CF
Certifyforce
Agentic Architecture & Orchestration27% of the exam

Spawning subagents and passing context

What crosses the boundary when a coordinator spawns a subagent with the Task tool (now named Agent): only the delegation prompt, so findings, constraints and quality criteria must be written into it. Covers AgentDefinition fields, forks as the exception, structured hand-offs that keep source metadata, and parallel spawning in one response.

Official exam objective
Configure subagent invocation, context passing, and spawning

~8 min read · 13 practice questions ·

What the exam expects

  • Explain the Task tool (now named Agent) as the spawn mechanism, and state both the exam guide's allowedTools claim and today's behaviour.
  • Predict what a subagent can and cannot see, and put findings, constraints and rules where it will actually read them.
  • Configure an AgentDefinition (description, prompt, tool restrictions, skills, model) and know which definition wins on a name clash.
  • Pass context as structured records that keep content separate from metadata, and require structured final reports.
  • Spawn independent subagents in parallel from one coordinator response, and write goal-and-criteria delegations instead of procedures.

Key terms: Agent tool (the tool a coordinator calls to spawn a subagent; formerly Task); delegation prompt (the Agent tool's prompt string, the only parent-to-subagent channel); AgentDefinition (programmatic subagent configuration); fork (a subagent that copies the conversation instead of starting fresh); content/metadata separation (findings kept apart from source URL, document and page).

Key facts

Spawn tool name
Detail
Agent since v2.1.63; Task(...) still works as an alias
Source
Create custom subagents (opens in a new tab)
Parent-to-subagent channel
Detail
Only the Agent tool's prompt string
Source
Subagents in the SDK (opens in a new tab)
Not inherited
Detail
Parent history, tool results, system prompt; skill content not listed in skills
Source
Subagents in the SDK (opens in a new tab)
Subagent-to-parent channel
Detail
Final message only; intermediate tool calls stay inside
Source
Subagents in the SDK (opens in a new tab)
Required AgentDefinition fields
Detail
description and prompt
Source
Subagents in the SDK (opens in a new tab)
Name clash
Detail
Programmatic agents override filesystem agents
Source
Subagents in the SDK (opens in a new tab)
Fork
Detail
Copies the conversation; start with /subtask or subagent_type: "fork"
Source
Create custom subagents (opens in a new tab)Subagents in the SDK (opens in a new tab)
allowed_tools
Detail
Auto-approves only; Agent does not ask before running
Source
Configure permissions (opens in a new tab)
Sources for this section (3)

Check yourself

  • Why must a coordinator restate a data-handling rule in the subagent's own prompt?
  • When is a fork the right way to start a subagent?
  • Which AgentDefinition fields are required?
  • Why can't the coordinator read the URLs a subagent fetched unless the report lists them?
  • How should a coordinator run four independent investigations fastest?

The questions are free to read. Answers with their sources and self-grading are included with the pack.

Keep reading with the Claude Certified Architect - Foundations pack

  • Core concepts
  • Exam traps
  • Worked scenario

The pack includes:

  • All 30 lessons of the Claude Certified Architect - Foundations guide, with reading progress
  • Interactive decision walks and step-by-step sequence steppers
  • Inline practice questions and one-click topic practice (13 questions for this lesson)
  • An official source cited for every claim
  • An AI tutor that answers from the guide
  • Unlimited timed practice exams
Get the packAlready own it? Sign in

Practice for the Claude Certified Architect - Foundations exam

Certifyforce is an independent study resource and is not affiliated with, endorsed by or sponsored by Anthropic. Claude is a trademark of Anthropic.