Abstract
This proposal defines a standard interface for registering formal governance proofs
of autonomous agents on-chain. Unlike existing agent registries (ERC-8122, ERC-8107,
ERC-8160) that track identity and coordination, this standard tracks behavioral
compliance — cryptographically binding an agent’s runtime decisions to Z3-verified
mathematical invariants stored as immutable attestations.
A compliant agent produces a GovernanceProof for each execution cycle.
These proofs are anchored on-chain, enabling any protocol to verify that an
autonomous agent operated within formally defined safety boundaries — without
trusting the agent’s owner or operator.
Motivation
Autonomous agents are increasingly managing significant on-chain value. Current
infrastructure answers who an agent is (registries) and what it can do
(capability declarations). No standard answers the critical question:
Did this agent act within its formally declared safety constraints, and can that
be proven immutably?
This gap creates systemic risk for DeFi protocols, DAOs, and any system delegating
decisions to autonomous agents. A rogue or misconfigured agent can drain funds,
distort governance, or produce cascading failures — with no on-chain audit trail
of constraint violations.
DOF-MESH (Deterministic Oversight Framework — Multi-chain Execution Safety Hub)
has been running this model in production on Avalanche C-Chain since Q1 2026,
with 1,772+ agents indexed and two agents in live production. This proposal
formalizes the learnings into a reusable standard.
Specification
Core Invariants
Every compliant implementation MUST enforce two mathematical invariants: