AgentBank uses multiple independent security layers to protect user funds and maintain protocol integrity.Documentation Index
Fetch the complete documentation index at: https://0xcaptain.mintlify.app/llms.txt
Use this file to discover all available pages before exploring further.
Verifiable AI execution
Every agent decision is cryptographically verifiable:- LLM runs inside an Intel SGX enclave (Phala Cloud or Marlin Oyster)
- The TEE produces a hardware-signed attestation quote
TEEAttestationVerifier(0x51E52dCBD0FBfaDaDB43ad1EB1Ea0d3A79f128c3) verifies the quote on-chainLLMReasoningRegistrystores a hash chain of all reasoning steps
Circuit breaker
TheCircuitBreaker auto-pauses the protocol when thresholds are breached:
- TVL drop exceeding the configured drawdown limit
- Oracle price deviation beyond safe bounds
- Anomalous block activity
Agent slashing and insurance
Analysts stake MNT at registration. Slashing occurs when signal accuracy falls below 40% over 4 epochs:- Slashed MNT flows to the
InsurancePool - The
InsurancePoolreimburses user losses from agent failures
Commit-reveal signals
CommitRevealSignal (0x2A7D252D0bFF31eC1098FF642C0934b7124a5A33) enforces a 5-block delay between signal commitment and reveal (~1 minute on Mantle), preventing analysts from front-running their own signals.
Anti-sybil
AntiSybilGuard (0x2fc6e0987bF58F5A0Dc76801A9556Ab62bD42049) requires World ID or Gitcoin Passport verification for analyst registration, preventing reputation farming through fake identities.
Timelock
All governance parameter changes go throughAgentBankTimelock with a 48-hour delay, giving users time to review changes and exit before they take effect.
Contract immutability
ABNKTokenandVotingEscroware immutable — they cannot be upgraded- Upgradeable contracts use UUPS proxy with a governance vote + 7-day timelock
- Peripheral contracts (strategies, adapters) are swappable via registry