Agentic Architecture and Orchestration
Free Claude Certified Architect - Foundations practice — 6 questions on Agentic Architecture and Orchestration, with explanations. No sign-up. Full 12-question mixed test →
A coordinator agent spawns three subagents to research three independent competitors in parallel. The architect wants to avoid context pollution between subagents while still letting the coordinator synthesize a final comparison. Which design is correct per Claude's agentic architecture?
Subagents operate in isolated context windows and return a final condensed report to the coordinator, which prevents context pollution while allowing parallel work and synthesis at the coordinator level.
An architect wants Claude Code to prevent any `rm -rf` Bash command from executing at all, not just flag it afterward. Which hook configuration achieves this?
PreToolUse hooks run before a tool call executes; exiting with status code 2 blocks the call and returns the hook's output to Claude as feedback, which is the only option that stops the command before it runs.
An architect is deep into a long Claude Code session and wants to try a risky alternate refactoring approach without losing the original conversation state if it fails. What should they do?
Resuming a specific session and forking it creates a new branch of the conversation, preserving the original session's state so the risky approach can be abandoned without side effects.
A regulated workflow requires a verification gate between code generation and deployment that must never be skipped, even if the model judges it unnecessary. Which architecture best enforces this?
Enforced multi-step workflows rely on structural mechanisms outside the model's discretion, such as tool-level gating or hooks that block subsequent steps until verification completes, guaranteeing the step cannot be skipped.
Task B requires the completed output of Task A, and Task C requires the completed output of Task B. Which multi-agent orchestration pattern correctly handles this dependency chain?
When tasks have strict data dependencies, the coordinator must run them in sequence, feeding each subagent's output forward as the next subagent's input, rather than parallelizing independent work.
In the gather-context, act, verify agent loop, what is the correct architectural response when the verify step detects a failure, such as a failing test?
The agent loop is designed to be iterative: a verification failure becomes new context that drives another cycle of gathering context and acting, continuing until the work passes verification or a bounded retry limit is hit.
Ready for the real thing?
The full course has two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed answer explanations.
Start my full course on Udemy →