course contents
Architect Foundations · 20% of the exam
claude code configuration and workflows
share
Domain 3 is about the files around Claude Code: what Claude reads before you type anything, which instructions reach your teammates, and when a task should be planned first.
You are using Claude Code to accelerate software development. Your team runs it every day for code generation, refactoring, debugging and documentation, in a monorepo with packages/web, packages/api and packages/infra that everyone clones from version control. You maintain the team's CLAUDE.md files, rules, custom slash commands and skills, and you decide when a task needs plan mode and when direct execution is enough.
The configuration surface you are responsible for:
Each of the six sections below is a different way that happens, from an instruction that sits in a file a teammate never receives to a CI run that cannot see what it needs.
Start hereThe words this domain is built on
- CLAUDE.md
- A file of standing instructions that Claude Code loads by itself, at launch from the launch directory upward, and from a subdirectory when Claude reads a file there.
- Rule (
.claude/rules/) - An instruction file in
.claude/rules/that loads at launch or, with apathsfield, only when Claude reads a matching file. - glob
- A filename pattern such as
**/*.test.tsx, which a rule lists inpaths. - Skill (
SKILL.md) - A folder holding a
SKILL.mdfile, whose one-line description is always in context while its body loads on demand. - Slash command
- A markdown file in a
commands/directory whose body becomes the prompt when you type its name, such as/review. Claude Code has since merged commands into skills. - Scope
- Where a configuration file lives: project scope reaches everyone who clones, and user scope stays on your machine.
- Plan mode
- A permission mode in which Claude proposes a plan but edits nothing until you approve it.
- Frontmatter
- The YAML block between
---lines that configures a rule or skill. - Context cost
- Tokens loaded before your first message, which anything loaded at launch costs in every session.
- Subagent
- A separate Claude session with its own context window, which returns only its result.
- Explore subagent
- A built-in, read-only subagent that searches code and returns a summary.
- hook and MCP
- A hook is your code, which Claude Code runs at a fixed point such as before a tool call, and MCP, the Model Context Protocol, connects it to outside tools.
--resumeand--continue- Flags that reopen an earlier session with its full history, which 3.6 says a CI review must not use.
~/.claude/ for you alone. When it must be in context decides the file. | In every session | The root CLAUDE.md, or a rule with no paths field. |
| When Claude works inside one directory | A CLAUDE.md in that directory, or a rule whose glob names it. |
| For one file type, wherever it lives | A rule whose paths glob matches that type. |
| When someone runs it | A skill, or a slash command. |