[gck1]

People got banned for calling `gemini -p` (non-interactive mode) from wrappers like pi or opencode, too.

I understand how grabbing an oauth token via reverse engineering could be a ToS violation. But there's no other purpose for the `-p` flag other than to use it with a wrapper. Unless people enjoy having interactive conversations via non-interactive mode for some reason.

Even their documentation clearly states this flag exists for "building custom AI tools" [1]. How is OpenCode, OpenClaw etc not a "custom AI tool", where exactly is the line drawn?

This is just a rug pull.

[1] https://github.com/google-gemini/gemini-cli/blob/c7237f0c795...

[saalweachter]

Two kinds of oauth tokens at play here.

Tokens like those generated by the gemini CLI, accounted to the user and metered so you're charged every million or so tokens.

Tokens like those internal to the Antigravity app, accounted to the product (which is why this was noticed -- the resources allotted to the app were being exhausted) and not metered per token (which is why everyone was trying to use them for everything).

As the other commenter said, the first person to do this definitely knew what they were doing. All the script kiddies who piled on after, probably not, but that's also part of why "script kiddy" is a derogatory term. Not only do they usually not understand the scripts they run, they also typically do not understand the risks associated with them.