開Sign

People keep losing money because they don't know what their wallets are signing. ByBit lost 1.4 Billion dollars. Ledger made a standard called ERC7730 to help describe contract actions in English so they the wallet can display them to the user. But it uses a centralized repo controlled by them. Other hardware wallets don't want to use that. We built a tool for decentralized on-chain curation of ERC7730 specs using the reality.eth optimistic oracle and Kleros, and we forked the Ledger ERC7730 creation tool to link it up to the on-chain system. We also built an AI bot to detect bad ERC7730 submissions and challenge them. Anyone can run their own bot and improve on it and make money by detecting bad submissions. Kai-Sign is a platform where users can create and verify ERC7730 metadata.

The workflow involves:

1. Users build ERC7730 metadata specifications.
2. These specifications are sent to Reality.eth, a crowdsourced verification system using an escalation game, backstopped by Kleros.
3. After passing verification, the metadata is curated and displayed on a single page.

This approach ensures trusted and verified metadata through decentralized consensus mechanisms.

Kai-Sign was built to address a critical problem in the blockchain ecosystem: the lack of clear transaction descriptions leading to security vulnerabilities.

We forked Ledger's ERC7730 creation tool but enhanced it with a decentralized on-chain curation system using reality.eth and Kleros.

Our frontend leverages Next.js 15, TypeScript, and TailwindCSS to create a responsive user interface, with ethers.js and web3.js handling blockchain interactions.

For the backend, we implemented a FastAPI server that generates ERC7730 descriptors and interfaces with our AI evaluation system.

The smart contract infrastructure, written in Solidity and deployed using Foundry, consists of the KaiSign contract that manages the lifecycle of ERC7730 specifications through Reality.eth's crowdsourced verification mechanism.

A key innovation in our project is the AI bot challenger system powered by Google's Gemini API. This Python-based bot continuously monitors new ERC7730 submissions, analyzes them for potential security issues, and automatically challenges suspicious specifications by submitting transactions with increasing bond amounts.

All specifications are stored on IPFS, with only the content hashes stored on-chain, significantly reducing storage costs while maintaining decentralization principles.

We integrated MultiBaas from Curvegrid to provide efficient blockchain data access, using custom event queries and JWT authentication for monitoring verification status updates in real-time.

Moreover, we have another branch on the GitHub which we put the smart contract on Celo and the result will be curated by KaiSign Contract.

Our architecture creates a complete pipeline from ERC7730 creation to verification: users submit specifications through our frontend, the metadata is stored on IPFS and registered on-chain, then both our AI bot and human participants can verify or challenge submissions through Reality.eth's escalation game mechanism.

Disputed cases are ultimately resolved through Kleros, ensuring a fully decentralized arbitration process. By replacing Ledger's centralized repository with this trustless approach, we've created a more secure, transparent system for transaction descriptions that any wallet provider can leverage, potentially preventing losses like ByBit's $1.4 billion incident.

The entire project demonstrates how AI, smart contracts, and decentralized verification can combine to solve real blockchain security challenges.