Q3 of 60 — CCA Practice Exam

Multi-Agent Research System

Your multi-agent research system uses Claude Code agent teams, each spawning five subagents to parallelize literature synthesis across topic areas. After the first week, the team lead reports that token costs are 4× the projected estimate, and /usage confirms the overage is driven by long-running subagent sessions. Examining the spawn prompts, you find each one includes a 2,000-token context block describing the full research methodology, shared terminology glossary, and team communication norms — all of which are already documented in the project's CLAUDE.md. What is the most effective change to reduce token costs without reducing research throughput?

A. Switch from five subagents to two subagents per team to reduce the number of active context windows, then rely on those agents to cover more topic areas per session.

B. Strip the methodology, glossary, and communication norms from the spawn prompts and let CLAUDE.md carry that shared context, since teammates load it automatically at spawn time.

C. Upgrade subagents from Sonnet to the most capable model so each agent completes its work faster, reducing the total wall-clock time that context windows remain open.

D. Add an explicit instruction in each spawn prompt telling subagents to summarize and discard context aggressively after completing each subtask to reduce token accumulation.