roadmap
- 01
- 02
- 03
- 04
- 05
- 06
- 07
- 08
Use Claude Code, Anthropic's agentic CLI for software development.
188 of 188 questions in your plan: 0 known, 0 in progress, 0 due, 0 skipped, 188 new. 0 cut, outside your plan.
Interviews seldom set aside time for Claude Code alone. The tool surfaces when the talk turns to how a team writes software with an agent, and the interviewer's seat decides the angle. An application team wants to hear how you work with it day to day: what belongs in a project's CLAUDE.md and what stays out of it, when a long session should be cleared rather than compacted, why a pass in plan mode before a large edit saves a rewrite. A platform or developer-experience team asks how you would roll it out across an engineering org: which settings layer wins when a developer's file and a managed policy disagree, how a hook stops a shell command before it runs, how the same agent runs headless in a pipeline to review pull requests. Security reviewers ask the narrowest question of all: what a repository you have just cloned can make the agent do, through the instructions it carries, the MCP servers it declares or the hooks it configures, and what sandboxing and deny rules leave out of reach. Forward deployed engineers meet all three angles, often in one customer meeting, because the job is fitting an agent into someone else's codebase and controls. Candidates for that job should also work through the [Forward Deployed Engineer roadmap](/roadmaps/forward-deployed-engineer), where Claude Code is one stop among the other subjects that loop covers.
Drill the plan question by question; it is not a manual to read in order. If you have used the tool without ever configuring it, start with the [Claude Code primer](/topics/ai-claude-code) on its hub page, and reopen it each time your answer names a feature but not the part of the agent's loop it changes. A junior should first be able to explain [memory and context](/topics/ai-claude-code-context): which CLAUDE.md files are loaded when a session starts, what fills the window as it runs and what compaction gives up. Most other answers assume it. Middle-level questions turn to changing the agent's behavior deterministically rather than by asking, and [hooks](/topics/ai-claude-code-hooks) are where that happens: which event fires before a tool call, what the handler reads on standard input, and how its exit status lets the call through or blocks it. Senior candidates reach [security and governance](/topics/ai-claude-code-security) early, because a team rollout is judged by what the agent cannot do: the sandbox boundary, injected instructions in a file it reads, a managed policy no developer can override. The tool changes from release to release, so drill with a scratch repository open and test any precedence rule or flag a question turns on before trusting your memory of it. For a role that only uses the tool, raising importance to must know usually trims enough; for one that will administer it for others, choose by topics under study order, so hooks and governance arrive whole instead of scattered by frequency.
Interviews seldom set aside time for Claude Code alone. The tool surfaces when the talk turns to how a team writes software with an agent, and the interviewer's seat decides the angle. An application team wants to hear how you work with it day to day: what belongs in a project's CLAUDE.md and what stays out of it, when a long session should be cleared rather than compacted, why a pass in plan mode before a large edit saves a rewrite. A platform or developer-experience team asks how you would roll it out across an engineering org: which settings layer wins when a developer's file and a managed policy disagree, how a hook stops a shell command before it runs, how the same agent runs headless in a pipeline to review pull requests.
Security reviewers ask the narrowest question of all: what a repository you have just cloned can make the agent do, through the instructions it carries, the MCP servers it declares or the hooks it configures, and what sandboxing and deny rules leave out of reach.
Forward deployed engineers meet all three angles, often in one customer meeting, because the job is fitting an agent into someone else's codebase and controls. Candidates for that job should also work through the Forward Deployed Engineer roadmap, where Claude Code is one stop among the other subjects that loop covers.
Drill the plan question by question; it is not a manual to read in order. If you have used the tool without ever configuring it, start with the Claude Code primer on its hub page, and reopen it each time your answer names a feature but not the part of the agent's loop it changes.
A junior should first be able to explain memory and context: which CLAUDE.md files are loaded when a session starts, what fills the window as it runs and what compaction gives up. Most other answers assume it. Middle-level questions turn to changing the agent's behavior deterministically rather than by asking, and hooks are where that happens: which event fires before a tool call, what the handler reads on standard input, and how its exit status lets the call through or blocks it. Senior candidates reach security and governance early, because a team rollout is judged by what the agent cannot do: the sandbox boundary, injected instructions in a file it reads, a managed policy no developer can override.
The tool changes from release to release, so drill with a scratch repository open and test any precedence rule or flag a question turns on before trusting your memory of it. For a role that only uses the tool, raising importance to must know usually trims enough; for one that will administer it for others, choose by topics under study order, so hooks and governance arrive whole instead of scattered by frequency.