aicert.study

5 CCA-F Practice Questions: Claude Code Configuration & Workflows (with Explanations)

July 8, 2026

5 CCA-F Practice Questions: Claude Code Configuration & Workflows (with Explanations)

Five real, previously unpublished CCA-F practice questions for Claude Code Configuration & Workflows (20% of the exam). Try to answer each one yourself before revealing the explanation — that's how you actually find the gaps in your understanding.

Practice the full domain-filtered set on AICert Study →


Question 1

A developer is about to ask Claude Code to migrate an authentication library that touches 45 files across three packages. Which mode should they start in?

  • A. bypassPermissions mode so Claude can edit all 45 files quickly without prompts.
  • B. Plan mode, so Claude explores the code and proposes a migration plan before making changes.
  • C. Default mode, since plan mode is only for read-only investigation and a migration is a write task.
  • D. dontAsk mode to make the run fully non-interactive.
Show answer & explanation

Correct answer: B

Plan mode is designed for exactly this case: a large, multi-file change with architectural implications where multiple valid approaches exist. In plan mode Claude reads files and explores but does not edit, then presents a plan you can review (or refine with Ultraplan) before approving execution. This prevents costly rework on a 45-file migration.

  • bypassPermissions mode is incorrect because it disables safety checks and is intended only for isolated containers/VMs; rushing a 45-file migration with no oversight is risky.
  • Default mode, since plan mode is only for read-only investigation and a migration is a write task is incorrect. Plan mode is specifically about deferring writes during a write task: you explore in plan mode, then approve and switch to edit mode for execution.
  • dontAsk mode is incorrect because it auto-denies anything outside the allow rules, making interactive design work impractical; it is for locked-down CI, not exploratory architectural work.

Share this question on LinkedIn →


Question 2

You are wiring Claude Code into a GitHub Actions workflow to run automated PR reviews. Which flag is the documented way to run Claude Code non-interactively so the job exits when the prompt is processed?

  • A. --interactive=false
  • B. -p (or --print)
  • C. --ci-mode
  • D. --detach
Show answer & explanation

Correct answer: B

The -p / --print flag is the documented non-interactive mode: Claude processes the prompt, prints the response to stdout, and exits without waiting for input. It is the canonical entry point for CI/CD scripts and the Agent SDK.

  • --interactive=false is incorrect; there is no such flag in the documented CLI.
  • --ci-mode is incorrect; no such flag exists. The intent is achieved through -p combined with flags like --output-format json and permission flags.
  • --detach is incorrect; there is no --detach flag. Background sessions are launched with --bg, which is a different concept (a long-lived background agent, not a one-shot exit-on-done run).

Share this question on LinkedIn →


Question 3

Your CI job calls Claude Code to review pull requests and the next step posts inline PR comments. The downstream script expects a stable machine-parseable payload. Which flag combination should the job use?

  • A. -p with --output-format json and --json-schema describing the expected fields.
  • B. -p with --verbose, parsing the human-readable transcript with regex.
  • C. Interactive mode with --resume so the next CI step can attach to the same session.
  • D. Pipe the output through claude --teleport to convert it to JSON.
Show answer & explanation

Correct answer: A

For CI integrations that consume the output programmatically, run with -p, set --output-format json, and (when you need a stable shape) pass --json-schema. The schema enforces the structure of the structured_output field so a downstream script can map findings to inline PR comments without ad-hoc parsing.

  • -p with --verbose, parsing with regex is incorrect: free-form text is unstable across runs and Claude versions; regex parsing is exactly what --json-schema replaces.
  • Interactive mode with --resume is incorrect: interactive mode waits for stdin and is not appropriate for CI. --resume picks up an existing session by ID and is unrelated to structured CI output.
  • claude --teleport is incorrect: --teleport resumes a web session in a local terminal; it does not convert output to JSON.

Share this question on LinkedIn →


Question 4

Your monorepo has 30+ packages, each with its own testing conventions. Some packages use Vitest with snapshot tests, others use Playwright e2e. You want Claude to follow the right conventions only when editing test files in each area. What is the best approach?

  • A. Add every testing convention to the root CLAUDE.md so it is always loaded.
  • B. Place a CLAUDE.md inside each package; nested CLAUDE.md files load when Claude reads files in those subdirectories. Optionally complement with .claude/rules/ files scoped by paths glob (e.g., **/*.test.tsx) for conventions that span the codebase.
  • C. Tell developers to manually paste the relevant convention into chat at the start of each session.
  • D. Encode every convention into a single Skill called /testing, expecting Claude to read it for every file edit.
Show answer & explanation

Correct answer: B

Nested CLAUDE.md files load on-demand when Claude reads files in those subdirectories, which is the documented mechanism for directory-scoped conventions. For conventions that span the codebase (like 'all *.test.tsx files'), .claude/rules/ files with a paths glob frontmatter load only when matching files are read, which is more efficient than a directory-level CLAUDE.md.

  • Add every testing convention to the root CLAUDE.md is incorrect: it bloats the always-loaded context and causes Vitest rules to interfere when editing Playwright tests.
  • Tell developers to manually paste the relevant convention is incorrect: it is unreliable and defeats the purpose of persistent project memory.
  • Single /testing skill is incorrect: a skill is invoked on demand (typed or auto-invoked by description match), not on every file edit. For per-file-type conventions, path-scoped rules or nested CLAUDE.md are the right primitives.

Share this question on LinkedIn →


Question 5

You want Claude to automatically apply Terraform-specific conventions only when editing files under terraform/. The conventions should not be in context for any other work. What is the most appropriate configuration?

  • A. Create .claude/rules/terraform.md with YAML frontmatter paths: ["terraform/**/*"].
  • B. Append the Terraform conventions to the root CLAUDE.md.
  • C. Configure a PreToolUse hook to inject the rules on every Bash call.
  • D. Save the rules to ~/.claude/rules/terraform.md with no frontmatter.
Show answer & explanation

Correct answer: A

Path-scoped rules are exactly this: a markdown file under .claude/rules/ with a paths field in the YAML frontmatter that uses glob patterns. The rule is loaded into context only when Claude reads files matching the pattern, keeping unrelated sessions lean and ensuring Terraform conventions are not noise outside that scope.

  • Append the Terraform conventions to the root CLAUDE.md is incorrect: root CLAUDE.md loads on every session regardless of what you are doing, defeating the targeting goal.
  • PreToolUse hook on every Bash call is incorrect: hooks enforce shell-side behavior at lifecycle events; they are not the documented mechanism for path-scoped contextual instructions, and would inject rules on unrelated Bash calls too.
  • ~/.claude/rules/terraform.md with no frontmatter is incorrect: rules without a paths field load unconditionally, and the user-level path means they would apply to every project on the machine, not just this one.

Share this question on LinkedIn →


Want more? Our free bank has hundreds of Claude Code Configuration & Workflows questions. Practice them all →

Ready to practice for Claude Certified Architect — Foundations (CCA-F)?

Try a free sample simulado and see how you score, domain by domain.

Try the practice exam