Cresnex IntentLock
Security and outcome enforcement for autonomous onchain agents.
Demo
About
Cresnex IntentLock is an open-source security layer for autonomous and AI-agent-controlled smart accounts. It lets an owner authorize an agent using signed EIP-712 intents and binds execution to explicit constraints such as spend limits, recipient restrictions, minimum output, token approval limits, expiry, and replay protection. Unsafe outcomes can be rolled back while preserving violation evidence and agent containment state. This is a Continuity Track project. The core IntentLock prototype existed before ETHOnline 2026. During the hackathon, I am extending it with new context-aware security functionality and sponsor integrations, while keeping the pre-hackathon repository history public so the new work can be reviewed separately.
How it's made
IntentLock is built with Solidity and Foundry for the smart-contract security layer, OpenZeppelin primitives for standard security components, and a Next.js/TypeScript frontend using wagmi and viem for wallet and contract interaction. Owners sign structured EIP-712 intents that bind an authorized agent, call data, nonce, validity period, and execution policy. The contracts execute permitted calls in an isolated flow and verify post-execution financial constraints before allowing the outcome to persist. Foundry tests cover signatures, access control, nonce replay, spend/output limits, approval containment, rollback, strikes, quarantine, and recovery. During ETHOnline 2026, I am adding the hackathon-specific integration layer rather than presenting the existing prototype as new work. My primary focus is using live blockchain data as a load-bearing risk input for IntentLock decisions, with additional sponsor integrations only where they materially improve agent authorization or execution safety.