Five checks. One rule.
Each check verifies one thing before your agent acts, and returns only the answer two rival AIs agree on. Together they cover cost proof before a trade, token and contract truth, and scam and hidden-prompt defence.
- Two labs must agreeA model from OpenAI and one from Anthropic answer blind.
- Pay only for agreed answersNo agreement means no charge.
- A signed receiptEvery answer is signed, with the facts behind it.
- Base, Arc and Robinhood ChainPay per check in USDC or USDG. No account.
Trade Check
Know exactly what your agent is about to sign.
$0.50per check, or $1.50 for Medium inputs
What it catches
- A swap with no real price limit, or one that moves the price too far
- An approval or permit that grants too much: an unlimited amount or an unknown spender
- A fee far above the network rate
- A router, spender or token posing as the genuine contract
- A recipient made to look like a familiar address, the trick behind address poisoning
How it works. Send the exact unsigned transaction or EIP-712 permit. SECONDED reads the chain from two independent sources that must agree, simulates the transaction, decodes the call and quotes the trade against live pools. Two rival models then judge those verified facts.
Use it when your agent is about to sign a trade, an approval, a transfer or a permit on Base, Arc or Robinhood Chain.
Inputs and scope in the docsToken Check
Spot a bad token before your agent touches it.
$0.50per check
What it catches
- A honeypot that lets you buy but not sell, caught by simulating a sale
- A heavy tax on every sale
- A token impersonating another
- Mint, freeze or upgrade powers still in someone else’s hands
- Contract source code that is not verified
How it works. SECONDED reads the token’s contract and state from two independent sources, simulates a sale and checks the source code. Two rival models judge the verified facts.
Use it when your agent meets a new token on Base, Arc or Robinhood Chain.
Inputs and scope in the docsStock Token Check
Trade the real stock token, not a copy.
$0.50per check
What it catches
- A look-alike that copies the ticker and name, caught by matching Robinhood’s official list by chain and address, never by ticker alone
- A token whose trading is halted, or that is inactive
- A price out of line with the reference price
How it works. SECONDED verifies the token from two independent sources and compares your agent’s intended price with the reference. Two rival models judge the verified facts. It checks the token; it never recommends a trade.
Use it when your agent is about to buy, sell or accept a stock token on Robinhood Chain.
Inputs and scope in the docsAgent Registry Check
Know who your agent is dealing with.
$0.50per check
What it catches
- An agent with no ERC-8004 registration
- A claimed owner or wallet that does not match the registry
- Revoked feedback, alongside what others have recorded about the agent
How it works. SECONDED reads the agent’s registry entry and feedback records from two independent sources, instead of taking the agent’s word for who it is. Two rival models judge the verified facts.
Use it when your agent is about to pay, hire or take instructions from another agent on Base, Arc or Robinhood Chain.
Inputs and scope in the docsScam Check
Don’t let a message talk your agent into anything.
$0.50per check; $1.50 Medium, $2.50 Large
What it catches
- Phishing and fake support contacts
- Prize lures and urgent requests to send funds
- Links to newly registered domains or freshly issued certificates
- Domains that imitate official ones
How it works. Two rival models each classify the message as benign, suspicious or malicious, with facts SECONDED looks up about every domain it links to: domain age, certificate age and look-alikes.
Use it when your agent receives a message it might act on.
Inputs and scope in the docsComing next.
In final testing
Hidden Prompt Check
Finds instructions hidden in the pages and files your agent is about to read.
In testing
Code Review
Two rival AIs review code before your agent runs it.