View Full Anthropic CCA-F Exam Dumps and Practice Test Dumps.

 

Question 61

An agent needs to perform a task that requires information from two independent APIs. Neither API depends on the result of the other. What is an efficient orchestration approach?

1. Call both APIs concurrently when supported, then combine their results.
2. Call the first API repeatedly before calling the second.
3. Ignore one API and use only the other.
4. Make the second API depend artificially on the first.

Correct Answer: 1

Explanation

When two tool calls are independent, they can often be executed concurrently to reduce overall latency. The orchestration layer can initiate both calls, wait for their results, and then provide the combined information to the next stage of the workflow. There is no benefit in artificially creating a dependency between independent operations. Repeatedly calling one API also wastes resources and may increase cost or rate-limit pressure. Ignoring one API could leave the agent without required information. Parallel execution should still include appropriate timeout, error handling, authorization, and rate-limit controls.

Question 62

A developer wants Claude to use a company-specific terminology consistently across a project. What is the most appropriate approach?

1. Provide clear project-specific terminology guidance in the relevant instructions or project documentation.
2. Expect Claude to infer all terminology from its general training.
3. Randomly change terminology throughout the project.
4. Remove domain-specific terms from the application.

Correct Answer: 1

Explanation

Project-specific terminology should be explicitly documented so Claude has reliable guidance when working on the project. Instructions can define preferred names, abbreviations, product terminology, naming conventions, and terms that should be avoided. Relying entirely on general model knowledge may result in terminology that does not match the organization’s conventions. Randomly changing terms creates inconsistency, while removing useful domain terminology reduces clarity. Maintaining these conventions in project documentation also makes them easier for developers and other tools to follow consistently across tasks.

Question 63

A tool requires a date in the format YYYY-MM-DD, but Claude sometimes supplies dates in natural-language formats. What is the best way to improve reliability?

1. Accept every possible format without validation.
2. Clearly define the required format in the tool schema and validate the input.
3. Remove the date parameter.
4. Ask Claude to guess the expected format.

Correct Answer: 2

Explanation

The required date format should be explicitly documented in the tool schema and enforced through application-side validation. This gives Claude a clear contract for constructing the tool call while ensuring that invalid values do not reach the backend. Accepting arbitrary formats may cause inconsistent behavior or parsing failures. Removing the date parameter would eliminate necessary information rather than solve the formatting problem. Asking Claude to guess the format introduces unnecessary ambiguity. Schema descriptions, validation, and useful error messages together provide a stronger and more predictable interface.

Question 64

An agent repeatedly asks the user for information that was already provided earlier in the same workflow. What should the developer investigate first?

1. Whether relevant state or context is being preserved and passed between steps.
2. Whether the model should be given random instructions.
3. Whether all previous information should be deleted.
4. Whether the temperature should always be set to maximum.

Correct Answer: 1

Explanation

Repeatedly asking for information can indicate that important state is not being preserved between workflow stages. The application should inspect how conversation context, tool results, and structured workflow state are passed to subsequent model calls. If information is available but not included where needed, the agent may reasonably behave as though it does not know it. Deleting previous information would worsen the issue. Random instructions and extreme temperature settings do not address state management. Reliable workflows explicitly preserve and retrieve important information instead of assuming every stage automatically has access to it.

Question 65

A company wants to allow an agent to read customer records but not modify them. Which design best follows this requirement?

1. Give the agent full read/write database credentials.
2. Expose only a read-only customer lookup capability.
3. Give the agent administrator access and rely on instructions.
4. Allow arbitrary SQL queries against the production database.

Correct Answer: 2

Explanation

A read-only tool provides a narrower capability that directly matches the required task. It prevents the agent from making modifications even if an incorrect tool request is generated. Full database credentials or administrator privileges grant capabilities beyond what the task requires. Arbitrary SQL access is particularly broad and can introduce unnecessary security and reliability risks. This design follows least privilege by exposing only the operation needed by the agent. The read-only interface should still enforce authentication, authorization, input validation, and appropriate logging.

Question 66

An agent receives an API error indicating that a request exceeded the service’s rate limit. What response is generally appropriate?

1. Immediately send unlimited additional requests.
2. Ignore the error.
3. Apply an appropriate backoff and retry policy when the operation is safely retryable.
4. Permanently disable the API.

Correct Answer: 3

Explanation

Rate-limit errors generally indicate that the application should temporarily reduce request frequency. A controlled retry strategy with appropriate backoff can allow the service to recover before another attempt is made. The application should also respect any retry guidance provided by the service and establish a maximum number of attempts. Unlimited retries can worsen the rate-limit problem. Ignoring the error prevents useful recovery, while permanently disabling the API may be unnecessary. Retry behavior should also distinguish safely repeatable operations from operations that could create duplicate side effects.

Question 67

A developer wants to make a tool’s purpose immediately understandable to Claude. Which tool name is most descriptive?

