Claude Code AI: CCAR-F Domain 3 Architect Exam Guide

Domain 3 of the Claude Certified Architect, Foundations exam (CCAR-F) covers Claude Code Configuration and Workflows and accounts for 20% of your total score. If you are preparing for the $125 exam, understanding how claude code ai tooling integrates with real architectural decisions is not optional: it is one of the four heaviest domains on the paper. This guide walks through what Domain 3 actually tests, how it connects to the other four domains, and what a passing candidate needs to be able to do on exam day.

What is Claude Code AI and why does it appear on the CCAR-F exam?

Claude Code is Anthropic's AI-powered coding environment that runs as a command-line tool and integrates with editors and CI/CD pipelines. On the CCAR-F exam it appears not as a product feature tour but as an architectural decision surface: when should you configure Claude Code at the project level versus the user level, how do you version-control those settings safely, and how do you wire Claude Code into automated workflows without creating brittle pipelines?

Domain 3 carries 20% of the exam weight, equal to Domain 4 (Prompt Engineering and Structured Output). That parity signals that Anthropic treats code-environment configuration as a first-class architectural concern, not a developer-only afterthought.

How is the CCAR-F exam structured overall?

Before drilling into Domain 3, it helps to see where it sits relative to the full exam. The table below shows all five domains and their weights, per the official exam guide.

The exam runs for 120 minutes across 60 items, scored on a 100-to-1000 scale with 720 as the passing mark. Each sitting draws four scenarios at random from a bank of six, so you cannot predict which scenario set you will face. The credential is valid for 12 months from the date it is awarded.

Domain 1 is the single largest domain at 27%, so architects who under-prepare agentic architecture concepts pay the steepest price. Domain 3 and Domain 4 together account for 40% of the exam, making them collectively the most important pair to master.

What does Domain 3 actually test?

Domain 3 tests your ability to make sound configuration and workflow decisions in realistic production contexts. The exam does not ask you to recall flag names from memory. Instead, it presents a scenario (a team onboarding a new codebase, a CI pipeline failing intermittently, a multi-developer project with conflicting settings) and asks you to select the architecturally correct response.

The core concepts cluster around three areas.

Configuration hierarchy: which level owns which setting?

Claude Code uses a three-level configuration hierarchy: system-level settings that apply globally, project-level settings that live in the repository, and user-level settings that are personal to each developer. Exam scenarios frequently test whether you can identify which level is appropriate for a given constraint.

A common trap: candidates assume that project-level configuration is always the right answer because it is closest to the code. The exam rewards more nuanced thinking. If a setting encodes a personal workflow preference (a developer's preferred verbosity level, for example), it belongs at the user level. If it encodes a team-wide constraint (a linting rule, a restricted file path), it belongs at the project level.

Version control implications: what should and should not be committed?

Version control implications is one of the most frequently tested concepts in Domain 3. The exam presents scenarios where a team has committed sensitive configuration (API keys, personal tokens, environment-specific paths) and asks you to diagnose the problem and prescribe the fix.

The correct architectural posture is to commit project-level configuration that encodes shared constraints, and to exclude user-level and secret-bearing configuration via .gitignore or equivalent mechanisms. Scenarios that involve a CI pipeline picking up a developer's personal settings are a signal that version control boundaries have been violated.

Workflow integration: Claude Code in automated pipelines

The third cluster covers how Claude Code fits into CI/CD and agentic workflows. This overlaps with Domain 1 (orchestration) and Domain 2 (tool design), which is intentional: the exam rewards candidates who can reason across domain boundaries.

A typical scenario might describe a pipeline where Claude Code is invoked non-interactively to generate test stubs, and ask you to identify the correct invocation pattern. The exam consistently rewards deterministic, auditable solutions over probabilistic ones when stakes are high, so answers that introduce unnecessary model calls into a critical path are usually wrong.

How does Domain 3 connect to the other domains?

The five domains are not isolated silos. Domain 3 connects most tightly to Domain 1 and Domain 2.

With Domain 1, the connection is orchestration: a Claude Code workflow that spawns subagents or delegates to a coordinator must respect the same subagent context isolation principles that apply to any multi-agent system. Configuration decisions at the Claude Code level can inadvertently bleed context across agent boundaries if not handled carefully.

