Practice — Claude Code Workflows and Agent Teams (12 questions)

For each scenario below, name the orchestration primitive you'd reach for — a plain subagent, a dynamic workflow, an agent team, or "none of these, just work in the conversation directly" — and justify the choice in terms of who needs to hold the plan, whether workers need to talk to each other, and how large the task is.

1. "Read through this 4,000-line log file and tell me which requests returned a 500."
2. "Every one of our ~300 React components under src/components/ still imports from the old legacy-ui package. Migrate each one to the new ui-kit package, working file by file, and verify each migration compiles."
3. "We keep getting intermittent test failures in CI and nobody can reproduce them locally. I want several different theories chased down at once — flaky timing, shared test state, a real race condition in the code — with whoever's investigating actively trying to shoot down the others' theories."
4. "Rename the fetchUser function to getUserById and update its one call site in src/api/users.ts."

A saved workflow migrates database access code file by file. Its script fans out with pipeline() over eight files, starting one agent per file in this order as soon as each previous agent's request has been dispatched: orders.ts, users.ts, products.ts, reviews.ts, payments.ts, shipping.ts, inventory.ts, refunds.ts.

By the time you stop the run, orders.ts, users.ts, and products.ts have all finished successfully. reviews.ts is still running. payments.ts, shipping.ts, and inventory.ts have also already finished (they were faster, smaller files). refunds.ts hadn't started yet.

1. When you resume this run, which agents return cached results without rerunning, and which agents rerun? Name each file explicitly and explain the rule that determines the outcome for each one.
2. A teammate on your team argues this is a bug: "payments.ts, shipping.ts, and inventory.ts already finished — rerunning them wastes money for no reason." Explain why this is how resume is designed to work, not a bug, referencing what the alternative would have to guarantee to avoid it.
3. Propose a concrete change to how this specific workflow is written that would reduce the cost of a future stop-and-resume, without changing what the workflow ultimately accomplishes.

You want Claude Code to interview you about a large new feature — asking about implementation, UI/UX, edge cases, and tradeoffs — before anything gets built.

What should the resulting workflow look like, per this subject?

You're running a Claude Code session in Manual permission mode and launch a dynamic workflow that spawns 20 subagents to edit files across the repo.

What mode do those spawned subagents' file edits actually run under?

A workflow fans out four agents via pipeline(), starting them in order: A, B, C, D. You stop the run while B is still in progress — A, C, and D had all already finished by that point.

On resume, which agents actually rerun?

You ask Claude Code to spawn several teammates to debate competing hypotheses about a bug. No team forms — Claude just runs ordinary subagents instead.

What's the most likely reason?