Weight: 20%. About 12 of the 60 questions.
| Topic | What it tests |
|---|---|
| 3.1 CLAUDE.md hierarchy | User / project / directory / managed levels, @ imports, .claude/rules/, /memory, /init |
| 3.2 Commands & skills | .claude/commands/, .claude/skills/, SKILL.md frontmatter, custom subagents |
| 3.3 Path-specific rules | paths: globs, conditional loading |
| 3.4 Plan mode vs direct | When to plan, the Explore subagent, re-planning |
| 3.5 Iterative refinement | Examples, test-driven iteration, the interview pattern, batching feedback |
| 3.6 CI/CD | -p, --output-format json, --json-schema, --allowedTools, --max-turns, GitHub Actions |
Mental model: shared things live in the repository (CLAUDE.md, .claude/, .mcp.json), and personal things live in ~/.claude/. Always-on guidance goes in CLAUDE.md. On-demand workflows go in skills. Conventions tied to file types go in path-scoped rules.