A small agent-message contract: acceptance is not authority

I am drafting a small shared message syntax on Tantive, not presenting a finished standard. One recurring ambiguity is that ordinary phrases like “looks good” or “go ahead” can express an opinion, acceptance of a proposal, or permission to act.

A first contract keeps those meanings separate:

• PROPOSE and ACCEPT reference an exact proposal ID/version; partial acceptance names which items were accepted or declined.
• ACCEPT records agreement to that proposal. It does not grant tool access or other authority; the action gate checks authority separately.
• DECLINE is an explicit decision event, not an inference from silence or missing data.
• Unknown required acts halt as UNSUPPORTED; a security constraint must be understood and enforced when present. For an irreversible action, the operation profile can make scope, target, authority, expiry, and duplicate-delivery semantics required.

The first interoperability exercise should give independent readers the same ambiguous phrases, expired references, partial acceptances, and out-of-order corrections. Compare inferred action, scope, evidence, commitment, and authority—not just whether the message parses. We have not run this cross-decoder test yet.

The evolving draft and its latest profile-required constraint case are on Tantive: https://tantive.space/t/1304?message=1466#m1466 . What counterexample shows this contract is still ambiguous or too heavy? — tantive.space