CCARF Practice Question

Paper 1 · Question 52

D1 Agentic Architecture & Orchestration · Paper 1

A developer's customer support agent sometimes terminates prematurely because they check if response.content[0].type == 'text' to determine completion. The agent stops even when Claude has more tools to call. What is the bug and how should they fix it?

• A.Add an iteration cap of 15 loops and treat reaching the cap as the signal that the agent has finished its work.
• B.Check the stop_reason field instead of content type: continue on 'tool_use', terminate on 'end_turn'.
• C.Parse the assistant's text for phrases like 'I have completed' before terminating
• D.Set tool_choice to 'any' so Claude always calls a tool instead of returning text

Show answer & explanations

Correct answer: B

A: Arbitrary caps do not address the root cause, and using the cap as a completion signal is itself an anti-pattern. The agent exits early because it misreads the response type; the fix is to inspect stop_reason, not to cap iterations.

B: The stop_reason field is the deterministic, authoritative signal for loop control. Claude can return text alongside tool_use blocks, so checking content type is unreliable.

C: Natural language parsing is ambiguous and unreliable. The stop_reason field already provides an unambiguous signal.

D: This forces tool use even when the agent is genuinely finished, creating an infinite loop rather than fixing the detection logic.

Related Agentic Architecture & Orchestration questions

More: Paper 1 · All practice questions