CCAR-P Advanced Agentic Architecture Practice Question
You are operating a Claude-based support agent that must follow a strict refund policy: refunds over $500 require a manager approval code that is only obtainable through a separate internal API. The agent has access to a `get_manager_code` tool, but in production it occasionally issues refunds above $500 without calling the tool. Which architectural change most reliably prevents this?
⚠ Common exam trap
The trap here is assuming that stronger prompting or lower temperature can enforce a hard business rule, when only deterministic interception outside the model can guarantee it.
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Add a pre-tool-use hook that inspects the refund amount and blocks any refund tool call above $500 unless a valid manager code is present in the session state.
Hard business constraints must be enforced deterministically outside the model. A pre-tool-use hook inspects the pending refund call and rejects any amount above $500 that lacks a validated manager code, making the violation impossible rather than improbable. Prompting, temperature, and token limits only shift probabilities and cannot guarantee compliance with a policy that carries financial or legal consequences.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Expand the system prompt to describe the $500 threshold in bold and add two few-shot examples of correct refund handling.
Why it's wrong here
Prompt engineering improves the probability of correct behavior but never guarantees it; the model can still skip the approval call under distribution shift or long context. A regulatory-style rule needs deterministic enforcement, not persuasive text. Examples and emphasis reduce but do not eliminate the failure, so this is insufficient here.
- ✗
Increase the max_tokens setting so the agent has more room to reason about whether approval is needed.
Why it's wrong here
Token budget affects how much the model can generate, not whether it respects a business rule. A larger budget may even give more room for an incorrect chain of reasoning. It does nothing to guarantee the approval API is consulted, so it cannot prevent the over-threshold refund from being issued.
- ✓
Add a pre-tool-use hook that inspects the refund amount and blocks any refund tool call above $500 unless a valid manager code is present in the session state.
Why this is correct
A pre-tool-use hook runs deterministically before the tool executes, so it can inspect the refund payload and reject any call over $500 that lacks a verified manager code. This enforces the policy outside the model's probabilistic reasoning, which is exactly where a hard business rule belongs. It makes violation architecturally impossible rather than merely unlikely.
- ✗
Lower the model temperature to 0 so the agent produces deterministic decisions.
Why it's wrong here
Temperature 0 makes token selection more deterministic but does not guarantee policy compliance; the model can still consistently choose the wrong path. It cannot inject the missing business constraint that a refund above $500 must be gated by the approval API. Determinism is not correctness, so this fails to enforce the rule.
About these practice questions
This CCAR-P question is part of Courseiva's 262-question bank — original exam-style content with full explanations and wrong-answer analysis, never real exam questions or exam dumps. Learn why practice questions differ from exam dumps →
JA
Written and reviewed by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
Last reviewed September 2026 · checked against the official Anthropic exam blueprint
This CCAR-P practice question is part of Courseiva's free Anthropic certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the CCAR-P exam.