What is ERC-8004? The Ethereum agent identity standard, explained

ERC-8004, titled "Trustless Agents," is an open Ethereum standard proposed August 13, 2025 by Marco De Rossi (MetaMask), Davide Crapis (Ethereum Foundation), Jordan Ellis (Google), and Erik Reppel (Coinbase): Solidus did not write any part of it. It answers a question that gets harder to ignore every month autonomous agents spend trading, hiring other agents, and paying for API calls on-chain: when you interact with an agent, who's actually behind it, and does it have a track record worth trusting, without taking the agent's own word for it?

What ERC-8004 defines

The standard specifies three on-chain registries a smart contract or another agent can query.

An Identity Registry mints each agent an ERC-721 "passport", an NFT whose tokenURI points to an off-chain agent card describing what the agent is and who controls it. The NFT gives the agent a portable, censorship-resistant handle that survives independent of any single platform.

A Reputation Registry lets clients who've actually used an agent leave on-chain feedback about it.

A Validation Registry lets a third party attach cryptographic proof that a specific agent action was checked and found valid.

Put together, the three registries let a counterparty look up an agent's record instead of trusting its self-description.

Who actually built it

ERC-8004 was formally proposed on Ethereum on August 13, 2025, co-authored by Marco De Rossi (MetaMask), Davide Crapis (Ethereum Foundation), Jordan Ellis (Google), and Erik Reppel (Coinbase). It was refined with input from more than 100 contributors across the ecosystem, including ENS, EigenLayer, and The Graph.

Solidus is not on that list of authors and is not a co-author of the standard. Solidus is an integrator, one of a field of implementers building against registries someone else specified and shipped.

What the standard leaves open

ERC-8004 deliberately leaves two questions unanswered.

First: which DID method anchors an agent's identity. The Identity Registry's tokenURI points to an off-chain agent card, and that card can name any identity scheme the issuer chooses, the standard doesn't pick one. Second: what evidence format satisfies the Validation Registry. The registry defines a slot for a cryptographic validation proof; it doesn't define the proof's shape, and as of this writing the upstream standard hasn't published a canonical Validation Registry contract address for implementers to target.

Both are genuinely open design questions, not gaps that any single implementer, Solidus included, has closed on the field's behalf. Different projects will answer them differently, and that's the standard working as intended: a shared registry layer with room for different identity and evidence schemes underneath it.

Solidus's own answer to the first question

Solidus's own passport is one worked answer to the DID-method question, not the only possible one. (One honest footnote if you fetch that card yourself: its own JSON now names the registration correctly, but it carried a stale pre-registration status line for a stretch after the fact. The registration is real and dated 2026-07-04; the card text lagged it, and nothing caught that but a human reading it.)

That's Solidus's answer to "which DID method," built with the @solidus-network/agent-identity SDK. It says nothing about what the right answer is for anyone else's agent: a different implementer could anchor to a different DID method entirely and satisfy the same Identity Registry slot. The full command sequence for verifying this passport yourself, the eth_calls, the exact outputs, no trust required, lives on How to verify our ERC-8004 agent passport yourself, not here. The Validation Registry question, the second open one, is not something Solidus has an answer to yet either; that gets its own page at ERC-8004 Validation Registry.

Composes, doesn't invent

To restate plainly: ERC-8004 is MetaMask, the Ethereum Foundation, Google, Coinbase, and 100+ contributors' standard. Solidus composes with it as an integrator, filling in one of the two questions it leaves open, using its own agent passport as a worked example rather than a definition.