Skip to main content

AgentPolicy

@totemsdk/agent-policy

@totemsdk/agent-policy / AgentPolicy

Interface: AgentPolicy

Policy evaluated by the Totem wallet layer — NEVER by the agent.

The wallet implements this interface to decide whether to auto-sign or route to the user. The AI never has access to the policy implementation.

Methods​

canAutoApprove()​

canAutoApprove(proposal): Promise<boolean>

Return true if the wallet should sign the intent without user interaction. Implementations typically check risk, amount thresholds, and known agents.

Parameters​

proposal​

AgentProposal

Returns​

Promise<boolean>

commit()?​

optional commit(operationId): Promise<void>

Parameters​

operationId​

string

Returns​

Promise<void>

release()?​

optional release(operationId): Promise<void>

Parameters​

operationId​

string

Returns​

Promise<void>

requiresUserApproval()​

requiresUserApproval(proposal): Promise<boolean>

Return true if the wallet must show a user-approval UI before signing. Generally the complement of canAutoApprove, but may have independent logic (e.g. always require approval for settlements regardless of risk).

Parameters​

proposal​

AgentProposal

Returns​

Promise<boolean>

reserve()?​

optional reserve(proposal): Promise<PolicyEvalResult>

Optional reservation lifecycle for stateful policies.

The execution boundary calls reserve before executing, then commit after a successful execution or release on any failure path. When implemented, quota is only consumed through reserve/commit — a read-only evaluate / canAutoApprove never touches the limits.

Parameters​

proposal​

AgentProposal

Returns​

Promise<PolicyEvalResult>