This quiz is interactive and needs JavaScript. Here are the questions for reference:
1. Which stop_reason value is the only one that continues the agentic loop?
- A)
end_turn - B)
tool_use - C)
pause_turn - D)
max_tokens
2. A response returns stop_reason: "pause_turn". What is the correct action?
- A) Treat it as a failure and abort the run
- B) Build a
tool_result and send it back - C) Resend the returned conversation unmodified to continue
- D) Reduce
max_tokens and retry the original request
3. Which of these tools is client-executed even though it uses an Anthropic-defined schema?
- A)
web_search - B)
code_execution - C)
bash - D)
web_fetch
4. Which situation is the best fit for spawning subagents?
- A) A simple single-file edit
- B) A sequence of steps that each depend on the previous result
- C) Independent, parallelizable workstreams that benefit from isolated context
- D) Work that must continuously share mutable state
5. For coding and agentic work, which effort level does the guidance recommend?
- A)
low - B)
medium - C)
high - D)
xhigh
6. While extended thinking is active, which tool_choice values are compatible?
- A)
any and tool - B)
auto and none - C)
any and none - D) All four values
7. To auto-approve subagent invocation without a permission prompt in Claude Code, what must you do?
- A) Set
run_in_background to false - B) Include
Agent in allowedTools - C) Add a
parent_tool_use_id to the request - D) Set
tool_choice: any
8. In a programmatic AgentDefinition (the agents param), which fields are required?
- A)
name and description - B)
description and prompt - C)
prompt and tools - D)
model and prompt
9. On Opus 4.8 / Opus 4.7, setting a manual budget_tokens for thinking returns a 400 error; you should migrate to effort.
10. Interleaved thinking is automatic on Opus 4.6+ / Sonnet 4.6 with adaptive thinking, while Opus 4.5 / Sonnet 4.5 and earlier Claude 4 need a beta header.
11. When you return a tool_result while thinking is active, what must you include alongside it, and why can't you modify it?
12. Name the two decomposition strategies contrasted for breaking work into steps, and state which one lets the agent decide the steps at runtime.
13. Given a multi-agent research system with one orchestrator coordinating several specialist subagents in isolated context, what is this pattern called — and what is the only parent→child input channel?
14. Given a long-running code-generation task that will exceed a single context window, which strategy best preserves progress across windows?