AI is fast. That is exactly why these matter.
01
Never paste a password or API key into the chat.
Chats get saved. Keys leak.
say this insteadPut my key in a .env file and never show it on screen.
02
Never say yes without reading.
Yes to the wrong delete is forever. Slow down on words like delete, remove, force and reset.
say this insteadBefore you change anything, tell me what it does in one sentence.
03
Never let it message a customer without your yes.
A text or email that goes out cannot come back.
say this insteadDraft it. Do not send. Wait for me to say send.
04
Never run a command you can't explain.
Especially one you copied from a random website.
say this insteadWhat does this command do, and is it safe?
05
Never start a big change without a save point.
No save point means no undo.
say this insteadMake a save point first, then make the change.
06happened to me
Never trust "done" without checking.
A robot told me a job succeeded. It had sent nothing.
say this insteadShow me proof it worked, from the place it was supposed to land.
07
Never let AI do money math in its head.
It can sound sure and still be wrong.
say this insteadDo the math in code and show your work.
08happened to me
Never let it guess facts that change.
Prices, install steps and laws change. Old answers sound just as confident.
say this insteadSearch the web and show me your sources.
09
Never turn on an automation without an off switch.
Things that run on their own keep running when you are asleep.
say this insteadBefore this runs on its own, show me how to turn it off, and send the first one only to me.
10happened to me
Never run two AI windows on the same project blind.
They can overwrite each other's work.
say this insteadGet the latest version before you change anything.
11
Never turn off the yes and no questions on day one.
Claude Code can run without asking you first. That is for later, once you know what it does.
say this insteadAsk me before every change.
12
Never put secrets in a public project.
Projects on GitHub can be public. Anyone can read them.
say this insteadCheck that no keys or passwords are in anything we publish.
13
Never keep one chat going forever.
Everything you say stays in its memory and uses up tokens. Old instructions pile up and leak into new work. Long chats get slower, pricier and sloppier.
say this insteadWrite me a short handoff note. Then I'll type /clear and we start fresh.
14
Never explain the same rule twice.
If it matters every time, save it once. Then every fresh chat already knows it.
say this insteadSave this as a rule in my CLAUDE.md file so you always follow it.
15
Never start without saying what done looks like.
Without a finish line, AI keeps going past what you wanted, or stops too early.
say this insteadHere is what done looks like: [one sentence]. Stop there and show me.