Cheat sheet

Everything we used today, on one page. Print it or keep it open in a tab.

Start and stop

Do this To
cd into your project folder, then claude Start a session in that folder. Claude can read and change files there.
claude --version Check it’s installed
claude doctor Check your install and settings when something’s off
/exit, or twice Leave the session
claude --continue Pick up your last session in this folder

Keys

Key What it does

Stop Claude mid-step, keeping the work so far, so you can redirect it
(empty input) Open the rewind menu: go back to an earlier point

Change the permission mode (see below). On Windows, if it does nothing:
(Mac Terminal), (iTerm2), (Windows) Paste a screenshot into your prompt

Open the plan, or your prompt, in a text editor to edit it

Show the full details of what Claude did
! at the start Run a shell command yourself, such as ! open dashboard.html
@ Point at a file, such as @README.md

Permission modes

The bottom of the screen shows the current mode. cycles through them.

Mode Claude… Use it when
⏵⏵ auto mode on (the default) Acts without asking; a separate safety check reviews each action and blocks risky ones You trust the direction of the task
⏸ manual mode on Asks before edits and most commands; reads freely You want to see every step
⏵⏵ accept edits on Edits files, and runs file commands like mkdir, mv and rm in the folder, without asking; asks before other commands Edits are routine
⏸ plan mode on Reads and proposes a plan, changes nothing Anything that matters: read the plan first

/plan <what you want> jumps straight into plan mode. When the plan is ready you choose Yes, and use auto mode, Yes, manually approve edits, or No, keep planning (then type what to change).

Commands we used

Command What it does
/statusline <what to show> Choose what the line under the input box shows: model, folder, effort, context used. /statusline remove it turns it off
/plan <task> Plan first, change nothing until you approve
/goal <condition> Keep working until the condition is true; a separate model checks after each turn. Give it a finish line Claude can prove, and a limit like “stop after 5 turns”
/memory See and edit what Claude remembers: your CLAUDE.md files and its own notes
/init Draft a CLAUDE.md for this folder
/rewind Go back to before a prompt: restore the files, the conversation, or both
/clear Start a fresh conversation (memory files stay)
/skills List the skills Claude can use here
/reload-skills Pick up a skill you created or changed in a skills folder that’s new this session
/artifacts List pages you’ve published, and copy their links
/usage See how much of your plan’s usage you’ve used
/context See how full the conversation is
/btw <question> Ask a side question without derailing the work
/help, /powerup Help, and short interactive lessons on features
/doctor Diagnose and fix setup problems

Memory: two kinds

You write it Claude writes it
What CLAUDE.md in your project folder Auto memory: notes Claude saves as it learns your project
How Ask: “Add this to CLAUDE.md: …”, or run /init for a first draft On by default; you’ll see “Saved 1 memory”
Where Your folder, next to your files ~/.claude/projects/<project>/memory/
Good for Rules and facts Claude can’t know: definitions, deadlines, house style Things it figured out along the way

Keep CLAUDE.md short. Every line is read at the start of every session.

Skills (also called custom commands)

A skill is a folder with a SKILL.md file of instructions. It becomes a command named after the folder.

.claude/skills/house-style/SKILL.md   →   /house-style
  • You can run it by typing /house-style.
  • Claude can also pick it up on its own when its description matches what you asked for.
  • Put it in your project’s .claude/skills/ to keep it with the project, or in ~/.claude/skills/ to use it everywhere.
  • The old .claude/commands/name.md files still work and do the same thing.

The easiest way to make one: ask Claude. “Create a skill called … that …”

Habits worth keeping

  1. Plan before anything that matters. Read the plan for assumptions you didn’t make.
  2. Write down what it can’t know: definitions, dates, rules, all in CLAUDE.md.
  3. Give it a way to check its work: a number you trust, a test, a screenshot, a /goal.
  4. Check one number yourself. Looking right isn’t being right.
  5. Know your undo. /rewind restores Claude’s file edits but not what shell commands did (deleting or moving files). Work on copies, and use git for anything real.
  6. Mind the data. Only give it data you’re allowed to share with an AI service, and follow NC State’s rules for sensitive data.
  7. Start fresh between tasks with /clear. Long, tangled sessions get worse, not better.

Checked against Claude Code 2.1.299 on October 14, 2026. The official docs are the latest word.