Free CCAR-F Exam Braindumps
Claude Certified Architect - Foundations Questions 21–25 (page: 6)

Updated On: 17-Nov-2026

Your publish article tool calls an external CMS API that occasionally returns transient errors (network timeouts,
503s) and non-transient errors (403 permission denied, 422 validation failure). Currently, every error is returned directly to the agent, which leads to the agent retrying non-transient errors and wasting turns on failures that will never succeed. How should you partition error-handling responsibility between the tool implementation and the agent?

  1. Handle all errors inside the tool: Implement retries with exponential backoff for every error type, and only surface a failure to the agent after a fixed number of retry attempts have been exhausted.
  2. Handle transient errors (timeouts, 503s) with automatic retries inside the tool implementation, and surface non-transient errors (permission denied, validation fallures) to the agent with descriptive messages so it can take corrective action.
  3. Surface all errors to the agent immediately with detailed context, and let the agent decide which errors to retry and how many times-keeping the tool implementation stateless and simple.
  4. Implement a universal error handler that catches all exceptions and returns a generic "tool unavailable-try again later" message, shielding the agent from error complexity.

Answer(s): B

Explanation:

A: Handle all errors inside the tool: Implement retries with exponential backoff for every error type, and only surface a failure to the agent after a fixed number of retry attempts have been exhausted. Incorrect.
This wastes time retrying non-transient errors (e.g., 403, 422) that will never succeed and hides useful feedback from the agent.
B: Handle transient errors (timeouts, 503s) with automatic retries inside the tool implementation, and surface non-transient errors (permission denied, validation failures) to the agent with descriptive messages so it can take corrective action. Correct.
This cleanly separates responsibility:
Tool handles recoverable/transient issues automatically Agent receives actionable errors it can fix (permissions, input validation)
C: Surface all errors to the agent immediately with detailed context, and let the agent decide which errors to retry and how many times—keeping the tool implementation stateless and simple. Incorrect.
This pushes retry logic to the agent, leading to inefficient behavior and wasted turns .
D: Implement a universal error handler that catches all exceptions and returns a generic "tool unavailable —try again later" message, shielding the agent from error complexity. Incorrect.
This removes critical detail, preventing the agent from taking corrective actions when possible.



Your remove_team_member tool uses a dry_run: boolean parameter for previewing impacts before execution. Production monitoring shows the agent bypasses the preview step in 15% of calls by calling with dry_run=false directly. You need to ensure every removal is preceded by a preview that the user explicitly confirms.
What is the most reliable approach?

  1. Add server-side validation that permits dry_run=false only when a dry_run=true call with identical parameters occurred within the past 60 seconds.
  2. Replace with two tools: preview_remove_member returns impact details and a single-use confirmation token; execute_remove_member requires that token, binding execution to the specific previewed action.
  3. Annotate the tool as requiring confirmation and configure the orchestration layer to prompt the user for approval before forwarding any calls to annotated tools.
  4. Add detailed instructions and few-shot examples to the tool description requiring the agent to always call with dry_run=true first and wait for user confirmation before calling with dry_run=false.

Answer(s): B

Explanation:

A: Add server-side validation that permits dry_run=false only when a dry_run=true call with identical parameters occurred within the past 60 seconds. Incorrect.
This approach is brittle because it depends on timing and does not guarantee that the user actually reviewed or confirmed the preview.
B: Replace with two tools: preview_remove_member returns impact details and a single-use confirmation token; execute_remove_member requires that token, binding execution to the specific previewed action. Correct.
This enforces the correct workflow at the system level by requiring a valid preview step and tying execution to an explicit confirmation, making bypass impossible.
C: Annotate the tool as requiring confirmation and configure the orchestration layer to prompt the user for approval before forwarding any calls to annotated tools. Incorrect.
This depends on orchestration behavior and is not strictly enforced, so it can still be bypassed or misconfigured.
D: Add detailed instructions and few-shot examples to the tool description requiring the agent to always call with dry_run=true first and wait for user confirmation before calling with dry_run=false. Incorrect.
Instruction-based approaches are not reliable for enforcement, as demonstrated by the existing bypass rate.



Your expense reimbursement agent processes employee requests using a process reimbursement tool. Company policy requires that reimbursements above $500 must be approved before funds are disbursed. The agent handles hundreds of requests daily, and you need the threshold enforcement to be tamper-proof regardless of how the agent is prompted ensures the $500 approval threshold cannot be bypassed?

  1. The process reimbursement tool accepts an approved by manager parameter. The system prompt instructs the agent to only set this to true after confirming that a manager approved the request. A nightly audit script reviews all reimbursements where approved by manager was set to true.
  2. Provide two tools: auto reimburse (hard-coded limit of $500) and manager approval. Include detailed system prompt instructions telling the agent to check the amount and use the appropriate tool. Add a Post ToolUse hook that logs which tool was called for auditing.
  3. The process reimbursement tool accepts amount and details, and internally enforces the threshold; amounts <$500 are auto-disbursed and the tool returns a success confirmation. Amounts >$500 cause the tool to create a pending approval request and return a status indicating manager review is pending.
  4. Implement the threshold check in a PreToolUse hook that inspects the amount parameter before process reimbursement executes. If the amount exceeds $500, the hook modifies the context to add a requires approval: true flag, which the tool checks before disbursing.

