Anthropic · CCAR-F
Spawning subagents and passing context
CCAR-F · Lesson 3 of 18 · 15:30
YouTube loads only after you press play. This chapter opens at its timestamp in the full course. Watch on YouTube ↗
In this lesson
Prepare for the Claude Certified Architect – Foundations exam, on the exam guide version 1.0, effective July 2026. In this lesson: spawning subagents with the Task tool, now called the Agent tool, allowedTools, AgentDefinition descriptions, prompts, and tool restrictions, forks, passing complete findings with their metadata, parallel subagent calls in one response, and coordinator prompts built on goals and quality criteria. Includes three original practice questions.
Preview the topics
- Kestrel’s coordinator never spawns any subagent, although four subagent definitions are configured. What is the most likely cause?
- Kestrel’s synthesis agent cites sources that don’t match the claims, and sometimes no source at all. The coordinator passes it a short summary of what the other agents found. What is the best fix?
- The coordinator needs three independent searches. Today, it spawns one search, waits, then spawns the next. What should change?