Claude Code in Practice

What this pack asks

The free questions in this pack. Answer choices and explanations appear as you play.

• A 25-person product team has adopted Claude Code for feature work. Engineers currently paste large, multi-file feature requests in one go; roughly a third of the time the implementation is technically sound but solves a different problem than intended, and reviewers only discover this at pull-request time. The team lead wants course correction to happen before any files are edited, insists on no new plugins, servers, or custom modes, and refuses to make engineers manually decompose every feature into dozens of small prompts. Which approach BEST meets these requirements?

  Answer this one →

• A security-conscious fintech maintains a 1.2M-line monorepo where hundreds of files change daily. The platform team wants an in-terminal coding assistant that can answer "where is X handled?" questions and then make edits. Compliance forbids shipping source code to any external store or third-party index, the two-person platform team has no capacity to run and re-sync a retrieval pipeline, and answers must reflect the repository as it exists at the moment of the query. Which approach to codebase understanding BEST satisfies all three constraints?

  Answer this one →

• A developer-tools startup built an internal coding agent by calling a hosted Claude model directly through the API with one large instruction prompt and no other scaffolding. The agent stalls on multi-step tasks, loses track of earlier steps in long sessions, and cannot reliably act on the repository. Leadership has ruled out fine-tuning or hosting custom weights, wants measurable improvement on the current model generation within one quarter, and has only three engineers who cannot maintain a training or retraining pipeline. Which approach BEST meets these requirements?

  Answer this one →