What your AI coding agent is asking for.
By the Olomni team · Updated
Every command an AI asks you to approve can be said in one everyday sentence: what it wants to do and what happens if you say yes. For example, rm -rf is “delete a folder with everything inside it” and git push --force is “replace the version saved online”. This guide translates the most common ones and says which can be undone.
Deleting
rm -rf folder→ delete a folder with everything inside it. If the project uses Git, Olo saves a copy first so you can go back.git clean -fd→ delete new files you haven’t saved yet. They can’t be recovered.DROP TABLEorsupabase db reset→ delete data or your whole database. It can’t be undone.
Publishing and uploading
git push→ save your work online.git push --force→ replace the version saved online: it can erase work, yours or other people’s.vercel deploy --prodorsupabase functions deploy→ publish changes to your real app.npm publish→ publish your code for anyone to use.
Installing and running
npm install -g→ change programs your whole computer uses.curl … | sh→ run a program downloaded from the internet.sudo→ do something as your computer’s administrator.
Keys and access
- Reading
.sshor a.envfile → reading a private key or a file with passwords. chmod→ change who can use a file.
These are the sentences Olo shows on each permission; the exact command stays behind “See exactly what it asked”.
Frequently asked questions
What does rm -rf mean?
Delete a folder with everything inside it, without asking. If there is no copy, it can’t be recovered.
What does git push --force mean?
Replace the version saved online with yours. It can erase work that was already uploaded, including other people’s.
What is --dangerously-skip-permissions?
A Claude Code option that skips permission prompts: the AI acts without asking. With it nobody reviews what gets deleted or published; that is why Olo prefers to explain each permission instead of skipping them.