用 OpenAI Dots 规则屏给钱包操作设权限的实盘做法

Mason @Mason33xbt原推发布于 2026.10.21

在 X 上打开原推

WHAT THE F*CK: HIS AI AGENT CAN DUMP HIS MEME BAGS AT 3AM WITHOUT ASKING. IT CAN'T BUY A SINGLE ONE. YOURS IS SET UP BACKWARDS, AND THAT IS HOW WALLETS GET EMPTIED. A 21-year-old runs a one-person desk with one OpenAI dot watching his bags 24/7. When OpenAI shipped dots, they shipped a rule screen with them, and every action gets one label: allow, require approval or block. Most people skip that screen, and he spent more time on it than on the bot. The chain is dot watches → rules label the action → allowed ones run, a buy waits for his tap, a transfer out never happens. His rules, ready to copy: reading wallets and dev history is allowed, so the dot digs all night without bothering him → selling an open bag is allowed, because when a whale dumps at 3am he wants out, not a notification → buying a token needs approval, one tap from him every single time → moving Solana out of the wallet is blocked → typing a seed phrase is blocked → changing a password stays with him, OpenAI built that one in. Exits protect money, so they run free. Entries risk money, so they need a human. A transfer out can end everything, so nobody gets it, not even a perfect agent. The honest part: approval only works if the human reading it is awake and sober, and a tired yes at 4am is still a yes. Write the rules before the agent touches a wallet, because after that you are just hoping. The full terminal is free and open source: http://github.com/Mason333xbt/sol-terminal Steal this: exits run free, buys need one human yes, transfers out are blocked. One-person desk, the dot does about 90% of the work and his only job is the yes on a buy. Ready to copy. Save this. Follow for the rules nobody writes.

这条是引用着另一条写的：x.com/mason33xbt/status/2106439961117348281