A coordinator spawns a synthesis subagent, which replies that it has no search results to work with, although the search subagent finished. What is wrong?
Why B is the answer
Subagents have isolated context, so the findings must be included in the prompt
Subagents do not inherit the coordinator's history or share memory between runs. The coordinator has to pass the findings explicitly.
In detail
- Each subagent runs in its own context window, with its own system prompt and tool access. That isolation is the reason subagents are useful, because it keeps bulky work out of the main conversation.
- A fresh subagent starts without the conversation history or the files already read. The search subagent finishing does not put its results anywhere the synthesis subagent can see.
- The coordinator is the only part that has both. It receives the search summary, so it has to include those findings in the task it gives the synthesis subagent.
- Decide what each subagent needs before you start it, and put exactly that in its task.
Use case
An illustrative scenario, not a real incident.
A research coordinator runs a search subagent, then asks a writer subagent to "draft the report". The writer has no sources, so it says it has nothing to work with. Pasting the findings into the writer's task fixes it.
See it in different ways
flowchart LR A["Coordinator"] -->|"Task plus findings"| B["Research subagent"] B -->|"Summary"| A A -->|"Task plus summary"| C["Synthesis subagent"] C -->|"Report"| A
sequenceDiagram participant C as Coordinator participant S as Search subagent participant R as Synthesis subagent C->>S: Task S-->>C: Findings C->>R: Task with the findings included R-->>C: Final report
sequenceDiagram participant P0 as Coordinator participant P1 as Search subagent participant P2 as Synthesis subagent P0->>P1: Search the topic P1-->>P0: Findings P0->>P2: Write this up, with the findings included P2-->>P0: Finished report
Key point: A subagent knows only what its task tells it.
Why the other options are wrong
- A. The synthesis subagent needs the web search tool
- The results already exist. Giving the subagent search would repeat the work instead of passing it along.
- C. The coordinator needs a larger context window
- The coordinator already has the findings. A larger window changes nothing about what the subagent receives.
- D. The search subagent should have written to shared memory automatically
- Nothing is shared automatically. If you want shared storage you have to build it and pass the location in the task.