18 Claude Code commands that save you time and money

Opus and Fable are the strongest models, and the most expensive. But most of what Claude does in a session is routine: reading files, making small edits, running tests. That work doesn’t need the top model.

With the advisor on, you run a cheaper model as your main one. When it reaches a decision that matters, like choosing an approach, getting stuck on an error that keeps coming back, or checking its work before saying it’s done, it consults the stronger model. The advisor reads the whole conversation and sends back guidance.

/model sonnet/advisor opus

Run /advisor on its own to pick from a list, and /advisor off to turn it off. Your choice is saved for future sessions. When Claude consults it, you’ll see an Advising line; press Ctrl+O to read what the advisor said.

Good pairings

• Sonnet + Opus advisor: the sensible default for most projects.
• Haiku + Opus advisor: the cheapest setup that still plans well.
• Sonnet + Fable advisor: Fable’s judgment at the decision points, if your plan includes Fable.

Normally Claude finishes a turn and hands control back to you, even when the job is half done. With a goal set, a separate model checks after every turn whether your condition is met. If not, Claude starts the next turn on its own.

/goal npm test exits 0 and no test file is modified, or stop after 20 turns

Setting a goal starts work immediately, so you don’t need a separate prompt. Run /goal on its own to see progress, turns used and tokens spent. Run /goal clear to stop early.

Write a condition Claude can prove

The checker doesn’t run commands or read files. It only reads the conversation. So write a condition that shows up in Claude’s output:

• One measurable end state. “All tests pass”, not “the app works well”.
• How to prove it. “npm run build exits 0”.
• What must not change. Without “no test file is modified”, the quickest way to pass a test is to edit the test.
• A limit. Add “or stop after 20 turns” so a stuck goal doesn’t run all afternoon.

Sometimes you want a teammate to see how a bug was tracked down, or you want ChatGPT or Gemini to review what Claude did. Copying a long terminal session by hand is painful. /export does it in one step.

/export                      # choose: copy to clipboard, or save to a file/export login-bug-session.txt  # write straight to this file

Claude Code saves a checkpoint of your files every time you send a prompt. Run /rewind, or press Esc twice with an empty prompt, and you’ll see a list of your earlier prompts. Pick one, then choose what to restore:

• Code and conversation: as if the later messages never happened.
• Conversation only: keep the files, go back in the chat.
• Code only: undo the file changes, keep the discussion so Claude remembers what didn’t work.

Checkpoints are kept with the session, so rewind still works after you close and resume it.

You’re deep into a session and need to know one small thing: which file had the config, what a function was called, why Claude chose one approach over another. Asking normally adds that to the history, and every extra message makes the session longer.

/btw which file did we put the Stripe webhook in?

The answer pops up in an overlay. Press Space, Enter or Esc to dismiss it, and the main conversation carries on as if you never asked. It works even while Claude is busy, and it’s cheap.

In plan mode, Claude reads your code and writes a plan, but doesn’t edit anything. You read the plan, push back, and only then let it build. This is the single best habit for bigger changes.

/plan add Google sign-in to the app

You can also press Shift+Tab to cycle into plan mode.

/init scans your project and writes a CLAUDE.md file: how to run it, how it’s organised, which commands test and build it. Claude reads that file at the start of every session, so it already knows the basics.

Then add your own rules to it, like “use Tailwind, not inline styles” or “never touch the payments folder”. Run /memory to open and edit it.

Claude can only hold so much at once. That space, the context window, fills up with your messages, files it read, tool output, your CLAUDE.md and any connected tools. When it’s nearly full, Claude starts losing track of earlier details.

/context shows what is using that space and suggests what to trim. If one connected tool or a huge file is taking a big share, you’ll see it straight away.

When /context shows the window filling up and you’re not done, /compact replaces the history with a summary. You can tell it what matters most:

/compact keep the database schema and the list of bugs still open

Finished with one task and moving to another? Use /clear instead. It starts a fresh conversation but keeps your CLAUDE.md.

You’re halfway through and want to try a riskier approach, but you don’t want to lose the session as it stands. /branch makes a copy of the conversation at this point and switches you into it. The original stays as it was, and /resume takes you back to it.

/branch try-redis-cache

AI-written code works, but it often skips the boring safety parts: checking who is logged in, cleaning what users type, keeping keys out of the browser. /security-review reads the changes on your current branch and flags risks like injection, broken login checks and leaked data.

/security-review

Passing tests and a clean build don’t prove a feature works. The button can still do nothing, or the page can still be blank. /verify has Claude build your app, run it, use the change the way a person would, and report what actually happened.

/verify

If you’re trying the advisor setup from section 01, this is how you check it’s saving money. /usage shows what the session has cost so far and, on a paid Claude plan, how much of your limit is left and what’s using it up. /cost opens the same screen.

/goal keeps going until a condition is met. /loop repeats on a clock instead. It’s for anything you’d otherwise check by hand every few minutes.

/loop 5m check if the deploy finished and tell me if it failed/loop check the error logs and fix anything new

Leave out the interval, like the second example, and Claude decides how long to wait between runs.

Claude is halfway through a long task and you have to go. /remote-control (or /rc) makes the session available on claude.ai, so you can follow along, answer Claude’s questions and send new prompts from your phone or another computer.

/insights reads your recent sessions on this computer and builds a report you open in the browser: which projects you work in, how you use Claude Code, where things tend to go wrong, and features you haven’t tried. It’s the quickest way to find your own bad habits.

If you click “yes” to git status or ls twenty times a day, this one is for you. It scans your past sessions for read-only commands you keep approving and adds them to an allow list, so Claude stops asking.

Selecting text in a terminal picks up line breaks and spacing that weren’t in the answer. /copy copies Claude’s last reply to your clipboard exactly as written.

/copy     # the last answer/copy 2   # the answer before that

If the answer has code blocks, a picker lets you copy just the one you need. Press w in the picker to save it to a file instead.