Initialising agent
Takes a raw address, ENS name, or a ticker you were shown. Determines what it actually is on-chain — token, proxy, EOA, or safe — and on which chain, then pins the exact contract behind the symbol rather than trusting the symbol.
Pulls the state that decides whether you can ever get out: who holds mint, burn, pause and blocklist authority, whether the implementation is upgradeable and by whom, what the transfer path does beyond moving a balance, and how ownership is distributed.
Runs the trade you were about to run — buy, hold, then sell — against forked state. A honeypot that hides behind clean-looking source has to reveal itself here: reverting sells, punitive tax on exit, balances that shrink after the block, approvals that move more than you offered.
Compares the contract against the official issuer set for tokenised assets, so a symbol that looks like a real Stock Token but was deployed by an unrelated wallet is named as impersonation. Oracle freshness and pause state are read directly, along with Uniswap depth and how locked it really is.
Weights every signal into one credibility score from 0 to 100 and writes the verdict in plain language — what is wrong, how badly, and what would have to change for the score to move. Each line links back to the evidence. The agent never signs, never holds funds, never asks for an approval.
26 signals, one score, the reasoning attached. Read-only — the agent never signs and never asks for an approval.