AI 原生开放教材

外观

The Context Course系统课程·英文原文 ↗原件 ↓仓库 ↗

Quiz 2: Multi-Agent Workflows ​

Test your ability to design and implement multi-agent workflows.

Question 1: When is the overhead of subagents worth it? ​

选项

  • A. Always spawn as many subagents as possible for maximum parallelism
  • B. Use subagents when tasks are independent and parallelizable, and the overhead is justified by the task complexity
  • C. Subagents are only for research tasks; never use for implementation
  • D. Subagents should always share the same context window as the parent

答案解析

  • A — More subagents means more token usage, latency, and coordination overhead. Use subagents selectively.
  • B（正确答案） — Correct! Subagent workflows consume more tokens than single-agent runs. Only use them when the parallelism benefit outweighs the overhead.
  • C — False. Worker subagents can implement fixes, write code, and make changes. Codex even has a built-in 'worker' agent type for this.
  • D — Subagents run as isolated instances with their own context. They communicate results back to the parent, not share context directly.

Question 2: How do parent agents share context with subagents? ​

选项

  • A. Subagents can directly access the parent agent's memory and variables
  • B. Pass small context in the task description; for large data, write to a file and reference the path
  • C. Use global variables that all subagents can read
  • D. There's no way to share data between parent and subagent

答案解析

  • A — False. Subagents are isolated. They receive context from the parent via their task description and report results back.
  • B（正确答案） — Correct! Subagents can't access parent memory. Use task descriptions for small context and files for larger data.
  • C — Subagents are separate instances. Global variables don't cross agent boundaries.
  • D — False. The parent passes context in the task description, and subagents report results back. File system can also be used.

Question 3: How should you design for subagent failure? ​

选项

  • A. If a subagent fails, the entire workflow always fails
  • B. Design for failure: use timeouts, handle partial results, and have fallback strategies
  • C. Assume all subagents will succeed; don't add error handling since it's just overhead
  • D. If a subagent fails, immediately spawn 5 more to retry the same task

答案解析

  • A — Not necessarily. Good design handles failures gracefully — retry, skip, or use partial results.
  • B（正确答案） — Correct! Subagents can timeout, produce errors, or return unexpected results. Robust workflows handle these cases.
  • C — Risky. Network issues, token limits, and unexpected errors are common in multi-agent workflows.
  • D — Wasteful. Diagnose the failure first, then retry selectively. Brute-force retries waste tokens.

Question 4: How do Claude Code and Codex expose subagents? ​

选项

  • A. Claude Code: use /agent spawn. Codex: use codex-agent spawn. They're identical.
  • B. Claude Code: conversational invocation + custom agents in .claude/agents/. Codex: natural language requests + custom agents as TOML files in .codex/agents/.
  • C. Claude Code doesn't support subagents. Only Codex does.
  • D. Both platforms use the exact same subagent API and configuration format

答案解析

  • A — Neither command exists. Claude Code uses conversational invocation; Codex uses natural language requests to spawn agents.
  • B（正确答案） — Correct! Both platforms use natural language to trigger subagent creation, but configure custom agents differently — Claude Code with markdown files, Codex with TOML files.
  • C — False. Claude Code has robust subagent support via the Agent tool, conversational invocation, and custom agents.
  • D — No. Claude Code uses markdown-based agent definitions; Codex uses TOML. The invocation style is conversational for both but with different capabilities.

Question 5: What built-in agent types does Codex provide? ​

选项

  • A. Codex has three built-in agent types: default (general), worker (implementation), and explorer (read-heavy codebase exploration)
  • B. Codex has no built-in agent types; you must always define custom agents
  • C. Codex only supports one agent at a time; there's no parallelism
  • D. Codex subagents require a separate codex-agent binary to manage

答案解析

  • A（正确答案） — Correct! These three built-in types cover the most common subagent use cases without requiring custom agent definitions.
  • B — False. Codex ships with default, worker, and explorer agents that can be used immediately.
  • C — False. Codex supports up to max_threads (default 6) concurrent agent threads.
  • D — False. Subagents are managed through Codex itself — use /agent to switch between threads, or ask Codex in natural language.

Summary ​

If you got 4-5 correct, you're ready to design multi-agent workflows without reaching for subagents unnecessarily. If you missed several, review the platform-specific invocation patterns and failure-handling guidance.

Key Takeaways ​

  • Use subagents when the task shape justifies isolation or parallelism
  • Pass only the context a child agent needs, and use files when the payload is too large for a prompt
  • Design for retries, partial results, and platform-specific invocation details from the start

Next Steps ​

You've finished Unit 4: Subagents. Next up is Unit 5, where you'll add deterministic lifecycle hooks around those same agent workflows.