A proposed standard · v1
Clear the name before you mint the token.
A short, adoptable convention for naming AI agents. Five checks, in order, each with a plain pass condition. Meant to be cited, forked, and built into the tools that register agents.
The rule
An agent name should pass five checks before registration: Resolvable (unique across registries and name services), Lawful (clear of live US trademarks), Speakable (unambiguous by voice), Cultural (no offensive-term match across markets), and Distinct (measurable distance from names already registered). The first two are gating; the rest inform.
The five checks
The convention, in order
01
Resolvablegatingunique across registries & name services
Pass: the exact name resolves to no existing agent identity on any known ERC-8004 registry or agent name service.
02
Lawfulgatingclear of live US trademarks
Pass: no confusingly similar live US trademark in a class the agent operates in.
How that’s judged →
03
Speakablevoice-invocation disambiguation
Pass: the name is unambiguous when spoken — it won’t be confused with a common word or another agent by a voice interface.
04
Culturaloffensive-term screen · 20 markets
Pass: no exact match to a known offensive term across the 20-market panel.
05
Distinctphonetic distance from the field
Pass: sufficient phonetic and string distance from the names already registered, so the agent stays findable and un-confusable.
Why a convention
Standards are how a young category agrees with itself
The agent-identity layer arrived faster than its conventions. Registries mint names first-come, with nothing checking the five things above. A shared convention gives builders, registrars, and tooling a common bar to point at — and makes “did you clear it?” a question with a definite answer. This is v1; it’s meant to be argued with and improved.