Domain 3 – Claude Code (3.1%)

• Claude Code Operation (3.1%) – CLAUDE.md hierarchy, settings.json, repository init

CCDV-F Domain 3 · Task 3.1

What You Need to Know

Claude Code reads configuration from files on disk and flags on the command line. Components: rules, skills, commands, agents, agent memory. Scopes: managed, user, project, local.

Five Components, Five Places on Disk

Component	Location	Notes
Rules	.claude/rules/, ~/.claude/rules/	Markdown constraints; optional paths frontmatter for file-scoped rules
Skills	.claude/skills/<name>/SKILL.md, ~/.claude/skills/	Procedures; Claude invokes by description match or /skill-name. Precedence: enterprise > personal > project
Commands	Built-in slash commands + bundled skills + custom (merged into skills)	.claude/commands/deploy.md and .claude/skills/deploy/SKILL.md both create /deploy; skill wins name clash
Agents	.claude/agents/, ~/.claude/agents/	Subagents; own context window; frontmatter requires name and description
Agent Memory	~/.claude/projects/<project>/memory/	Auto memory; MEMORY.md index loads at start (capped 200 lines / 25KB); topic files on demand

CLAUDE.md Concatenates, settings.json Overrides

CLAUDE.md scopes (broadest first): Managed policy → User (~/.claude/CLAUDE.md) → Project (./CLAUDE.md or ./.claude/CLAUDE.md) → Local (./CLAUDE.local.md, gitignored).

Claude Code concatenates discovered files into context — they do not override each other. More specific files land after broader ones. Walks up the directory tree from the working directory. @path/to/import pulls in more (max depth 4 hops). /init analyses the codebase and writes a starting CLAUDE.md.

Settings precedence (highest to lowest): managed → command line arguments → local .claude/settings.local.json → project .claude/settings.json → user ~/.claude/settings.json.

Managed settings cannot be overridden. List settings (e.g. permissions) merge across files — a local file can add but never remove a shared entry.

Settings Rules vs CLAUDE.md

“Settings rules are enforced by the client regardless of what Claude decides to do. CLAUDE.md instructions shape Claude’s behavior but are not a hard enforcement layer.” Conventions → CLAUDE.md. Anything that must hold whatever the model decides → deny rule or hook. Deny rules block in every permission mode, including bypassPermissions.

Permission Modes

A permission mode decides which actions Claude may take without asking. A cloned repository cannot grant itself capability — allow rules take effect only after the workspace trust dialog is accepted. Deny and ask rules apply regardless.

Sessions

Sessions save continuously as local JSONL transcripts at ~/.claude/projects/<project>/<session-id>.jsonl. Resume via --continue / -c, --resume, or /resume. Sessions created by claude -p or the Agent SDK stay out of the picker. A resumed session brings back the conversation and the model; permission mode restoration is limited. Launch flags do not survive a resume — pass them again or move into a settings file.

Key Distinctions

• Skills name-clash precedence runs the opposite way to settings (personal beats project for skills; project beats personal for settings/agents)
• CLAUDE.md is what you write for Claude; auto memory is what Claude writes for itself; neither enforces
• A sentence in CLAUDE.md is a request; a deny rule is a control

Exam Traps

Focus on precedence chains (settings vs CLAUDE.md vs skills), tracked vs local files, the enforcement distinction between settings rules and CLAUDE.md instructions, and session resume behaviour.