Answer(s): C

Explanation:

A: The process reimbursement tool accepts an approved by manager parameter. The system prompt instructs the agent to only set this to true after confirming that a manager approved the request. A nightly audit script reviews all reimbursements where approved by manager was set to true. Incorrect.
This relies on the agent following instructions and post-hoc auditing, which is not tamper-proof and allows bypass at execution time.
B: Provide two tools: auto reimburse (hard-coded limit of $500) and manager approval. Include detailed system prompt instructions telling the agent to check the amount and use the appropriate tool. Add a Post ToolUse hook that logs which tool was called for auditing. Incorrect.
Again depends on agent behavior and correct tool selection. Logging helps auditing but does not prevent misuse.
C: The process reimbursement tool accepts amount and details, and internally enforces the threshold; amounts <$500 are auto-disbursed and the tool returns a success confirmation. Amounts >$500 cause the tool to create a pending approval request and return a status indicating manager review is pending. Correct.
This enforces the rule inside the tool itself , making it impossible to bypass regardless of how the agent is prompted.
D: Implement the threshold check in a PreToolUse hook that inspects the amount parameter before process reimbursement executes. If the amount exceeds $500, the hook modifies the context to add a requires approval: true flag, which the tool checks before disbursing. Incorrect.
PreToolUse hooks can be bypassed or misconfigured and still rely on downstream logic. Enforcement should reside directly within the tool for full reliability.



Your order management system requires tools for three distinct operations: issuing refunds (requires amount and reason), canceling orders (requires reason), and res (requires shipping address). Each operation shares an order id parameter but has different additional requirements. You notice during testing that with your current frequently omits required parameters or includes irrelevant ones.
What design change will most effectively improve parameter accuracy?

  1. Split into three separate tools (each defining only the parameters required for that specific operation.
  2. Keep one unified tool with all parameters marked optional, but add few-shot examples in the system prompt showing correct parameter combinations for each operation.
  3. Keep one unified tool but add JSON Schema if-then-else conditionals to enforce that parameters like amount are required only when the operation type is "refund".
  4. Keep one unified tool with a nested operation object parameter whose internal structure varies by operation type, documented in the tool description.

Answer(s): A

Explanation:

A: Split into three separate tools (e.g., issue_refund, cancel_order, reship_order), each defining only the parameters required for that specific operation. Correct.
This reduces ambiguity and ensures the agent only sees relevant parameters per operation , leading to much higher accuracy.
B: Keep one unified tool with all parameters marked optional, but add few-shot examples in the system prompt showing correct parameter combinations for each operation. Incorrect.
Examples help, but the schema remains ambiguous, so errors will still occur.
C: Keep one unified tool but add JSON Schema if-then-else conditionals to enforce that parameters like amount are required only when the operation type is "refund". Incorrect.
While technically valid, this increases complexity and is less reliable than simply separating tools.
D: Keep one unified tool with a nested operation object parameter whose internal structure varies by operation type, documented in the tool description. Incorrect.
This adds complexity and cognitive load, making it harder for the agent to consistently provide correct parameters.



Your portfolio value tool returns the total value of a user's investment portfolio. You're deciding between returning a structured JSON object with explicit fields versus returning information as a formatted text string.
What is the primary advantage of using structured output with defined fields?

  1. Structured JSON consumes significantly fewer tokens than natural language, substantially reducing API costs.
  2. The agent can reliably extract specific values without parsing free form text, reducing errors in subsequent operations.
  3. Structured JSON is processed deterministically by the model, significantly improving accuracy when extracting values.
  4. JSON schemas automatically validate that the underlying API returned correct data before the agent processes it.

Answer(s): B

Explanation:

A: Structured JSON consumes significantly fewer tokens than natural language, substantially reducing API costs. Incorrect.
Token usage depends on the content; JSON is not inherently more compact than text and may sometimes use more tokens.
B: The agent can reliably extract specific values without parsing free form text, reducing errors in subsequent operations. Correct.
Structured output provides clear, predictable fields , making it easy for the agent to use the data accurately in downstream steps.
C: Structured JSON is processed deterministically by the model, significantly improving accuracy when extracting values. Incorrect.
The model is still probabilistic; JSON improves structure, but not deterministic processing.
D: JSON schemas automatically validate that the underlying API returned correct data before the agent processes it. Incorrect.
Schemas define structure, but they do not guarantee correctness of the actual data returned by the API.



Viewing page 6 of 31
Viewing questions 26 - 30 out of 149 questions


Post your Comments and Discuss Anthropic CCAR-F exam prep with other Community members:

AI Tutor AI Tutor 👋 I’m here to help!