Claude Code Configuration Exam: Domain 3 Deep Dive

Domain 3 of the CCAR-F exam, Claude Code Configuration and Workflows, carries a 20% weighting and is one of the most concrete sections to prepare for. Unlike the probabilistic reasoning tested in other domains, the claude code configuration exam questions reward candidates who understand deterministic hierarchy rules, file placement, and version-control implications. This guide covers every concept you need, in the order the exam tests it.

What is Domain 3 and why does it carry 20% of the exam?

Domain 3 accounts for 20 of every 100 scaled-score points on the CCAR-F. Per the official exam guide, it tests practical judgment about configuring Claude Code for teams and production workflows, not recall of syntax. That framing matters: every scenario in this domain asks you to choose between competing configuration approaches given a specific constraint (team size, security posture, repository structure).

The domain sits alongside four others in the 60-item, 120-minute exam. For context, here is the full domain weighting:

Domains 3 and 4 are tied at 20% each, making them the joint second-largest investment of your study time after Domain 1.

How does the CLAUDE.md configuration hierarchy work?

The three-level configuration hierarchy is the single most-tested concept in Domain 3. Understanding it precisely is non-negotiable.

Claude Code reads configuration files at three scopes, each overriding the one above it for the files it governs:

1. User-level: ~/.claude/CLAUDE.md on the developer's machine. Applies to every project that developer opens. Suitable for personal preferences (preferred language, editor conventions) that should not be shared with the team.
2. Project-level: CLAUDE.md at the repository root. Committed to version control. Applies to every file in the project for every team member who pulls the repo.
3. Directory-level: CLAUDE.md files nested inside subdirectories. Apply only to files within that subtree, overriding the project-level file for those paths.

Our Claude Code Configuration and Workflows concept library maps all 30 task statements in this domain to atomic study cards, including the three-level configuration hierarchy in detail.

The exam consistently presents scenarios where a team wants to enforce a rule for one subdirectory (say, a payments/ module with stricter linting) without affecting the rest of the codebase. The correct answer is always a directory-level CLAUDE.md inside payments/, not a global rule with exceptions.

What are path-scoped rules and how do you write them?

Path-scoped rules let you attach YAML frontmatter to a CLAUDE.md file so that its instructions activate only for files matching a glob pattern. This is the mechanism that makes directory-level configuration precise.

A minimal example:

---
globs:
  - "src/payments/**"
  - "src/billing/**"
---
Always require two decimal places for monetary values.
Never log raw card numbers, even in debug output.

The frontmatter globs key accepts standard glob syntax. Claude Code evaluates the active file path against each pattern before deciding whether to apply the rules in that CLAUDE.md. If no file is active (for example, during a planning step), the project-level file governs.

Key exam traps around path-scoped rules:

• A rule in a directory-level file without frontmatter applies to all files in that subtree, not just matched globs. Frontmatter is optional but adds precision.
• Conflicting rules between a directory-level file and the project-level file resolve in favour of the more specific (directory-level) file for paths it covers.
• User-level configuration is never committed to version control. Exam scenarios that ask you to enforce a rule for the whole team always require a project-level or directory-level file, not a user-level one.

What are the version control implications of each configuration scope?

The version control implications of configuration scope are tested directly in Domain 3 scenarios. The rule is straightforward:

The exam frequently presents a scenario where a developer has configured a personal preference (for example, always using British English) in their user-level file, and asks whether a new team member will inherit that preference. The answer is no: user-level files are machine-local and never travel with the repository.

Conversely, when a security team wants to enforce a rule that no secrets appear in generated code, the correct placement is the project-level CLAUDE.md so that every contributor and every CI runner that clones the repo picks it up automatically.

How do commands and skills fit into Domain 3?

Beyond CLAUDE.md, Domain 3 tests two additional configuration artefacts: commands and skills.

Commands live in .claude/commands/ and are reusable prompt templates that team members invoke with a slash prefix. A command file is a Markdown file whose body becomes the prompt sent to Claude when a developer types /command-name. Commands are committed to version control, making them a team-wide asset.

.claude/
  commands/
    review-pr.md
    generate-tests.md
  rules/
    payments.md

Skills (sometimes called skill.md files) extend Claude Code's capabilities for a specific domain. The exam distinguishes between skills and CLAUDE.md instructions: skills teach Claude how to do something (a procedure), while CLAUDE.md instructions tell Claude what constraints to observe (a policy). Mixing the two in the same file is an anti-pattern the exam penalises.

