CCA Prep
CCAR-FUA

Lesson 1.3Lesson 3 of 30

Spawning subagents and passing context

Configure subagent invocation, context passing, and spawning

Spawning subagents and passing contextCoordinatorCoordinator history(not inherited)Task promptGoal + qualitycriteriaPrior findingsSource metadata: url,doc, pageSubagent (freshcontext)Several Task calls in one response = parallel

Key points

  1. 1Subagents are spawned with the Task tool, so the coordinator's allowedTools must include "Task". (Newer Claude Code docs call it the Agent tool; the exam guide says Task.)
  2. 2A subagent starts with an empty context: no coordinator history, no shared memory. Everything it needs goes into its prompt.
  3. 3Pass prior findings in a structured form that keeps content separate from metadata (URL, document, page), so citations survive.
  4. 4Parallel work = several Task calls in one coordinator response. Prompts state the goal and quality bar, not step-by-step procedures.
What the exam guide checks here

Objective wording from the exam guide (English).

  • 1.3.K1Task tool spawns subagents; coordinator's allowedTools must include "Task"
  • 1.3.K2Subagent context must be passed explicitly in the prompt; no automatic inheritance or shared memory
  • 1.3.K3AgentDefinition: description, system prompt, tool restrictions per subagent type
  • 1.3.K4Fork-based session management to explore divergent approaches from a shared baseline
  • 1.3.S1Include complete prior findings directly in the subagent's prompt
  • 1.3.S2Use structured formats separating content from metadata (URLs, document names, pages) to keep attribution
  • 1.3.S3Spawn parallel subagents by emitting multiple Task calls in one coordinator response
  • 1.3.S4Coordinator prompts state goals and quality criteria, not step-by-step procedures

Lock it in

Answer scenario questions on this topic. Every option comes with an explanation.

Practise 1.3 · 23 questions