Security
Security is part of the platform design, not an optional feature. YieldSeeker combines isolated Agent Wallets, restricted execution, on-chain validation, emergency controls, and user exit rights.
No system eliminates all risk. You should understand the controls and limitations before depositing funds.
Security through layers
Section titled “Security through layers”No single control is expected to catch every failure on its own. Each layer below assumes the one before it could fail, and is designed to still hold:
- Wallet isolation — your Agent Wallet is scoped to you; assets are never pooled with other users.
- Adapter-only execution — an agent can only reach approved adapters and registered targets, never an arbitrary contract call.
- On-chain parameter validation — adapters validate the asset, recipient, vault, or route before a call is made, so a permitted function cannot be redirected to an attacker-controlled destination.
- Registry and emergency controls — a target or adapter can be removed, disabled, or paused across every wallet that uses the registry.
- User exit rights — the wallet owner can always withdraw supported tokens and ETH directly, independent of the interface or server.
These layers are described individually below and in more technical detail in AgentWalletKit integration.
You own your Agent Wallet
Section titled “You own your Agent Wallet”Each agent uses an isolated smart wallet associated with one user. Assets are not pooled between users, and the wallet owner retains the ability to withdraw supported assets.
The YieldSeeker interface is a way to operate the wallet. It is not the source of ownership. If the interface is unavailable, the owner can still use the wallet’s on-chain withdrawal functions for supported tokens and ETH with suitable wallet tooling.
Agents cannot make arbitrary calls
Section titled “Agents cannot make arbitrary calls”The platform does not give an agent unrestricted access to every contract. Automated actions must use approved adapters and registered protocol targets.
This matters because allowing a function such as swap is not enough on its own: a malicious call could otherwise redirect the result to an attacker. The execution framework validates the target and relevant parameters on-chain.
Protection against server compromise
Section titled “Protection against server compromise”YieldSeeker’s server signs operations for agents, so a compromised server is a serious incident. The wallet and registry controls are designed to limit what a compromised server can do:
- It may attempt operations through approved adapters.
- It cannot use the disabled arbitrary-call path.
- It cannot use unregistered adapters or targets.
- Adapter validation is designed to keep assets and receipts in the Agent Wallet.
- Emergency controls can pause adapter execution across wallets.
These controls reduce the impact of a compromise; they do not make the platform or its integrations risk-free.
Emergency controls and upgrades
Section titled “Emergency controls and upgrades”The Adapter Registry can remove targets, disable adapters, or pause operations when an integration or operator presents a threat. Administrative changes and approved wallet implementation changes are subject to the controls described in the deployed contracts.
Some wallet configuration is cached for execution efficiency. A configuration change may require the wallet to synchronize from the Factory, while the registry pause provides an immediate execution stop.
Audits and open source
Section titled “Audits and open source”YieldSeeker’s contract system is open source and has been audited by Nethermind. Independent audits and source code are useful evidence, but an audit is not a guarantee that future code, integrations, or operations are safe.
- AgentWalletKit explains the reusable security model.
- YieldSeeker contracts contain the implementation and deployment material.
- Nethermind audit reports contain the published audit documents.
What security does not cover
Section titled “What security does not cover”You can still lose value through market movements, protocol exploits, bad liquidity, oracle failures, unsupported assets, operational errors, or an agent strategy performing poorly. Withdrawals can also be limited by the underlying protocol’s liquidity or state.
See Risks and limitations and Recovery and emergencies.
Technical details
Section titled “Technical details”For the contract-level lifecycle, ERC-4337 authorization, Factory deployment, adapter delegatecalls, registry mappings, operator synchronization, and permissions, see the developer AgentWalletKit guide.