Agent wallet — spending policy
Spending policy is a planned protocol and wallet safety boundary, not a usable product feature. No released Stele, browser-wallet, desktop-wallet, or supported local-MCP flow currently lets a user create an agent wallet, register a policy, or execute a transaction under policy enforcement.
Do not fund an agent on the assumption that local caps, prompts, or configuration files are consensus-enforced. A client-side limit is not sufficient: another client or a compromised process could bypass it.
Target contract
Section titled “Target contract”The intended design may include per-transaction and rolling caps, counterparty and asset allowlists, expiry, revocation, and separately approved exceptions. Those fields, error codes, dispute windows, and UI flows are not a released contract and may change.
Before this page can be promoted from Planned, release evidence must show:
- consensus-side rejection of transactions outside the registered policy;
- an SDK contract that builds and verifies the exact policy and transaction payload;
- a wallet review surface that displays network, asset, amount, recipient, expiry, and fees;
- race-safe cap accounting, fresh state rereads, deterministic simulation, and replay protection;
- independent security review and an end-to-end test against the released binaries.
Stele boundary
Section titled “Stele boundary”Stele economic writes and hosted signing are disabled. The hosted MCP remains keyless. The existing local exact-three lyth_mcp Stele profile has no wallet or signing authority. Any future local economic profile would have to use a dedicated agent wallet and could not reuse a person’s primary wallet, but that dedicated-wallet lifecycle and economic execution path are unavailable today.