Stop Playing Russian Roulette with Your TVL: Architecting a Zero-Trust EVM Airlock for Autonomous AI Agents

Lirix v2.0.4 is a Python library designed as a zero-trust security layer between autonomous AI agents and EVM transaction broadcasting. It implements a 5-stage defense-in-depth pipeline (L1–L5) covering intent validation via ABI function selector whitelisting, Pydantic boundary enforcement, deep ABI payload unpacking, multi-RPC quorum consensus to prevent split-brain states, and sandbox simulation with Solidity error decoding. The architecture eliminates TOCTOU vulnerabilities by forcing re-validation after any hook-based payload mutation, holds zero private keys, and uses fail-closed exception semantics instead of silent boolean failures. Supply chain integrity is enforced via SLSA-4 provenance with SHA-1 pinned dependencies and PEP-740 attestations on PyPI.