CCAR-F Practice Test 21 — 15 Questions

Question 1

When implementing a Tool Use (function calling) loop with Claude, which TWO steps are required for the application to successfully complete a multi-step task involving external data?

Accepted Answer

Detecting the 'tool_use' stop_reason in the API response. In an agentic loop, the application must manage the transition between the assistant's request to use a tool and the provision of the tool's output back to the model. This iterative process requires detecting the specific stop reason and then formatting the external data correctly so Claude can process it in the next turn.

Answer

Hard-coding the tool results directly into the system prompt

Answer

Using a temperature setting of 0.0 to ensure tool syntax is correct

Answer

Manually parsing the model's text output using regular expressions

Question 2

When designing a multi-agent system where Claude acts as a 'Router,' how does the 'Lost in the Middle' phenomenon specifically impact context reliability, and how should it be mitigated?

Accepted Answer

It degrades retrieval accuracy for data in the center of the prompt.. The 'Lost in the Middle' phenomenon describes how LLMs tend to perform better on information found at the very beginning or end of a prompt compared to the middle. In a routing scenario, critical decision-making logic or classification criteria might be missed if they are buried. Mitigation involves strategic placement of instructions and using structural markers to guide the model's focus.

Answer

It causes the model to forget the system prompt entirely.

Answer

It results in higher costs due to redundant token processing.

Answer

It prevents the model from generating any output after 100k tokens.

Question 3

Which of the following is the recommended approach for handling PII (Personally Identifiable Information) in prompts?

Accepted Answer

Redact the PII at the application layer before sending the request to the API.. Always sanitize PII before it reaches the model to protect user privacy and comply with data governance policies. Anthropic recommends that sensitive data be redacted or masked upstream. Relying on the model to ignore or redact PII is not a secure architectural pattern, as models are not deterministic security filters. Protecting data at the application layer ensures compliance and minimizes security risks within the LLM pipeline.

Answer

Tell the model in the system prompt to ignore and delete any PII found.

Answer

Use a specialized system prompt that instructs the model to encrypt the data.

Answer

Send the PII but add a disclaimer in the system prompt about data usage.

Question 4

Which 'stop_reason' does the Anthropic API return when the model has finished its current thought process and is waiting for the user to provide the results of a tool execution?

Accepted Answer

stop_reason: "tool_use". The 'tool_use' stop reason is the specific signal from the model that it has reached a point where it cannot proceed without external information. This is the trigger for the application to stop waiting for text and start executing the functions defined in the 'tool_use' content blocks.

Answer

stop_reason: "end_turn"

Answer

stop_reason: "max_tokens"

Answer

stop_reason: "stop_sequence"

Question 5

The Model Context Protocol (MCP) defines three primary primitives that servers can expose to clients. Which THREE of the following are the core primitives of MCP?

Accepted Answer

Resources: Read-only data sources such as files, database tables, or API responses.. MCP is built on a small set of powerful primitives that allow for standardized data and capability exchange. Understanding these three components—Resources, Tools, and Prompts—is essential for any architect designing MCP-compliant systems. Each primitive serves a distinct purpose in providing context, capabilities, and reusable interaction patterns to the model.

Answer

Workflows: Sequential chains of tool calls managed automatically by the MCP server without host intervention.

Answer

Models: The specific LLM versions that are hosted directly within the MCP server for local processing.

Question 6

A company wants to ensure their Claude-powered application remains reliable even when Anthropic releases new model versions. Which TWO strategies should the architect implement?

Accepted Answer

Pin the application to a specific model version (e.g., claude-3-5-sonnet-20240620).. Model versioning is a key aspect of reliability. To prevent unexpected changes in behavior, architects should pin their applications to specific model versions rather than using generic aliases. Additionally, maintaining a robust evaluation suite allows the team to test new versions before migrating production traffic to them.

Answer

Always use the 'latest' model alias to get the most recent fixes.

Answer

Randomly sample 10% of traffic to test new models in production.

Answer

Reduce the system prompt complexity for newer models.

Question 7

What is the recommended method for handling large tool outputs in MCP?

Accepted Answer

Return a summary and a link/ID to fetch the full data via a resource endpoint.. MCP is not designed to pass massive amounts of data directly within the JSON response, as this can crash the connection or exceed token limits. The architect should return a reference or a summary, and provide a secondary resource-based retrieval mechanism if the full data is needed. This approach maintains the efficiency of the MCP protocol and ensures the system remains responsive, avoiding the performance bottlenecks associated with large data payloads.

Answer

Base64 encode the large output and embed it directly in the tool result.

Answer

Truncate the output arbitrarily to fit the available context window.

Answer

Split the response into multiple sequential tool calls.

Question 8

Refer to the exhibit. Why might the model struggle with this request?

Accepted Answer

The prompt does not provide the source text for the summary.. The prompt lacks sufficient grounding. Asking the model to summarize a 500-page book without providing the content relies on the model's internal memory of the book, which may be incomplete or hallucinated. To achieve reliability, you must provide the source text (e.g., via RAG) or, if the book is well-known, provide high-quality reference material within the context to ensure the summary is based on verifiable facts, not stochastic recall.

