CCAR-F scenarios › Multi-Agent Research System · Question 21 of 60
Scenario 3 · Multi-Agent Research System
You're building a research system on the Claude Agent SDK in which a coordinator agent delegates work to specialized subagents: one searches the web, one analyzes documents, one synthesizes findings, and one generates reports. Given a topic, the system produces a thorough report with citations.
Agentic Architecture & Orchestration
The synthesis subagent's reports often include generic statements that appear in none of the gathered sources and leave out findings the web search and document analysis subagents clearly returned. The coordinator's Task call to the synthesis subagent reads only: "Synthesize the findings on the topic we've been researching." What is the most likely cause and fix?
Show answer and explanation
Answer: A — Subagents don't inherit the coordinator's context, so synthesis never saw the findings; put the complete search and analysis outputs in its Task prompt.
Subagents run with isolated context and don't inherit the coordinator's conversation history, so context has to be passed explicitly, including the complete findings from prior agents (task 1.3). Without the findings, the subagent can only produce generic content. Forbidding unsourced claims (b) doesn't give it the sources it needs. A larger model (c) still has nothing to synthesize. Giving synthesis web search tools (d) duplicates another subagent's role and invites cross-specialization misuse instead of fixing the context handoff.
More from this scenario
- For a broad topic, the coordinator delegates to four web search subagents covering…
- Users submit both broad questions ("How is climate policy affecting shipping costs?")…
- On the topic "remote work and urban real estate," the coordinator sends the same broad…
- To keep the configuration simple, every subagent's AgentDefinition grants the same 16…
- The document analysis subagent has one tool, process_document(url, mode), described as…