Claude Code Configuration & Workflows on CCAR-F: twelve items, six task statements, three decisions

Claude Code Configuration & Workflows on CCAR-F is Domain 3 of the architect foundations blueprint: 20 percent of 60 items, about twelve questions over six published task statements, roughly two each. Three of the six exam scenarios name it as a primary domain and four of the guide's twelve sample questions sit in it. Most items reduce to one of three decisions: whether a setting travels with the repository or stays on one machine, whether an instruction loads always, on a path match or on demand, and whether a task earns plan mode before execution.

What Claude Code Configuration & Workflows on CCAR-F weighs

Section 4 of the official guide gives Domain 3 a weight of 20 percent, behind only Agentic Architecture at 27. On a 60-item paper that is twelve items, about 24 minutes at two minutes per item, or two items per task statement. In Section 5, three of the six published scenarios list the CCAR-F Claude Code domain under Primary domains: Code Generation with Claude Code, Developer Productivity with Claude, and Claude Code for Continuous Integration; the full map is in CCAR-F exam scenarios.

Source: Claude Certified Architect Foundations Exam Guide v1.0, Sections 4, 5, 6 and 9, from the Partner Academy certification page. Checked against the official sources on 17 September 2026. Twelve items is the allocation that matches the weight, not a guaranteed count.

Six task statements, and what each one names

The table keeps to what Section 6 prints under each statement. The documentation column is the page for current behaviour; the exam is written to the guide, which moves more slowly.

MCP server scoping is Task Statement 2.4 and belongs to tool design and MCP on CCAR-F; whether an instruction file gives a tendency or a guarantee is developed in Claude Code configuration for the exams.

Shared through the repository, or personal to one machine

The first decision runs through 3.1 and 3.2 and is the easiest pair of CCAR-F Domain 3 items once you see it. Anything teammates must receive lives in the repository under .claude/ and arrives through version control; anything only you want lives under ~/.claude/, invisible to everyone else. The guide's own diagnostic for 3.1 is a new team member not receiving an instruction: it sits in a user-level file. The CCAR-F CLAUDE.md hierarchy question adds one level, a subdirectory CLAUDE.md for work in that directory.

Always, when a path matches, or when invoked

The second decision separates 3.1, 3.3 and 3.2 by when an instruction enters context. A CLAUDE.md loads at the start of every session, so it carries universal standards and stays lean; @import keeps it modular, and /memory shows which memory files loaded. A file in .claude/rules/ with a paths list in its frontmatter loads only when Claude edits a matching file, which the guide prefers over subdirectory CLAUDE.md files for a file type scattered across the tree, such as every test file. A skill or command loads when invoked, so task-specific workflows belong there, and context: fork keeps a verbose skill's output out of the main conversation.

Plan mode, direct execution and refinement

Task Statement 3.4 is a judgment with named examples on both sides. Plan mode is for architectural decisions, multi-file changes and problems with several valid approaches; the guide cites restructuring a monolith into microservices and a library migration touching dozens of files. Direct execution is for a well-scoped change with a clear cause, such as a single-file fix with a stack trace. Plan the investigation, then execute directly, with verbose discovery delegated to the Explore subagent so the main context survives to implementation.

Task Statement 3.5 is the part candidates skip because it reads as prompting advice. Four techniques are named: two or three concrete input and output examples when prose produces inconsistent results; tests first, then iterating on the failures; the interview pattern, where Claude asks the questions before implementing in an unfamiliar domain; and one detailed message when fixes interact against sequential messages when they are independent. The tempting wrong answer is a longer prose specification.

Unattended: what 3.6 expects you to know

The CI statement is the most spelling-heavy in the domain. The -p flag, also written --print, runs Claude Code non-interactively so a pipeline does not hang waiting for input. Adding --output-format json with --json-schema returns findings the pipeline can parse. CLAUDE.md is how the CI run learns the project: testing standards, fixtures and review criteria belong there. Two behavioural rules complete it: an independent instance reviews code better than the session that wrote it, and a re-run after new commits should see the prior findings so it reports only what is new.

A worked item written for this post

Give the domain a fifth of your preparation, the spellings learned early and the three decisions rehearsed on items. The Claude certification study guide covers weighting a plan by blueprint share; the which Claude certification guide confirms that Claude Certified Architect Foundations is the right paper, since the CCDV-F Claude Code domain is 3.1 percent. The free CCAR-F mock is written to the five-domain guide; the practice exams guide lists what else exists.

Key takeaways

• Domain 3 is 20 percent of 60 items, about twelve questions, over six task statements at roughly two items each.
• Repository or home directory. Files under .claude/ travel with the repo; files under ~/.claude/ stay with you. A teammate missing an instruction has it on the wrong side.
• Always, on a path match, or on demand. CLAUDE.md for standards, a paths rule for scattered file types, a skill or command for task-specific work.
• Plan mode for architecture and many files; direct execution for a scoped fix.
• Unattended means -p, JSON output against a schema, project context in CLAUDE.md, and an independent instance for review.