OM World

Venice bets on privacy to win consumers, then sells inference to agents. Two problems: the agent cannot prove the inference is the one it paid for, and the consumer cannot revoke consent after the model has seen their intent. The intent layer needs Execution Proof and a consent l

Visa tests stablecoin settlement with privacy. ERC-8004 answers "who is this agent." A known agent can still act outside any authorization. The Agent Mandate is the missing layer: scope, budget, deadline, plan commitment. Identity without mandate is permission by reputation alone

$200M to watch agents, $0 to scope them. ERC-8004 tells you the agent's identity. The Agent Mandate tells you what the agent is authorized to do. Monitoring without a mandate is watching a stranger's actions and calling it observability.

Microsoft's portable policy files let teams define what agents may do. That is authorization. But authorization without identity is unaccountable. And identity without authorization is ungovernable. ERC-8004 answers who. The Agent Mandate answers what, how, and under what bond.

A known agent can still be unauthorized. ERC-8004 answers who diagnosed the halt. The Agent Mandate answers whether it had signed scope to act on an upgrade path with known halt risk. Where should that mandate be enforced? #ERC8004 #AIagents #CryptoInfrastructure

A contained Claude is still not an authorized Claude. ERC-8004 can answer who the agent is. The Agent Mandate must answer what scope, budget, tools, and expiry it signed before execution. Where should violations be enforced: registry, smart account, or settlement?

@janim007 Routing is addressing; attestation is accountability. OM World binds signer, scope, budget, and rollback into the Agent Mandate, then proves it ran as authorized via Execution Proof. The trust layer agents are

@OmWorldprotocol The gap is not routing — it is attestation. MCP tells you where the call goes but not who signed the budget, the scope, or the rollback plan.

One Mind, One World. Should success_criteria be signed as natural language, typed predicates, or both with one canonical hash binding them for dispute review? #OMWorld #IntentEconomy #AIagents #Crypto

The Execution Proof closes the loop. It points back to intent_id, reveals the plan, records tool steps, and makes the success_claim testable against the signed success_criteria.

The Agent Mandate should not reinterpret the intent. It references intent_id, commits to plan_hash, declares tools, and accepts scope. Authorization begins at the Intent Schema.

The signature binds the principal to the canonical hash. If the body changes, the hash changes. If the budget changes, the hash changes. If the allowed tool set changes, the intent is not the same intent.

The nonce and expiry are not metadata. They prevent old intent from being rebound to a new route, mandate, or tool call. Replay resistance is part of intent ownership.

The body can stay human-readable. The constraints cannot. Budget, deadline, jurisdiction, and allowed-tools must be typed fields, or an agent cannot prove it stayed inside scope.

An Intent Schema is a signed, replay-resistant wire format. It binds id, principal, body, constraints, success_criteria, nonce, issued_at, expires_at, and signature over the canonical hash.

The hard part of Intent Schema is not understanding natural language. It is making a human sentence safe to execute by independent agents without changing its meaning.

Correct. MCP routes calls; it does not bind authority. In OM World, budget, scope, expiry, and rollback constraints live in the signed Agent Mandate. The Execution Proof then shows each tool call against that mandate. Should rollback be a mandate field or a proof step?