The .claude/rules/ directory is the canonical location for path-scoped rule files when you want to keep them separate from the main CLAUDE.md. Files here can carry YAML frontmatter identical to directory-level CLAUDE.md files.

What configuration strategy should teams use in production?

The exam rewards proportionate, deterministic solutions. For Domain 3 scenarios, that translates into a clear decision tree:

1. Is the rule personal to one developer? Use user-level ~/.claude/CLAUDE.md.
2. Is the rule team-wide and applies to the whole codebase? Use project-level CLAUDE.md at the repo root.
3. Is the rule team-wide but applies only to a subsystem? Use a directory-level CLAUDE.md or a file in .claude/rules/ with matching frontmatter globs.
4. Is the rule a reusable prompt template? Use .claude/commands/.
5. Is the rule a procedural capability? Use a skill file.

This decision tree appears implicitly in every Domain 3 scenario. The exam does not reward memorising file names; it rewards understanding which scope solves the stated problem without over-engineering.

For teams managing multiple repositories, the project-level file is the only portable artefact. User-level files must be documented separately (for example, in a developer onboarding guide) because they cannot be distributed via version control.

How does Domain 3 connect to the other four domains?

Configuration is not isolated. The exam draws cross-domain connections that candidates frequently miss.

Domain 1 (Agentic Architecture, 27%) tests how Claude Code sessions interact with multi-agent systems. A well-configured CLAUDE.md can constrain which tools a sub-agent is permitted to call, reducing the risk of agentic loop anti-patterns that cause runaway tool use. The hub-and-spoke architecture pattern, for instance, often relies on project-level configuration to define the coordinator's permitted actions.

Domain 2 (Tool Design and MCP Integration, 18%) intersects with Domain 3 when MCP servers are configured via environment variable expansion in CLAUDE.md or in MCP config files. The exam may present a scenario where an MCP server credential is hardcoded in a project-level file that gets committed to a public repository, asking you to identify the security failure and the correct fix (environment variable expansion, not inline secrets).

Domain 4 (Prompt Engineering, 20%) connects to Domain 3 through the prompt engineering and structured output concepts that live in CLAUDE.md system-prompt sections. A project-level file that includes a structured output schema enforces consistent JSON formatting across every session without requiring each developer to re-specify it.

Domain 5 (Context Management, 15%) is affected by configuration when CLAUDE.md files grow large enough to consume significant context budget. The exam may ask whether a 4,000-token CLAUDE.md is appropriate for a project with a 200k-token context window. The answer depends on the ratio: a 2% context cost for persistent, team-wide rules is generally proportionate; a 40% cost for rarely-used personal preferences is not.

What scenario types appear most often in Domain 3?

Each CCAR-F sitting draws four scenarios at random from a bank of six. Domain 3 scenarios cluster around three recurring situation types:

1. Scope mismatch: A rule is placed at the wrong level (user instead of project, or project instead of directory). The question asks you to identify the failure mode and the correct placement.
2. Version control conflict: A configuration change causes unexpected behaviour for a team member who pulled the latest commit. The question asks you to trace the root cause to a specific file and scope.
3. Modular organisation: A large CLAUDE.md has become unwieldy. The question asks you to restructure it using imports, directory-level files, and .claude/rules/ without changing the effective behaviour.

For the third type, the exam rewards splitting a monolithic file into purpose-specific files over adding more content to the root CLAUDE.md. The principle is the same one that governs good software architecture: single responsibility, minimal scope.

How should you allocate study time across Domain 3?

Given the 20% weight, Domain 3 represents roughly 12 items in a 60-item exam (the exact count varies because each sitting draws from a randomised item bank). Our concept library at /concepts maps 174 atomic concepts to all five domains; the Domain 3 subset covers the configuration hierarchy, version control implications, commands, skills, and path-scoped rules.

A practical study allocation for Domain 3:

We recommend working through the three-level configuration hierarchy concept card first, then the version control implications card, before attempting any practice scenarios. The hierarchy is the foundation; everything else builds on it.

The CCAR-F exam costs $125 per attempt and is scored on a 100-to-1000 scale with a passing score of 720. Treating Domain 3 as a high-confidence domain (because its rules are deterministic and learnable) is one of the most reliable ways to build the scaled-score buffer you need across the full exam.