CCAR-P - Integration (19% of the exam) - Section 3.1

Evaluate tool/agent configuration for capability bloat.

Reviewing an agent's tool set for capabilities its role does not need. Excess tools widen the attack surface and degrade tool selection. The guide's sample item turns on least privilege: removing an unneeded capability beats logging it, confirming it or switching to a larger model.

least privilegetool set scopingattack surfacetool selection accuracy

Practice question for this objective

Free sampleIntegrationmedium

A software company's internal IT helpdesk agent answers staff questions about laptops, access requests and VPN faults. Its weekly score on a fixed 300-question evaluation set fell from 91 percent to 77 percent between two runs. The change log for that week shows the model version, system prompt and IT knowledge base unchanged, and one platform change: the agent's six scoped tools were replaced by a shared enterprise bundle of 58 tools also used by the finance, HR and facilities agents. In 62 of the 69 newly failing cases, the agent's first tool call went to a system outside IT. What should the architect investigate first?

ARerun the evaluation on a larger model tier, since a stronger model should pick the right tool even from a long tool list.
BCompare the tools the agent now holds with the six its role uses, since the bundle swap is the only change in the failing window. Correct
CCheck the IT knowledge base index for stale entries, since retrieval faults are a frequent cause of a sudden accuracy drop.
DEnlarge the evaluation set before acting, since a 14-point drop on 300 questions could be sampling noise rather than a real change.

When an agent's accuracy drops after its tool set is widened, investigate the tool set first, because overlapping tools from other roles degrade tool selection. Tool selection accuracy depends on how many candidate tools the model must choose between and how much they overlap. Replacing six scoped tools with a 58-tool shared bundle gave the helpdesk agent dozens of tools from other departments, and the failing transcripts show it calling them first. With the model, prompt and knowledge base held constant, the tool set change is the only candidate cause, so it is the first thing to confirm before scoping the agent back to its role.

Why A is wrong: This is tempting because larger models are often better at tool selection. It is wrong as a first step because the model did not change, so a model swap would mask rather than explain a regression caused by the tool set, and it adds cost without removing the 52 tools the role does not need.

Why B is correct: Correct. The 'what changed' evidence points to one change, and the failure pattern of first calls going to non-IT systems is the signature of wrong-tool selection in a bloated tool set. Confirming that the extra 52 tools are what the agent is reaching for leads directly to scoping it back to its six.

Why C is wrong: This is tempting because retrieval and indexing are a common source of confident-but-wrong answers. It is wrong here because the knowledge base was unchanged and most failures began with a call to a system outside IT, before any IT retrieval took place.

Why D is wrong: This is tempting because small evaluation sets can produce misleading swings. It is wrong because on a fixed set of 300 questions a drop from 91 to 77 percent is far larger than run-to-run noise, and the 69 new failures share one clear pattern.

See more CCAR-P practice questions, answers explained.

Exam traps in Integration

Answers that look right on this material and are not. Each one is a distractor from a different question in the CCAR-P bank for this domain.

Keep both tools but require the clinician to approve each order or cancellation in a confirmation dialog before it runs
Why it is wrong: A confirmation step feels safe because a human sees every write before it lands. It is wrong because it is a compensating control on a capability the role does not need, it adds exactly the workflow interruptions clinicians said would end the pilot, and approval fatigue tends to turn confirmations into reflexive clicks.
The confirmation dialog shows too little about each write, so caseworkers need fuller detail on screen before they can approve.
Why it is wrong: This is tempting because a richer dialog can improve the quality of human review. It is wrong because it keeps a compensating control around capabilities the drafting role does not need; the structural fix is to remove the write tools, after which there is nothing to confirm.
Add the lookup tool behind an approval workflow so the data custodian signs off on each identified lookup the agent makes
Why it is wrong: Custodian sign-off is tempting because it keeps a responsible person in the loop for each identified lookup. It is wrong because the agent's approval does not cover identified data at all, so granting the capability breaches the governance boundary regardless of who signs off, and it rebuilds a process that already exists.

Examworthy is not affiliated with or endorsed by Anthropic. Original, blueprint-aligned practice material only.