1. run
2. data
3. process
4. get_customer_order_status

Correct Answer: 4

Explanation

Descriptive tool names provide semantic information that helps Claude distinguish capabilities. A name such as get_customer_order_status clearly communicates that the tool retrieves the status of a customer’s order. Generic names such as run, data, or process provide little information about what the operation actually does. Tool descriptions should complement the name by explaining appropriate use cases, parameters, constraints, and expected output. Clear naming reduces ambiguity, especially when multiple tools are available in the same agent environment.

Question 68

An agent can trigger a payment operation, but the payment amount comes from user-provided text. What should happen before the payment is submitted?

1. The application should validate the amount and relevant transaction parameters.
2. The agent should submit any amount it generates.
3. The application should remove all validation to reduce latency.
4. The model should be allowed to modify the amount after submission.

Correct Answer: 1

Explanation

Financial transaction parameters should be validated before execution. The application can verify that the amount is correctly formatted, within permitted limits, associated with the correct currency and account, and consistent with applicable business rules. User-provided text should not automatically be treated as a trusted transaction instruction without validation. Allowing the model to modify a payment after submission can introduce serious inconsistencies. Removing validation simply to reduce latency is inappropriate for a sensitive operation. Deterministic application checks provide an important control around model-generated tool requests.

Question 69

A team wants to determine why an agent failed a task involving multiple tool calls. Which trace would provide the most useful information?

1. Only the final answer
2. Only the user’s original prompt
3. The sequence of model decisions, tool calls, arguments, results, and errors
4. Only the application startup logs

Correct Answer: 3

Explanation

A complete execution trace can reveal where an agent workflow failed. It can show which tool was selected, what arguments were generated, whether the tool succeeded, what result was returned, and how the workflow proceeded afterward. Looking only at the final answer hides the intermediate events that may have caused the failure. The original prompt is useful but does not show runtime behavior. Startup logs generally provide little information about individual agent decisions. Structured tracing therefore provides a much clearer view of multi-step agent execution and supports debugging and performance analysis.

Question 70

A developer needs to ensure that a tool cannot be called without a required account_id. Which implementation provides the strongest protection?

1. Mention the requirement only in a comment.
2. Make account_id required in the schema and enforce validation before execution.
3. Let the model invent an account ID if it is missing.
4. Execute the tool with an empty value.

Correct Answer: 2

Explanation

Required parameters should be represented explicitly in the tool schema and validated by the application before the underlying operation runs. This creates a clear interface contract and prevents malformed requests from reaching the backend. A comment may help developers but does not necessarily constrain model-generated calls. Allowing the model to invent an identifier could result in accessing the wrong account. Executing with an empty value can produce errors or unintended behavior. Strong schema definitions combined with deterministic validation provide a more reliable mechanism for protecting required inputs.

Question 71

A research agent retrieves information from multiple sources that disagree with each other. What should the agent ideally do?

1. Automatically assume the newest-looking statement is correct.
2. Hide the disagreement from the user.
3. Preserve source attribution and represent the conflicting information appropriately.
4. Select a source randomly.

Correct Answer: 3

Explanation

When credible sources disagree, preserving source attribution allows the agent and user to understand where each claim originated. The workflow can identify the disagreement, compare the evidence, and communicate uncertainty rather than silently selecting an arbitrary answer. Automatically choosing a source without justification can produce misleading results. Hiding disagreements removes useful information from the user. Random selection has no reliable relationship to source quality. Source provenance and evidence tracking are therefore important components of research-oriented agent architectures.

Question 72

A developer is creating a tool that allows an agent to update a customer’s phone number. Which design is most appropriate?

1. Use a narrowly scoped update tool with explicit parameters and authorization checks.
2. Give the agent unrestricted database access.
3. Allow arbitrary SQL commands.
4. Allow the agent to update any customer without authentication.

Correct Answer: 1

Explanation

A narrowly scoped tool exposes only the capability required by the workflow. An update-phone-number tool can define the customer identifier and new phone number as explicit parameters while allowing the application to validate authorization and input format. Unrestricted database access and arbitrary SQL provide far more power than necessary and increase security risks. Allowing updates without authentication could expose customer information or permit unauthorized modifications. Narrow tools also make testing and auditing easier because their intended behavior is well defined.

Question 73

A team is designing a workflow in which an agent may make several attempts to solve a task. Which control helps prevent excessive execution?

1. An unlimited iteration loop
2. A maximum number of iterations or execution budget
3. Removing all tool results
4. Allowing the model to continue forever

Correct Answer: 2

Explanation

An execution budget or maximum iteration count provides a deterministic boundary for agent loops. It prevents an unsuccessful workflow from consuming unlimited model calls, tool requests, time, or other resources. The application can terminate the workflow when the limit is reached and return an appropriate status or ask for human intervention. Unlimited loops create reliability and cost risks. Removing tool results would prevent the agent from making informed decisions but would not provide proper termination. Execution limits are an important part of production agent orchestration.

Question 74

An application needs to distinguish between a temporary API outage and an invalid user request. Why is this distinction important?

1. Both conditions should always be retried indefinitely.
2. The application may use different recovery strategies for each type of failure.
3. The distinction has no effect on agent behavior.
4. Invalid requests should always be executed again.

