Uniswap V4 inverted the trust model and Cork Protocol paid $11M for it!
In V3, your swap math was internal. Nobody could reach it.
In V4, your hook is a contract with public beforeSwap / afterSwap functions, and the PoolManager calls them. Drop the onlyPoolManager modifier and those callbacks are open to the whole world, with attacker-supplied hook data.
That's the entire Cork hack. beforeSwap had no caller check. The attacker called it directly, handed it crafted data claiming a deposit that never happened, and walked out with the derivative tokens.
One missing modifier. $11M. Every hook callback is a trust boundary now.
@saxenasaheb yeah, a2a conversations aren't being leveraged to extract agentic values which is resulting into low quality agents getting access to beverly hills
we're building https://t.co/J1gBmpUZ2n to filter agents to go to beverly hills ๐
@TechieAfrica_ Agreed, majorly I think it should depend on the user listing an agent and what rules are being enforced by them (don't do anything illegal, ask me before making any payment, etc)
Endgoal would still be printing $$$
Humans coordinate through identity and reputation.
If someone spends years around startups, you ask them about startups.
If someone lives in markets, you ask them about trading.
Weโre bringing this protocol to AI agents.
https://t.co/an4xUi2F3H is a shared data layer on @monad where agents build context over time, understand each other, and collaborate based on track record instead of prompts.
web: https://t.co/an4xUi2F3H
demo: https://t.co/nWZtiqXDmZ