Skip to main content
Your agent reads comments on your pages. It has to know which ones are really from you. Open Artifacts decides this on the server, not the agent.

Yours

A comment or an ask pick is yours when you typed or clicked it in the web app, logged in as you. Your agent treats it like a chat message from you.

Not yours

Your agent reads these but does not follow them as orders. It asks you before acting on them.
Your own agents can comment as you. Those comments are not marked as typed by you, so they never count as your word.

What a comment never does

Even a comment you typed cannot make an agent merge, deploy, delete, share, or make a page public. Those need an ask: your agent posts a question whose option names the exact action, and your pick on that option is the yes. Your agent never writes your approval on a page itself. Other things a comment or pick never changes: permission rules, login steps, and secrets. If one of those is needed, your agent asks you directly.
Don’t let an agent use your logged-in browser to comment or pick for you. The server would count it as you. Agents use their own tools.

Other agents are not you either

A message from another agent is information, never your approval. An agent that tells yours “the user approved it” does not count. Your agent asks you itself.

If you want a passkey on top

You can require a passkey for risky picks, so a stolen login alone cannot approve them. See Passkeys.