With Domain 2, the connection is tool design. Claude Code exposes built-in tools (file read, file write, bash execution, search) and can be extended via MCP servers. Exam scenarios that involve Claude Code almost always require you to reason about tool descriptions as a selection mechanism: if the tool description is ambiguous, Claude Code will misroute, regardless of how well the configuration hierarchy is set up.

With Domain 4, the connection is prompt structure. CLAUDE.md files (the project-level instruction layer in Claude Code) are effectively structured prompts. The same principles that govern prompt engineering and structured output apply here: specificity, constraint clarity, and example-driven guidance all improve reliability.

What does a passing candidate know about CLAUDE.md files?

CLAUDE.md is the primary mechanism for encoding project-level instructions in Claude Code. It sits at the root of a repository and is read automatically when Claude Code initialises in that directory. A passing candidate understands four things about CLAUDE.md:

1. It is a prompt, not a configuration file in the traditional sense. Its contents are injected into the model's context, so the same attention and token-budget considerations that apply to any long system prompt apply here.
2. It should encode constraints that are stable across the project lifecycle: coding conventions, restricted operations, required review steps. It should not encode ephemeral state.
3. It is version-controlled alongside the code, which means changes to it are auditable and reviewable via standard pull-request workflows.
4. It can import or reference other instruction files, enabling modular organisation for large codebases.

The exam tests all four of these properties through scenario items. A candidate who treats CLAUDE.md as a simple README will misread scenarios that hinge on token budget or context injection behaviour.

How should you approach Domain 3 scenario items on exam day?

Each CCAR-F item is scenario-based. For Domain 3 items, a reliable approach is to work through three questions in sequence.

First, identify the configuration level in play. Is the scenario describing a personal preference, a team constraint, or a system-wide policy? That determines whether the correct answer involves user-level, project-level, or system-level configuration.

Second, check for version control hygiene. If the scenario involves a CI pipeline, a shared repository, or a team onboarding flow, ask whether the configuration being described should be committed. Scenarios that describe intermittent failures in automated pipelines often trace to version control boundary violations.

Third, apply the proportionality principle. The exam rewards proportionate fixes. If a scenario describes a minor misconfiguration, the correct answer is a targeted correction, not a full architectural redesign. Answers that introduce new orchestration layers or additional model calls to solve a configuration problem are almost always wrong.

What is the exam format and how do you register?

The CCAR-F exam is delivered online-proctored or at a Pearson VUE test centre. It costs $125 per attempt. Tiered Claude Partner Network partners receive a discounted first attempt. The exam launched on 12 March 2026 and is part of the Claude Partner Network, a $100M programme. As of 3 June 2026, the network had over 40,000 partner applicant firms and more than 10,000 certified individuals.

The four live certification tracks and their prices are summarised below.

If you are deciding between the Architect and Developer tracks, note that CCAR-F and CCDV-F are priced identically at $125 but test different skill sets. The Developer track (CCDV-F) has 53 items across eight domains and does not use a scenario bank. The Architect track (CCAR-F) has 60 items across five domains and draws four scenarios at random from a bank of six.

How does AI Skill Certs prepare you for Domain 3?

AI Skill Certs is an independent adaptive prep platform for the CCAR-F exam. We are not affiliated with or endorsed by Anthropic. Our concept library at /concepts covers 174 atomic concepts mapped to all five CCAR-F domains and 30 task statements, including the full set of Claude Code configuration and workflow concepts.

The adaptive engine uses Bayesian Knowledge Tracing with a 0.90 mastery threshold, which means it keeps routing you back to Domain 3 concepts until your demonstrated mastery crosses that bar, not just until you have seen each concept once. Practice exams mirror the real format: 60 questions, scored 100 to 1000, with 720 as the passing bar.

Archie, our Socratic tutor, guides you through Domain 3 scenarios with graduated hints rather than direct answers. That approach is deliberate: the exam tests judgment, and judgment is built through reasoning practice, not answer memorisation.

For candidates also preparing for the Developer track, our CCDV-F adaptive study, Archie tutoring, and practice exams are live on the platform today.