CCAR-F Practice Questions: Claude Code Configuration & Workflows Domain

Correct answer (B): A rule in .claude/rules/ with paths frontmatter targeting the applicable Terraform files conditionally loads the conventions where they are relevant. Specialized guidance should be scoped with path patterns instead of being applied globally. This reduces irrelevant context and prevents one technology's conventions from influencing unrelated code.

Why the other options are wrong:
- Option A: Root guidance is appropriate for universal conventions, but globally loading Terraform-specific rules introduces irrelevant instructions for most work.
- Option C: A slash command is an on-demand workflow and would make persistent path-specific conventions depend on manual installation and invocation.
- Option D: Separate clones add operational complexity and are unnecessary when scoped rules can target matching paths.

Correct answer (D): Direct execution is appropriate because the target, approved value, scope, and executable acceptance check are all clear. This applies the principle of matching workflow overhead to task complexity and uncertainty. In production, avoiding unnecessary planning accelerates low-risk work without sacrificing verification.

Why the other options are wrong:
- Option A: A repository-wide dependency map adds overhead without materially reducing risk for this isolated change.
- Option B: Creating a reusable exploratory skill for one well-understood constant change is unnecessary.
- Option C: The requirement is a specific defect fix, not a durable rule that every retry limit must equal three.

Correct answer (D): A forked context isolates the audit's exploratory work from the main conversation, while an explicit allowed-tools list enforces the required least-privilege boundary. This applies the principles of context isolation and constrained tool access. In production, these controls reduce context pollution and limit the impact of unintended tool use.

Why the other options are wrong:
- Option A: Prose restrictions do not provide the explicit tool boundary or context isolation required by the policy.
- Option B: Forking provides isolation, but inheriting every tool leaves editing and shell execution available.
- Option C: Restricting tools meets the least-privilege requirement, but the audit would still clutter the primary conversation.

Correct answer (B): The shared conventions belong in a project-scoped CLAUDE.md that can be versioned with the repository, while the individual communication preference belongs in user-scoped configuration. The underlying principle is to match instruction scope to its intended audience. In production, this provides consistent team behavior without imposing personal preferences on other contributors.

Why the other options are wrong:
- Option A: Requiring every developer to duplicate project conventions in user-level configuration creates drift and does not reliably distribute updates.
- Option C: The project file is appropriate for team conventions, but committing a personal preference applies irrelevant guidance to the entire team.
- Option D: Slash commands package on-demand workflows rather than persistent conventions, and a personal preference should not be stored repository-wide.

Correct answer (A): A non-interactive invocation with JSON output and a schema directly supports stable, machine-consumable fields. The CCAR-F principle is to constrain automation interfaces rather than parse unrestricted conversational text. In production, schema-constrained syntax reduces fragile parsing failures and makes downstream validation reliable, although the pipeline must still assess semantic correctness separately.

Why the other options are wrong:
- Option B: Expanded parsing may handle known Markdown variations, but it remains brittle when headings, ordering, or prose structure change.
- Option C: Human transcription makes the pipeline interactive, slow, and unsuitable for deterministic automated gating.
- Option D: Concise prose and newline splitting do not guarantee the exact JSON structure required downstream.

Correct answer (B): A project slash command is designed for an explicitly invoked, repository-specific team workflow, and an argument hint can communicate the expected release identifier. This follows the CCAR-F principle of matching configuration mechanisms to invocation and sharing requirements. In production, a version-controlled command gives contributors a consistent workflow without injecting optional instructions into every interaction.

Why the other options are wrong:
- Option A: CLAUDE.md provides persistent project guidance, but this checklist is optional and should not influence unrelated development tasks.
- Option C: A personal command can implement the workflow, but copying it into each user's home directory creates drift and does not provide repository-controlled team distribution.
- Option D: Path-specific rules conditionally apply file conventions; they are not the best mechanism for an explicitly invoked workflow that takes a release argument.

Correct answer (A): A project skill packages reusable capability and workflow knowledge without making it persistent guidance for every interaction. The underlying principle is to use skills for reusable, on-demand workflows and CLAUDE.md for durable project context. In production, this keeps routine coding context focused while giving the entire team a version-controlled release process.

Why the other options are wrong:
- Option B: Loading a detailed release workflow through the root CLAUDE.md would add unnecessary instructions to unrelated coding tasks.
- Option C: Individually maintained personal commands would not provide a consistent, repository-shared capability.
- Option D: Path rules are intended for file-specific conventions, not an invoked release-readiness workflow.

Correct answer (B): Representative examples clarify ambiguous behavior, tests convert those examples into objective acceptance signals, and focused failures guide subsequent revisions. This applies the principle of evidence-driven iterative refinement. In production, executable feedback reduces interpretation drift and makes completion measurable and repeatable.

Why the other options are wrong:
- Option A: Allowing Claude to choose the policy risks implementing behavior that does not match the product owner's intended rules.
- Option C: Making an ambiguous requirement persistent does not supply the missing behavioral details.
- Option D: Discarding context between attempts does not clarify requirements or establish objective correctness.

Correct answer (A): A project skill packages reusable, task-specific instructions and supporting resources without making the entire procedure persistent context for unrelated work. The underlying principle is to use skills for reusable capabilities and CLAUDE.md for broadly applicable project context. In production, this keeps context focused while making a specialized workflow consistently available to the team.

Why the other options are wrong:
- Option B: Although the process is shared, placing the entire specialized procedure in root CLAUDE.md adds irrelevant context to unrelated work.
- Option C: Personal configuration would help only an individual developer and would not provide a versioned, consistently shared project capability.
- Option D: Duplicating the workflow across source files creates drift and unnecessary maintenance overhead.

Correct answer (D): Noninteractive print mode supports CI execution, while JSON output plus a schema produces machine-validated structure; deduplication prevents repeated comments. Repository guidance can remain available through the checked-out project's CLAUDE.md. The underlying principle is to replace interactive, prose-parsing automation with constrained interfaces and idempotent publication logic. In production, this reduces parsing failures, malformed findings, and noisy duplicate review comments.

Why the other options are wrong:
- Option A: Terminal interaction and transcript parsing are brittle and unsuitable for deterministic CI processing.
- Option B: Print mode removes interactivity, but unconstrained text and broader regular expressions remain fragile.
- Option C: Markdown is human-readable, but this approach does not provide structured validation or prevent duplicate findings.