking.

Answer

Exfiltrating environment variables to remote servers.

Question 13

An agent developer is building a workflow where an agent must query a private database. To ensure the agent only performs read operations, which component of the Anthropic Agent SDK pattern should be prioritized?

Accepted Answer

A strictly defined tool schema limited to read-only functions.. Tool definitions are the primary mechanism for restricting agent capabilities. By explicitly defining the schema and purpose of the tool, the developer enforces the principle of least privilege. The agent only executes what is provided in the tools list, making it vital to sanitize inputs and restrict access permissions at the API level rather than relying solely on natural language instructions within the system prompt.

Answer

A complex system prompt with strict negative constraints.

Answer

The Agent SDK's built-in session state management.

Answer

The model's internal safety filtering layer.

Question 14

A developer wants to reduce latency and costs for a high-traffic application that sends a large, static set of instructions in every request. Which API feature should they implement to achieve this?

Accepted Answer

Prompt Caching using cache_control.. Prompt Caching is a specialized feature designed to optimize performance for repetitive content. By marking specific parts of the prompt as cacheable, developers can significantly reduce the processing time for the 'prefill' phase and take advantage of lower pricing for cached tokens. This is particularly effective for large system prompts or complex context documents.

Answer

Batch API processing.

Answer

Top-k sampling reduction.

Answer

System prompt compression.

Question 15

Which TWO of the following are essential components of a 'tool_use' block generated by Claude?

Accepted Answer

A unique 'id' string used to match results to the specific request.. A 'tool_use' block must contain specific identifiers so that the SDK knows which tool to run and how to track its result. The 'id' uniquely identifies the specific call in the conversation history, while the 'name' maps the call to the actual function or service defined in the agent's configuration.

Answer

The full source code of the function being called.

Answer

The 'timestamp' of when the model decided to use the tool.

Answer

The 'confidence_score' the model assigns to this specific tool call.

Question 16

Which of the following scenarios describes the most effective use of Prompt Caching for cost management?

Accepted Answer

An AI assistant that references a 50-page technical manual for every user query.. Prompt caching is most effective when a large amount of static information is reused across many different requests. It allows the model to 'remember' the prefix of a prompt, significantly reducing the cost of processing that prefix in subsequent calls. Identifying workloads with high prefix overlap is key to maximizing the financial benefits of this feature in production environments.

Answer

A chatbot where every user query is unique and no history is maintained.

Answer

A translation service that processes single words one at a time.

Answer

A daily report generator that uses a completely different dataset every morning.

Question 17

Refer to the exhibit. An application monitoring system captures this response from the Anthropic API. Which strategy is the most mechanically sound approach for the application to take to resolve this specific error and continue processing?

Accepted Answer

Implementing an exponential backoff algorithm before retrying the request.. Rate limit errors occur when a developer exceeds the allocated requests per minute or tokens per minute for their tier. Handling these gracefully via exponential backoff is a fundamental skill for API integration. This approach ensures that the application doesn't overwhelm the server further, allowing the rate limit bucket to refill naturally while maintaining the best possible user experience.

Answer

Immediately resending the request with a new API key to bypass the limits.

Answer

Switching the model parameter to a larger model like Claude 3 Opus.

Answer

Increasing the max_tokens parameter to ensure the request is prioritized.

Question 18

An agent using the SDK encounters an error where it generates a tool call for a tool that does not exist in its configuration. Which adjustment to the agent's setup is most likely to resolve this hallucination?

Accepted Answer

Refining the system prompt to clearly define the boundaries of available tools.. Hallucinating non-existent tools usually occurs when the model is aware of a capability but hasn't been given the structured definition for it, or when the system prompt is ambiguous. Refining the system prompt to explicitly list the available tools and their purposes helps ground the model's reasoning in the provided technical context.

Answer

Enabling 'force_tool_use' to ensure it only uses one of the existing tools.

Answer

Switching the model from Claude 3.5 Sonnet to Claude 3 Haiku.

Answer

Increasing the max_tokens parameter to allow for more reasoning.

Question 19

A developer wants to integrate Claude Code into a CI/CD pipeline. Why is this generally discouraged or difficult to achieve directly?

Accepted Answer

The tool is designed for interactive human sessions.. Claude Code is primarily an interactive, human-in-the-loop tool. Its design relies on terminal input, natural language prompts, and real-time user validation, which are not conducive to non-interactive CI/CD automation. CI/CD pipelines require deterministic, automated scripts. Attempting to force an interactive agent into such environments leads to unpredictable behavior, race conditions, and high costs, as the agent is not optimized for the standard, scripted environment of automated build and deployment systems.

Answer

The tool is not licensed for use in CI/CD pipelines.

Answer

The tool requires an internet connection for every step.

Answer

It cannot execute shell commands reliably.

Question 20

A developer is concerned about the 'token overhead' when using system prompts. How does Anthropic charge for tokens included in the system prompt compared to tokens in the user message?

Accepted Answer

They are billed at the same standard input token rate as the user message.. Understanding the billing of different prompt components is essential for cost management. Anthropic treats all input tokens equally, regardless of their role in the message (system, user, or assistant). This means that a large system prompt containing complex instructions or few-shot examples will incur the same per-token cost as the actual user query or any historical context provided in the conversation.

Answer

System prompt tokens are billed at a premium rate because they influence the model more.

Answer

System prompt tokens are free for the first 500 tokens in every request.

Answer

System prompts are only billed if they exceed the 200,000-token context limit.