CCAR-P Advanced Agentic Architecture Practice Question
You are architecting a Claude agent that orchestrates a long-running approval workflow spanning hours or days, where human reviewers may intervene between steps. Which TWO mechanisms are necessary to keep the workflow correct across these interruptions? (Choose two.)
⚠ Common exam trap
The trap here is believing that reconstructing context from summaries is equivalent to persisted state, when only durable checkpoints preserve pending tool calls and exact arguments.
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
✓
Persist the full message history and tool state to durable storage after each turn so the workflow can resume from the exact checkpoint.
Long-running workflows with human-in-the-loop interruptions require durable checkpointing to resume exact state and idempotency keys to make retries safe. Together they ensure that an interruption or crash neither loses the reasoning chain nor duplicates side effects such as approvals or notifications. Stateless summarization, token limits, and disabling tools all fail to provide either guarantee.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Set a very short max_tokens on every turn to reduce the chance of context drift across days.
Why it's wrong here
Limiting output length does not address durability or duplication; it merely truncates reasoning. Over a multi-day workflow, the risk is lost state and repeated side effects, not verbose output. A short token budget could even cut off a critical tool call, so it actively harms correctness rather than helping it.
- ✓
Persist the full message history and tool state to durable storage after each turn so the workflow can resume from the exact checkpoint.
Why this is correct
Durable checkpointing is essential because the process may be suspended or crash between human interventions. Saving message history and tool state lets the workflow resume mid-conversation without re-running side effects or losing prior reasoning. Without it, a restart would either duplicate actions or force the agent to begin again, breaking the approval chain.
- ✗
Disable tool use entirely during human review windows and rely on the reviewer to re-enter all prior context manually.
Why it's wrong here
Disabling tools does not preserve workflow state and pushes an unreasonable burden onto the human reviewer, who would have to reconstruct machine context by hand. It also risks inconsistent re-entry and lost audit data. The workflow needs automated continuity, not manual reconstruction, so this fails the scenario.
- ✓
Implement idempotency keys on all side-effecting tool calls so retries after an interruption do not duplicate approvals or notifications.
Why this is correct
Idempotency keys ensure that if a step is retried after a crash or a resumed session, the downstream system recognizes the duplicate and does not execute the effect twice. In an approval workflow, double-sending a notification or double-recording an approval is unacceptable. This complements durable state by making recovery safe even when a call was in flight.
- ✗
Use a stateless design where each invocation rebuilds context by summarizing all prior decisions into the system prompt.
Why it's wrong here
Rebuilding context through summarization is lossy and cannot represent pending tool calls or exact argument values. For an approval workflow, losing a pending tool_use or an exact amount would corrupt the audit trail. Stateless reconstruction also grows the prompt unboundedly and risks dropping legally relevant detail, so it is the opposite of what this scenario needs.
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.