You are now the
compromised agent.
Below is a real NEAR private key, published on purpose. Pick an attack and sign it. Every attempt is a genuine transaction sent to NEAR testnet from your browser — nothing on this page is simulated.
key ed25519:dmA2aFjMVuTGSfkcAHLavZs7zXRhFjraP3xF6pZgLH7gdPVJvn9aMazSovmpmB1CPvLa3EAhJupSHoE1zuaAbQb
permission request_spend, request_spend_ft @ agent-allowance.testnet This key cannot attach tokens and cannot call any other contract on NEAR. The account holds no full-access key, so no permission can ever be added to it. That is why publishing it is safe — and why it is the whole argument.
Reading the live policy from chain…
Two gates, and what each one is for.
Gate 1 refuses the key
Six of these attacks never reach a block. A NEAR function-call key names one contract and a list of methods, and can never attach a deposit. Send anything else and the network rejects the transaction outright — the vault does not run, no fee is charged, and there is no transaction hash to show you.
Gate 2 refuses the request
The rest are calls the key is allowed to make. The vault compares the recipient with an allowlist and the amount with a per-payment cap and a rolling daily cap. Your wording never enters the decision — the contract does not read English — but it is recorded on-chain beside the refusal.
The same attacks, from a terminal.
Nothing to install — the script signs with the Python standard library and exits non-zero if any attack gets through.