CCAR-P Advanced Agentic Architecture Practice Question
A production agent uses the Claude Messages API with extended thinking enabled to solve multi-constraint scheduling problems. The agent must preserve its reasoning across several tool calls within a single user turn. A developer notices that after the second tool call, the model appears to forget earlier constraints it had already reasoned about. Which change best preserves the reasoning chain across tool calls?
⚠ Common exam trap
The trap here is treating thinking budget as a memory setting, when the real requirement is to preserve the original thinking blocks in the conversation history across tool calls.
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
✓
Pass the full assistant message, including thinking blocks, back into the conversation history on each subsequent request.
Extended thinking stores reasoning in thinking blocks that are part of the assistant message. When an agent makes multiple tool calls within one turn, each follow-up request must include the prior assistant message with its thinking blocks intact. Omitting them causes the model to lose the constraints it already reasoned through, producing the observed regression after the second tool call.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✓
Pass the full assistant message, including thinking blocks, back into the conversation history on each subsequent request.
Why this is correct
With extended thinking, the reasoning is carried in thinking blocks attached to assistant messages. To keep that reasoning available across tool calls in the same turn, the orchestrator must return the complete assistant message, thinking blocks included, in the next request's messages array. Stripping them discards the model's prior reasoning and forces it to re-derive constraints, which explains the apparent forgetting after the second tool call.
- ✗
Disable extended thinking for subsequent requests once the first tool call has returned, since the constraints are already known.
Why it's wrong here
Turning off extended thinking mid-turn removes the mechanism that maintains structured reasoning and can cause the model to answer without the benefit of prior deliberation. The constraints were derived through thinking, so disabling it discards the very context that needs to persist. It also creates an inconsistent request pattern within a single turn, which can confuse orchestration logic that expects thinking blocks to be present.
- ✗
Increase the thinking budget_tokens value so the model has more room to reason on each request.
Why it's wrong here
The thinking budget controls how much internal reasoning the model may generate in a single response, not whether prior reasoning is retained across requests. If earlier thinking blocks are dropped from the conversation history, a larger budget only produces fresh reasoning that may contradict or omit previously established constraints. The retention problem is about message history, not about per-turn reasoning capacity.
- ✗
Summarize the thinking blocks into a single text note and append it as a user message before the next tool call.
Why it's wrong here
Summarizing discards the structured reasoning and replaces it with a lossy paraphrase. The model may then treat the summary as external instruction rather than its own prior reasoning, and nuance about constraints can be lost. It also breaks the expected message structure for extended thinking. Preserving the original thinking blocks is more faithful and avoids introducing summarization errors into the reasoning chain.
About these practice questions
Courseiva writes every CCAR-P question from scratch — 262 in total, each with an explanation and a wrong-answer breakdown. None are copied from real exams or 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.