Each scenario sets up a realistic production context, then asks a cluster of questions about it. The exam draws four of the six at random, so prepare for all of them. Our workshop's six labs mirror these scenarios.
Scenario 1 — Customer Support Resolution Agent
A support agent with tools to look up customers and orders, process refunds and escalate to a human. Tests: escalation logic, enforcing tool order (verify the customer before refunding — in code, not in the prompt), and handling messages that contain several issues.
Scenario 2 — Code Generation with Claude Code
A team using Claude Code in its everyday workflow. Tests: the CLAUDE.md hierarchy, path-specific rules, plan mode versus direct execution, and delegating exploration to a subagent. See Domain 2.
Scenario 3 — Multi-Agent Research System
A coordinator dispatching to research, analysis, synthesis and report subagents. Tests: passing context explicitly, spawning in parallel, and propagating errors with coverage gaps instead of hiding them. See Domain 1.
Scenario 4 — Developer Productivity with Claude
An agent built with the Agent SDK to explore codebases and automate tasks. Tests: choosing the right built-in tool (Grep, Glob, Read, Edit, Write), MCP resource design, and persisting findings in a scratchpad.
Scenario 5 — Claude Code for Continuous Integration
Claude Code inside a pipeline for review and test generation. Tests: the -p flag, JSON output flags, and using independent sessions for unbiased review.
Scenario 6 — Structured Data Extraction
Pulling structured data out of documents at scale. Tests: tool_choice, nullable schema fields, validation-retry loops, batch strategy and routing low-confidence fields to human review. See Domain 3.
How to use this page
For each scenario, ask yourself: what would I enforce in code, what context does each agent need passed explicitly, which tools would I expose, and where could context get lost? If you can answer those four for all six, you are in good shape.