Claude Code — Quick Reference
Claude Code Configuration & Workflows · 20% of the exam. Density over narrative — this is for review, not first-pass learning.
CLAUDE.md hierarchy
Enterprise → user → project → local. All applicable levels are merged into context — a more specific file adds to the others, it does not replace them.
Slash command vs skill
Slash command: user types it explicitly. Skill: Claude decides on its own to invoke it based on relevance. If the scenario needs autonomous judgment, it's a skill.
Hooks vs prompted instructions
Use a hook when behavior must be deterministic and non-negotiable (e.g. always run linter before commit). A prompted instruction alone is not reliable enough for hard requirements.
Plan mode
Use when changes are risky, hard to reverse, or need human sign-off before execution. It is read-only until explicitly approved.
Permissions design
Grant the narrowest permission set that still lets the workflow complete — not blanket allow, not blanket ask.
Headless / CI use
Non-interactive runs need permissions configured up front; there's no interactive prompt to fall back on mid-run.