CCA Study Tips for Domain 3: Claude Code Configuration & Workflows

Study Domain 3 (Claude Code Configuration & Workflows, 20% of the CCA-F) by reading the 26 articles in task order over four sessions, creating the real files as you go, and drilling scenario questions that ask “which file, which mechanism, which mode”. Domain 3 rewards hands-on time more than any other domain, so this plan builds a small drill around each session.

For what the domain covers and why it is worth a fifth of the exam, see CCA Domain 3 Explained. This post is only about how to study it.

What order should I read the Domain 3 articles in?

Use the task order on the Domain 3 study guide page, with a short drill after each session:

Session	Articles	Drill
1	Task 3.1: Put Team Standards Where the Team Can Actually Get Them, Mentioning a File Is Not the Same as Loading It, Twelve Unscoped Rules Files Waste 78% of Your Context Budget, /memory Tells You What Is Actually Loaded	Create a root .claude/CLAUDE.md with one @import, then run /memory and confirm what loaded
2	Tasks 3.2 and 3.3: Commands Are Prompts. Skills Are Configured Workflows., Three Fields That Turn a Prompt Into a Safe Workflow, fork isolation, personal skills, CLAUDE.md Is Always-On. Skills Are On-Demand., 86% of Your Rules Context Is Wasted Without Path Globs, path rules, File TYPE or File LOCATION	Write one command, one skill with allowed-tools, and one rules file with paths:
3	Tasks 3.4 and 3.5: 55% Rollback Rate When You Skip Planning, If You Already Know the Fix, Just Fix It, Explore sub-agent, Plan → Explore → Direct, examples over text, tests first, interview mode, issue classification	Run one task in plan mode and one directly; note the difference in what happens before the first edit
4	Task 3.6: Without -p, Your CI Job Hangs, Schema Flag → 99%, CLAUDE.md Is Your CI Agent’s Only Source of Project Standards, self-review isolation, incremental review, test dedup	Run claude -p "..." --output-format json once and look at the output shape

If Claude Code is new to you, read the Foundations pieces first: Claude Code: A CLI Tool, Not an IDE, Three Levels of CLAUDE.md, and Four Permission Modes.

What should I memorise, and what should I reason about?

Memorise (these are file-location and syntax facts):

• ~/.claude/CLAUDE.md is personal, never shared. .claude/CLAUDE.md is project, version-controlled. Subdirectory CLAUDE.md loads only for edits under it. All active levels stack.
• @path/to/file.md imports; a plain mention does not. @~/... resolves to each user’s home and silently loads nothing for people who lack the file.
• Rules files in .claude/rules/ need paths: globs; ** matches any depth; overlapping globs both load.
• Commands are plain markdown in .claude/commands/. Skills are SKILL.md in .claude/skills/<name>/ and support context: fork, allowed-tools, argument-hint.
• context: fork isolates conversation memory, not the filesystem (Fork Isolates Memory, Not the Filesystem).
• CI: claude -p "prompt" for non-interactive; --output-format json for structure. There is no --batch, --no-interactive, or CLAUDE_CI variable.
• /memory shows what is loaded now.

Reason about:

• Type or location? File type (all tests, all Terraform) means a glob rule; a location (everything under database/) means a directory CLAUDE.md.
• Command or skill? Only a skill if you need isolation, tool restriction, or argument prompting; otherwise a command is simpler and faster.
• CLAUDE.md or skill? Always-on rule versus on-demand workflow.
• Plan mode or direct? Uncertainty, not file count. Multiple valid approaches or unfamiliar code means plan; a known fix means direct.
• Describe or show? When the format matters, show a concrete example (Text Descriptions Hit an Ambiguity Ceiling at 63%. Examples Reach 91%.); when correctness matters, write tests first (Tests First: 3 Iterations to 15/15).
• One issue or several? Interacting changes go together, independent ones separately (Classify Before Submitting).
• Same-session review or independent? Independent, with prior comments and existing tests supplied (Claude Reviewing Its Own Code Is Barely a Review, 60% Duplicate Test Suggestions).

What practice cadence works for Domain 3?

• After each session, do 15 to 20 Domain 3 practice questions. Before revealing, name the mechanism: level, import, glob, skill frontmatter, mode, or CI flag. Most Domain 3 questions collapse to one of those six.
• After session 4, do a 40-question block. Domain 3 has more distinct facts than Domains 1 or 2, so a longer block is useful for finding the ones that have not stuck.
• Review mistakes every few days. The recurring ones are user-level versus project-level, section headers mistaken for scoping, and picking a fictional CI flag. Each is a two-minute reread.
• Final week: the free CCA-F mock exam will include roughly 12 Domain 3 questions among 60 (confirm current weights on the official Anthropic / Pearson VUE page before you register).

Self-check questions before moving on

1. Standards live in a senior engineer’s ~/.claude/CLAUDE.md. What happens when a new hire clones the repo?
2. CLAUDE.md says “follow docs/style.md”. Is style.md in context?
3. A React rule and a testing rule both match Button.test.tsx. Which loads?
4. A /migrate-db skill ran DROP TABLE despite an instruction not to. What frontmatter was missing?
5. Which mode for “add caching to the API layer” in a codebase you have never seen? Which for a stack trace pointing at line 42?
6. A large repo makes plan mode fill the context. What do you delegate, and to what? (120 Files Explored, 12% Context Used)
7. A CI job hangs and exits 137. What flag is missing?
8. Prompt says “output JSON” and it works 73% of the time. What is the reliable alternative?

Mistakes worth reviewing twice

• Believing ”## Backend” as a heading in a root file scopes rules (Section Headers Do Not Enforce Scope — Path Rules Do).
• Making everything a skill and paying fork latency on two-second operations.
• Naming a personal skill the same as a project skill (Personal Skills Live in ~/.claude/skills/ — Name Them Differently).
• Skipping the interview step on a large task and rebuilding three times (Three Rounds of Rework Because Nobody Asked About Deployment Topology).
• Assuming the CI agent knows anything your local session knew.
• Repeating review comments because the reviewer never saw prior comments (62% of Review Comments Were Duplicates).

Next step

Start session 1 on the Domain 3 study guide page with a real repo open, and finish the Domain 3 practice set by the end of the week. For a deeper look at the CLAUDE.md piece specifically, read How to Write an Effective CLAUDE.md File. When your accuracy holds, sit the free CCA-F mock exam.