Skip to content

CCAR-F · Domain 3: Claude Code Configuration & Workflows · 20% of the exam

Task 3.4: Determine when to use plan mode vs direct execution

Plan mode for large, multi-file or architectural changes with more than one valid approach, direct execution for small well-scoped fixes, and the Explore subagent for keeping verbose discovery out of the main context.

Study it

  • Plan mode, direct execution and the Explore subagent

    Partly covered by: Context Windows

Sample questions

Try each one before opening the answer. Every option is explained, with the Anthropic documentation page that proves it.

Question 1 · choose 1

You need to replace a date library across about 60 files. There are two plausible strategies, a thin wrapper module or direct calls to the new library, and parts of the codebase use the old library in unusual ways. How should you run this in Claude Code?

  1. AStart in plan mode to explore the usages and settle the strategy, then approve the plan and let Claude implement it
  2. BDelegate the 60 files to parallel subagents, each converting its files in whichever way it judges best
  3. CHave Claude convert one typical file each way in acceptEdits mode, then pick the strategy from those two diffs
  4. DWrite the wrapper module yourself, then have Claude convert all 60 files to use it directly
Show the answer and why
  • AStart in plan mode to explore the usages and settle the strategy, then approve the plan and let Claude implement it

    Correct

    Plan mode separates exploration and design from editing, which suits multi-file work with competing approaches; execution follows the approved plan.

  • BDelegate the 60 files to parallel subagents, each converting its files in whichever way it judges best

    Incorrect

    Each subagent would pick its own strategy, so the codebase could end up with both approaches and no one decision behind them.

  • CHave Claude convert one typical file each way in acceptEdits mode, then pick the strategy from those two diffs

    Incorrect

    One typical file does not show the unusual usages that decide between the strategies, and edits start before the approach is settled.

  • DWrite the wrapper module yourself, then have Claude convert all 60 files to use it directly

    Incorrect

    This commits to one strategy before anyone has explored how the unusual usages would fit it.

Use plan mode for large, multi-file changes with more than one valid approach, and combine it with execution: explore and plan the migration, approve the plan, then implement it.

Question 2 · choose 1

A production stack trace points to a single function that crashes when an optional discount field is missing. The fix is a null check in that function plus a unit test. How should you use Claude Code?

  1. AAsk Claude to interview you about requirements and write a spec file before implementing
  2. BEnter plan mode first so that Claude can explore the codebase and propose alternative designs
  3. CAsk Claude to make the fix directly, giving it the stack trace and the unit test it should add
  4. DHave the Explore subagent survey every module that touches discounts before any change is made
Show the answer and why
  • AAsk Claude to interview you about requirements and write a spec file before implementing

    Incorrect

    The requirement is already clear from the stack trace; there is nothing to gather.

  • BEnter plan mode first so that Claude can explore the codebase and propose alternative designs

    Incorrect

    Planning adds overhead that a change you can describe in one sentence does not need.

  • CAsk Claude to make the fix directly, giving it the stack trace and the unit test it should add

    Correct

    A well-understood, single-file change with a clear cause is what direct execution is for.

  • DHave the Explore subagent survey every module that touches discounts before any change is made

    Incorrect

    A codebase-wide survey is unnecessary when the stack trace already identifies the function.

Choose direct execution for simple, well-scoped changes with a clear cause; the extra steps in the other options add overhead without reducing risk here.

Question 3 · choose 1

In plan mode, Claude needs to understand a large part of your codebase before proposing a plan. Where does that research happen?

  1. AIn the main conversation, which reads every file directly
  2. BNowhere; plan mode only plans from the prompt text
  3. CIn a background session that you must start yourself
  4. DIn the built-in Plan subagent, in its own context window
Show the answer and why
  • AIn the main conversation, which reads every file directly

    Incorrect

    Delegation keeps that output in a separate context.

  • BNowhere; plan mode only plans from the prompt text

    Incorrect

    Plan mode reads files and explores before planning.

  • CIn a background session that you must start yourself

    Incorrect

    Plan mode handles the delegation.

  • DIn the built-in Plan subagent, in its own context window

    Correct

    Plan mode delegates research to the Plan subagent so exploration stays out of the main conversation.