Correct Answer: 2

Explanation

Different failure types require different handling strategies. A temporary service outage may justify a limited retry with backoff, while an invalid user request may require correction or additional information rather than another identical attempt. Treating both conditions identically can waste resources or repeatedly submit invalid requests. Structured error categories help the application and agent determine the appropriate next step. This separation also improves observability because teams can distinguish infrastructure failures from input or business-logic problems when analyzing production behavior.

Question 75

A developer wants an agent to summarize a document while preserving important numerical values exactly. Which approach can improve reliability?

1. Explicitly instruct the model to preserve required values and validate critical fields afterward.
2. Tell the model that numbers are not important.
3. Remove numerical information from the document.
4. Randomly change the values before summarization.

Correct Answer: 1

Explanation

Important numerical information should be explicitly identified as data that must be preserved, and critical values can be checked programmatically after generation when practical. This is particularly useful for financial figures, dates, measurements, identifiers, or other values where small changes can affect meaning. Removing numbers or changing them before processing destroys important information. Simply relying on a general instruction without validation may still allow occasional transcription errors. Combining precise instructions with deterministic checks provides stronger protection for critical structured information.

Question 76

A team is building an agent that can access several internal services. What is an important reason to keep each service capability behind a well-defined tool interface?

1. It makes permissions, validation, and behavior easier to control.
2. It guarantees that the model can never make mistakes.
3. It eliminates the need for authentication.
4. It allows unrestricted access to every service.

Correct Answer: 1

Explanation

Well-defined tool interfaces create clear boundaries between the agent and external services. Each interface can specify allowed operations, required parameters, authorization requirements, validation rules, and expected responses. This makes the system easier to secure, test, monitor, and maintain. Tool boundaries do not guarantee that the model will never make mistakes, but they can limit the impact of incorrect decisions. Authentication and authorization remain necessary. The goal is not unrestricted access but controlled capabilities that match the tasks the agent is expected to perform.

Question 77

An agent needs to ask a user for clarification because two possible interpretations of the request would result in different actions. What is the safest behavior?

1. Pick one interpretation silently.
2. Execute both actions automatically.
3. Ask a focused clarification question before performing the consequential action.
4. Ignore the ambiguity.

Correct Answer: 3

Explanation

When ambiguity could lead to different consequential actions, asking a focused clarification question helps prevent unintended side effects. The agent should explain the specific information needed and present the relevant alternatives when appropriate. Silently choosing an interpretation can cause the system to perform an action the user did not intend. Executing both possibilities may create unnecessary or harmful side effects. Ignoring ambiguity does not resolve the underlying uncertainty. Clarification is particularly important for financial transactions, data changes, communications, and other operations that may be difficult to reverse.

Question 78

A developer wants to reduce the risk of duplicate records when an agent retries a create operation after a timeout. Which technique can help?

1. Use idempotency controls or unique request identifiers where supported.
2. Retry the operation indefinitely without tracking requests.
3. Remove all transaction identifiers.
4. Create multiple copies intentionally.

Correct Answer: 1

Explanation

Retries can create duplicate side effects when the initial request may have succeeded but the response was lost or delayed. Idempotency keys or unique request identifiers allow the receiving service to recognize repeated requests and avoid performing the same operation multiple times. Indefinite retries increase the risk of duplicates and resource consumption. Removing identifiers makes it harder to detect repeated operations. Intentionally creating multiple copies does not solve the reliability problem. Idempotency is particularly valuable for payment, order creation, account updates, and other operations where duplicate execution has significant consequences.

Question 79

A team wants to protect sensitive information that may appear in agent logs. Which practice is appropriate?

1. Log every secret in plain text.
2. Apply appropriate redaction or filtering before sensitive information is stored in logs.
3. Disable all security controls.
4. Share production logs publicly for debugging.

Correct Answer: 2

Explanation

Logs can contain sensitive information such as credentials, personal data, tokens, or private business information. Appropriate redaction and filtering can prevent unnecessary exposure while still retaining useful operational information. Logging every secret in plain text creates a serious security risk. Disabling security controls does not address the problem, and publicly sharing production logs can expose confidential information. Logging systems should be designed with data minimization, access controls, retention policies, and appropriate redaction. Debugging requirements should be balanced with privacy and security considerations.

Question 80

A developer is designing an agent that can both retrieve information and modify external systems. Why should these capabilities generally be treated differently?

1. Retrieval and modification always have exactly the same risk.
2. Modification can create external side effects, so it may require stronger authorization and confirmation controls.
3. Retrieval should always be prohibited.
4. Modification should never be logged.

Correct Answer: 2

Explanation

Read operations generally retrieve information, while modification operations can change external state and may be difficult to reverse. Because of this difference, write actions often require stronger controls such as authorization checks, parameter validation, confirmation, approval workflows, transaction limits, audit logging, and rollback mechanisms. Retrieval may still involve sensitive information and therefore requires appropriate access controls, but its risk profile can differ from a destructive or consequential write operation. Treating every capability identically can lead to either excessive restrictions or insufficient safeguards. Risk-based controls help match protections to the potential impact of each action.