A subagent does not inherit the coordinator's conversation. Whatever it needs must be passed explicitly in its prompt. Build a coordinator that dispatches two subagents, then break the handoff on purpose so you can recognise the symptom on the exam and in production.
What to build
- Build a coordinator that delegates a research task to two subagents with different remits.
- Give each subagent a self-contained brief: objective, constraints, and the exact output shape.
- Scope tools per subagent — the reader gets read tools only, the writer gets write tools.
- Have each subagent return a structured summary, not a transcript.
- Now remove half the context from one brief and observe the degradation.
- Add a synthesis step in the coordinator that reconciles the two summaries.
Done when
- Each subagent brief is complete enough to execute in isolation, with no pronouns pointing at coordinator context.
- Tool grants differ per subagent and you can justify each grant.
- Subagents return structured results the coordinator can merge without re-reading their work.
- You have observed and can describe what an under-briefed subagent produces.
If you want to go further
- Add a third subagent and measure how coordinator synthesis cost scales.
The trap this exercise teaches
Writing a subagent prompt that references "the file we discussed" or "the plan above". The subagent has no such context, and the failure is quiet — it produces confident, unrelated work.
Before you start
The theory behind this build is covered in Subagent prompting and tool scoping. If any step below is unfamiliar, read that first — the exercise assumes it. Primary source: Claude Docs — Subagents.
Check yourself against the exam
This exercise sits in Domain 1, which is 27% of the CCAR‑F exam. Once you have built it, run a domain drill in the study app and see whether the questions read differently.