Skip to content

This page was translated from the Korean original with AI and may contain errors.

Chapter 12. Using It Like an Expert — Pick What Fits the Task, and Stack One at a Time as Needed ​

Advanced Read this once you are comfortable with the earlier chapters and want to use it even better.

One-line summary: An expert is not someone who uses many features, but someone who picks the feature that fits the size of the task. Don't set up everything from the start. Add one thing at a time when the same inconvenience repeats. This chapter wraps up with a weekly routine for running the "Organizing Tool" you built in Stage 3.

This chapter is based on the Claude Code official docs checked on October 31, 2026. The universal principles of "using it like an expert" that work with any AI are in Stage 1, Chapter 11. Here, we cover only how to pick and stack features in Claude Code.

Picking to Fit the Size of the Task ​

For this kind of taskUse thisWhere you learned it
One question, one piece of writingChat (Chat in the Claude app)Stage 2
Reading and editing files in your folderCode tab sessionChapter 2
A big task with several ways to do itStart with a method in Plan (계획) modeChapter 5
Rules you keep repeating every timeCLAUDE.mdChapter 6
The same steps you do oftenSkill (/name)Chapter 7
A task to run on its own at a set timeScheduled taskChapter 8
One-line check or automationclaude -p in the terminalChapter 10
A task you must be able to undo if it goes wrongRewind + Git historyChapter 4

The official docs also explain features by dividing them into "what is always visible every session (CLAUDE.md)," "what you call when needed (skills)," and "what runs on its own at set times."[1] The bigger the feature you use, the more you can do, but you also have more to check.

Think of one of your own tasks and find the matching row in the table above. For example, "make a weekly review every Friday evening" fits the "A task to run on its own at a set time" row, so it is a scheduled task. If no row fits, a single chat or Code tab session is usually enough.

Quick check

Every Friday, you organize your review notes following the exact same steps. Which feature is the best fit?

Stack One at a Time Instead of Building Everything at Once ​

You don't need to set up everything from the start. The official docs say each feature has a signal telling you "add it when this happens."[2]

When this happensAdd this
It gets the same rule or command wrong twiceWrite it in CLAUDE.md
You keep starting tasks with the same requestSave it as a skill
You paste the same steps for the third timeSave it as a skill
You keep copying material from a browserConnect that service with a connector
Something must happen every time without askingHook (a command that runs automatically at a set moment)

Source: Claude Code official docs (partial excerpt)[2:1]

The last two rows of the table are features this guide does not cover in depth.

  • A connector links an outside service such as Gmail or Calendar to Claude (Chapter 7). Once connected, Claude reads that content without you copying and pasting it.
  • A hook is a command that runs on its own when a set moment comes, such as "every time a file is edited." For example, you could have it run a spell check every time Claude edits a file. When the same thing repeats often enough to need a hook, look it up in the official docs then.

The organizing tool in this guide grew in this exact order. Long requests in Chapter 2 → templates in Chapter 5 → CLAUDE.md in Chapter 6 → skills in Chapter 7. You experienced the inconvenience first, and built the tool next.

You use the same signals when fixing what you already built. If the same mistake comes up again, don't just fix it once in chat and move on—fix CLAUDE.md. If there is a step you keep fixing by hand, fix the skill.[3]

Quick check

Which approach to setting up features matches the official docs' recommendation?

Habits You Can Drop Now ​

You may see tricks like the ones below in AI tips articles online. They were needed for older models, but now they can actually get in the way. If you are new, just know that you don't need to follow them. From what the official docs point out, here are the ones that apply to readers of this guide.[4]

Old trickNow
Adding "IMPORTANT!" or ALL CAPS to important instructionsIt follows a plainly written single line too. Piling on emphasis makes it over-apply the rule
Pushing it with "Don't stop, do it to the end"It usually finishes an agreed task to the end. Give it check criteria (Chapter 5) instead
Writing a long folder structure into CLAUDE.mdClaude reads the folder itself. Write down only what isn't in the folder

Source: Claude Code official docs[4:1]

If a rule must never be broken, use another mechanism instead of emphasis. The official docs say to turn unbendable rules into a hook or a deny rule.[4:2] For readers of this guide, protecting your work with permission modes and Git history from Chapter 4 is plenty to start with.

Quick check

Claude occasionally misses a rule in CLAUDE.md. What is the right way to handle this with current models?

Weekly Organizing Routine ​

Try running the tool you built in Stage 3 each week like this.

  1. Collect: Drop your meeting, study, and daily reflection notes into inbox as you go. Remove personal information.
  2. Organize: Run /meeting-notes, /study-notes, and on Fridays /weekly-review (or a scheduled task).
  3. Check: Read the self-check results and "Items to Check" that Claude shows you, and match numbers, owners, and deadlines against the originals.
  4. Record: When you are happy with it, save it as a commit (Chapter 4).
  5. Fix: If you pointed out the same thing twice, update CLAUDE.md or the skill.

It doesn't need to be perfect from the start. Even if you run it once a week and fix just one inconvenient thing, in a few weeks it will become a tool that fits you just right.

Quick check

While running your organizer every week, you end up giving Claude the exact same correction twice. In this guide's routine, what should you do next?

Summary ​

  • Pick to fit the size of the task: Chat for questions, Code for file work, Plan first for big tasks, CLAUDE.md and skills for repetition, scheduled tasks for set times.
  • Don't set up everything from the start. Add one at a time when the same inconvenience repeats, and fix what you already built using the same signals.
  • Give a plain single line and check criteria instead of emphasis or pushing. Protect must-follow rules with mechanisms like permissions and Git.
  • Run Collect → Organize → Check → Record → Fix every week.

Glossary ​

TermOne-line meaning
HookA command that runs automatically at set moments
RoutineA work habit you run in the same order each week (different from Routines, which appears in Chapter 8 as the old name for scheduled tasks)

Finished this chapter?

Progress is saved only in this browser. See your progress on the full table of contents.

Next chapter: You have finished the Stage 3 Claude Code guide. To move the same organizing tool over to OpenAI's Codex, continue to the Stage 3 · Codex Guide. You can also pick a chapter to review from the Stage 3 · Claude Code Table of Contents, or check what you have learned so far with the AI Knowledge Level Test.

References (4)
  1. Claude Code official docs, "Extend Claude Code". "Features range from always-on context that Claude sees every session, to on-demand capabilities you or Claude can invoke, to background automation that runs on specific events." code.claude.com (checked on 2026-10-31) ↩︎

  2. Claude Code official docs, "Extend Claude Code". "You don't need to configure everything up front. Each feature has a recognizable trigger, and most teams add them in roughly this order" / "Claude gets a convention or command wrong twice" / "You keep typing the same prompt to start a task" / "You paste the same playbook or multi-step procedure into chat for the third time" / "You keep copying data from a browser tab Claude can't see" / "You want something to happen every time without asking" code.claude.com (checked on 2026-10-31) ↩︎

  3. Claude Code official docs, "Extend Claude Code". "A repeated mistake or a recurring review comment is a CLAUDE.md edit, not a one-off correction in chat. A workflow you keep tweaking by hand is a skill that needs another revision." code.claude.com (checked on 2026-10-31) ↩︎

  4. Claude Code official docs, "Best practices for Claude Code". "Current ones follow a plainly worded line, and piling on emphasis mostly makes Claude over-apply the rule. If it's a rule that can't bend, turn it into a hook or a deny rule." / "Current models usually carry on through an agreed task without being told" / "Giving Claude a check to pass is different, and still worth doing." / "Claude reads the repo itself. Only write down what isn't in there." code.claude.com (checked on 2026-10-31) ↩︎