During plan mode, Claude delegates codebase research to the read-only Plan subagent, keeping exploration output in a separate context window.

Question 4 · choose 1

You ask Claude Code to find where one environment variable is read, and it hands the search to the Explore subagent. What can Claude specify to keep this lookup fast?

  1. AA "quick" thoroughness level instead of "very thorough"
  2. BThe full transcript of the main conversation
  3. CA plan file that Explore must approve first
  4. DA write permission so that Explore can save its findings
Show the answer and why
  • AA "quick" thoroughness level instead of "very thorough"

    Correct

    Explore takes a thoroughness level: quick, medium or very thorough.

  • BThe full transcript of the main conversation

    Incorrect

    Explore runs in its own context and does not need the history.

  • CA plan file that Explore must approve first

    Incorrect

    Explore does not approve plans.

  • DA write permission so that Explore can save its findings

    Incorrect

    Explore is read-only.

When invoking Explore, Claude chooses a thoroughness level so that targeted lookups stay quick and broad analyses go deeper.

Question 5 · choose 1

You are working in a normal-mode Claude Code session. You now want to switch into plan mode and have Claude start on your question about how to restructure the queue workers, all from the prompt. What is the simplest way?

  1. AAsk the Explore subagent the question first and wait for its summary
  2. BType /plan followed by the question
  3. CRun /compact before asking the question
  4. DAdd "do not edit files" to CLAUDE.md, then ask the question
Show the answer and why
  • AAsk the Explore subagent the question first and wait for its summary

    Incorrect

    Explore searches the codebase and returns a summary; it does not put the session into plan mode or produce a plan to approve.

  • BType /plan followed by the question

    Correct

    /plan enters plan mode from the prompt, and a description after it starts that task right away.

  • CRun /compact before asking the question

    Incorrect

    Compaction frees context; it does not change the permission mode.

  • DAdd "do not edit files" to CLAUDE.md, then ask the question

    Incorrect

    CLAUDE.md is guidance for every session, not a mode switch, and it is not enforced.

/plan enters plan mode directly from the prompt; pass a description, such as /plan restructure the queue workers, to enter plan mode and start on that task immediately.

Question 6 · choose 1

Your forked research skill uses agent: Explore, and its findings ignore a naming convention that is documented only in your CLAUDE.md. Why?

  1. AExplore starts without CLAUDE.md, so the skill must say it
  2. BExplore only reads files that were modified today
  3. CForked skills never receive any project files
  4. DThe convention must be written in uppercase to be noticed
Show the answer and why
  • AExplore starts without CLAUDE.md, so the skill must say it

    Correct

    The built-in Explore and Plan agents skip CLAUDE.md to keep their startup small.

  • BExplore only reads files that were modified today

    Incorrect

    There is no such restriction.

  • CForked skills never receive any project files

    Incorrect

    They can read files; they just start without CLAUDE.md here.

  • DThe convention must be written in uppercase to be noticed

    Incorrect

    The file is not loaded at all, whatever its wording.

Explore and Plan start without CLAUDE.md and the git status snapshot; put whatever such a task needs into its prompt or SKILL.md.

Question 7 · choose 1

Your team wants every terminal session in a sensitive infrastructure repository to start in plan mode, so that work begins with analysis. What should you configure?

  1. AA skill named plan that developers invoke first
  2. BdefaultMode "plan" in .claude/settings.json
  3. CA PreToolUse hook that logs every edit
  4. DAn instruction in CLAUDE.md to plan before editing
Show the answer and why
  • AA skill named plan that developers invoke first

    Incorrect

    A skill does not set the session's permission mode.

  • BdefaultMode "plan" in .claude/settings.json

    Correct

    defaultMode sets the starting permission mode for the project's sessions.

  • CA PreToolUse hook that logs every edit

    Incorrect

    Logging edits does not change the mode a session starts in.

  • DAn instruction in CLAUDE.md to plan before editing

    Incorrect

    That is advisory and does not change the mode.

To make plan mode the default for a project's terminal sessions, set defaultMode to "plan" in .claude/settings.json.

Practise domain 3 →Practise all domains →