S3 · scenario119 questions · 33 free
Multi-agent research (S3)
Coordinator + subagents researching topics into cited reports.
The Multi-agent Research scenario (S3) is a coordinator delegating to subagents that gather, evaluate, and synthesize sources into a cited report. It's the canonical test of orchestration plus reliability — fan-out work that has to come back trustworthy.
Expect questions on coordinator–subagent design, context passing and spawning, propagating errors across agents, and preserving provenance and uncertainty when many sources are merged into a single answer.
This scenario spans 18 subtopic areas, covered by 119 practice questions across 30 easy, 60 medium, and 29 hard items.
Sample question · free
The coordinator agent calls the report generation subagent via the Task tool. The API response comes back with stop_reason 'end_turn'. What should the coordinator do?
What's covered
Subtopic areas in Multi-agent research, drawn from the exam blueprint:
1.1Design and implement agentic loops for autonomous task execution91.2Orchestrate multi-agent systems with coordinator-subagent patterns101.3Configure subagent invocation, context passing, and spawning91.4Implement multi-step workflows with enforcement and handoff patterns71.5Apply Agent SDK hooks for tool call interception and data normalization101.6Design task decomposition strategies for complex workflows51.7Manage session state, resumption, and forking102.1Design effective tool interfaces with clear descriptions and boundaries92.2Implement structured error responses for MCP tools82.3Distribute tools appropriately across agents and configure tool choice92.4Integrate MCP servers into Claude Code and agent workflows72.5Select and apply built-in tools (Read, Write, Edit, Bash, Grep, Glob) effectively65.1Manage conversation context to preserve critical information across long interactions45.2Design effective escalation and ambiguity resolution patterns45.3Implement error propagation strategies across multi-agent systems35.4Manage context effectively in large codebase exploration35.5Design human review workflows and confidence calibration25.6Preserve information provenance and handle uncertainty in multi-source synthesis4