Key points
- The Task tool spawns subagents; the coordinator's
allowedToolsmust includeTask. - Subagents do not inherit parent context or share memory between invocations.
- An AgentDefinition sets a subagent type's description, system prompt, and tool restrictions.
- Fork-based session management branches a shared analysis into independent explorations.
- Multiple Task calls in one coordinator response run in parallel; calls across turns run one after another.
- Prompts that state goals and quality criteria let subagents adapt; rigid step lists break on unexpected input.
Skills the exam expects
- Include complete findings from earlier agents directly in the next agent's prompt.
- Pass context as structured data that separates content from metadata (source URL, document name, page number) to preserve attribution.
- Launch parallel subagents by emitting several Task calls in a single response.
- Write coordinator prompts around research goals and quality criteria rather than procedures.
Common misconceptions
- "The synthesis agent can see what the search agent found." It sees only what the coordinator passes in.
- "A summary of findings is enough." Compressed handoffs lose detail and attribution.
- "Task calls on successive turns still run concurrently." They run sequentially.
Watch for
Coordinators missing Task in allowedTools, free-text handoffs that drop sources, and step-by-step subagent prompts.
Practice task 1.3
57 practice items cover this task in CCA Prep Studio: flashcards, matching, scenario questions and explain-it-back exercises, each tracked toward your mastery score.