CCAR-P Advanced Agentic Architecture Practice Question
Your team operates a Claude agent that triages inbound customer support tickets. After several weeks in production, the agent begins approving refunds above the policy ceiling and citing outdated policy text. Investigation shows the system prompt embeds a policy document that was updated three weeks ago, but the deployed prompt was never regenerated. Which architectural practice most directly prevents this class of failure?
⚠ Common exam trap
The trap here is treating the symptom as a prompt-engineering problem, when the real defect is that authoritative, frequently changing content was hard-coded into the prompt.
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
✓
Move policy content out of the system prompt and retrieve the current policy at runtime via a versioned tool or retrieval step, recording the version used in each decision.
Stale behavior traced to a frozen prompt is best solved by externalizing the mutable knowledge and fetching it at decision time from a versioned source. That way policy updates propagate immediately, and logging the retrieved version makes each action auditable. Compression, randomness, or peer review all leave the embedded copy as the de facto source of truth.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✓
Move policy content out of the system prompt and retrieve the current policy at runtime via a versioned tool or retrieval step, recording the version used in each decision.
Why this is correct
Retrieving policy at runtime decouples the agent's behavior from a frozen prompt snapshot, so updates take effect without redeploying the prompt. Recording the retrieved version makes each decision reproducible and lets auditors see which policy text governed a given refund. This directly addresses staleness because the agent always reads the current authoritative source rather than an embedded copy.
- ✗
Add a second Claude agent that reviews the first agent's refund decisions and overrides any that appear inconsistent with its own recollection of company policy.
Why it's wrong here
A reviewer agent built from the same stale prompt inherits the same outdated policy, so it cannot reliably catch drift. Adding a second model also doubles latency and cost without establishing an authoritative source. The failure is a data freshness problem, and a peer review loop does not refresh the underlying policy text, so the override logic would be as wrong as the original decision.
- ✗
Increase the model's temperature slightly so the agent explores alternative interpretations of the policy text it has memorized.
Why it's wrong here
Raising temperature makes outputs more varied, which would worsen compliance drift rather than correct a stale embedded document. The root cause is that the prompt contains outdated policy, not that the model is too deterministic. More randomness increases the chance of inconsistent refund decisions and does nothing to refresh the source of truth, so it cannot resolve the stale-prompt defect.
- ✗
Shorten the system prompt by summarizing the policy into a compact bullet list, which reduces the chance that outdated sections remain embedded.
Why it's wrong here
Summarization compresses whatever text is currently embedded, so if the source is stale the summary is stale too. It may even lose nuance that governs edge cases like refund ceilings. The problem is not prompt length but the absence of a live link to the authoritative policy, so a shorter static prompt still drifts the moment policy changes.
About these practice questions
One of 262 original CCAR-P practice questions on Courseiva, each with a full explanation and wrong-answer analysis — not exam dumps or protected exam content. 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.