Claude Code — Glossary
Claude Code Configuration & Workflows · 20% of the exam. 10 terms.
CLAUDE.md
A memory file automatically pulled into context documenting project conventions, commands, and constraints.
Exam context: Multiple CLAUDE.md files can exist at different scopes and are all included — none of them get silently dropped.
CLAUDE.md hierarchy
Enterprise, user, project, and local CLAUDE.md files are merged together, with more specific scopes layered on top of general ones.
Exam context: A common wrong answer assumes the most specific file replaces the others — it's additive, not a full override.
settings.json
Project- or user-level configuration for permissions, hooks, environment variables, and other Claude Code behavior.
Exam context: Know which settings are safe to check into a repo (project-shared) versus which belong in a local, git-ignored file.
Permissions
Allow / deny / ask rules controlling which tools and commands Claude Code can run without prompting the user.
Exam context: Scenario questions test picking the narrowest permission that still lets the workflow function.
Slash command
A reusable, user-invoked prompt template stored as a Markdown file and triggered explicitly with /name.
Exam context: The user has to type it — contrast with a skill, which Claude can invoke on its own judgment.
Skill
A packaged, discoverable capability — instructions plus optional scripts — that Claude can invoke autonomously when it judges it relevant.
Exam context: If a scenario needs Claude to decide when to use it, that rules out a slash command as the right mechanism.
Hook
A shell command that fires automatically on a lifecycle event (e.g. PreToolUse, PostToolUse, Stop) to enforce policy deterministically.
Exam context: Hooks exist specifically because prompted instructions ('please always run tests first') aren't reliably deterministic.
Plan mode
A read-only mode where Claude researches and proposes an approach before any file is edited, requiring explicit approval to proceed.
Exam context: Correct answer for scenarios needing human sign-off before risky or hard-to-reverse changes.
Subagents (Claude Code)
Separately configured agents with their own system prompt and tool access, delegated a scoped piece of work by the main agent.
Exam context: Distinguish product-level Claude Code subagents from the general agentic-architecture subagent pattern in Domain 1 — related, not identical.
CI/CD integration
Running Claude Code non-interactively (headless mode) inside a pipeline, e.g. for automated review or fix-and-PR workflows.
Exam context: Headless runs still need explicit permission configuration — they don't get an interactive prompt to fall back on.