Agentleash

AI agents that hold money are one prompt injection away from losing it. AgentLeash lets an owner write spending rules in plain English (per-payment limit, daily limit, approved recipients, expiry) and enforces them on Arbitrum with a smart contract. The agent can't override the rules, and the owner has a kill switch. Demo: a prompt-injected agent tries to pay an attacker and the chain blocks it. 8 Foundry tests, deployed on Arbitrum Sepolia.

설명

AgentLeash is a spending leash for AI agents, enforced on Arbitrum. The problem: an AI agent with a wallet is one prompt injection away from sending funds to an attacker. The solution: the owner writes rules in plain English (for example "Max $50 per day, $5 per payment, expires in 2 days"). Those rules are written on-chain, and a smart contract enforces them: per-payment limit, rolling daily limit, approved recipients only, expiry, and an owner kill switch. The agent can't change or bypass any of it. Demo: a prompt-injected agent tries to pay an attacker and the chain blocks it with a plain-English reason. A payment over the limit is blocked, then allowed after the owner raises the limit. The kill switch freezes everything. Links - GitHub: https://github.com/BenCoy09/agentleash - AgentLeash contract (Arbitrum Sepolia): 0x313856e11c3dd317c19c9FF8B95623c05Bf673d6 - TestUSDC (Arbitrum Sepolia): 0x98B2feAf9ee9FAbA8f1812c3001e343679A73928

해커톤 진행 상황

Built from scratch during the Buildathon: the AgentLeash Solidity contract, 8 passing Foundry tests (injection attack, per-payment and daily limits, expiry, kill switch, access control), deployment to Arbitrum Sepolia, and a live demo page where the owner sets rules in plain English and the agent's payments are allowed or blocked on-chain. Next: LLM-based rule parsing, Robinhood Chain support, multi-agent budgets.

Agentleash

비디오

기술 스택

설명

해커톤 진행 상황

자금 모금 상태