Answer

The request is too simple for the model.

Answer

The model cannot handle 500 pages of text.

Answer

The request for 3 sentences is too restrictive.

Question 9

Refer to the exhibit. The model output is a list, but it also includes introductory text like 'Here is the summary you requested:'. How can you ensure the output contains ONLY the list?

Accepted Answer

Add a negative constraint: 'Do not include conversational filler or introductory text.'. Models are naturally helpful and often include conversational filler. To force precise, non-conversational output, you must explicitly constrain the model in the system prompt by stating 'Output only the bulleted list; do not include any conversational filler or introductory text.' This is vital for downstream processing where unexpected chatty prefixes would cause integration errors, and it ensures the output strictly matches the expected schema for your application.

Answer

Set the temperature to 0.

Answer

Truncate the output in the application code.

Answer

Use a regex filter on the input document.

Question 10

Which of the following is an advantage of using XML tags in Anthropic prompts?

Accepted Answer

It helps the model differentiate between instructions and data.. XML tags provide a clear way to structure input, allowing the model to distinguish between instructions, examples, and data. Because Claude is trained to respect these delimiters, using them improves the model's ability to parse complex prompts. This architectural clarity reduces ambiguity, makes prompts easier to maintain, and leads to significantly more consistent performance, especially when dealing with complex data extraction or multi-step reasoning tasks.

Answer

It guarantees that the output will be in JSON format.

Answer

It automatically reduces the token usage of the prompt.

Answer

It is required for the Anthropic API to function.

Question 11

In an Evaluator-Optimizer agentic workflow, which THREE components are essential for creating a successful iterative refinement loop for high-quality content generation?

Accepted Answer

A generator model to produce the initial draft. The Evaluator-Optimizer pattern relies on a feedback loop where quality is improved through critique. This requires a generator to produce the initial work, a distinct set of evaluation criteria or a separate model to provide feedback, and a mechanism to feed that feedback back into the generator for the next iteration.

Answer

A router to bypass the evaluation for simple tasks

Answer

A parallel worker to execute the feedback simultaneously

Question 12

A security audit of an agentic architecture reveals a risk of 'Indirect Prompt Injection' where the agent reads a malicious email and then uses its 'delete_account' tool on itself. Which architectural change best mitigates this risk?

Accepted Answer

Moving sensitive tools like 'delete_account' to a 'Human-in-the-Loop' approval workflow.. Indirect prompt injection occurs when a model treats untrusted data (like an email) as a set of instructions. To prevent this, architects must separate the data from the instructions, use strict tool schemas, and implement 'Human-in-the-Loop' for high-impact tools to ensure that the agent's actions are always verified.

Answer

Adding a sentence to the system prompt telling Claude to ignore instructions in emails.

Answer

Implementing a separate 'Sandboxed Guardrail' agent to pre-scan all incoming data for commands.

Answer

Encoding all email text into a format like Base64 before showing it to the agent.

Question 13

A team wants to optimize their Claude Code workflow for a large monorepo. Which TWO of the following actions improve performance and context relevance?

Accepted Answer

Use .claudeignore to exclude build artifacts and node_modules.. Efficient monorepo management requires reducing the noise within the context window. By using a .claudeignore file, you prevent the agent from parsing irrelevant configuration or binary files. Additionally, focusing the working directory or using selective indexing ensures that the model operates only on the relevant sub-projects. These strategies significantly reduce token usage and improve the latency of context-aware operations, leading to a much more responsive developer experience.

Answer

Increase the max_tokens limit to account for the large codebase.

Answer

Always set temperature to 1.0 to maximize creativity in code.

Answer

Enable verbose logging to monitor every token sent to the API.

Question 14

When designing a multi-step tool process, which architecture most effectively prevents the model from getting stuck in an infinite loop?

Accepted Answer

Use a state machine to track progress and enforce a maximum step count.. Implementing a clear step-tracking mechanism or a 'max-iterations' limit in the orchestration logic prevents infinite loops. By forcing the model to complete a task within a predefined number of steps, the architect ensures that the process eventually terminates. This design pattern is critical for autonomous agents, ensuring that even if the model fails to reach a state transition, the overall system remains responsive and under control.

Answer

Enable auto-retry logic on every tool call.

Answer

Rely on the model's internal reasoning capability to detect and break loops.

Answer

Require human intervention for every tool call in the sequence.

Question 15

Refer to the exhibit. This technique, where the assistant's response is started with a specific tag, is known as 'prefilling.' How does this technique primarily improve reliability for structured output?

Accepted Answer

It steers the model to follow a specific logical and structural path.. Prefilling the assistant's response is a powerful technique for guiding Claude's behavior. By starting the response with a specific tag like <thought>, you force the model to enter a specific 'mode' of operation immediately. This ensures the model follows the desired format from the first token, which is essential for reliably parsing outputs in automated pipelines.

Answer

It reduces the token cost by skipping the model's initial reasoning.

Answer

It bypasses the system prompt's safety filters for technical code.

Answer

It allows the model to access external tools without a tool definition.