Claude Certified Architect – Foundations Practice Question 12: Subagent Isolated Context
Claude Certified Architect, Foundations · Test 2 · question 2 of 10
An orchestrator built with the Claude Agent SDK investigates a failing deployment. After the main agent reads the error logs and decides the cause is a missing environment variable in config/prod.yaml, it delegates "fix the problem" to a code-fixing subagent through the Agent tool. The subagent reports back that it cannot tell what problem to fix and starts re-exploring the whole repository. What is the most likely cause, and the right fix?
Show answer and explanation
Answer:
C is correct. A subagent runs in its own isolated context window with no parent conversation. The only content passed from parent to subagent is the Agent tool's prompt string, so the parent must put file paths, error messages and decisions into that prompt. Isolation is a feature: the subagent's intermediate tool calls do not clutter the parent's context, and only its final message returns.
A is wrong because subagents do not share the parent's history (only a fork of the current conversation does, which is a different mechanism), and nothing is corrupted. B is wrong because model size is irrelevant; even the same model cannot see turns that were never given to it. D is wrong because there is no description-keyed result cache that returns empty responses.